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     switch (dsa->dsa.depth.func) {
  47     case PIPE_FUNC_NEVER:
  48     case PIPE_FUNC_EQUAL:
  49     case PIPE_FUNC_NOTEQUAL:
  50     case PIPE_FUNC_ALWAYS:
  51     default:
  52         /* Guess MAX for uncertain cases. */
  53     case PIPE_FUNC_LESS:
  54     case PIPE_FUNC_LEQUAL:
  55         return HIZ_FUNC_MAX;
  56
  57     case PIPE_FUNC_GREATER:
  58     case PIPE_FUNC_GEQUAL:
  59         return HIZ_FUNC_MIN;
  60     }
  61 }
  62
  63 /* Return what's used for the depth test (either minimum or maximum). */
  64 static unsigned r300_get_sc_hz_max(struct r300_context *r300)
  65 {
  66     struct r300_dsa_state *dsa = r300->dsa_state.state;
  67     unsigned func = dsa->dsa.depth.func;
  68
  69     return func >= PIPE_FUNC_GREATER ? R300_SC_HYPERZ_MAX : R300_SC_HYPERZ_MIN;
  70 }
  71
  72 static boolean r300_is_hiz_func_valid(struct r300_context *r300)
  73 {
  74     struct r300_dsa_state *dsa = r300->dsa_state.state;
  75     unsigned func = dsa->dsa.depth.func;
  76
  77     if (r300->hiz_func == HIZ_FUNC_NONE)
  78         return TRUE;
  79
  80     /* func1 is less/lessthan */
  81     if (r300->hiz_func == HIZ_FUNC_MAX &&
  82         (func == PIPE_FUNC_GEQUAL || func == PIPE_FUNC_GREATER))
  83         return FALSE;
  84
  85     /* func1 is greater/greaterthan */
  86     if (r300->hiz_func == HIZ_FUNC_MIN &&
  87         (func == PIPE_FUNC_LESS   || func == PIPE_FUNC_LEQUAL))
  88         return FALSE;
  89
  90     return TRUE;
  91 }
  92
  93 static boolean r300_dsa_stencil_op_not_keep(struct pipe_stencil_state *s)
  94 {
  95     return s->enabled && (s->fail_op != PIPE_STENCIL_OP_KEEP ||
  96                           s->zfail_op != PIPE_STENCIL_OP_KEEP);
  97 }
  98
  99 static boolean r300_hiz_allowed(struct r300_context *r300)
 100 {
 101     struct r300_dsa_state *dsa = r300->dsa_state.state;
 102     struct r300_screen *r300screen = r300->screen;
 103
 104     if (r300_fragment_shader_writes_depth(r300_fs(r300)))
 105         return FALSE;
 106
 107     if (r300->query_current)
 108         return FALSE;
 109
 110     /* If the depth function is inverted, HiZ must be disabled. */
 111     if (!r300_is_hiz_func_valid(r300))
 112         return FALSE;
 113
 114     /* if stencil fail/zfail op is not KEEP */
 115     if (r300_dsa_stencil_op_not_keep(&dsa->dsa.stencil[0]) ||
 116         r300_dsa_stencil_op_not_keep(&dsa->dsa.stencil[1]))
 117         return FALSE;
 118
 119     if (dsa->dsa.depth.enabled) {
 120         /* if depth func is EQUAL pre-r500 */
 121         if (dsa->dsa.depth.func == PIPE_FUNC_EQUAL && !r300screen->caps.is_r500)
 122             return FALSE;
 123
 124         /* if depth func is NOTEQUAL */
 125         if (dsa->dsa.depth.func == PIPE_FUNC_NOTEQUAL)
 126             return FALSE;
 127     }
 128     return TRUE;
 129 }
 130
 131 static void r300_update_hyperz(struct r300_context* r300)
 132 {
 133     struct r300_hyperz_state *z =
 134         (struct r300_hyperz_state*)r300->hyperz_state.state;
 135     struct pipe_framebuffer_state *fb =
 136         (struct pipe_framebuffer_state*)r300->fb_state.state;
 137     struct r300_dsa_state *dsa = r300->dsa_state.state;
 138     struct r300_resource *zstex =
 139             fb->zsbuf ? r300_resource(fb->zsbuf->texture) : NULL;
 140
 141     z->gb_z_peq_config = 0;
 142     z->zb_bw_cntl = 0;
 143     z->sc_hyperz = R300_SC_HYPERZ_ADJ_2;
 144     z->flush = 0;
 145
 146     if (r300->cbzb_clear) {
 147         z->zb_bw_cntl |= R300_ZB_CB_CLEAR_CACHE_LINE_WRITE_ONLY;
 148         return;
 149     }
 150
 151     if (!zstex || !r300->hyperz_enabled)
 152         return;
 153
 154     /* Set the size of ZMASK tiles. */
 155     if (zstex->tex.zcomp8x8[fb->zsbuf->u.tex.level]) {
 156         z->gb_z_peq_config |= R300_GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_8_8;
 157     }
 158
 159     /* R500-specific features and optimizations. */
 160     if (r300->screen->caps.is_r500) {
 161         z->zb_bw_cntl |= R500_PEQ_PACKING_ENABLE |
 162                          R500_COVERED_PTR_MASKING_ENABLE;
 163     }
 164
 165     /* Setup decompression if needed. No other HyperZ setting is required. */
 166     if (r300->zmask_decompress) {
 167         z->zb_bw_cntl |= R300_FAST_FILL_ENABLE |
 168                          R300_RD_COMP_ENABLE;
 169         return;
 170     }
 171
 172     /* Do not set anything if depth and stencil tests are off. */
 173     if (!dsa->dsa.depth.enabled &&
 174         !dsa->dsa.stencil[0].enabled &&
 175         !dsa->dsa.stencil[1].enabled) {
 176         assert(!dsa->dsa.depth.writemask);
 177         return;
 178     }
 179
 180     /* Zbuffer compression. */
 181     if (r300->zmask_in_use && !r300->locked_zbuffer) {
 182         z->zb_bw_cntl |= R300_FAST_FILL_ENABLE |
 183                          R300_RD_COMP_ENABLE |
 184                          R300_WR_COMP_ENABLE;
 185     }
 186
 187     /* HiZ. */
 188     if (r300->hiz_in_use && !r300->locked_zbuffer) {
 189         /* HiZ cannot be used under some circumstances. */
 190         if (!r300_hiz_allowed(r300)) {
 191             /* If writemask is disabled, the HiZ memory will not be changed,
 192              * so we can keep its content for later. */
 193             if (dsa->dsa.depth.writemask) {
 194                 r300->hiz_in_use = FALSE;
 195             }
 196             return;
 197         }
 198         DBG(r300, DBG_HYPERZ, "r300: Z-func: %i\n", dsa->dsa.depth.func);
 199
 200         /* Set the HiZ function if needed. */
 201         if (r300->hiz_func == HIZ_FUNC_NONE) {
 202             r300->hiz_func = r300_get_hiz_func(r300);
 203         }
 204
 205         /* Setup the HiZ bits. */
 206         z->zb_bw_cntl |= R300_HIZ_ENABLE |
 207                 (r300->hiz_func == HIZ_FUNC_MIN ? R300_HIZ_MIN : R300_HIZ_MAX);
 208
 209         z->sc_hyperz |= R300_SC_HYPERZ_ENABLE |
 210                         r300_get_sc_hz_max(r300);
 211
 212         if (r300->screen->caps.is_r500) {
 213             z->zb_bw_cntl |= R500_HIZ_EQUAL_REJECT_ENABLE;
 214         }
 215     }
 216 }
 217
 218 /*****************************************************************************/
 219 /* The ZTOP state                                                            */
 220 /*****************************************************************************/
 221
 222 static boolean r300_dsa_writes_stencil(
 223         struct pipe_stencil_state *s)
 224 {
 225     return s->enabled && s->writemask &&
 226            (s->fail_op  != PIPE_STENCIL_OP_KEEP ||
 227             s->zfail_op != PIPE_STENCIL_OP_KEEP ||
 228             s->zpass_op != PIPE_STENCIL_OP_KEEP);
 229 }
 230
 231 static boolean r300_dsa_writes_depth_stencil(
 232         struct pipe_depth_stencil_alpha_state *dsa)
 233 {
 234     /* We are interested only in the cases when a depth or stencil value
 235      * can be changed. */
 236
 237     if (dsa->depth.enabled && dsa->depth.writemask &&
 238         dsa->depth.func != PIPE_FUNC_NEVER)
 239         return TRUE;
 240
 241     if (r300_dsa_writes_stencil(&dsa->stencil[0]) ||
 242         r300_dsa_writes_stencil(&dsa->stencil[1]))
 243         return TRUE;
 244
 245     return FALSE;
 246 }
 247
 248 static boolean r300_dsa_alpha_test_enabled(
 249         struct pipe_depth_stencil_alpha_state *dsa)
 250 {
 251     /* We are interested only in the cases when alpha testing can kill
 252      * a fragment. */
 253
 254     return dsa->alpha.enabled && dsa->alpha.func != PIPE_FUNC_ALWAYS;
 255 }
 256
 257 static void r300_update_ztop(struct r300_context* r300)
 258 {
 259     struct r300_ztop_state* ztop_state =
 260         (struct r300_ztop_state*)r300->ztop_state.state;
 261     uint32_t old_ztop = ztop_state->z_buffer_top;
 262
 263     /* This is important enough that I felt it warranted a comment.
 264      *
 265      * According to the docs, these are the conditions where ZTOP must be
 266      * disabled:
 267      * 1) Alpha testing enabled
 268      * 2) Texture kill instructions in fragment shader
 269      * 3) Chroma key culling enabled
 270      * 4) W-buffering enabled
 271      *
 272      * The docs claim that for the first three cases, if no ZS writes happen,
 273      * then ZTOP can be used.
 274      *
 275      * (3) will never apply since we do not support chroma-keyed operations.
 276      * (4) will need to be re-examined (and this comment updated) if/when
 277      * Hyper-Z becomes supported.
 278      *
 279      * Additionally, the following conditions require disabled ZTOP:
 280      * 5) Depth writes in fragment shader
 281      * 6) Outstanding occlusion queries
 282      *
 283      * This register causes stalls all the way from SC to CB when changed,
 284      * but it is buffered on-chip so it does not hurt to write it if it has
 285      * not changed.
 286      *
 287      * ~C.
 288      */
 289
 290     /* ZS writes */
 291     if (r300_dsa_writes_depth_stencil(r300->dsa_state.state) &&
 292            (r300_dsa_alpha_test_enabled(r300->dsa_state.state) ||  /* (1) */
 293             r300_fs(r300)->shader->info.uses_kill)) {              /* (2) */
 294         ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
 295     } else if (r300_fragment_shader_writes_depth(r300_fs(r300))) { /* (5) */
 296         ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
 297     } else if (r300->query_current) {                              /* (6) */
 298         ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
 299     } else {
 300         ztop_state->z_buffer_top = R300_ZTOP_ENABLE;
 301     }
 302     if (ztop_state->z_buffer_top != old_ztop)
 303         r300_mark_atom_dirty(r300, &r300->ztop_state);
 304 }
 305
 306 void r300_update_hyperz_state(struct r300_context* r300)
 307 {
 308     r300_update_ztop(r300);
 309
 310     if (r300->hyperz_state.dirty) {
 311         r300_update_hyperz(r300);
 312     }
 313 }