Subversion Repositories Kolibri OS

Rev

Blame | Last modification | View Log | RSS feed

  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. #include "r300_reg.h"
  27.  
  28. #include "util/u_format.h"
  29. #include "util/u_half.h"
  30. #include "util/u_pack_color.h"
  31. #include "util/u_surface.h"
  32.  
  33. enum r300_blitter_op /* bitmask */
  34. {
  35.     R300_STOP_QUERY         = 1,
  36.     R300_SAVE_TEXTURES      = 2,
  37.     R300_SAVE_FRAMEBUFFER   = 4,
  38.     R300_IGNORE_RENDER_COND = 8,
  39.  
  40.     R300_CLEAR         = R300_STOP_QUERY,
  41.  
  42.     R300_CLEAR_SURFACE = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER,
  43.  
  44.     R300_COPY          = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER |
  45.                          R300_SAVE_TEXTURES | R300_IGNORE_RENDER_COND,
  46.  
  47.     R300_BLIT          = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER |
  48.                          R300_SAVE_TEXTURES,
  49.  
  50.     R300_DECOMPRESS    = R300_STOP_QUERY | R300_IGNORE_RENDER_COND,
  51. };
  52.  
  53. static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op op)
  54. {
  55.     if ((op & R300_STOP_QUERY) && r300->query_current) {
  56.         r300->blitter_saved_query = r300->query_current;
  57.         r300_stop_query(r300);
  58.     }
  59.  
  60.     /* Yeah we have to save all those states to ensure the blitter operation
  61.      * is really transparent. The states will be restored by the blitter once
  62.      * copying is done. */
  63.     util_blitter_save_blend(r300->blitter, r300->blend_state.state);
  64.     util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state);
  65.     util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref));
  66.     util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
  67.     util_blitter_save_fragment_shader(r300->blitter, r300->fs.state);
  68.     util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
  69.     util_blitter_save_viewport(r300->blitter, &r300->viewport);
  70.     util_blitter_save_scissor(r300->blitter, r300->scissor_state.state);
  71.     util_blitter_save_sample_mask(r300->blitter, *(unsigned*)r300->sample_mask.state);
  72.     util_blitter_save_vertex_buffer_slot(r300->blitter, r300->vertex_buffer);
  73.     util_blitter_save_vertex_elements(r300->blitter, r300->velems);
  74.  
  75.     if (op & R300_SAVE_FRAMEBUFFER) {
  76.         util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
  77.     }
  78.  
  79.     if (op & R300_SAVE_TEXTURES) {
  80.         struct r300_textures_state* state =
  81.             (struct r300_textures_state*)r300->textures_state.state;
  82.  
  83.         util_blitter_save_fragment_sampler_states(
  84.             r300->blitter, state->sampler_state_count,
  85.             (void**)state->sampler_states);
  86.  
  87.         util_blitter_save_fragment_sampler_views(
  88.             r300->blitter, state->sampler_view_count,
  89.             (struct pipe_sampler_view**)state->sampler_views);
  90.     }
  91.  
  92.     if (op & R300_IGNORE_RENDER_COND) {
  93.         /* Save the flag. */
  94.         r300->blitter_saved_skip_rendering = r300->skip_rendering+1;
  95.         r300->skip_rendering = FALSE;
  96.     } else {
  97.         r300->blitter_saved_skip_rendering = 0;
  98.     }
  99. }
  100.  
  101. static void r300_blitter_end(struct r300_context *r300)
  102. {
  103.     if (r300->blitter_saved_query) {
  104.         r300_resume_query(r300, r300->blitter_saved_query);
  105.         r300->blitter_saved_query = NULL;
  106.     }
  107.  
  108.     if (r300->blitter_saved_skip_rendering) {
  109.         /* Restore the flag. */
  110.         r300->skip_rendering = r300->blitter_saved_skip_rendering-1;
  111.     }
  112. }
  113.  
  114. static uint32_t r300_depth_clear_cb_value(enum pipe_format format,
  115.                                           const float* rgba)
  116. {
  117.     union util_color uc;
  118.     util_pack_color(rgba, format, &uc);
  119.  
  120.     if (util_format_get_blocksizebits(format) == 32)
  121.         return uc.ui[0];
  122.     else
  123.         return uc.us | (uc.us << 16);
  124. }
  125.  
  126. static boolean r300_cbzb_clear_allowed(struct r300_context *r300,
  127.                                        unsigned clear_buffers)
  128. {
  129.     struct pipe_framebuffer_state *fb =
  130.         (struct pipe_framebuffer_state*)r300->fb_state.state;
  131.  
  132.     /* Only color clear allowed, and only one colorbuffer. */
  133.     if ((clear_buffers & ~PIPE_CLEAR_COLOR) != 0 || fb->nr_cbufs != 1 || !fb->cbufs[0])
  134.         return FALSE;
  135.  
  136.     return r300_surface(fb->cbufs[0])->cbzb_allowed;
  137. }
  138.  
  139. static boolean r300_fast_zclear_allowed(struct r300_context *r300,
  140.                                         unsigned clear_buffers)
  141. {
  142.     struct pipe_framebuffer_state *fb =
  143.         (struct pipe_framebuffer_state*)r300->fb_state.state;
  144.  
  145.     return r300_resource(fb->zsbuf->texture)->tex.zmask_dwords[fb->zsbuf->u.tex.level] != 0;
  146. }
  147.  
  148. static boolean r300_hiz_clear_allowed(struct r300_context *r300)
  149. {
  150.     struct pipe_framebuffer_state *fb =
  151.         (struct pipe_framebuffer_state*)r300->fb_state.state;
  152.  
  153.     return r300_resource(fb->zsbuf->texture)->tex.hiz_dwords[fb->zsbuf->u.tex.level] != 0;
  154. }
  155.  
  156. static uint32_t r300_depth_clear_value(enum pipe_format format,
  157.                                        double depth, unsigned stencil)
  158. {
  159.     switch (format) {
  160.         case PIPE_FORMAT_Z16_UNORM:
  161.         case PIPE_FORMAT_X8Z24_UNORM:
  162.             return util_pack_z(format, depth);
  163.  
  164.         case PIPE_FORMAT_S8_UINT_Z24_UNORM:
  165.             return util_pack_z_stencil(format, depth, stencil);
  166.  
  167.         default:
  168.             assert(0);
  169.             return 0;
  170.     }
  171. }
  172.  
  173. static uint32_t r300_hiz_clear_value(double depth)
  174. {
  175.     uint32_t r = (uint32_t)(CLAMP(depth, 0, 1) * 255.5);
  176.     assert(r <= 255);
  177.     return r | (r << 8) | (r << 16) | (r << 24);
  178. }
  179.  
  180. static void r300_set_clear_color(struct r300_context *r300,
  181.                                  const union pipe_color_union *color)
  182. {
  183.     struct pipe_framebuffer_state *fb =
  184.         (struct pipe_framebuffer_state*)r300->fb_state.state;
  185.     union util_color uc;
  186.  
  187.     memset(&uc, 0, sizeof(uc));
  188.     util_pack_color(color->f, fb->cbufs[0]->format, &uc);
  189.  
  190.     if (fb->cbufs[0]->format == PIPE_FORMAT_R16G16B16A16_FLOAT ||
  191.         fb->cbufs[0]->format == PIPE_FORMAT_R16G16B16X16_FLOAT) {
  192.         /* (0,1,2,3) maps to (B,G,R,A) */
  193.         r300->color_clear_value_gb = uc.h[0] | ((uint32_t)uc.h[1] << 16);
  194.         r300->color_clear_value_ar = uc.h[2] | ((uint32_t)uc.h[3] << 16);
  195.     } else {
  196.         r300->color_clear_value = uc.ui[0];
  197.     }
  198. }
  199.  
  200. DEBUG_GET_ONCE_BOOL_OPTION(hyperz, "RADEON_HYPERZ", FALSE)
  201.  
  202. /* Clear currently bound buffers. */
  203. static void r300_clear(struct pipe_context* pipe,
  204.                        unsigned buffers,
  205.                        const union pipe_color_union *color,
  206.                        double depth,
  207.                        unsigned stencil)
  208. {
  209.     /* My notes about Zbuffer compression:
  210.      *
  211.      * 1) The zbuffer must be micro-tiled and whole microtiles must be
  212.      *    written if compression is enabled. If microtiling is disabled,
  213.      *    it locks up.
  214.      *
  215.      * 2) There is ZMASK RAM which contains a compressed zbuffer.
  216.      *    Each dword of the Z Mask contains compression information
  217.      *    for 16 4x4 pixel tiles, that is 2 bits for each tile.
  218.      *    On chips with 2 Z pipes, every other dword maps to a different
  219.      *    pipe. On newer chipsets, there is a new compression mode
  220.      *    with 8x8 pixel tiles per 2 bits.
  221.      *
  222.      * 3) The FASTFILL bit has nothing to do with filling. It only tells hw
  223.      *    it should look in the ZMASK RAM first before fetching from a real
  224.      *    zbuffer.
  225.      *
  226.      * 4) If a pixel is in a cleared state, ZB_DEPTHCLEARVALUE is returned
  227.      *    during zbuffer reads instead of the value that is actually stored
  228.      *    in the zbuffer memory. A pixel is in a cleared state when its ZMASK
  229.      *    is equal to 0. Therefore, if you clear ZMASK with zeros, you may
  230.      *    leave the zbuffer memory uninitialized, but then you must enable
  231.      *    compression, so that the ZMASK RAM is actually used.
  232.      *
  233.      * 5) Each 4x4 (or 8x8) tile is automatically decompressed and recompressed
  234.      *    during zbuffer updates. A special decompressing operation should be
  235.      *    used to fully decompress a zbuffer, which basically just stores all
  236.      *    compressed tiles in ZMASK to the zbuffer memory.
  237.      *
  238.      * 6) For a 16-bit zbuffer, compression causes a hung with one or
  239.      *    two samples and should not be used.
  240.      *
  241.      * 7) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
  242.      *    to avoid needless decompression.
  243.      *
  244.      * 8) Fastfill must not be used if reading of compressed Z data is disabled
  245.      *    and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
  246.      *    i.e. it cannot be used to compress the zbuffer.
  247.      *
  248.      * 9) ZB_CB_CLEAR does not interact with zbuffer compression in any way.
  249.      *
  250.      * - Marek
  251.      */
  252.  
  253.     struct r300_context* r300 = r300_context(pipe);
  254.     struct pipe_framebuffer_state *fb =
  255.         (struct pipe_framebuffer_state*)r300->fb_state.state;
  256.     struct r300_hyperz_state *hyperz =
  257.         (struct r300_hyperz_state*)r300->hyperz_state.state;
  258.     uint32_t width = fb->width;
  259.     uint32_t height = fb->height;
  260.     uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
  261.  
  262.     /* Use fast Z clear.
  263.      * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
  264.     if (buffers & PIPE_CLEAR_DEPTHSTENCIL) {
  265.         boolean zmask_clear, hiz_clear;
  266.  
  267.         /* If both depth and stencil are present, they must be cleared together. */
  268.         if (fb->zsbuf->texture->format == PIPE_FORMAT_S8_UINT_Z24_UNORM &&
  269.             (buffers & PIPE_CLEAR_DEPTHSTENCIL) != PIPE_CLEAR_DEPTHSTENCIL) {
  270.             zmask_clear = FALSE;
  271.             hiz_clear = FALSE;
  272.         } else {
  273.             zmask_clear = r300_fast_zclear_allowed(r300, buffers);
  274.             hiz_clear = r300_hiz_clear_allowed(r300);
  275.         }
  276.  
  277.         /* If we need Hyper-Z. */
  278.         if (zmask_clear || hiz_clear) {
  279.             /* Try to obtain the access to Hyper-Z buffers if we don't have one. */
  280.             if (!r300->hyperz_enabled &&
  281.                 (r300->screen->caps.is_r500 || debug_get_option_hyperz())) {
  282.                 r300->hyperz_enabled =
  283.                     r300->rws->cs_request_feature(r300->cs,
  284.                                                 RADEON_FID_R300_HYPERZ_ACCESS,
  285.                                                 TRUE);
  286.                 if (r300->hyperz_enabled) {
  287.                    /* Need to emit HyperZ buffer regs for the first time. */
  288.                    r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
  289.                 }
  290.             }
  291.  
  292.             /* Setup Hyper-Z clears. */
  293.             if (r300->hyperz_enabled) {
  294.                 if (zmask_clear) {
  295.                     hyperz_dcv = hyperz->zb_depthclearvalue =
  296.                         r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
  297.  
  298.                     r300_mark_atom_dirty(r300, &r300->zmask_clear);
  299.                     r300_mark_atom_dirty(r300, &r300->gpu_flush);
  300.                     buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
  301.                 }
  302.  
  303.                 if (hiz_clear) {
  304.                     r300->hiz_clear_value = r300_hiz_clear_value(depth);
  305.                     r300_mark_atom_dirty(r300, &r300->hiz_clear);
  306.                     r300_mark_atom_dirty(r300, &r300->gpu_flush);
  307.                 }
  308.                 r300->num_z_clears++;
  309.             }
  310.         }
  311.     }
  312.  
  313.     /* Use fast color clear for an AA colorbuffer.
  314.      * The CMASK is shared between all colorbuffers, so we use it
  315.      * if there is only one colorbuffer bound. */
  316.     if ((buffers & PIPE_CLEAR_COLOR) && fb->nr_cbufs == 1 && fb->cbufs[0] &&
  317.         r300_resource(fb->cbufs[0]->texture)->tex.cmask_dwords) {
  318.         /* Try to obtain the access to the CMASK if we don't have one. */
  319.         if (!r300->cmask_access) {
  320.             r300->cmask_access =
  321.                 r300->rws->cs_request_feature(r300->cs,
  322.                                               RADEON_FID_R300_CMASK_ACCESS,
  323.                                               TRUE);
  324.         }
  325.  
  326.         /* Setup the clear. */
  327.         if (r300->cmask_access) {
  328.             /* Pair the resource with the CMASK to avoid other resources
  329.              * accessing it. */
  330.             if (!r300->screen->cmask_resource) {
  331.                 pipe_mutex_lock(r300->screen->cmask_mutex);
  332.                 /* Double checking (first unlocked, then locked). */
  333.                 if (!r300->screen->cmask_resource) {
  334.                     /* Don't reference this, so that the texture can be
  335.                      * destroyed while set in cmask_resource.
  336.                      * Then in texture_destroy, we set cmask_resource to NULL. */
  337.                     r300->screen->cmask_resource = fb->cbufs[0]->texture;
  338.                 }
  339.                 pipe_mutex_unlock(r300->screen->cmask_mutex);
  340.             }
  341.  
  342.             if (r300->screen->cmask_resource == fb->cbufs[0]->texture) {
  343.                 r300_set_clear_color(r300, color);
  344.                 r300_mark_atom_dirty(r300, &r300->cmask_clear);
  345.                 r300_mark_atom_dirty(r300, &r300->gpu_flush);
  346.                 buffers &= ~PIPE_CLEAR_COLOR;
  347.             }
  348.         }
  349.     }
  350.     /* Enable CBZB clear. */
  351.     else if (r300_cbzb_clear_allowed(r300, buffers)) {
  352.         struct r300_surface *surf = r300_surface(fb->cbufs[0]);
  353.  
  354.         hyperz->zb_depthclearvalue =
  355.                 r300_depth_clear_cb_value(surf->base.format, color->f);
  356.  
  357.         width = surf->cbzb_width;
  358.         height = surf->cbzb_height;
  359.  
  360.         r300->cbzb_clear = TRUE;
  361.         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
  362.     }
  363.  
  364.     /* Clear. */
  365.     if (buffers) {
  366.         /* Clear using the blitter. */
  367.         r300_blitter_begin(r300, R300_CLEAR);
  368.         util_blitter_clear(r300->blitter, width, height, 1,
  369.                            buffers, color, depth, stencil);
  370.         r300_blitter_end(r300);
  371.     } else if (r300->zmask_clear.dirty ||
  372.                r300->hiz_clear.dirty ||
  373.                r300->cmask_clear.dirty) {
  374.         /* Just clear zmask and hiz now, this does not use the standard draw
  375.          * procedure. */
  376.         /* Calculate zmask_clear and hiz_clear atom sizes. */
  377.         unsigned dwords =
  378.             r300->gpu_flush.size +
  379.             (r300->zmask_clear.dirty ? r300->zmask_clear.size : 0) +
  380.             (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
  381.             (r300->cmask_clear.dirty ? r300->cmask_clear.size : 0) +
  382.             r300_get_num_cs_end_dwords(r300);
  383.  
  384.         /* Reserve CS space. */
  385.         if (dwords > (RADEON_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
  386.             r300_flush(&r300->context, RADEON_FLUSH_ASYNC, NULL);
  387.         }
  388.  
  389.         /* Emit clear packets. */
  390.         r300_emit_gpu_flush(r300, r300->gpu_flush.size, r300->gpu_flush.state);
  391.         r300->gpu_flush.dirty = FALSE;
  392.  
  393.         if (r300->zmask_clear.dirty) {
  394.             r300_emit_zmask_clear(r300, r300->zmask_clear.size,
  395.                                   r300->zmask_clear.state);
  396.             r300->zmask_clear.dirty = FALSE;
  397.         }
  398.         if (r300->hiz_clear.dirty) {
  399.             r300_emit_hiz_clear(r300, r300->hiz_clear.size,
  400.                                 r300->hiz_clear.state);
  401.             r300->hiz_clear.dirty = FALSE;
  402.         }
  403.         if (r300->cmask_clear.dirty) {
  404.             r300_emit_cmask_clear(r300, r300->cmask_clear.size,
  405.                                   r300->cmask_clear.state);
  406.             r300->cmask_clear.dirty = FALSE;
  407.         }
  408.     } else {
  409.         assert(0);
  410.     }
  411.  
  412.     /* Disable CBZB clear. */
  413.     if (r300->cbzb_clear) {
  414.         r300->cbzb_clear = FALSE;
  415.         hyperz->zb_depthclearvalue = hyperz_dcv;
  416.         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
  417.     }
  418.  
  419.     /* Enable fastfill and/or hiz.
  420.      *
  421.      * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
  422.      * looks if zmask/hiz is in use and programs hardware accordingly. */
  423.     if (r300->zmask_in_use || r300->hiz_in_use) {
  424.         r300_mark_atom_dirty(r300, &r300->hyperz_state);
  425.     }
  426. }
  427.  
  428. /* Clear a region of a color surface to a constant value. */
  429. static void r300_clear_render_target(struct pipe_context *pipe,
  430.                                      struct pipe_surface *dst,
  431.                                      const union pipe_color_union *color,
  432.                                      unsigned dstx, unsigned dsty,
  433.                                      unsigned width, unsigned height)
  434. {
  435.     struct r300_context *r300 = r300_context(pipe);
  436.  
  437.     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
  438.     util_blitter_clear_render_target(r300->blitter, dst, color,
  439.                                      dstx, dsty, width, height);
  440.     r300_blitter_end(r300);
  441. }
  442.  
  443. /* Clear a region of a depth stencil surface. */
  444. static void r300_clear_depth_stencil(struct pipe_context *pipe,
  445.                                      struct pipe_surface *dst,
  446.                                      unsigned clear_flags,
  447.                                      double depth,
  448.                                      unsigned stencil,
  449.                                      unsigned dstx, unsigned dsty,
  450.                                      unsigned width, unsigned height)
  451. {
  452.     struct r300_context *r300 = r300_context(pipe);
  453.     struct pipe_framebuffer_state *fb =
  454.         (struct pipe_framebuffer_state*)r300->fb_state.state;
  455.  
  456.     if (r300->zmask_in_use && !r300->locked_zbuffer) {
  457.         if (fb->zsbuf->texture == dst->texture) {
  458.             r300_decompress_zmask(r300);
  459.         }
  460.     }
  461.  
  462.     /* XXX Do not decompress ZMask of the currently-set zbuffer. */
  463.     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
  464.     util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
  465.                                      dstx, dsty, width, height);
  466.     r300_blitter_end(r300);
  467. }
  468.  
  469. void r300_decompress_zmask(struct r300_context *r300)
  470. {
  471.     struct pipe_framebuffer_state *fb =
  472.         (struct pipe_framebuffer_state*)r300->fb_state.state;
  473.  
  474.     if (!r300->zmask_in_use || r300->locked_zbuffer)
  475.         return;
  476.  
  477.     r300->zmask_decompress = TRUE;
  478.     r300_mark_atom_dirty(r300, &r300->hyperz_state);
  479.  
  480.     r300_blitter_begin(r300, R300_DECOMPRESS);
  481.     util_blitter_custom_clear_depth(r300->blitter, fb->width, fb->height, 0,
  482.                                     r300->dsa_decompress_zmask);
  483.     r300_blitter_end(r300);
  484.  
  485.     r300->zmask_decompress = FALSE;
  486.     r300->zmask_in_use = FALSE;
  487.     r300_mark_atom_dirty(r300, &r300->hyperz_state);
  488. }
  489.  
  490. void r300_decompress_zmask_locked_unsafe(struct r300_context *r300)
  491. {
  492.     struct pipe_framebuffer_state fb;
  493.  
  494.     memset(&fb, 0, sizeof(fb));
  495.     fb.width = r300->locked_zbuffer->width;
  496.     fb.height = r300->locked_zbuffer->height;
  497.     fb.zsbuf = r300->locked_zbuffer;
  498.  
  499.     r300->context.set_framebuffer_state(&r300->context, &fb);
  500.     r300_decompress_zmask(r300);
  501. }
  502.  
  503. void r300_decompress_zmask_locked(struct r300_context *r300)
  504. {
  505.     struct pipe_framebuffer_state saved_fb;
  506.  
  507.     memset(&saved_fb, 0, sizeof(saved_fb));
  508.     util_copy_framebuffer_state(&saved_fb, r300->fb_state.state);
  509.     r300_decompress_zmask_locked_unsafe(r300);
  510.     r300->context.set_framebuffer_state(&r300->context, &saved_fb);
  511.     util_unreference_framebuffer_state(&saved_fb);
  512.  
  513.     pipe_surface_reference(&r300->locked_zbuffer, NULL);
  514. }
  515.  
  516. bool r300_is_blit_supported(enum pipe_format format)
  517. {
  518.     const struct util_format_description *desc =
  519.         util_format_description(format);
  520.  
  521.     return desc->layout == UTIL_FORMAT_LAYOUT_PLAIN ||
  522.            desc->layout == UTIL_FORMAT_LAYOUT_S3TC ||
  523.            desc->layout == UTIL_FORMAT_LAYOUT_RGTC;
  524. }
  525.  
  526. /* Copy a block of pixels from one surface to another. */
  527. static void r300_resource_copy_region(struct pipe_context *pipe,
  528.                                       struct pipe_resource *dst,
  529.                                       unsigned dst_level,
  530.                                       unsigned dstx, unsigned dsty, unsigned dstz,
  531.                                       struct pipe_resource *src,
  532.                                       unsigned src_level,
  533.                                       const struct pipe_box *src_box)
  534. {
  535.     struct pipe_screen *screen = pipe->screen;
  536.     struct r300_context *r300 = r300_context(pipe);
  537.     struct pipe_framebuffer_state *fb =
  538.         (struct pipe_framebuffer_state*)r300->fb_state.state;
  539.     unsigned src_width0 = r300_resource(src)->tex.width0;
  540.     unsigned src_height0 = r300_resource(src)->tex.height0;
  541.     unsigned dst_width0 = r300_resource(dst)->tex.width0;
  542.     unsigned dst_height0 = r300_resource(dst)->tex.height0;
  543.     unsigned layout;
  544.     struct pipe_box box, dstbox;
  545.     struct pipe_sampler_view src_templ, *src_view;
  546.     struct pipe_surface dst_templ, *dst_view;
  547.  
  548.     /* Fallback for buffers. */
  549.     if ((dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) ||
  550.         !r300_is_blit_supported(dst->format)) {
  551.         util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
  552.                                   src, src_level, src_box);
  553.         return;
  554.     }
  555.  
  556.     /* Can't read MSAA textures. */
  557.     if (src->nr_samples > 1 || dst->nr_samples > 1) {
  558.         return;
  559.     }
  560.  
  561.     /* The code below changes the texture format so that the copy can be done
  562.      * on hardware. E.g. depth-stencil surfaces are copied as RGBA
  563.      * colorbuffers. */
  564.  
  565.     util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
  566.     util_blitter_default_src_texture(&src_templ, src, src_level);
  567.  
  568.     layout = util_format_description(dst_templ.format)->layout;
  569.  
  570.     /* Handle non-renderable plain formats. */
  571.     if (layout == UTIL_FORMAT_LAYOUT_PLAIN &&
  572.         (!screen->is_format_supported(screen, src_templ.format, src->target,
  573.                                       src->nr_samples,
  574.                                       PIPE_BIND_SAMPLER_VIEW) ||
  575.          !screen->is_format_supported(screen, dst_templ.format, dst->target,
  576.                                       dst->nr_samples,
  577.                                       PIPE_BIND_RENDER_TARGET))) {
  578.         switch (util_format_get_blocksize(dst_templ.format)) {
  579.             case 1:
  580.                 dst_templ.format = PIPE_FORMAT_I8_UNORM;
  581.                 break;
  582.             case 2:
  583.                 dst_templ.format = PIPE_FORMAT_B4G4R4A4_UNORM;
  584.                 break;
  585.             case 4:
  586.                 dst_templ.format = PIPE_FORMAT_B8G8R8A8_UNORM;
  587.                 break;
  588.             case 8:
  589.                 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UNORM;
  590.                 break;
  591.             default:
  592.                 debug_printf("r300: copy_region: Unhandled format: %s. Falling back to software.\n"
  593.                              "r300: copy_region: Software fallback doesn't work for tiled textures.\n",
  594.                              util_format_short_name(dst_templ.format));
  595.         }
  596.         src_templ.format = dst_templ.format;
  597.     }
  598.  
  599.     /* Handle compressed formats. */
  600.     if (layout == UTIL_FORMAT_LAYOUT_S3TC ||
  601.         layout == UTIL_FORMAT_LAYOUT_RGTC) {
  602.         assert(src_templ.format == dst_templ.format);
  603.  
  604.         box = *src_box;
  605.         src_box = &box;
  606.  
  607.         dst_width0 = align(dst_width0, 4);
  608.         dst_height0 = align(dst_height0, 4);
  609.         src_width0 = align(src_width0, 4);
  610.         src_height0 = align(src_height0, 4);
  611.         box.width = align(box.width, 4);
  612.         box.height = align(box.height, 4);
  613.  
  614.         switch (util_format_get_blocksize(dst_templ.format)) {
  615.         case 8:
  616.             /* one 4x4 pixel block has 8 bytes.
  617.              * we set 1 pixel = 4 bytes ===> 1 block corrensponds to 2 pixels. */
  618.             dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
  619.             dst_width0 = dst_width0 / 2;
  620.             src_width0 = src_width0 / 2;
  621.             dstx /= 2;
  622.             box.x /= 2;
  623.             box.width /= 2;
  624.             break;
  625.         case 16:
  626.             /* one 4x4 pixel block has 16 bytes.
  627.              * we set 1 pixel = 4 bytes ===> 1 block corresponds to 4 pixels. */
  628.             dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
  629.             break;
  630.         }
  631.         src_templ.format = dst_templ.format;
  632.  
  633.         dst_height0 = dst_height0 / 4;
  634.         src_height0 = src_height0 / 4;
  635.         dsty /= 4;
  636.         box.y /= 4;
  637.         box.height /= 4;
  638.     }
  639.  
  640.     /* Fallback for textures. */
  641.     if (!screen->is_format_supported(screen, dst_templ.format,
  642.                                      dst->target, dst->nr_samples,
  643.                                      PIPE_BIND_RENDER_TARGET) ||
  644.         !screen->is_format_supported(screen, src_templ.format,
  645.                                      src->target, src->nr_samples,
  646.                                      PIPE_BIND_SAMPLER_VIEW)) {
  647.         assert(0 && "this shouldn't happen, update r300_is_blit_supported");
  648.         util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
  649.                                   src, src_level, src_box);
  650.         return;
  651.     }
  652.  
  653.     /* Decompress ZMASK. */
  654.     if (r300->zmask_in_use && !r300->locked_zbuffer) {
  655.         if (fb->zsbuf->texture == src ||
  656.             fb->zsbuf->texture == dst) {
  657.             r300_decompress_zmask(r300);
  658.         }
  659.     }
  660.  
  661.     dst_view = r300_create_surface_custom(pipe, dst, &dst_templ, dst_width0, dst_height0);
  662.     src_view = r300_create_sampler_view_custom(pipe, src, &src_templ, src_width0, src_height0);
  663.  
  664.     u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
  665.              abs(src_box->depth), &dstbox);
  666.  
  667.     r300_blitter_begin(r300, R300_COPY);
  668.     util_blitter_blit_generic(r300->blitter, dst_view, &dstbox,
  669.                               src_view, src_box, src_width0, src_height0,
  670.                               PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL);
  671.     r300_blitter_end(r300);
  672.  
  673.     pipe_surface_reference(&dst_view, NULL);
  674.     pipe_sampler_view_reference(&src_view, NULL);
  675. }
  676.  
  677. static boolean r300_is_simple_msaa_resolve(const struct pipe_blit_info *info)
  678. {
  679.     unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
  680.     unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
  681.  
  682.     return info->src.resource->nr_samples > 1 &&
  683.            info->dst.resource->nr_samples <= 1 &&
  684.            info->dst.resource->format == info->src.resource->format &&
  685.            info->dst.resource->format == info->dst.format &&
  686.            info->src.resource->format == info->src.format &&
  687.            !info->scissor_enable &&
  688.            info->mask == PIPE_MASK_RGBA &&
  689.            dst_width == info->src.resource->width0 &&
  690.            dst_height == info->src.resource->height0 &&
  691.            info->dst.box.x == 0 &&
  692.            info->dst.box.y == 0 &&
  693.            info->dst.box.width == dst_width &&
  694.            info->dst.box.height == dst_height &&
  695.            info->src.box.x == 0 &&
  696.            info->src.box.y == 0 &&
  697.            info->src.box.width == dst_width &&
  698.            info->src.box.height == dst_height &&
  699.            (r300_resource(info->dst.resource)->tex.microtile != RADEON_LAYOUT_LINEAR ||
  700.             r300_resource(info->dst.resource)->tex.macrotile[info->dst.level] != RADEON_LAYOUT_LINEAR);
  701. }
  702.  
  703. static void r300_simple_msaa_resolve(struct pipe_context *pipe,
  704.                                      struct pipe_resource *dst,
  705.                                      unsigned dst_level,
  706.                                      unsigned dst_layer,
  707.                                      struct pipe_resource *src,
  708.                                      enum pipe_format format)
  709. {
  710.     struct r300_context *r300 = r300_context(pipe);
  711.     struct r300_surface *srcsurf, *dstsurf;
  712.     struct pipe_surface surf_tmpl;
  713.     struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
  714.  
  715.     memset(&surf_tmpl, 0, sizeof(surf_tmpl));
  716.     surf_tmpl.format = format;
  717.     srcsurf = r300_surface(pipe->create_surface(pipe, src, &surf_tmpl));
  718.  
  719.     surf_tmpl.format = format;
  720.     surf_tmpl.u.tex.level = dst_level;
  721.     surf_tmpl.u.tex.first_layer =
  722.     surf_tmpl.u.tex.last_layer = dst_layer;
  723.     dstsurf = r300_surface(pipe->create_surface(pipe, dst, &surf_tmpl));
  724.  
  725.     /* COLORPITCH should contain the tiling info of the resolve buffer.
  726.      * The tiling of the AA buffer isn't programmable anyway. */
  727.     srcsurf->pitch &= ~(R300_COLOR_TILE(1) | R300_COLOR_MICROTILE(3));
  728.     srcsurf->pitch |= dstsurf->pitch & (R300_COLOR_TILE(1) | R300_COLOR_MICROTILE(3));
  729.  
  730.     /* Enable AA resolve. */
  731.     aa->dest = dstsurf;
  732.     r300->aa_state.size = 8;
  733.     r300_mark_atom_dirty(r300, &r300->aa_state);
  734.  
  735.     /* Resolve the surface. */
  736.     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
  737.     util_blitter_custom_color(r300->blitter, &srcsurf->base, NULL);
  738.     r300_blitter_end(r300);
  739.  
  740.     /* Disable AA resolve. */
  741.     aa->dest = NULL;
  742.     r300->aa_state.size = 4;
  743.     r300_mark_atom_dirty(r300, &r300->aa_state);
  744.  
  745.     pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL);
  746.     pipe_surface_reference((struct pipe_surface**)&dstsurf, NULL);
  747. }
  748.  
  749. static void r300_msaa_resolve(struct pipe_context *pipe,
  750.                               const struct pipe_blit_info *info)
  751. {
  752.     struct r300_context *r300 = r300_context(pipe);
  753.     struct pipe_screen *screen = pipe->screen;
  754.     struct pipe_resource *tmp, templ;
  755.     struct pipe_blit_info blit;
  756.  
  757.     assert(info->src.level == 0);
  758.     assert(info->src.box.z == 0);
  759.     assert(info->src.box.depth == 1);
  760.     assert(info->dst.box.depth == 1);
  761.  
  762.     if (r300_is_simple_msaa_resolve(info)) {
  763.         r300_simple_msaa_resolve(pipe, info->dst.resource, info->dst.level,
  764.                                  info->dst.box.z, info->src.resource,
  765.                                  info->src.format);
  766.         return;
  767.     }
  768.  
  769.     /* resolve into a temporary texture, then blit */
  770.     memset(&templ, 0, sizeof(templ));
  771.     templ.target = PIPE_TEXTURE_2D;
  772.     templ.format = info->src.resource->format;
  773.     templ.width0 = info->src.resource->width0;
  774.     templ.height0 = info->src.resource->height0;
  775.     templ.depth0 = 1;
  776.     templ.array_size = 1;
  777.     templ.usage = PIPE_USAGE_DEFAULT;
  778.     templ.flags = R300_RESOURCE_FORCE_MICROTILING;
  779.  
  780.     tmp = screen->resource_create(screen, &templ);
  781.  
  782.     /* resolve */
  783.     r300_simple_msaa_resolve(pipe, tmp, 0, 0, info->src.resource,
  784.                              info->src.format);
  785.  
  786.     /* blit */
  787.     blit = *info;
  788.     blit.src.resource = tmp;
  789.     blit.src.box.z = 0;
  790.  
  791.     r300_blitter_begin(r300, R300_BLIT | R300_IGNORE_RENDER_COND);
  792.     util_blitter_blit(r300->blitter, &blit);
  793.     r300_blitter_end(r300);
  794.  
  795.     pipe_resource_reference(&tmp, NULL);
  796. }
  797.  
  798. static void r300_blit(struct pipe_context *pipe,
  799.                       const struct pipe_blit_info *blit)
  800. {
  801.     struct r300_context *r300 = r300_context(pipe);
  802.     struct pipe_framebuffer_state *fb =
  803.         (struct pipe_framebuffer_state*)r300->fb_state.state;
  804.     struct pipe_blit_info info = *blit;
  805.  
  806.     /* The driver supports sRGB textures but not framebuffers. Blitting
  807.      * from sRGB to sRGB should be the same as blitting from linear
  808.      * to linear, so use that, This avoids incorrect linearization.
  809.      */
  810.     if (util_format_is_srgb(info.src.format)) {
  811.       info.src.format = util_format_linear(info.src.format);
  812.       info.dst.format = util_format_linear(info.dst.format);
  813.     }
  814.  
  815.     /* MSAA resolve. */
  816.     if (info.src.resource->nr_samples > 1 &&
  817.         !util_format_is_depth_or_stencil(info.src.resource->format)) {
  818.         r300_msaa_resolve(pipe, &info);
  819.         return;
  820.     }
  821.  
  822.     /* Can't read MSAA textures. */
  823.     if (info.src.resource->nr_samples > 1) {
  824.         return;
  825.     }
  826.  
  827.     /* Blit a combined depth-stencil resource as color.
  828.      * S8Z24 is the only supported stencil format. */
  829.     if ((info.mask & PIPE_MASK_S) &&
  830.         info.src.format == PIPE_FORMAT_S8_UINT_Z24_UNORM &&
  831.         info.dst.format == PIPE_FORMAT_S8_UINT_Z24_UNORM) {
  832.         if (info.dst.resource->nr_samples > 1) {
  833.             /* Cannot do that with MSAA buffers. */
  834.             info.mask &= ~PIPE_MASK_S;
  835.             if (!(info.mask & PIPE_MASK_Z)) {
  836.                 return;
  837.             }
  838.         } else {
  839.             /* Single-sample buffer. */
  840.             info.src.format = PIPE_FORMAT_B8G8R8A8_UNORM;
  841.             info.dst.format = PIPE_FORMAT_B8G8R8A8_UNORM;
  842.             if (info.mask & PIPE_MASK_Z) {
  843.                 info.mask = PIPE_MASK_RGBA; /* depth+stencil */
  844.             } else {
  845.                 info.mask = PIPE_MASK_B; /* stencil only */
  846.             }
  847.         }
  848.     }
  849.  
  850.     /* Decompress ZMASK. */
  851.     if (r300->zmask_in_use && !r300->locked_zbuffer) {
  852.         if (fb->zsbuf->texture == info.src.resource ||
  853.             fb->zsbuf->texture == info.dst.resource) {
  854.             r300_decompress_zmask(r300);
  855.         }
  856.     }
  857.  
  858.     r300_blitter_begin(r300, R300_BLIT |
  859.                        (info.render_condition_enable ? 0 : R300_IGNORE_RENDER_COND));
  860.     util_blitter_blit(r300->blitter, &info);
  861.     r300_blitter_end(r300);
  862. }
  863.  
  864. static void r300_flush_resource(struct pipe_context *ctx,
  865.                                 struct pipe_resource *resource)
  866. {
  867. }
  868.  
  869. void r300_init_blit_functions(struct r300_context *r300)
  870. {
  871.     r300->context.clear = r300_clear;
  872.     r300->context.clear_render_target = r300_clear_render_target;
  873.     r300->context.clear_depth_stencil = r300_clear_depth_stencil;
  874.     r300->context.resource_copy_region = r300_resource_copy_region;
  875.     r300->context.blit = r300_blit;
  876.     r300->context.flush_resource = r300_flush_resource;
  877. }
  878.