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
  26 #include "util/u_format.h"
  27 #include "util/u_pack_color.h"
  28
  29 enum r300_blitter_op /* bitmask */
  30 {
  31     R300_CLEAR         = 1,
  32     R300_CLEAR_SURFACE = 2,
  33     R300_COPY          = 4
  34 };
  35
  36 static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op op)
  37 {
  38     if (r300->query_current) {
  39         r300->blitter_saved_query = r300->query_current;
  40         r300_stop_query(r300);
  41     }
  42
  43     /* Yeah we have to save all those states to ensure the blitter operation
  44      * is really transparent. The states will be restored by the blitter once
  45      * copying is done. */
  46     util_blitter_save_blend(r300->blitter, r300->blend_state.state);
  47     util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state);
  48     util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref));
  49     util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
  50     util_blitter_save_fragment_shader(r300->blitter, r300->fs.state);
  51     util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
  52     util_blitter_save_viewport(r300->blitter, &r300->viewport);
  53     util_blitter_save_clip(r300->blitter, (struct pipe_clip_state*)r300->clip_state.state);
  54     util_blitter_save_vertex_elements(r300->blitter, r300->velems);
  55     util_blitter_save_vertex_buffers(r300->blitter, r300->vertex_buffer_count,
  56                                      r300->vertex_buffer);
  57
  58     if (op & (R300_CLEAR_SURFACE | R300_COPY))
  59         util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
  60
  61     if (op & R300_COPY) {
  62         struct r300_textures_state* state =
  63             (struct r300_textures_state*)r300->textures_state.state;
  64
  65         util_blitter_save_fragment_sampler_states(
  66             r300->blitter, state->sampler_state_count,
  67             (void**)state->sampler_states);
  68
  69         util_blitter_save_fragment_sampler_views(
  70             r300->blitter, state->sampler_view_count,
  71             (struct pipe_sampler_view**)state->sampler_views);
  72     }
  73 }
  74
  75 static void r300_blitter_end(struct r300_context *r300)
  76 {
  77     if (r300->blitter_saved_query) {
  78         r300_resume_query(r300, r300->blitter_saved_query);
  79         r300->blitter_saved_query = NULL;
  80     }
  81 }
  82
  83 static uint32_t r300_depth_clear_cb_value(enum pipe_format format,
  84                                           const float* rgba)
  85 {
  86     union util_color uc;
  87     util_pack_color(rgba, format, &uc);
  88
  89     if (util_format_get_blocksizebits(format) == 32)
  90         return uc.ui;
  91     else
  92         return uc.us | (uc.us << 16);
  93 }
  94
  95 static boolean r300_cbzb_clear_allowed(struct r300_context *r300,
  96                                        unsigned clear_buffers)
  97 {
  98     struct pipe_framebuffer_state *fb =
  99         (struct pipe_framebuffer_state*)r300->fb_state.state;
 100     struct r300_surface *surf = r300_surface(fb->cbufs[0]);
 101     unsigned bpp;
 102
 103     /* Only color clear allowed, and only one colorbuffer. */
 104     if (clear_buffers != PIPE_CLEAR_COLOR || fb->nr_cbufs != 1)
 105         return FALSE;
 106
 107     /* The colorbuffer must be point-sampled. */
 108     if (surf->base.texture->nr_samples > 1)
 109         return FALSE;
 110
 111     bpp = util_format_get_blocksizebits(surf->base.format);
 112
 113     /* ZB can only work with the two pixel sizes. */
 114     if (bpp != 16 && bpp != 32)
 115         return FALSE;
 116
 117     /* If the midpoint ZB offset is not aligned to 2048, it returns garbage
 118      * with certain texture sizes. Macrotiling ensures the alignment. */
 119     if (!r300_texture(surf->base.texture)->mip_macrotile[surf->base.level])
 120         return FALSE;
 121
 122     return TRUE;
 123 }
 124
 125 /* Clear currently bound buffers. */
 126 static void r300_clear(struct pipe_context* pipe,
 127                        unsigned buffers,
 128                        const float* rgba,
 129                        double depth,
 130                        unsigned stencil)
 131 {
 132     /* My notes about fastfill:
 133      *
 134      * 1) Only the zbuffer is cleared.
 135      *
 136      * 2) The zbuffer must be micro-tiled and whole microtiles must be
 137      *    written. If microtiling is disabled, it locks up.
 138      *
 139      * 3) There is Z Mask RAM which contains a compressed zbuffer and
 140      *    it interacts with fastfill. We should figure out how to use it
 141      *    to get more performance.
 142      *    This is what we know about the Z Mask:
 143      *
 144      *       Each dword of the Z Mask contains compression information
 145      *       for 16 4x4 pixel blocks, that is 2 bits for each block.
 146      *       On chips with 2 Z pipes, every other dword maps to a different
 147      *       pipe.
 148      *
 149      * 4) ZB_DEPTHCLEARVALUE is used to clear the zbuffer and the Z Mask must
 150      *    be equal to 0. (clear the Z Mask RAM with zeros)
 151      *
 152      * 5) For 16-bit zbuffer, compression causes a hung with one or
 153      *    two samples and should not be used.
 154      *
 155      * 6) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
 156      *    to avoid needless decompression.
 157      *
 158      * 7) Fastfill must not be used if reading of compressed Z data is disabled
 159      *    and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
 160      *    i.e. it cannot be used to compress the zbuffer.
 161      *
 162      * 8) ZB_CB_CLEAR does not interact with fastfill in any way.
 163      *
 164      * - Marek
 165      */
 166
 167     struct r300_context* r300 = r300_context(pipe);
 168     struct pipe_framebuffer_state *fb =
 169         (struct pipe_framebuffer_state*)r300->fb_state.state;
 170     struct r300_hyperz_state *hyperz =
 171         (struct r300_hyperz_state*)r300->hyperz_state.state;
 172     uint32_t width = fb->width;
 173     uint32_t height = fb->height;
 174
 175     /* Enable CBZB clear. */
 176     if (r300_cbzb_clear_allowed(r300, buffers)) {
 177         struct r300_surface *surf = r300_surface(fb->cbufs[0]);
 178
 179         hyperz->zb_depthclearvalue =
 180                 r300_depth_clear_cb_value(surf->base.format, rgba);
 181
 182         width = surf->cbzb_width;
 183         height = surf->cbzb_height;
 184
 185         r300->cbzb_clear = TRUE;
 186         r300_mark_fb_state_dirty(r300, R300_CHANGED_CBZB_FLAG);
 187     }
 188
 189     /* Clear. */
 190     r300_blitter_begin(r300, R300_CLEAR);
 191     util_blitter_clear(r300->blitter,
 192                        width,
 193                        height,
 194                        fb->nr_cbufs,
 195                        buffers, rgba, depth, stencil);
 196     r300_blitter_end(r300);
 197
 198     /* Disable CBZB clear. */
 199     if (r300->cbzb_clear) {
 200         r300->cbzb_clear = FALSE;
 201         r300_mark_fb_state_dirty(r300, R300_CHANGED_CBZB_FLAG);
 202     }
 203
 204     /* XXX this flush "fixes" a hardlock in the cubestorm xscreensaver */
 205     if (r300->flush_counter == 0)
 206         pipe->flush(pipe, 0, NULL);
 207 }
 208
 209 /* Clear a region of a color surface to a constant value. */
 210 static void r300_clear_render_target(struct pipe_context *pipe,
 211                                      struct pipe_surface *dst,
 212                                      const float *rgba,
 213                                      unsigned dstx, unsigned dsty,
 214                                      unsigned width, unsigned height)
 215 {
 216     struct r300_context *r300 = r300_context(pipe);
 217
 218     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 219     util_blitter_clear_render_target(r300->blitter, dst, rgba,
 220                                      dstx, dsty, width, height);
 221     r300_blitter_end(r300);
 222 }
 223
 224 /* Clear a region of a depth stencil surface. */
 225 static void r300_clear_depth_stencil(struct pipe_context *pipe,
 226                                      struct pipe_surface *dst,
 227                                      unsigned clear_flags,
 228                                      double depth,
 229                                      unsigned stencil,
 230                                      unsigned dstx, unsigned dsty,
 231                                      unsigned width, unsigned height)
 232 {
 233     struct r300_context *r300 = r300_context(pipe);
 234
 235     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 236     util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
 237                                      dstx, dsty, width, height);
 238     r300_blitter_end(r300);
 239 }
 240
 241 /* Copy a block of pixels from one surface to another using HW. */
 242 static void r300_hw_copy_region(struct pipe_context* pipe,
 243                                 struct pipe_resource *dst,
 244                                 struct pipe_subresource subdst,
 245                                 unsigned dstx, unsigned dsty, unsigned dstz,
 246                                 struct pipe_resource *src,
 247                                 struct pipe_subresource subsrc,
 248                                 unsigned srcx, unsigned srcy, unsigned srcz,
 249                                 unsigned width, unsigned height)
 250 {
 251     struct r300_context* r300 = r300_context(pipe);
 252
 253     r300_blitter_begin(r300, R300_COPY);
 254     util_blitter_copy_region(r300->blitter, dst, subdst, dstx, dsty, dstz,
 255                              src, subsrc, srcx, srcy, srcz, width, height,
 256                              TRUE);
 257     r300_blitter_end(r300);
 258 }
 259
 260 /* Copy a block of pixels from one surface to another. */
 261 static void r300_resource_copy_region(struct pipe_context *pipe,
 262                                       struct pipe_resource *dst,
 263                                       struct pipe_subresource subdst,
 264                                       unsigned dstx, unsigned dsty, unsigned dstz,
 265                                       struct pipe_resource *src,
 266                                       struct pipe_subresource subsrc,
 267                                       unsigned srcx, unsigned srcy, unsigned srcz,
 268                                       unsigned width, unsigned height)
 269 {
 270     enum pipe_format old_format = dst->format;
 271     enum pipe_format new_format = old_format;
 272
 273     if (!pipe->screen->is_format_supported(pipe->screen,
 274                                            old_format, src->target,
 275                                            src->nr_samples,
 276                                            PIPE_BIND_RENDER_TARGET |
 277                                            PIPE_BIND_SAMPLER_VIEW, 0) &&
 278         util_format_is_plain(old_format)) {
 279         switch (util_format_get_blocksize(old_format)) {
 280             case 1:
 281                 new_format = PIPE_FORMAT_I8_UNORM;
 282                 break;
 283             case 2:
 284                 new_format = PIPE_FORMAT_B4G4R4A4_UNORM;
 285                 break;
 286             case 4:
 287                 new_format = PIPE_FORMAT_B8G8R8A8_UNORM;
 288                 break;
 289             case 8:
 290                 new_format = PIPE_FORMAT_R16G16B16A16_UNORM;
 291                 break;
 292             default:
 293                 debug_printf("r300: surface_copy: Unhandled format: %s. Falling back to software.\n"
 294                              "r300: surface_copy: Software fallback doesn't work for tiled textures.\n",
 295                              util_format_short_name(old_format));
 296         }
 297     }
 298
 299     if (old_format != new_format) {
 300         dst->format = new_format;
 301         src->format = new_format;
 302
 303         r300_texture_reinterpret_format(pipe->screen,
 304                                         dst, new_format);
 305         r300_texture_reinterpret_format(pipe->screen,
 306                                         src, new_format);
 307     }
 308
 309     r300_hw_copy_region(pipe, dst, subdst, dstx, dsty, dstz,
 310                         src, subsrc, srcx, srcy, srcz, width, height);
 311
 312     if (old_format != new_format) {
 313         dst->format = old_format;
 314         src->format = old_format;
 315
 316         r300_texture_reinterpret_format(pipe->screen,
 317                                         dst, old_format);
 318         r300_texture_reinterpret_format(pipe->screen,
 319                                         src, old_format);
 320     }
 321 }
 322
 323 void r300_init_blit_functions(struct r300_context *r300)
 324 {
 325     r300->context.clear = r300_clear;
 326     r300->context.clear_render_target = r300_clear_render_target;
 327     r300->context.clear_depth_stencil = r300_clear_depth_stencil;
 328     r300->context.resource_copy_region = r300_resource_copy_region;
 329 }