svga: Performance fixes
[mesa.git] / src / gallium / drivers / svga / svga_pipe_depthstencil.c
1 /**********************************************************
2 * Copyright 2008-2009 VMware, Inc. All rights reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 **********************************************************/
25
26 #include "pipe/p_defines.h"
27 #include "util/u_bitmask.h"
28 #include "util/u_inlines.h"
29 #include "util/u_math.h"
30 #include "util/u_memory.h"
31
32 #include "svga_context.h"
33 #include "svga_hw_reg.h"
34 #include "svga_cmd.h"
35
36
37 static inline unsigned
38 svga_translate_compare_func(unsigned func)
39 {
40 switch (func) {
41 case PIPE_FUNC_NEVER: return SVGA3D_CMP_NEVER;
42 case PIPE_FUNC_LESS: return SVGA3D_CMP_LESS;
43 case PIPE_FUNC_LEQUAL: return SVGA3D_CMP_LESSEQUAL;
44 case PIPE_FUNC_GREATER: return SVGA3D_CMP_GREATER;
45 case PIPE_FUNC_GEQUAL: return SVGA3D_CMP_GREATEREQUAL;
46 case PIPE_FUNC_NOTEQUAL: return SVGA3D_CMP_NOTEQUAL;
47 case PIPE_FUNC_EQUAL: return SVGA3D_CMP_EQUAL;
48 case PIPE_FUNC_ALWAYS: return SVGA3D_CMP_ALWAYS;
49 default:
50 assert(0);
51 return SVGA3D_CMP_ALWAYS;
52 }
53 }
54
55 static inline unsigned
56 svga_translate_stencil_op(unsigned op)
57 {
58 switch (op) {
59 case PIPE_STENCIL_OP_KEEP: return SVGA3D_STENCILOP_KEEP;
60 case PIPE_STENCIL_OP_ZERO: return SVGA3D_STENCILOP_ZERO;
61 case PIPE_STENCIL_OP_REPLACE: return SVGA3D_STENCILOP_REPLACE;
62 case PIPE_STENCIL_OP_INCR: return SVGA3D_STENCILOP_INCRSAT;
63 case PIPE_STENCIL_OP_DECR: return SVGA3D_STENCILOP_DECRSAT;
64 case PIPE_STENCIL_OP_INCR_WRAP: return SVGA3D_STENCILOP_INCR;
65 case PIPE_STENCIL_OP_DECR_WRAP: return SVGA3D_STENCILOP_DECR;
66 case PIPE_STENCIL_OP_INVERT: return SVGA3D_STENCILOP_INVERT;
67 default:
68 assert(0);
69 return SVGA3D_STENCILOP_KEEP;
70 }
71 }
72
73
74 /**
75 * Define a vgpu10 depth/stencil state object for the given
76 * svga depth/stencil state.
77 */
78 static void
79 define_depth_stencil_state_object(struct svga_context *svga,
80 struct svga_depth_stencil_state *ds)
81 {
82 assert(svga_have_vgpu10(svga));
83
84 ds->id = util_bitmask_add(svga->ds_object_id_bm);
85
86 /* spot check that these comparision tokens are the same */
87 STATIC_ASSERT(SVGA3D_COMPARISON_NEVER == SVGA3D_CMP_NEVER);
88 STATIC_ASSERT(SVGA3D_COMPARISON_LESS == SVGA3D_CMP_LESS);
89 STATIC_ASSERT(SVGA3D_COMPARISON_NOT_EQUAL == SVGA3D_CMP_NOTEQUAL);
90
91 /* Note: we use the ds->stencil[0].enabled value for both the front
92 * and back-face enables. If single-side stencil is used, we'll have
93 * set the back state the same as the front state.
94 */
95 SVGA_RETRY(svga, SVGA3D_vgpu10_DefineDepthStencilState
96 (svga->swc,
97 ds->id,
98 /* depth/Z */
99 ds->zenable,
100 ds->zwriteenable,
101 ds->zfunc,
102 /* Stencil */
103 ds->stencil[0].enabled, /*f|b*/
104 ds->stencil[0].enabled, /*f*/
105 ds->stencil[0].enabled, /*b*/
106 ds->stencil_mask,
107 ds->stencil_writemask,
108 /* front stencil */
109 ds->stencil[0].fail,
110 ds->stencil[0].zfail,
111 ds->stencil[0].pass,
112 ds->stencil[0].func,
113 /* back stencil */
114 ds->stencil[1].fail,
115 ds->stencil[1].zfail,
116 ds->stencil[1].pass,
117 ds->stencil[1].func));
118 }
119
120
121 static void *
122 svga_create_depth_stencil_state(struct pipe_context *pipe,
123 const struct pipe_depth_stencil_alpha_state *templ)
124 {
125 struct svga_context *svga = svga_context(pipe);
126 struct svga_depth_stencil_state *ds = CALLOC_STRUCT(svga_depth_stencil_state);
127
128 if (!ds)
129 return NULL;
130
131 /* Don't try to figure out CW/CCW correspondence with
132 * stencil[0]/[1] at this point. Presumably this can change as
133 * back/front face are modified.
134 */
135 ds->stencil[0].enabled = templ->stencil[0].enabled;
136 if (ds->stencil[0].enabled) {
137 ds->stencil[0].func = svga_translate_compare_func(templ->stencil[0].func);
138 ds->stencil[0].fail = svga_translate_stencil_op(templ->stencil[0].fail_op);
139 ds->stencil[0].zfail = svga_translate_stencil_op(templ->stencil[0].zfail_op);
140 ds->stencil[0].pass = svga_translate_stencil_op(templ->stencil[0].zpass_op);
141
142 /* SVGA3D has one ref/mask/writemask triple shared between front &
143 * back face stencil. We really need two:
144 */
145 ds->stencil_mask = templ->stencil[0].valuemask & 0xff;
146 ds->stencil_writemask = templ->stencil[0].writemask & 0xff;
147 }
148 else {
149 ds->stencil[0].func = SVGA3D_CMP_ALWAYS;
150 ds->stencil[0].fail = SVGA3D_STENCILOP_KEEP;
151 ds->stencil[0].zfail = SVGA3D_STENCILOP_KEEP;
152 ds->stencil[0].pass = SVGA3D_STENCILOP_KEEP;
153 }
154
155 ds->stencil[1].enabled = templ->stencil[1].enabled;
156 if (templ->stencil[1].enabled) {
157 assert(templ->stencil[0].enabled);
158 /* two-sided stencil */
159 ds->stencil[1].func = svga_translate_compare_func(templ->stencil[1].func);
160 ds->stencil[1].fail = svga_translate_stencil_op(templ->stencil[1].fail_op);
161 ds->stencil[1].zfail = svga_translate_stencil_op(templ->stencil[1].zfail_op);
162 ds->stencil[1].pass = svga_translate_stencil_op(templ->stencil[1].zpass_op);
163
164 ds->stencil_mask = templ->stencil[1].valuemask & 0xff;
165 ds->stencil_writemask = templ->stencil[1].writemask & 0xff;
166
167 if (templ->stencil[1].valuemask != templ->stencil[0].valuemask) {
168 pipe_debug_message(&svga->debug.callback, CONFORMANCE,
169 "two-sided stencil mask not supported "
170 "(front=0x%x, back=0x%x)",
171 templ->stencil[0].valuemask,
172 templ->stencil[1].valuemask);
173 }
174 if (templ->stencil[1].writemask != templ->stencil[0].writemask) {
175 pipe_debug_message(&svga->debug.callback, CONFORMANCE,
176 "two-sided stencil writemask not supported "
177 "(front=0x%x, back=0x%x)",
178 templ->stencil[0].writemask,
179 templ->stencil[1].writemask);
180 }
181 }
182 else {
183 /* back face state is same as front-face state */
184 ds->stencil[1].func = ds->stencil[0].func;
185 ds->stencil[1].fail = ds->stencil[0].fail;
186 ds->stencil[1].zfail = ds->stencil[0].zfail;
187 ds->stencil[1].pass = ds->stencil[0].pass;
188 }
189
190
191 ds->zenable = templ->depth.enabled;
192 if (ds->zenable) {
193 ds->zfunc = svga_translate_compare_func(templ->depth.func);
194 ds->zwriteenable = templ->depth.writemask;
195 }
196 else {
197 ds->zfunc = SVGA3D_CMP_ALWAYS;
198 }
199
200 ds->alphatestenable = templ->alpha.enabled;
201 if (ds->alphatestenable) {
202 ds->alphafunc = svga_translate_compare_func(templ->alpha.func);
203 ds->alpharef = templ->alpha.ref_value;
204 }
205 else {
206 ds->alphafunc = SVGA3D_CMP_ALWAYS;
207 }
208
209 if (svga_have_vgpu10(svga)) {
210 define_depth_stencil_state_object(svga, ds);
211 }
212
213 svga->hud.num_depthstencil_objects++;
214
215 SVGA_STATS_COUNT_INC(svga_screen(svga->pipe.screen)->sws,
216 SVGA_STATS_COUNT_DEPTHSTENCILSTATE);
217
218 return ds;
219 }
220
221
222 static void
223 svga_bind_depth_stencil_state(struct pipe_context *pipe, void *depth_stencil)
224 {
225 struct svga_context *svga = svga_context(pipe);
226
227 if (svga_have_vgpu10(svga)) {
228 /* flush any previously queued drawing before changing state */
229 svga_hwtnl_flush_retry(svga);
230 }
231
232 svga->curr.depth = (const struct svga_depth_stencil_state *)depth_stencil;
233 svga->dirty |= SVGA_NEW_DEPTH_STENCIL_ALPHA;
234 }
235
236
237 static void
238 svga_delete_depth_stencil_state(struct pipe_context *pipe, void *depth_stencil)
239 {
240 struct svga_context *svga = svga_context(pipe);
241 struct svga_depth_stencil_state *ds =
242 (struct svga_depth_stencil_state *) depth_stencil;
243
244 if (svga_have_vgpu10(svga)) {
245 svga_hwtnl_flush_retry(svga);
246
247 assert(ds->id != SVGA3D_INVALID_ID);
248
249 SVGA_RETRY(svga, SVGA3D_vgpu10_DestroyDepthStencilState(svga->swc,
250 ds->id));
251
252 if (ds->id == svga->state.hw_draw.depth_stencil_id)
253 svga->state.hw_draw.depth_stencil_id = SVGA3D_INVALID_ID;
254
255 util_bitmask_clear(svga->ds_object_id_bm, ds->id);
256 ds->id = SVGA3D_INVALID_ID;
257 }
258
259 FREE(depth_stencil);
260 svga->hud.num_depthstencil_objects--;
261 }
262
263
264 static void
265 svga_set_stencil_ref(struct pipe_context *pipe,
266 const struct pipe_stencil_ref *stencil_ref)
267 {
268 struct svga_context *svga = svga_context(pipe);
269
270 if (svga_have_vgpu10(svga)) {
271 /* flush any previously queued drawing before changing state */
272 svga_hwtnl_flush_retry(svga);
273 }
274
275 svga->curr.stencil_ref = *stencil_ref;
276
277 svga->dirty |= SVGA_NEW_STENCIL_REF;
278 }
279
280
281 static void
282 svga_set_sample_mask(struct pipe_context *pipe,
283 unsigned sample_mask)
284 {
285 struct svga_context *svga = svga_context(pipe);
286
287 svga->curr.sample_mask = sample_mask;
288
289 svga->dirty |= SVGA_NEW_BLEND; /* See emit_rss_vgpu10() */
290 }
291
292
293 static void
294 svga_set_min_samples(struct pipe_context *pipe, unsigned min_samples)
295 {
296 /* This specifies the minimum number of times the fragment shader
297 * must run when doing per-sample shading for a MSAA render target.
298 * For our SVGA3D device, the FS is automatically run in per-sample
299 * mode if it uses the sample ID or sample position registers.
300 */
301 }
302
303
304 void
305 svga_init_depth_stencil_functions(struct svga_context *svga)
306 {
307 svga->pipe.create_depth_stencil_alpha_state = svga_create_depth_stencil_state;
308 svga->pipe.bind_depth_stencil_alpha_state = svga_bind_depth_stencil_state;
309 svga->pipe.delete_depth_stencil_alpha_state = svga_delete_depth_stencil_state;
310
311 svga->pipe.set_stencil_ref = svga_set_stencil_ref;
312 svga->pipe.set_sample_mask = svga_set_sample_mask;
313 svga->pipe.set_min_samples = svga_set_min_samples;
314 }