2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
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:
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
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. */
24 #include "r300_context.h"
28 #include "util/u_format.h"
29 #include "util/u_mm.h"
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
38 /*****************************************************************************/
39 /* The HyperZ setup */
40 /*****************************************************************************/
42 static enum r300_hiz_func
r300_get_hiz_func(struct r300_context
*r300
)
44 struct r300_dsa_state
*dsa
= r300
->dsa_state
.state
;
46 switch (dsa
->dsa
.depth
.func
) {
49 case PIPE_FUNC_NOTEQUAL
:
50 case PIPE_FUNC_ALWAYS
:
52 /* Guess MAX for uncertain cases. */
54 case PIPE_FUNC_LEQUAL
:
57 case PIPE_FUNC_GREATER
:
58 case PIPE_FUNC_GEQUAL
:
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
)
66 struct r300_dsa_state
*dsa
= r300
->dsa_state
.state
;
67 unsigned func
= dsa
->dsa
.depth
.func
;
69 return func
>= PIPE_FUNC_GREATER
? R300_SC_HYPERZ_MAX
: R300_SC_HYPERZ_MIN
;
72 static boolean
r300_is_hiz_func_valid(struct r300_context
*r300
)
74 struct r300_dsa_state
*dsa
= r300
->dsa_state
.state
;
75 unsigned func
= dsa
->dsa
.depth
.func
;
77 if (r300
->hiz_func
== HIZ_FUNC_NONE
)
80 /* func1 is less/lessthan */
81 if (r300
->hiz_func
== HIZ_FUNC_MAX
&&
82 (func
== PIPE_FUNC_GEQUAL
|| func
== PIPE_FUNC_GREATER
))
85 /* func1 is greater/greaterthan */
86 if (r300
->hiz_func
== HIZ_FUNC_MIN
&&
87 (func
== PIPE_FUNC_LESS
|| func
== PIPE_FUNC_LEQUAL
))
93 static boolean
r300_dsa_stencil_op_not_keep(struct pipe_stencil_state
*s
)
95 return s
->enabled
&& (s
->fail_op
!= PIPE_STENCIL_OP_KEEP
||
96 s
->zfail_op
!= PIPE_STENCIL_OP_KEEP
);
99 static boolean
r300_hiz_allowed(struct r300_context
*r300
)
101 struct r300_dsa_state
*dsa
= r300
->dsa_state
.state
;
102 struct r300_screen
*r300screen
= r300
->screen
;
104 if (r300_fragment_shader_writes_depth(r300_fs(r300
)))
107 if (r300
->query_current
)
110 /* If the depth function is inverted, HiZ must be disabled. */
111 if (!r300_is_hiz_func_valid(r300
))
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]))
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
)
124 /* if depth func is NOTEQUAL */
125 if (dsa
->dsa
.depth
.func
== PIPE_FUNC_NOTEQUAL
)
131 static void r300_update_hyperz(struct r300_context
* r300
)
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
;
141 z
->gb_z_peq_config
= 0;
143 z
->sc_hyperz
= R300_SC_HYPERZ_ADJ_2
;
146 if (r300
->cbzb_clear
) {
147 z
->zb_bw_cntl
|= R300_ZB_CB_CLEAR_CACHE_LINE_WRITE_ONLY
;
151 if (!zstex
|| !r300
->hyperz_enabled
)
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
;
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
;
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
|
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
);
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
|
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
;
198 DBG(r300
, DBG_HYPERZ
, "r300: Z-func: %i\n", dsa
->dsa
.depth
.func
);
200 /* Set the HiZ function if needed. */
201 if (r300
->hiz_func
== HIZ_FUNC_NONE
) {
202 r300
->hiz_func
= r300_get_hiz_func(r300
);
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
);
209 z
->sc_hyperz
|= R300_SC_HYPERZ_ENABLE
|
210 r300_get_sc_hz_max(r300
);
212 if (r300
->screen
->caps
.is_r500
) {
213 z
->zb_bw_cntl
|= R500_HIZ_EQUAL_REJECT_ENABLE
;
218 /*****************************************************************************/
220 /*****************************************************************************/
222 static boolean
r300_dsa_writes_stencil(
223 struct pipe_stencil_state
*s
)
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
);
231 static boolean
r300_dsa_writes_depth_stencil(
232 struct pipe_depth_stencil_alpha_state
*dsa
)
234 /* We are interested only in the cases when a depth or stencil value
237 if (dsa
->depth
.enabled
&& dsa
->depth
.writemask
&&
238 dsa
->depth
.func
!= PIPE_FUNC_NEVER
)
241 if (r300_dsa_writes_stencil(&dsa
->stencil
[0]) ||
242 r300_dsa_writes_stencil(&dsa
->stencil
[1]))
248 static boolean
r300_dsa_alpha_test_enabled(
249 struct pipe_depth_stencil_alpha_state
*dsa
)
251 /* We are interested only in the cases when alpha testing can kill
254 return dsa
->alpha
.enabled
&& dsa
->alpha
.func
!= PIPE_FUNC_ALWAYS
;
257 static void r300_update_ztop(struct r300_context
* r300
)
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
;
263 /* This is important enough that I felt it warranted a comment.
265 * According to the docs, these are the conditions where ZTOP must be
267 * 1) Alpha testing enabled
268 * 2) Texture kill instructions in fragment shader
269 * 3) Chroma key culling enabled
270 * 4) W-buffering enabled
272 * The docs claim that for the first three cases, if no ZS writes happen,
273 * then ZTOP can be used.
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.
279 * Additionally, the following conditions require disabled ZTOP:
280 * 5) Depth writes in fragment shader
281 * 6) Outstanding occlusion queries
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
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
;
300 ztop_state
->z_buffer_top
= R300_ZTOP_ENABLE
;
302 if (ztop_state
->z_buffer_top
!= old_ztop
)
303 r300_mark_atom_dirty(r300
, &r300
->ztop_state
);
306 void r300_update_hyperz_state(struct r300_context
* r300
)
308 r300_update_ztop(r300
);
310 if (r300
->hyperz_state
.dirty
) {
311 r300_update_hyperz(r300
);