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 "util/u_inlines.h"
27#include "pipe/p_defines.h"
28#include "util/u_math.h"
29
30#include "svga_context.h"
31#include "svga_state.h"
32#include "svga_cmd.h"
33#include "svga_debug.h"
34
35
36/***********************************************************************
37 * Hardware state update
38 */
39
40
41static enum pipe_error
42emit_framebuffer( struct svga_context *svga,
43                  unsigned dirty )
44{
45   const struct pipe_framebuffer_state *curr = &svga->curr.framebuffer;
46   struct pipe_framebuffer_state *hw = &svga->state.hw_clear.framebuffer;
47   boolean reemit = svga->rebind.rendertargets;
48   unsigned i;
49   enum pipe_error ret;
50
51   /*
52    * We need to reemit non-null surface bindings, even when they are not
53    * dirty, to ensure that the resources are paged in.
54    */
55
56   for(i = 0; i < PIPE_MAX_COLOR_BUFS; ++i) {
57      if (curr->cbufs[i] != hw->cbufs[i] ||
58          (reemit && hw->cbufs[i])) {
59         if (svga->curr.nr_fbs++ > 8)
60            return PIPE_ERROR_OUT_OF_MEMORY;
61
62         ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_COLOR0 + i, curr->cbufs[i]);
63         if (ret != PIPE_OK)
64            return ret;
65
66         pipe_surface_reference(&hw->cbufs[i], curr->cbufs[i]);
67      }
68   }
69
70
71   if (curr->zsbuf != hw->zsbuf ||
72       (reemit && hw->zsbuf)) {
73      ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_DEPTH, curr->zsbuf);
74      if (ret != PIPE_OK)
75         return ret;
76
77      if (curr->zsbuf &&
78          curr->zsbuf->format == PIPE_FORMAT_S8_UINT_Z24_UNORM) {
79         ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_STENCIL, curr->zsbuf);
80         if (ret != PIPE_OK)
81            return ret;
82      }
83      else {
84         ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_STENCIL, NULL);
85         if (ret != PIPE_OK)
86            return ret;
87      }
88
89      pipe_surface_reference(&hw->zsbuf, curr->zsbuf);
90   }
91
92   svga->rebind.rendertargets = FALSE;
93
94   return PIPE_OK;
95}
96
97
98/*
99 * Rebind rendertargets.
100 *
101 * Similar to emit_framebuffer, but without any state checking/update.
102 *
103 * Called at the beginning of every new command buffer to ensure that
104 * non-dirty rendertargets are properly paged-in.
105 */
106enum pipe_error
107svga_reemit_framebuffer_bindings(struct svga_context *svga)
108{
109   struct pipe_framebuffer_state *hw = &svga->state.hw_clear.framebuffer;
110   unsigned i;
111   enum pipe_error ret;
112
113   assert(svga->rebind.rendertargets);
114
115   for (i = 0; i < MIN2(PIPE_MAX_COLOR_BUFS, 8); ++i) {
116      if (hw->cbufs[i]) {
117         ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_COLOR0 + i, hw->cbufs[i]);
118         if (ret != PIPE_OK) {
119            return ret;
120         }
121      }
122   }
123
124   if (hw->zsbuf) {
125      ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_DEPTH, hw->zsbuf);
126      if (ret != PIPE_OK) {
127         return ret;
128      }
129
130      if (hw->zsbuf &&
131          hw->zsbuf->format == PIPE_FORMAT_S8_UINT_Z24_UNORM) {
132         ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_STENCIL, hw->zsbuf);
133         if (ret != PIPE_OK) {
134            return ret;
135         }
136      }
137      else {
138         ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_STENCIL, NULL);
139         if (ret != PIPE_OK) {
140            return ret;
141         }
142      }
143   }
144
145   svga->rebind.rendertargets = FALSE;
146
147   return PIPE_OK;
148}
149
150
151struct svga_tracked_state svga_hw_framebuffer =
152{
153   "hw framebuffer state",
154   SVGA_NEW_FRAME_BUFFER,
155   emit_framebuffer
156};
157
158
159
160
161/***********************************************************************
162 */
163
164static enum pipe_error
165emit_viewport( struct svga_context *svga,
166               unsigned dirty )
167{
168   const struct pipe_viewport_state *viewport = &svga->curr.viewport;
169   struct svga_prescale prescale;
170   SVGA3dRect rect;
171   /* Not sure if this state is relevant with POSITIONT.  Probably
172    * not, but setting to 0,1 avoids some state pingponging.
173    */
174   float range_min = 0.0;
175   float range_max = 1.0;
176   float flip = -1.0;
177   boolean degenerate = FALSE;
178   boolean invertY = FALSE;
179   enum pipe_error ret;
180
181   float fb_width = svga->curr.framebuffer.width;
182   float fb_height = svga->curr.framebuffer.height;
183
184   float fx =        viewport->scale[0] * -1.0 + viewport->translate[0];
185   float fy = flip * viewport->scale[1] * -1.0 + viewport->translate[1];
186   float fw =        viewport->scale[0] * 2;
187   float fh = flip * viewport->scale[1] * 2;
188
189   memset( &prescale, 0, sizeof(prescale) );
190
191   /* Examine gallium viewport transformation and produce a screen
192    * rectangle and possibly vertex shader pre-transformation to
193    * get the same results.
194    */
195
196   SVGA_DBG(DEBUG_VIEWPORT,
197            "\ninitial %f,%f %fx%f\n",
198            fx,
199            fy,
200            fw,
201            fh);
202
203   prescale.scale[0] = 1.0;
204   prescale.scale[1] = 1.0;
205   prescale.scale[2] = 1.0;
206   prescale.scale[3] = 1.0;
207   prescale.translate[0] = 0;
208   prescale.translate[1] = 0;
209   prescale.translate[2] = 0;
210   prescale.translate[3] = 0;
211   prescale.enabled = TRUE;
212
213
214
215   if (fw < 0) {
216      prescale.scale[0] *= -1.0;
217      prescale.translate[0] += -fw;
218      fw = -fw;
219      fx =        viewport->scale[0] * 1.0 + viewport->translate[0];
220   }
221
222   if (fh < 0.0) {
223      prescale.translate[1] = fh - 1 + fy * 2;
224      fh = -fh;
225      fy -= fh;
226      prescale.scale[1] = -1.0;
227      invertY = TRUE;
228   }
229
230   if (fx < 0) {
231      prescale.translate[0] += fx;
232      prescale.scale[0] *= fw / (fw + fx);
233      fw += fx;
234      fx = 0;
235   }
236
237   if (fy < 0) {
238      if (invertY) {
239         prescale.translate[1] -= fy;
240      }
241      else {
242         prescale.translate[1] += fy;
243      }
244      prescale.scale[1] *= fh / (fh + fy);
245      fh += fy;
246      fy = 0;
247   }
248
249   if (fx + fw > fb_width) {
250      prescale.scale[0] *= fw / (fb_width - fx);
251      prescale.translate[0] -= fx * (fw / (fb_width - fx));
252      prescale.translate[0] += fx;
253      fw = fb_width - fx;
254
255   }
256
257   if (fy + fh > fb_height) {
258      prescale.scale[1] *= fh / (fb_height - fy);
259      if (invertY) {
260         float in = fb_height - fy;       /* number of vp pixels inside view */
261         float out = fy + fh - fb_height; /* number of vp pixels out of view */
262         prescale.translate[1] += fy * out / in;
263      }
264      else {
265         prescale.translate[1] -= fy * (fh / (fb_height - fy));
266         prescale.translate[1] += fy;
267      }
268      fh = fb_height - fy;
269   }
270
271   if (fw < 0 || fh < 0) {
272      fw = fh = fx = fy = 0;
273      degenerate = TRUE;
274      goto out;
275   }
276
277
278   /* D3D viewport is integer space.  Convert fx,fy,etc. to
279    * integers.
280    *
281    * TODO: adjust pretranslate correct for any subpixel error
282    * introduced converting to integers.
283    */
284   rect.x = fx;
285   rect.y = fy;
286   rect.w = fw;
287   rect.h = fh;
288
289   SVGA_DBG(DEBUG_VIEWPORT,
290            "viewport error %f,%f %fx%f\n",
291            fabs((float)rect.x - fx),
292            fabs((float)rect.y - fy),
293            fabs((float)rect.w - fw),
294            fabs((float)rect.h - fh));
295
296   SVGA_DBG(DEBUG_VIEWPORT,
297            "viewport %d,%d %dx%d\n",
298            rect.x,
299            rect.y,
300            rect.w,
301            rect.h);
302
303
304   /* Finally, to get GL rasterization rules, need to tweak the
305    * screen-space coordinates slightly relative to D3D which is
306    * what hardware implements natively.
307    */
308   if (svga->curr.rast->templ.gl_rasterization_rules) {
309      float adjust_x = 0.0;
310      float adjust_y = 0.0;
311
312      switch (svga->curr.reduced_prim) {
313      case PIPE_PRIM_LINES:
314         adjust_x = -0.5;
315         adjust_y = 0;
316         break;
317      case PIPE_PRIM_POINTS:
318      case PIPE_PRIM_TRIANGLES:
319         adjust_x = -0.5;
320         adjust_y = -0.5;
321         break;
322      }
323
324      if (invertY)
325         adjust_y = -adjust_y;
326
327      prescale.translate[0] += adjust_x;
328      prescale.translate[1] += adjust_y;
329      prescale.translate[2] = 0.5; /* D3D clip space */
330      prescale.scale[2]     = 0.5; /* D3D clip space */
331   }
332
333
334   range_min = viewport->scale[2] * -1.0 + viewport->translate[2];
335   range_max = viewport->scale[2] *  1.0 + viewport->translate[2];
336
337   /* D3D (and by implication SVGA) doesn't like dealing with zmax
338    * less than zmin.  Detect that case, flip the depth range and
339    * invert our z-scale factor to achieve the same effect.
340    */
341   if (range_min > range_max) {
342      float range_tmp;
343      range_tmp = range_min;
344      range_min = range_max;
345      range_max = range_tmp;
346      prescale.scale[2]     = -prescale.scale[2];
347   }
348
349   if (prescale.enabled) {
350      float H[2];
351      float J[2];
352      int i;
353
354      SVGA_DBG(DEBUG_VIEWPORT,
355               "prescale %f,%f %fx%f\n",
356               prescale.translate[0],
357               prescale.translate[1],
358               prescale.scale[0],
359               prescale.scale[1]);
360
361      H[0] = (float)rect.w / 2.0;
362      H[1] = -(float)rect.h / 2.0;
363      J[0] = (float)rect.x + (float)rect.w / 2.0;
364      J[1] = (float)rect.y + (float)rect.h / 2.0;
365
366      SVGA_DBG(DEBUG_VIEWPORT,
367               "H %f,%f\n"
368               "J %fx%f\n",
369               H[0],
370               H[1],
371               J[0],
372               J[1]);
373
374      /* Adjust prescale to take into account the fact that it is
375       * going to be applied prior to the perspective divide and
376       * viewport transformation.
377       *
378       * Vwin = H(Vc/Vc.w) + J
379       *
380       * We want to tweak Vwin with scale and translation from above,
381       * as in:
382       *
383       * Vwin' = S Vwin + T
384       *
385       * But we can only modify the values at Vc.  Plugging all the
386       * above together, and rearranging, eventually we get:
387       *
388       *   Vwin' = H(Vc'/Vc'.w) + J
389       * where:
390       *   Vc' = SVc + KVc.w
391       *   K = (T + (S-1)J) / H
392       *
393       * Overwrite prescale.translate with values for K:
394       */
395      for (i = 0; i < 2; i++) {
396         prescale.translate[i] = ((prescale.translate[i] +
397                                   (prescale.scale[i] - 1.0) * J[i]) / H[i]);
398      }
399
400      SVGA_DBG(DEBUG_VIEWPORT,
401               "clipspace %f,%f %fx%f\n",
402               prescale.translate[0],
403               prescale.translate[1],
404               prescale.scale[0],
405               prescale.scale[1]);
406   }
407
408out:
409   if (degenerate) {
410      rect.x = 0;
411      rect.y = 0;
412      rect.w = 1;
413      rect.h = 1;
414      prescale.enabled = FALSE;
415   }
416
417   if (memcmp(&rect, &svga->state.hw_clear.viewport, sizeof(rect)) != 0) {
418      ret = SVGA3D_SetViewport(svga->swc, &rect);
419      if(ret != PIPE_OK)
420         return ret;
421
422      memcpy(&svga->state.hw_clear.viewport, &rect, sizeof(rect));
423      assert(sizeof(rect) == sizeof(svga->state.hw_clear.viewport));
424   }
425
426   if (svga->state.hw_clear.depthrange.zmin != range_min ||
427       svga->state.hw_clear.depthrange.zmax != range_max)
428   {
429      ret = SVGA3D_SetZRange(svga->swc, range_min, range_max );
430      if(ret != PIPE_OK)
431         return ret;
432
433      svga->state.hw_clear.depthrange.zmin = range_min;
434      svga->state.hw_clear.depthrange.zmax = range_max;
435   }
436
437   if (memcmp(&prescale, &svga->state.hw_clear.prescale, sizeof prescale) != 0) {
438      svga->dirty |= SVGA_NEW_PRESCALE;
439      svga->state.hw_clear.prescale = prescale;
440   }
441
442   return PIPE_OK;
443}
444
445
446struct svga_tracked_state svga_hw_viewport =
447{
448   "hw viewport state",
449   ( SVGA_NEW_FRAME_BUFFER |
450     SVGA_NEW_VIEWPORT |
451     SVGA_NEW_RAST |
452     SVGA_NEW_REDUCED_PRIMITIVE ),
453   emit_viewport
454};
455
456
457/***********************************************************************
458 * Scissor state
459 */
460static enum pipe_error
461emit_scissor_rect( struct svga_context *svga,
462                   unsigned dirty )
463{
464   const struct pipe_scissor_state *scissor = &svga->curr.scissor;
465   SVGA3dRect rect;
466
467   rect.x = scissor->minx;
468   rect.y = scissor->miny;
469   rect.w = scissor->maxx - scissor->minx; /* + 1 ?? */
470   rect.h = scissor->maxy - scissor->miny; /* + 1 ?? */
471
472   return SVGA3D_SetScissorRect(svga->swc, &rect);
473}
474
475
476struct svga_tracked_state svga_hw_scissor =
477{
478   "hw scissor state",
479   SVGA_NEW_SCISSOR,
480   emit_scissor_rect
481};
482
483
484/***********************************************************************
485 * Userclip state
486 */
487
488static enum pipe_error
489emit_clip_planes( struct svga_context *svga,
490                  unsigned dirty )
491{
492   unsigned i;
493   enum pipe_error ret;
494
495   /* TODO: just emit directly from svga_set_clip_state()?
496    */
497   for (i = 0; i < SVGA3D_MAX_CLIP_PLANES; i++) {
498      /* need to express the plane in D3D-style coordinate space.
499       * GL coords get converted to D3D coords with the matrix:
500       * [ 1  0  0  0 ]
501       * [ 0 -1  0  0 ]
502       * [ 0  0  2  0 ]
503       * [ 0  0 -1  1 ]
504       * Apply that matrix to our plane equation, and invert Y.
505       */
506      float a = svga->curr.clip.ucp[i][0];
507      float b = svga->curr.clip.ucp[i][1];
508      float c = svga->curr.clip.ucp[i][2];
509      float d = svga->curr.clip.ucp[i][3];
510      float plane[4];
511
512      plane[0] = a;
513      plane[1] = b;
514      plane[2] = 2.0f * c;
515      plane[3] = d - c;
516
517      ret = SVGA3D_SetClipPlane(svga->swc, i, plane);
518      if(ret != PIPE_OK)
519         return ret;
520   }
521
522   return PIPE_OK;
523}
524
525
526struct svga_tracked_state svga_hw_clip_planes =
527{
528   "hw viewport state",
529   SVGA_NEW_CLIP,
530   emit_clip_planes
531};
532