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 if (!dsa
->dsa
.depth
.enabled
|| !dsa
->dsa
.depth
.writemask
)
49 switch (dsa
->dsa
.depth
.func
) {
52 case PIPE_FUNC_NOTEQUAL
:
53 case PIPE_FUNC_ALWAYS
:
57 case PIPE_FUNC_LEQUAL
:
60 case PIPE_FUNC_GREATER
:
61 case PIPE_FUNC_GEQUAL
:
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
)
73 struct r300_dsa_state
*dsa
= r300
->dsa_state
.state
;
74 unsigned func
= dsa
->dsa
.depth
.func
;
76 return func
>= PIPE_FUNC_GREATER
? R300_SC_HYPERZ_MAX
: R300_SC_HYPERZ_MIN
;
79 static boolean
r300_is_hiz_func_valid(struct r300_context
*r300
)
81 struct r300_dsa_state
*dsa
= r300
->dsa_state
.state
;
82 unsigned func
= dsa
->dsa
.depth
.func
;
84 if (r300
->hiz_func
== HIZ_FUNC_NONE
)
87 /* func1 is less/lessthan */
88 if (r300
->hiz_func
== HIZ_FUNC_MAX
&&
89 (func
== PIPE_FUNC_GEQUAL
|| func
== PIPE_FUNC_GREATER
))
92 /* func1 is greater/greaterthan */
93 if (r300
->hiz_func
== HIZ_FUNC_MIN
&&
94 (func
== PIPE_FUNC_LESS
|| func
== PIPE_FUNC_LEQUAL
))
100 static boolean
r300_dsa_stencil_op_not_keep(struct pipe_stencil_state
*s
)
102 return s
->enabled
&& (s
->fail_op
!= PIPE_STENCIL_OP_KEEP
||
103 s
->zfail_op
!= PIPE_STENCIL_OP_KEEP
);
106 static boolean
r300_can_hiz(struct r300_context
*r300
)
108 struct r300_dsa_state
*dsa
= r300
->dsa_state
.state
;
109 struct r300_screen
*r300screen
= r300
->screen
;
111 /* shader writes depth - no HiZ */
112 if (r300_fragment_shader_writes_depth(r300_fs(r300
))) /* (5) */
115 if (r300
->query_current
)
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]))
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
)
128 /* if depth func is NOTEQUAL */
129 if (dsa
->dsa
.depth
.func
== PIPE_FUNC_NOTEQUAL
)
135 static void r300_update_hyperz(struct r300_context
* r300
)
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
;
144 z
->gb_z_peq_config
= 0;
146 z
->sc_hyperz
= R300_SC_HYPERZ_ADJ_2
;
149 if (r300
->cbzb_clear
) {
150 z
->zb_bw_cntl
|= R300_ZB_CB_CLEAR_CACHE_LINE_WRITE_ONLY
;
155 !r300
->rws
->get_value(r300
->rws
, RADEON_VID_CAN_HYPERZ
))
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 |*/
164 if (!r300
->zmask_decompress
) {
165 z
->zb_bw_cntl
|= R300_WR_COMP_ENABLE
;
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
;
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
);
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. */
187 (r300
->hiz_func
== HIZ_FUNC_MIN
? R300_HIZ_MIN
: R300_HIZ_MAX
);
189 z
->sc_hyperz
|= R300_SC_HYPERZ_ENABLE
|
190 r300_get_sc_hz_max(r300
);
192 if (r300
->screen
->caps
.is_r500
) {
193 z
->zb_bw_cntl
|= R500_HIZ_EQUAL_REJECT_ENABLE
;
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
;
205 /*****************************************************************************/
207 /*****************************************************************************/
209 static boolean
r300_dsa_writes_stencil(
210 struct pipe_stencil_state
*s
)
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
);
218 static boolean
r300_dsa_writes_depth_stencil(
219 struct pipe_depth_stencil_alpha_state
*dsa
)
221 /* We are interested only in the cases when a depth or stencil value
224 if (dsa
->depth
.enabled
&& dsa
->depth
.writemask
&&
225 dsa
->depth
.func
!= PIPE_FUNC_NEVER
)
228 if (r300_dsa_writes_stencil(&dsa
->stencil
[0]) ||
229 r300_dsa_writes_stencil(&dsa
->stencil
[1]))
235 static boolean
r300_dsa_alpha_test_enabled(
236 struct pipe_depth_stencil_alpha_state
*dsa
)
238 /* We are interested only in the cases when alpha testing can kill
241 return dsa
->alpha
.enabled
&& dsa
->alpha
.func
!= PIPE_FUNC_ALWAYS
;
244 static void r300_update_ztop(struct r300_context
* r300
)
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
;
250 /* This is important enough that I felt it warranted a comment.
252 * According to the docs, these are the conditions where ZTOP must be
254 * 1) Alpha testing enabled
255 * 2) Texture kill instructions in fragment shader
256 * 3) Chroma key culling enabled
257 * 4) W-buffering enabled
259 * The docs claim that for the first three cases, if no ZS writes happen,
260 * then ZTOP can be used.
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.
266 * Additionally, the following conditions require disabled ZTOP:
267 * 5) Depth writes in fragment shader
268 * 6) Outstanding occlusion queries
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
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
;
287 ztop_state
->z_buffer_top
= R300_ZTOP_ENABLE
;
289 if (ztop_state
->z_buffer_top
!= old_ztop
)
290 r300_mark_atom_dirty(r300
, &r300
->ztop_state
);
293 void r300_update_hyperz_state(struct r300_context
* r300
)
295 r300_update_ztop(r300
);
297 if (r300
->hyperz_state
.dirty
) {
298 r300_update_hyperz(r300
);