515155d3ae8cf410ac6330f212935bf15eaa26d1
[mesa.git] / src / gallium / drivers / svga / svga_state_rss.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_format.h"
29 #include "util/u_inlines.h"
30 #include "util/u_memory.h"
31 #include "util/u_math.h"
32 #include "util/u_memory.h"
33
34 #include "svga_context.h"
35 #include "svga_screen.h"
36 #include "svga_state.h"
37 #include "svga_cmd.h"
38 #include "svga_format.h"
39 #include "svga_shader.h"
40
41
42 struct rs_queue {
43 unsigned rs_count;
44 SVGA3dRenderState rs[SVGA3D_RS_MAX];
45 };
46
47
48 #define EMIT_RS(svga, value, token) \
49 do { \
50 STATIC_ASSERT(SVGA3D_RS_##token < ARRAY_SIZE(svga->state.hw_draw.rs)); \
51 if (svga->state.hw_draw.rs[SVGA3D_RS_##token] != value) { \
52 svga_queue_rs(&queue, SVGA3D_RS_##token, value); \
53 svga->state.hw_draw.rs[SVGA3D_RS_##token] = value; \
54 } \
55 } while (0)
56
57 #define EMIT_RS_FLOAT(svga, fvalue, token) \
58 do { \
59 unsigned value = fui(fvalue); \
60 STATIC_ASSERT(SVGA3D_RS_##token < ARRAY_SIZE(svga->state.hw_draw.rs)); \
61 if (svga->state.hw_draw.rs[SVGA3D_RS_##token] != value) { \
62 svga_queue_rs(&queue, SVGA3D_RS_##token, value); \
63 svga->state.hw_draw.rs[SVGA3D_RS_##token] = value; \
64 } \
65 } while (0)
66
67
68 static inline void
69 svga_queue_rs(struct rs_queue *q, unsigned rss, unsigned value)
70 {
71 assert(q->rs_count < ARRAY_SIZE(q->rs));
72 q->rs[q->rs_count].state = rss;
73 q->rs[q->rs_count].uintValue = value;
74 q->rs_count++;
75 }
76
77
78 /* Compare old and new render states and emit differences between them
79 * to hardware. Simplest implementation would be to emit the whole of
80 * the "to" state.
81 */
82 static enum pipe_error
83 emit_rss_vgpu9(struct svga_context *svga, unsigned dirty)
84 {
85 struct svga_screen *screen = svga_screen(svga->pipe.screen);
86 struct rs_queue queue;
87 float point_size_min;
88
89 queue.rs_count = 0;
90
91 if (dirty & (SVGA_NEW_BLEND | SVGA_NEW_BLEND_COLOR)) {
92 const struct svga_blend_state *curr = svga->curr.blend;
93
94 EMIT_RS(svga, curr->rt[0].writemask, COLORWRITEENABLE);
95 EMIT_RS(svga, curr->rt[0].blend_enable, BLENDENABLE);
96
97 if (curr->rt[0].blend_enable) {
98 EMIT_RS(svga, curr->rt[0].srcblend, SRCBLEND);
99 EMIT_RS(svga, curr->rt[0].dstblend, DSTBLEND);
100 EMIT_RS(svga, curr->rt[0].blendeq, BLENDEQUATION);
101
102 EMIT_RS(svga, curr->rt[0].separate_alpha_blend_enable,
103 SEPARATEALPHABLENDENABLE);
104
105 if (curr->rt[0].separate_alpha_blend_enable) {
106 EMIT_RS(svga, curr->rt[0].srcblend_alpha, SRCBLENDALPHA);
107 EMIT_RS(svga, curr->rt[0].dstblend_alpha, DSTBLENDALPHA);
108 EMIT_RS(svga, curr->rt[0].blendeq_alpha, BLENDEQUATIONALPHA);
109 }
110 }
111 }
112
113 if (dirty & SVGA_NEW_BLEND_COLOR) {
114 uint32 color;
115 uint32 r = float_to_ubyte(svga->curr.blend_color.color[0]);
116 uint32 g = float_to_ubyte(svga->curr.blend_color.color[1]);
117 uint32 b = float_to_ubyte(svga->curr.blend_color.color[2]);
118 uint32 a = float_to_ubyte(svga->curr.blend_color.color[3]);
119
120 color = (a << 24) | (r << 16) | (g << 8) | b;
121
122 EMIT_RS(svga, color, BLENDCOLOR);
123 }
124
125 if (dirty & (SVGA_NEW_DEPTH_STENCIL_ALPHA | SVGA_NEW_RAST)) {
126 const struct svga_depth_stencil_state *curr = svga->curr.depth;
127 const struct svga_rasterizer_state *rast = svga->curr.rast;
128
129 if (!curr->stencil[0].enabled) {
130 /* Stencil disabled
131 */
132 EMIT_RS(svga, FALSE, STENCILENABLE);
133 EMIT_RS(svga, FALSE, STENCILENABLE2SIDED);
134 }
135 else if (curr->stencil[0].enabled && !curr->stencil[1].enabled) {
136 /* Regular stencil
137 */
138 EMIT_RS(svga, TRUE, STENCILENABLE);
139 EMIT_RS(svga, FALSE, STENCILENABLE2SIDED);
140
141 EMIT_RS(svga, curr->stencil[0].func, STENCILFUNC);
142 EMIT_RS(svga, curr->stencil[0].fail, STENCILFAIL);
143 EMIT_RS(svga, curr->stencil[0].zfail, STENCILZFAIL);
144 EMIT_RS(svga, curr->stencil[0].pass, STENCILPASS);
145
146 EMIT_RS(svga, curr->stencil_mask, STENCILMASK);
147 EMIT_RS(svga, curr->stencil_writemask, STENCILWRITEMASK);
148 }
149 else {
150 int cw, ccw;
151
152 /* Hardware frontwinding is always CW, so if ours is also CW,
153 * then our definition of front face agrees with hardware.
154 * Otherwise need to flip.
155 */
156 if (rast->templ.front_ccw) {
157 ccw = 0;
158 cw = 1;
159 }
160 else {
161 ccw = 1;
162 cw = 0;
163 }
164
165 /* Twoside stencil
166 */
167 EMIT_RS(svga, TRUE, STENCILENABLE);
168 EMIT_RS(svga, TRUE, STENCILENABLE2SIDED);
169
170 EMIT_RS(svga, curr->stencil[cw].func, STENCILFUNC);
171 EMIT_RS(svga, curr->stencil[cw].fail, STENCILFAIL);
172 EMIT_RS(svga, curr->stencil[cw].zfail, STENCILZFAIL);
173 EMIT_RS(svga, curr->stencil[cw].pass, STENCILPASS);
174
175 EMIT_RS(svga, curr->stencil[ccw].func, CCWSTENCILFUNC);
176 EMIT_RS(svga, curr->stencil[ccw].fail, CCWSTENCILFAIL);
177 EMIT_RS(svga, curr->stencil[ccw].zfail, CCWSTENCILZFAIL);
178 EMIT_RS(svga, curr->stencil[ccw].pass, CCWSTENCILPASS);
179
180 EMIT_RS(svga, curr->stencil_mask, STENCILMASK);
181 EMIT_RS(svga, curr->stencil_writemask, STENCILWRITEMASK);
182 }
183
184 EMIT_RS(svga, curr->zenable, ZENABLE);
185 if (curr->zenable) {
186 EMIT_RS(svga, curr->zfunc, ZFUNC);
187 EMIT_RS(svga, curr->zwriteenable, ZWRITEENABLE);
188 }
189
190 EMIT_RS(svga, curr->alphatestenable, ALPHATESTENABLE);
191 if (curr->alphatestenable) {
192 EMIT_RS(svga, curr->alphafunc, ALPHAFUNC);
193 EMIT_RS_FLOAT(svga, curr->alpharef, ALPHAREF);
194 }
195 }
196
197 if (dirty & SVGA_NEW_STENCIL_REF) {
198 EMIT_RS(svga, svga->curr.stencil_ref.ref_value[0], STENCILREF);
199 }
200
201 if (dirty & (SVGA_NEW_RAST | SVGA_NEW_NEED_PIPELINE)) {
202 const struct svga_rasterizer_state *curr = svga->curr.rast;
203 unsigned cullmode = curr->cullmode;
204
205 /* Shademode: still need to rearrange index list to move
206 * flat-shading PV first vertex.
207 */
208 EMIT_RS(svga, curr->shademode, SHADEMODE);
209
210 /* Don't do culling while the software pipeline is active. It
211 * does it for us, and additionally introduces potentially
212 * back-facing triangles.
213 */
214 if (svga->state.sw.need_pipeline)
215 cullmode = SVGA3D_FACE_NONE;
216
217 point_size_min = util_get_min_point_size(&curr->templ);
218
219 EMIT_RS(svga, cullmode, CULLMODE);
220 EMIT_RS(svga, curr->scissortestenable, SCISSORTESTENABLE);
221 EMIT_RS(svga, curr->multisampleantialias, MULTISAMPLEANTIALIAS);
222 EMIT_RS(svga, curr->lastpixel, LASTPIXEL);
223 EMIT_RS_FLOAT(svga, curr->pointsize, POINTSIZE);
224 EMIT_RS_FLOAT(svga, point_size_min, POINTSIZEMIN);
225 EMIT_RS_FLOAT(svga, screen->maxPointSize, POINTSIZEMAX);
226 EMIT_RS(svga, curr->pointsprite, POINTSPRITEENABLE);
227
228 /* Emit line state, when the device understands it */
229 if (screen->haveLineStipple)
230 EMIT_RS(svga, curr->linepattern, LINEPATTERN);
231 if (screen->haveLineSmooth)
232 EMIT_RS(svga, curr->antialiasedlineenable, ANTIALIASEDLINEENABLE);
233 if (screen->maxLineWidth > 1.0F)
234 EMIT_RS_FLOAT(svga, curr->linewidth, LINEWIDTH);
235 }
236
237 if (dirty & (SVGA_NEW_RAST |
238 SVGA_NEW_FRAME_BUFFER |
239 SVGA_NEW_NEED_PIPELINE)) {
240 const struct svga_rasterizer_state *curr = svga->curr.rast;
241 float slope = 0.0;
242 float bias = 0.0;
243
244 /* Need to modify depth bias according to bound depthbuffer
245 * format. Don't do hardware depthbias while the software
246 * pipeline is active.
247 */
248 if (!svga->state.sw.need_pipeline &&
249 svga->curr.framebuffer.zsbuf)
250 {
251 slope = curr->slopescaledepthbias;
252 bias = svga->curr.depthscale * curr->depthbias;
253 }
254
255 EMIT_RS_FLOAT(svga, slope, SLOPESCALEDEPTHBIAS);
256 EMIT_RS_FLOAT(svga, bias, DEPTHBIAS);
257 }
258
259 if (dirty & SVGA_NEW_FRAME_BUFFER) {
260 /* XXX: we only look at the first color buffer's sRGB state */
261 float gamma = 1.0f;
262 if (svga->curr.framebuffer.cbufs[0] &&
263 util_format_is_srgb(svga->curr.framebuffer.cbufs[0]->format)) {
264 gamma = 2.2f;
265 }
266 EMIT_RS_FLOAT(svga, gamma, OUTPUTGAMMA);
267 }
268
269 if (dirty & SVGA_NEW_RAST) {
270 /* bitmask of the enabled clip planes */
271 unsigned enabled = svga->curr.rast->templ.clip_plane_enable;
272 EMIT_RS(svga, enabled, CLIPPLANEENABLE);
273 }
274
275 if (queue.rs_count) {
276 SVGA3dRenderState *rs;
277
278 if (SVGA3D_BeginSetRenderState(svga->swc,
279 &rs,
280 queue.rs_count) != PIPE_OK)
281 goto fail;
282
283 memcpy(rs,
284 queue.rs,
285 queue.rs_count * sizeof queue.rs[0]);
286
287 SVGA_FIFOCommitAll(svga->swc);
288 }
289
290 return PIPE_OK;
291
292 fail:
293 /* XXX: need to poison cached hardware state on failure to ensure
294 * dirty state gets re-emitted. Fix this by re-instating partial
295 * FIFOCommit command and only updating cached hw state once the
296 * initial allocation has succeeded.
297 */
298 memset(svga->state.hw_draw.rs, 0xcd, sizeof(svga->state.hw_draw.rs));
299
300 return PIPE_ERROR_OUT_OF_MEMORY;
301 }
302
303
304 /** Returns a non-culling rasterizer state object to be used with
305 * point sprite.
306 */
307 static struct svga_rasterizer_state *
308 get_no_cull_rasterizer_state(struct svga_context *svga)
309 {
310 const struct svga_rasterizer_state *r = svga->curr.rast;
311 unsigned int aa_point = r->templ.point_smooth;
312
313 if (!svga->rasterizer_no_cull[aa_point]) {
314 struct pipe_rasterizer_state rast;
315
316 memset(&rast, 0, sizeof(rast));
317 rast.flatshade = 1;
318 rast.front_ccw = 1;
319 rast.point_smooth = r->templ.point_smooth;
320
321 /* All rasterizer states have the same half_pixel_center,
322 * bottom_edge_rule and clip_halfz values since they are
323 * constant for a context. If we ever implement
324 * GL_ARB_clip_control, the clip_halfz field would have to be observed.
325 */
326 rast.half_pixel_center = r->templ.half_pixel_center;
327 rast.bottom_edge_rule = r->templ.bottom_edge_rule;
328 rast.clip_halfz = r->templ.clip_halfz;
329
330 svga->rasterizer_no_cull[aa_point] =
331 svga->pipe.create_rasterizer_state(&svga->pipe, &rast);
332 }
333 return svga->rasterizer_no_cull[aa_point];
334 }
335
336
337 /** Returns a depth stencil state object with depth and stencil test disabled.
338 */
339 static struct svga_depth_stencil_state *
340 get_no_depth_stencil_test_state(struct svga_context *svga)
341 {
342 if (!svga->depthstencil_disable) {
343 struct pipe_depth_stencil_alpha_state ds = {{0}};
344 svga->depthstencil_disable =
345 svga->pipe.create_depth_stencil_alpha_state(&svga->pipe, &ds);
346 }
347 return svga->depthstencil_disable;
348 }
349
350
351 static enum pipe_error
352 emit_rss_vgpu10(struct svga_context *svga, unsigned dirty)
353 {
354 enum pipe_error ret = PIPE_OK;
355
356 svga_hwtnl_flush_retry(svga);
357
358 if (dirty & (SVGA_NEW_BLEND | SVGA_NEW_BLEND_COLOR)) {
359 const struct svga_blend_state *curr;
360 float blend_factor[4];
361
362 if (svga_has_any_integer_cbufs(svga)) {
363 /* Blending is not supported in integer-valued render targets. */
364 curr = svga->noop_blend;
365 blend_factor[0] =
366 blend_factor[1] =
367 blend_factor[2] =
368 blend_factor[3] = 0;
369 }
370 else {
371 curr = svga->curr.blend;
372
373 if (curr->blend_color_alpha) {
374 blend_factor[0] =
375 blend_factor[1] =
376 blend_factor[2] =
377 blend_factor[3] = svga->curr.blend_color.color[3];
378 }
379 else {
380 blend_factor[0] = svga->curr.blend_color.color[0];
381 blend_factor[1] = svga->curr.blend_color.color[1];
382 blend_factor[2] = svga->curr.blend_color.color[2];
383 blend_factor[3] = svga->curr.blend_color.color[3];
384 }
385 }
386
387 /* Set/bind the blend state object */
388 if (svga->state.hw_draw.blend_id != curr->id ||
389 svga->state.hw_draw.blend_factor[0] != blend_factor[0] ||
390 svga->state.hw_draw.blend_factor[1] != blend_factor[1] ||
391 svga->state.hw_draw.blend_factor[2] != blend_factor[2] ||
392 svga->state.hw_draw.blend_factor[3] != blend_factor[3] ||
393 svga->state.hw_draw.blend_sample_mask != svga->curr.sample_mask) {
394 ret = SVGA3D_vgpu10_SetBlendState(svga->swc, curr->id,
395 blend_factor,
396 svga->curr.sample_mask);
397 if (ret != PIPE_OK)
398 return ret;
399
400 svga->state.hw_draw.blend_id = curr->id;
401 svga->state.hw_draw.blend_factor[0] = blend_factor[0];
402 svga->state.hw_draw.blend_factor[1] = blend_factor[1];
403 svga->state.hw_draw.blend_factor[2] = blend_factor[2];
404 svga->state.hw_draw.blend_factor[3] = blend_factor[3];
405 svga->state.hw_draw.blend_sample_mask = svga->curr.sample_mask;
406 }
407 }
408
409 if (svga->disable_rasterizer) {
410 if (!svga->state.hw_draw.rasterizer_discard) {
411 struct svga_depth_stencil_state *ds;
412
413 /* If rasterization is to be disabled, disable depth and stencil
414 * testing as well.
415 */
416 ds = get_no_depth_stencil_test_state(svga);
417 if (ds->id != svga->state.hw_draw.depth_stencil_id) {
418 ret = SVGA3D_vgpu10_SetDepthStencilState(svga->swc, ds->id, 0);
419 if (ret != PIPE_OK)
420 return ret;
421
422 svga->state.hw_draw.depth_stencil_id = ds->id;
423 svga->state.hw_draw.stencil_ref = 0;
424 }
425 svga->state.hw_draw.rasterizer_discard = TRUE;
426 }
427 } else {
428 if ((dirty & (SVGA_NEW_DEPTH_STENCIL_ALPHA | SVGA_NEW_STENCIL_REF)) ||
429 svga->state.hw_draw.rasterizer_discard) {
430 const struct svga_depth_stencil_state *curr = svga->curr.depth;
431 unsigned curr_ref = svga->curr.stencil_ref.ref_value[0];
432
433 if (curr->id != svga->state.hw_draw.depth_stencil_id ||
434 curr_ref != svga->state.hw_draw.stencil_ref) {
435 /* Set/bind the depth/stencil state object */
436 ret = SVGA3D_vgpu10_SetDepthStencilState(svga->swc, curr->id,
437 curr_ref);
438 if (ret != PIPE_OK)
439 return ret;
440
441 svga->state.hw_draw.depth_stencil_id = curr->id;
442 svga->state.hw_draw.stencil_ref = curr_ref;
443 }
444 }
445
446 if (dirty & (SVGA_NEW_REDUCED_PRIMITIVE | SVGA_NEW_RAST)) {
447 const struct svga_rasterizer_state *rast;
448
449 if (svga->curr.reduced_prim == PIPE_PRIM_POINTS &&
450 svga->curr.gs && svga->curr.gs->wide_point) {
451
452 /* If we are drawing a point sprite, we will need to
453 * bind a non-culling rasterizer state object
454 */
455 rast = get_no_cull_rasterizer_state(svga);
456 }
457 else {
458 rast = svga->curr.rast;
459 }
460
461 if (svga->state.hw_draw.rasterizer_id != rast->id) {
462 /* Set/bind the rasterizer state object */
463 ret = SVGA3D_vgpu10_SetRasterizerState(svga->swc, rast->id);
464 if (ret != PIPE_OK)
465 return ret;
466 svga->state.hw_draw.rasterizer_id = rast->id;
467 }
468 }
469 svga->state.hw_draw.rasterizer_discard = FALSE;
470 }
471 return PIPE_OK;
472 }
473
474
475 static enum pipe_error
476 emit_rss(struct svga_context *svga, unsigned dirty)
477 {
478 if (svga_have_vgpu10(svga)) {
479 return emit_rss_vgpu10(svga, dirty);
480 }
481 else {
482 return emit_rss_vgpu9(svga, dirty);
483 }
484 }
485
486
487 struct svga_tracked_state svga_hw_rss =
488 {
489 "hw rss state",
490
491 (SVGA_NEW_BLEND |
492 SVGA_NEW_BLEND_COLOR |
493 SVGA_NEW_DEPTH_STENCIL_ALPHA |
494 SVGA_NEW_STENCIL_REF |
495 SVGA_NEW_RAST |
496 SVGA_NEW_FRAME_BUFFER |
497 SVGA_NEW_NEED_PIPELINE |
498 SVGA_NEW_FS |
499 SVGA_NEW_REDUCED_PRIMITIVE),
500
501 emit_rss
502 };