r300_hyperz.c revision b8e80941 
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
30/*
31  HiZ rules - taken from various docs
32   1. HiZ only works on depth values
33   2. Cannot HiZ if stencil fail or zfail is !KEEP
34   3. on R300/400, HiZ is disabled if depth test is EQUAL
35   4. comparison changes without clears usually mean disabling HiZ
36*/
37/*****************************************************************************/
38/* The HyperZ setup                                                          */
39/*****************************************************************************/
40
41static enum r300_hiz_func r300_get_hiz_func(struct r300_context *r300)
42{
43    struct r300_dsa_state *dsa = r300->dsa_state.state;
44
45    switch (dsa->dsa.depth.func) {
46    case PIPE_FUNC_NEVER:
47    case PIPE_FUNC_EQUAL:
48    case PIPE_FUNC_NOTEQUAL:
49    case PIPE_FUNC_ALWAYS:
50    default:
51        /* Guess MAX for uncertain cases. */
52    case PIPE_FUNC_LESS:
53    case PIPE_FUNC_LEQUAL:
54        return HIZ_FUNC_MAX;
55
56    case PIPE_FUNC_GREATER:
57    case PIPE_FUNC_GEQUAL:
58        return HIZ_FUNC_MIN;
59    }
60}
61
62/* Return what's used for the depth test (either minimum or maximum). */
63static unsigned r300_get_sc_hz_max(struct r300_context *r300)
64{
65    struct r300_dsa_state *dsa = r300->dsa_state.state;
66    unsigned func = dsa->dsa.depth.func;
67
68    return func >= PIPE_FUNC_GREATER ? R300_SC_HYPERZ_MAX : R300_SC_HYPERZ_MIN;
69}
70
71static boolean r300_is_hiz_func_valid(struct r300_context *r300)
72{
73    struct r300_dsa_state *dsa = r300->dsa_state.state;
74    unsigned func = dsa->dsa.depth.func;
75
76    if (r300->hiz_func == HIZ_FUNC_NONE)
77        return TRUE;
78
79    /* func1 is less/lessthan */
80    if (r300->hiz_func == HIZ_FUNC_MAX &&
81        (func == PIPE_FUNC_GEQUAL || func == PIPE_FUNC_GREATER))
82        return FALSE;
83
84    /* func1 is greater/greaterthan */
85    if (r300->hiz_func == HIZ_FUNC_MIN &&
86        (func == PIPE_FUNC_LESS   || func == PIPE_FUNC_LEQUAL))
87        return FALSE;
88
89    return TRUE;
90}
91
92static boolean r300_dsa_stencil_op_not_keep(struct pipe_stencil_state *s)
93{
94    return s->enabled && (s->fail_op != PIPE_STENCIL_OP_KEEP ||
95                          s->zfail_op != PIPE_STENCIL_OP_KEEP);
96}
97
98static boolean r300_hiz_allowed(struct r300_context *r300)
99{
100    struct r300_dsa_state *dsa = r300->dsa_state.state;
101    struct r300_screen *r300screen = r300->screen;
102
103    if (r300_fragment_shader_writes_depth(r300_fs(r300)))
104        return FALSE;
105
106    if (r300->query_current)
107        return FALSE;
108
109    /* If the depth function is inverted, HiZ must be disabled. */
110    if (!r300_is_hiz_func_valid(r300))
111        return FALSE;
112
113    /* if stencil fail/zfail op is not KEEP */
114    if (r300_dsa_stencil_op_not_keep(&dsa->dsa.stencil[0]) ||
115        r300_dsa_stencil_op_not_keep(&dsa->dsa.stencil[1]))
116        return FALSE;
117
118    if (dsa->dsa.depth.enabled) {
119        /* if depth func is EQUAL pre-r500 */
120        if (dsa->dsa.depth.func == PIPE_FUNC_EQUAL && !r300screen->caps.is_r500)
121            return FALSE;
122
123        /* if depth func is NOTEQUAL */
124        if (dsa->dsa.depth.func == PIPE_FUNC_NOTEQUAL)
125            return FALSE;
126    }
127    return TRUE;
128}
129
130static void r300_update_hyperz(struct r300_context* r300)
131{
132    struct r300_hyperz_state *z =
133        (struct r300_hyperz_state*)r300->hyperz_state.state;
134    struct pipe_framebuffer_state *fb =
135        (struct pipe_framebuffer_state*)r300->fb_state.state;
136    struct r300_dsa_state *dsa = r300->dsa_state.state;
137    struct r300_resource *zstex =
138            fb->zsbuf ? r300_resource(fb->zsbuf->texture) : NULL;
139
140    z->gb_z_peq_config = 0;
141    z->zb_bw_cntl = 0;
142    z->sc_hyperz = R300_SC_HYPERZ_ADJ_2;
143    z->flush = 0;
144
145    if (r300->cbzb_clear) {
146        z->zb_bw_cntl |= R300_ZB_CB_CLEAR_CACHE_LINE_WRITE_ONLY;
147        return;
148    }
149
150    if (!zstex || !r300->hyperz_enabled)
151        return;
152
153    /* Set the size of ZMASK tiles. */
154    if (zstex->tex.zcomp8x8[fb->zsbuf->u.tex.level]) {
155        z->gb_z_peq_config |= R300_GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_8_8;
156    }
157
158    /* R500-specific features and optimizations. */
159    if (r300->screen->caps.is_r500) {
160        z->zb_bw_cntl |= R500_PEQ_PACKING_ENABLE |
161                         R500_COVERED_PTR_MASKING_ENABLE;
162    }
163
164    /* Setup decompression if needed. No other HyperZ setting is required. */
165    if (r300->zmask_decompress) {
166        z->zb_bw_cntl |= R300_FAST_FILL_ENABLE |
167                         R300_RD_COMP_ENABLE;
168        return;
169    }
170
171    /* Do not set anything if depth and stencil tests are off. */
172    if (!dsa->dsa.depth.enabled &&
173        !dsa->dsa.stencil[0].enabled &&
174        !dsa->dsa.stencil[1].enabled) {
175        assert(!dsa->dsa.depth.writemask);
176        return;
177    }
178
179    /* Zbuffer compression. */
180    if (r300->zmask_in_use && !r300->locked_zbuffer) {
181        z->zb_bw_cntl |= R300_FAST_FILL_ENABLE |
182                         R300_RD_COMP_ENABLE |
183                         R300_WR_COMP_ENABLE;
184    }
185
186    /* HiZ. */
187    if (r300->hiz_in_use && !r300->locked_zbuffer) {
188        /* HiZ cannot be used under some circumstances. */
189        if (!r300_hiz_allowed(r300)) {
190            /* If writemask is disabled, the HiZ memory will not be changed,
191             * so we can keep its content for later. */
192            if (dsa->dsa.depth.writemask) {
193                r300->hiz_in_use = FALSE;
194            }
195            return;
196        }
197        DBG(r300, DBG_HYPERZ, "r300: Z-func: %i\n", dsa->dsa.depth.func);
198
199        /* Set the HiZ function if needed. */
200        if (r300->hiz_func == HIZ_FUNC_NONE) {
201            r300->hiz_func = r300_get_hiz_func(r300);
202        }
203
204        /* Setup the HiZ bits. */
205        z->zb_bw_cntl |= R300_HIZ_ENABLE |
206                (r300->hiz_func == HIZ_FUNC_MIN ? R300_HIZ_MIN : R300_HIZ_MAX);
207
208        z->sc_hyperz |= R300_SC_HYPERZ_ENABLE |
209                        r300_get_sc_hz_max(r300);
210
211        if (r300->screen->caps.is_r500) {
212            z->zb_bw_cntl |= R500_HIZ_EQUAL_REJECT_ENABLE;
213        }
214    }
215}
216
217/*****************************************************************************/
218/* The ZTOP state                                                            */
219/*****************************************************************************/
220
221static boolean r300_dsa_writes_stencil(
222        struct pipe_stencil_state *s)
223{
224    return s->enabled && s->writemask &&
225           (s->fail_op  != PIPE_STENCIL_OP_KEEP ||
226            s->zfail_op != PIPE_STENCIL_OP_KEEP ||
227            s->zpass_op != PIPE_STENCIL_OP_KEEP);
228}
229
230static boolean r300_dsa_writes_depth_stencil(
231        struct pipe_depth_stencil_alpha_state *dsa)
232{
233    /* We are interested only in the cases when a depth or stencil value
234     * can be changed. */
235
236    if (dsa->depth.enabled && dsa->depth.writemask &&
237        dsa->depth.func != PIPE_FUNC_NEVER)
238        return TRUE;
239
240    if (r300_dsa_writes_stencil(&dsa->stencil[0]) ||
241        r300_dsa_writes_stencil(&dsa->stencil[1]))
242        return TRUE;
243
244    return FALSE;
245}
246
247static boolean r300_dsa_alpha_test_enabled(
248        struct pipe_depth_stencil_alpha_state *dsa)
249{
250    /* We are interested only in the cases when alpha testing can kill
251     * a fragment. */
252
253    return dsa->alpha.enabled && dsa->alpha.func != PIPE_FUNC_ALWAYS;
254}
255
256static void r300_update_ztop(struct r300_context* r300)
257{
258    struct r300_ztop_state* ztop_state =
259        (struct r300_ztop_state*)r300->ztop_state.state;
260    uint32_t old_ztop = ztop_state->z_buffer_top;
261
262    /* This is important enough that I felt it warranted a comment.
263     *
264     * According to the docs, these are the conditions where ZTOP must be
265     * disabled:
266     * 1) Alpha testing enabled
267     * 2) Texture kill instructions in fragment shader
268     * 3) Chroma key culling enabled
269     * 4) W-buffering enabled
270     *
271     * The docs claim that for the first three cases, if no ZS writes happen,
272     * then ZTOP can be used.
273     *
274     * (3) will never apply since we do not support chroma-keyed operations.
275     * (4) will need to be re-examined (and this comment updated) if/when
276     * Hyper-Z becomes supported.
277     *
278     * Additionally, the following conditions require disabled ZTOP:
279     * 5) Depth writes in fragment shader
280     * 6) Outstanding occlusion queries
281     *
282     * This register causes stalls all the way from SC to CB when changed,
283     * but it is buffered on-chip so it does not hurt to write it if it has
284     * not changed.
285     *
286     * ~C.
287     */
288
289    /* ZS writes */
290    if (r300_dsa_writes_depth_stencil(r300->dsa_state.state) &&
291           (r300_dsa_alpha_test_enabled(r300->dsa_state.state) ||  /* (1) */
292            r300_fs(r300)->shader->info.uses_kill)) {              /* (2) */
293        ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
294    } else if (r300_fragment_shader_writes_depth(r300_fs(r300))) { /* (5) */
295        ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
296    } else if (r300->query_current) {                              /* (6) */
297        ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
298    } else {
299        ztop_state->z_buffer_top = R300_ZTOP_ENABLE;
300    }
301    if (ztop_state->z_buffer_top != old_ztop)
302        r300_mark_atom_dirty(r300, &r300->ztop_state);
303}
304
305void r300_update_hyperz_state(struct r300_context* r300)
306{
307    r300_update_ztop(r300);
308
309    if (r300->hyperz_state.dirty) {
310        r300_update_hyperz(r300);
311    }
312}
313