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 #include "r300_winsys.h"
  28
  29 #include "util/u_format.h"
  30 #include "util/u_mm.h"
  31
  32 /*
  33   HiZ rules - taken from various docs
  34    1. HiZ only works on depth values
  35    2. Cannot HiZ if stencil fail or zfail is !KEEP
  36    3. on R300/400, HiZ is disabled if depth test is EQUAL
  37    4. comparison changes without clears usually mean disabling HiZ
  38 */
  39 /*****************************************************************************/
  40 /* The HyperZ setup                                                          */
  41 /*****************************************************************************/
  42
  43 static enum r300_hiz_func r300_get_hiz_func(struct r300_context *r300)
  44 {
  45     struct r300_dsa_state *dsa = r300->dsa_state.state;
  46
  47     if (!dsa->dsa.depth.enabled || !dsa->dsa.depth.writemask)
  48         return HIZ_FUNC_NONE;
  49
  50     switch (dsa->dsa.depth.func) {
  51     case PIPE_FUNC_NEVER:
  52     case PIPE_FUNC_EQUAL:
  53     case PIPE_FUNC_NOTEQUAL:
  54     case PIPE_FUNC_ALWAYS:
  55         return HIZ_FUNC_NONE;
  56
  57     case PIPE_FUNC_LESS:
  58     case PIPE_FUNC_LEQUAL:
  59         return HIZ_FUNC_MAX;
  60
  61     case PIPE_FUNC_GREATER:
  62     case PIPE_FUNC_GEQUAL:
  63         return HIZ_FUNC_MIN;
  64     }
  65 }
  66
  67 /* Return what's used for the depth test (either minimum or maximum). */
  68 static unsigned r300_get_sc_hz_max(struct r300_context *r300)
  69 {
  70     struct r300_dsa_state *dsa = r300->dsa_state.state;
  71     unsigned func = dsa->dsa.depth.func;
  72
  73     return func >= PIPE_FUNC_GREATER ? R300_SC_HYPERZ_MAX : R300_SC_HYPERZ_MIN;
  74 }
  75
  76 static boolean r300_is_hiz_func_valid(struct r300_context *r300)
  77 {
  78     struct r300_dsa_state *dsa = r300->dsa_state.state;
  79     unsigned func = dsa->dsa.depth.func;
  80
  81     if (r300->hiz_func == HIZ_FUNC_NONE)
  82         return TRUE;
  83
  84     /* func1 is less/lessthan */
  85     if (r300->hiz_func == HIZ_FUNC_MAX &&
  86         (func == PIPE_FUNC_GEQUAL || func == PIPE_FUNC_GREATER))
  87         return FALSE;
  88
  89     /* func1 is greater/greaterthan */
  90     if (r300->hiz_func == HIZ_FUNC_MIN &&
  91         (func == PIPE_FUNC_LESS   || func == PIPE_FUNC_LEQUAL))
  92         return FALSE;
  93
  94     return TRUE;
  95 }
  96
  97 static boolean r300_dsa_stencil_op_not_keep(struct pipe_stencil_state *s)
  98 {
  99     return s->enabled && (s->fail_op != PIPE_STENCIL_OP_KEEP ||
 100                           s->zfail_op != PIPE_STENCIL_OP_KEEP);
 101 }
 102
 103 static boolean r300_can_hiz(struct r300_context *r300)
 104 {
 105     struct r300_dsa_state *dsa = r300->dsa_state.state;
 106     struct r300_screen *r300screen = r300->screen;
 107
 108     /* shader writes depth - no HiZ */
 109     if (r300_fragment_shader_writes_depth(r300_fs(r300))) /* (5) */
 110         return FALSE;
 111
 112     if (r300->query_current)
 113         return FALSE;
 114
 115     /* if stencil fail/zfail op is not KEEP */
 116     if (r300_dsa_stencil_op_not_keep(&dsa->dsa.stencil[0]) ||
 117         r300_dsa_stencil_op_not_keep(&dsa->dsa.stencil[1]))
 118         return FALSE;
 119
 120     if (dsa->dsa.depth.enabled) {
 121         /* if depth func is EQUAL pre-r500 */
 122         if (dsa->dsa.depth.func == PIPE_FUNC_EQUAL && !r300screen->caps.is_r500)
 123             return FALSE;
 124
 125         /* if depth func is NOTEQUAL */
 126         if (dsa->dsa.depth.func == PIPE_FUNC_NOTEQUAL)
 127             return FALSE;
 128     }
 129     return TRUE;
 130 }
 131
 132 static void r300_update_hyperz(struct r300_context* r300)
 133 {
 134     struct r300_hyperz_state *z =
 135         (struct r300_hyperz_state*)r300->hyperz_state.state;
 136     struct pipe_framebuffer_state *fb =
 137         (struct pipe_framebuffer_state*)r300->fb_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 ||
 152         !r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
 153         return;
 154
 155     /* Zbuffer compression. */
 156     if (r300->zmask_in_use && !r300->hyperz_locked) {
 157         z->zb_bw_cntl |= R300_FAST_FILL_ENABLE |
 158                          /*R300_FORCE_COMPRESSED_STENCIL_VALUE_ENABLE |*/
 159                          R300_RD_COMP_ENABLE;
 160
 161         if (!r300->zmask_decompress) {
 162             z->zb_bw_cntl |= R300_WR_COMP_ENABLE;
 163         }
 164     }
 165
 166     if (zstex->tex.zcomp8x8[fb->zsbuf->u.tex.level]) {
 167         z->gb_z_peq_config |= R300_GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_8_8;
 168     }
 169
 170     /* HiZ. */
 171     if (r300->hiz_in_use && !r300->hyperz_locked) {
 172         /* Set the HiZ function if needed. */
 173         if (r300->hiz_func == HIZ_FUNC_NONE) {
 174             r300->hiz_func = r300_get_hiz_func(r300);
 175         }
 176
 177         /* If the depth function is inverted, HiZ must be disabled. */
 178         if (!r300_is_hiz_func_valid(r300)) {
 179             r300->hiz_in_use = FALSE;
 180         } else if (r300_can_hiz(r300)) {
 181             /* Setup the HiZ bits. */
 182             z->zb_bw_cntl |=
 183                 R300_HIZ_ENABLE |
 184                 (r300->hiz_func == HIZ_FUNC_MIN ? R300_HIZ_MIN : R300_HIZ_MAX);
 185
 186             z->sc_hyperz |= R300_SC_HYPERZ_ENABLE |
 187                             r300_get_sc_hz_max(r300);
 188
 189             if (r300->screen->caps.is_r500) {
 190                 z->zb_bw_cntl |= R500_HIZ_FP_EXP_BITS_3 |
 191                                  R500_HIZ_EQUAL_REJECT_ENABLE;
 192             }
 193         }
 194     }
 195
 196     /* R500-specific features and optimizations. */
 197     if (r300->screen->caps.is_r500) {
 198         z->zb_bw_cntl |= R500_PEQ_PACKING_ENABLE |
 199                          R500_COVERED_PTR_MASKING_ENABLE;
 200     }
 201 }
 202
 203 /*****************************************************************************/
 204 /* The ZTOP state                                                            */
 205 /*****************************************************************************/
 206
 207 static boolean r300_dsa_writes_stencil(
 208         struct pipe_stencil_state *s)
 209 {
 210     return s->enabled && s->writemask &&
 211            (s->fail_op  != PIPE_STENCIL_OP_KEEP ||
 212             s->zfail_op != PIPE_STENCIL_OP_KEEP ||
 213             s->zpass_op != PIPE_STENCIL_OP_KEEP);
 214 }
 215
 216 static boolean r300_dsa_writes_depth_stencil(
 217         struct pipe_depth_stencil_alpha_state *dsa)
 218 {
 219     /* We are interested only in the cases when a depth or stencil value
 220      * can be changed. */
 221
 222     if (dsa->depth.enabled && dsa->depth.writemask &&
 223         dsa->depth.func != PIPE_FUNC_NEVER)
 224         return TRUE;
 225
 226     if (r300_dsa_writes_stencil(&dsa->stencil[0]) ||
 227         r300_dsa_writes_stencil(&dsa->stencil[1]))
 228         return TRUE;
 229
 230     return FALSE;
 231 }
 232
 233 static boolean r300_dsa_alpha_test_enabled(
 234         struct pipe_depth_stencil_alpha_state *dsa)
 235 {
 236     /* We are interested only in the cases when alpha testing can kill
 237      * a fragment. */
 238
 239     return dsa->alpha.enabled && dsa->alpha.func != PIPE_FUNC_ALWAYS;
 240 }
 241
 242 static void r300_update_ztop(struct r300_context* r300)
 243 {
 244     struct r300_ztop_state* ztop_state =
 245         (struct r300_ztop_state*)r300->ztop_state.state;
 246     uint32_t old_ztop = ztop_state->z_buffer_top;
 247
 248     /* This is important enough that I felt it warranted a comment.
 249      *
 250      * According to the docs, these are the conditions where ZTOP must be
 251      * disabled:
 252      * 1) Alpha testing enabled
 253      * 2) Texture kill instructions in fragment shader
 254      * 3) Chroma key culling enabled
 255      * 4) W-buffering enabled
 256      *
 257      * The docs claim that for the first three cases, if no ZS writes happen,
 258      * then ZTOP can be used.
 259      *
 260      * (3) will never apply since we do not support chroma-keyed operations.
 261      * (4) will need to be re-examined (and this comment updated) if/when
 262      * Hyper-Z becomes supported.
 263      *
 264      * Additionally, the following conditions require disabled ZTOP:
 265      * 5) Depth writes in fragment shader
 266      * 6) Outstanding occlusion queries
 267      *
 268      * This register causes stalls all the way from SC to CB when changed,
 269      * but it is buffered on-chip so it does not hurt to write it if it has
 270      * not changed.
 271      *
 272      * ~C.
 273      */
 274
 275     /* ZS writes */
 276     if (r300_dsa_writes_depth_stencil(r300->dsa_state.state) &&
 277            (r300_dsa_alpha_test_enabled(r300->dsa_state.state) ||  /* (1) */
 278             r300_fs(r300)->shader->info.uses_kill)) {              /* (2) */
 279         ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
 280     } else if (r300_fragment_shader_writes_depth(r300_fs(r300))) { /* (5) */
 281         ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
 282     } else if (r300->query_current) {                              /* (6) */
 283         ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
 284     } else {
 285         ztop_state->z_buffer_top = R300_ZTOP_ENABLE;
 286     }
 287     if (ztop_state->z_buffer_top != old_ztop)
 288         r300_mark_atom_dirty(r300, &r300->ztop_state);
 289 }
 290
 291 void r300_update_hyperz_state(struct r300_context* r300)
 292 {
 293     r300_update_ztop(r300);
 294
 295     if (r300->hyperz_state.dirty) {
 296         r300_update_hyperz(r300);
 297     }
 298 }