src/gallium/drivers/r300/r300_hyperz.c

   1 /*
   2  * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
   3  * Copyright 2009 Marek Olšák <maraeo@gmail.com>
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * on the rights to use, copy, modify, merge, publish, distribute, sub
   9  * license, and/or sell copies of the Software, and to permit persons to whom
  10  * the Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  20  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  21  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  22  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
  23
  24 #include "r300_context.h"
  25 #include "r300_reg.h"
  26 #include "r300_fs.h"
  27
  28 #include "util/u_format.h"
  29 #include "util/u_mm.h"
  30
  31 /*
  32   HiZ rules - taken from various docs
  33    1. HiZ only works on depth values
  34    2. Cannot HiZ if stencil fail or zfail is !KEEP
  35    3. on R300/400, HiZ is disabled if depth test is EQUAL
  36    4. comparison changes without clears usually mean disabling HiZ
  37 */
  38 /*****************************************************************************/
  39 /* The HyperZ setup                                                          */
  40 /*****************************************************************************/
  41
  42 static enum r300_hiz_func r300_get_hiz_func(struct r300_context *r300)
  43 {
  44     struct r300_dsa_state *dsa = r300->dsa_state.state;
  45
  46     if (!dsa->dsa.depth.enabled || !dsa->dsa.depth.writemask)
  47         return HIZ_FUNC_NONE;
  48
  49     switch (dsa->dsa.depth.func) {
  50     case PIPE_FUNC_NEVER:
  51     case PIPE_FUNC_EQUAL:
  52     case PIPE_FUNC_NOTEQUAL:
  53     case PIPE_FUNC_ALWAYS:
  54         return HIZ_FUNC_NONE;
  55
  56     case PIPE_FUNC_LESS:
  57     case PIPE_FUNC_LEQUAL:
  58         return HIZ_FUNC_MAX;
  59
  60     case PIPE_FUNC_GREATER:
  61     case PIPE_FUNC_GEQUAL:
  62         return HIZ_FUNC_MIN;
  63
  64     default:
  65         assert(0);
  66         return HIZ_FUNC_NONE;
  67     }
  68 }
  69
  70 /* Return what's used for the depth test (either minimum or maximum). */
  71 static unsigned r300_get_sc_hz_max(struct r300_context *r300)
  72 {
  73     struct r300_dsa_state *dsa = r300->dsa_state.state;
  74     unsigned func = dsa->dsa.depth.func;
  75
  76     return func >= PIPE_FUNC_GREATER ? R300_SC_HYPERZ_MAX : R300_SC_HYPERZ_MIN;
  77 }
  78
  79 static boolean r300_is_hiz_func_valid(struct r300_context *r300)
  80 {
  81     struct r300_dsa_state *dsa = r300->dsa_state.state;
  82     unsigned func = dsa->dsa.depth.func;
  83
  84     if (r300->hiz_func == HIZ_FUNC_NONE)
  85         return TRUE;
  86
  87     /* func1 is less/lessthan */
  88     if (r300->hiz_func == HIZ_FUNC_MAX &&
  89         (func == PIPE_FUNC_GEQUAL || func == PIPE_FUNC_GREATER))
  90         return FALSE;
  91
  92     /* func1 is greater/greaterthan */
  93     if (r300->hiz_func == HIZ_FUNC_MIN &&
  94         (func == PIPE_FUNC_LESS   || func == PIPE_FUNC_LEQUAL))
  95         return FALSE;
  96
  97     return TRUE;
  98 }
  99
 100 static boolean r300_dsa_stencil_op_not_keep(struct pipe_stencil_state *s)
 101 {
 102     return s->enabled && (s->fail_op != PIPE_STENCIL_OP_KEEP ||
 103                           s->zfail_op != PIPE_STENCIL_OP_KEEP);
 104 }
 105
 106 static boolean r300_can_hiz(struct r300_context *r300)
 107 {
 108     struct r300_dsa_state *dsa = r300->dsa_state.state;
 109     struct r300_screen *r300screen = r300->screen;
 110
 111     /* shader writes depth - no HiZ */
 112     if (r300_fragment_shader_writes_depth(r300_fs(r300))) /* (5) */
 113         return FALSE;
 114
 115     if (r300->query_current)
 116         return FALSE;
 117
 118     /* if stencil fail/zfail op is not KEEP */
 119     if (r300_dsa_stencil_op_not_keep(&dsa->dsa.stencil[0]) ||
 120         r300_dsa_stencil_op_not_keep(&dsa->dsa.stencil[1]))
 121         return FALSE;
 122
 123     if (dsa->dsa.depth.enabled) {
 124         /* if depth func is EQUAL pre-r500 */
 125         if (dsa->dsa.depth.func == PIPE_FUNC_EQUAL && !r300screen->caps.is_r500)
 126             return FALSE;
 127
 128         /* if depth func is NOTEQUAL */
 129         if (dsa->dsa.depth.func == PIPE_FUNC_NOTEQUAL)
 130             return FALSE;
 131     }
 132     return TRUE;
 133 }
 134
 135 static void r300_update_hyperz(struct r300_context* r300)
 136 {
 137     struct r300_hyperz_state *z =
 138         (struct r300_hyperz_state*)r300->hyperz_state.state;
 139     struct pipe_framebuffer_state *fb =
 140         (struct pipe_framebuffer_state*)r300->fb_state.state;
 141     struct r300_resource *zstex =
 142             fb->zsbuf ? r300_resource(fb->zsbuf->texture) : NULL;
 143
 144     z->gb_z_peq_config = 0;
 145     z->zb_bw_cntl = 0;
 146     z->sc_hyperz = R300_SC_HYPERZ_ADJ_2;
 147     z->flush = 0;
 148
 149     if (r300->cbzb_clear) {
 150         z->zb_bw_cntl |= R300_ZB_CB_CLEAR_CACHE_LINE_WRITE_ONLY;
 151         return;
 152     }
 153
 154     if (!zstex ||
 155         !r300->rws->get_value(r300->rws, RADEON_VID_CAN_HYPERZ))
 156         return;
 157
 158     /* Zbuffer compression. */
 159     if (r300->zmask_in_use && !r300->locked_zbuffer) {
 160         z->zb_bw_cntl |= R300_FAST_FILL_ENABLE |
 161                          /*R300_FORCE_COMPRESSED_STENCIL_VALUE_ENABLE |*/
 162                          R300_RD_COMP_ENABLE;
 163
 164         if (!r300->zmask_decompress) {
 165             z->zb_bw_cntl |= R300_WR_COMP_ENABLE;
 166         }
 167     }
 168
 169     if (zstex->tex.zcomp8x8[fb->zsbuf->u.tex.level]) {
 170         z->gb_z_peq_config |= R300_GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_8_8;
 171     }
 172
 173     /* HiZ. */
 174     if (r300->hiz_in_use && !r300->locked_zbuffer) {
 175         /* Set the HiZ function if needed. */
 176         if (r300->hiz_func == HIZ_FUNC_NONE) {
 177             r300->hiz_func = r300_get_hiz_func(r300);
 178         }
 179
 180         /* If the depth function is inverted, HiZ must be disabled. */
 181         if (!r300_is_hiz_func_valid(r300)) {
 182             r300->hiz_in_use = FALSE;
 183         } else if (r300_can_hiz(r300)) {
 184             /* Setup the HiZ bits. */
 185             z->zb_bw_cntl |=
 186                 R300_HIZ_ENABLE |
 187                 (r300->hiz_func == HIZ_FUNC_MIN ? R300_HIZ_MIN : R300_HIZ_MAX);
 188
 189             z->sc_hyperz |= R300_SC_HYPERZ_ENABLE |
 190                             r300_get_sc_hz_max(r300);
 191
 192             if (r300->screen->caps.is_r500) {
 193                 z->zb_bw_cntl |= R500_HIZ_EQUAL_REJECT_ENABLE;
 194             }
 195         }
 196     }
 197
 198     /* R500-specific features and optimizations. */
 199     if (r300->screen->caps.is_r500) {
 200         z->zb_bw_cntl |= R500_PEQ_PACKING_ENABLE |
 201                          R500_COVERED_PTR_MASKING_ENABLE;
 202     }
 203 }
 204
 205 /*****************************************************************************/
 206 /* The ZTOP state                                                            */
 207 /*****************************************************************************/
 208
 209 static boolean r300_dsa_writes_stencil(
 210         struct pipe_stencil_state *s)
 211 {
 212     return s->enabled && s->writemask &&
 213            (s->fail_op  != PIPE_STENCIL_OP_KEEP ||
 214             s->zfail_op != PIPE_STENCIL_OP_KEEP ||
 215             s->zpass_op != PIPE_STENCIL_OP_KEEP);
 216 }
 217
 218 static boolean r300_dsa_writes_depth_stencil(
 219         struct pipe_depth_stencil_alpha_state *dsa)
 220 {
 221     /* We are interested only in the cases when a depth or stencil value
 222      * can be changed. */
 223
 224     if (dsa->depth.enabled && dsa->depth.writemask &&
 225         dsa->depth.func != PIPE_FUNC_NEVER)
 226         return TRUE;
 227
 228     if (r300_dsa_writes_stencil(&dsa->stencil[0]) ||
 229         r300_dsa_writes_stencil(&dsa->stencil[1]))
 230         return TRUE;
 231
 232     return FALSE;
 233 }
 234
 235 static boolean r300_dsa_alpha_test_enabled(
 236         struct pipe_depth_stencil_alpha_state *dsa)
 237 {
 238     /* We are interested only in the cases when alpha testing can kill
 239      * a fragment. */
 240
 241     return dsa->alpha.enabled && dsa->alpha.func != PIPE_FUNC_ALWAYS;
 242 }
 243
 244 static void r300_update_ztop(struct r300_context* r300)
 245 {
 246     struct r300_ztop_state* ztop_state =
 247         (struct r300_ztop_state*)r300->ztop_state.state;
 248     uint32_t old_ztop = ztop_state->z_buffer_top;
 249
 250     /* This is important enough that I felt it warranted a comment.
 251      *
 252      * According to the docs, these are the conditions where ZTOP must be
 253      * disabled:
 254      * 1) Alpha testing enabled
 255      * 2) Texture kill instructions in fragment shader
 256      * 3) Chroma key culling enabled
 257      * 4) W-buffering enabled
 258      *
 259      * The docs claim that for the first three cases, if no ZS writes happen,
 260      * then ZTOP can be used.
 261      *
 262      * (3) will never apply since we do not support chroma-keyed operations.
 263      * (4) will need to be re-examined (and this comment updated) if/when
 264      * Hyper-Z becomes supported.
 265      *
 266      * Additionally, the following conditions require disabled ZTOP:
 267      * 5) Depth writes in fragment shader
 268      * 6) Outstanding occlusion queries
 269      *
 270      * This register causes stalls all the way from SC to CB when changed,
 271      * but it is buffered on-chip so it does not hurt to write it if it has
 272      * not changed.
 273      *
 274      * ~C.
 275      */
 276
 277     /* ZS writes */
 278     if (r300_dsa_writes_depth_stencil(r300->dsa_state.state) &&
 279            (r300_dsa_alpha_test_enabled(r300->dsa_state.state) ||  /* (1) */
 280             r300_fs(r300)->shader->info.uses_kill)) {              /* (2) */
 281         ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
 282     } else if (r300_fragment_shader_writes_depth(r300_fs(r300))) { /* (5) */
 283         ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
 284     } else if (r300->query_current) {                              /* (6) */
 285         ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
 286     } else {
 287         ztop_state->z_buffer_top = R300_ZTOP_ENABLE;
 288     }
 289     if (ztop_state->z_buffer_top != old_ztop)
 290         r300_mark_atom_dirty(r300, &r300->ztop_state);
 291 }
 292
 293 void r300_update_hyperz_state(struct r300_context* r300)
 294 {
 295     r300_update_ztop(r300);
 296
 297     if (r300->hyperz_state.dirty) {
 298         r300_update_hyperz(r300);
 299     }
 300 }