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/format/u_format.h"
29#include "util/u_inlines.h"
30#include "util/u_math.h"
31#include "util/u_memory.h"
32#include "tgsi/tgsi_parse.h"
33
34#include "svga_context.h"
35#include "svga_cmd.h"
36#include "svga_debug.h"
37#include "svga_resource_texture.h"
38#include "svga_surface.h"
39#include "svga_sampler_view.h"
40
41
42static inline unsigned
43translate_wrap_mode(unsigned wrap)
44{
45   switch (wrap) {
46   case PIPE_TEX_WRAP_REPEAT:
47      return SVGA3D_TEX_ADDRESS_WRAP;
48   case PIPE_TEX_WRAP_CLAMP:
49      return SVGA3D_TEX_ADDRESS_CLAMP;
50   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
51      /* Unfortunately SVGA3D_TEX_ADDRESS_EDGE not respected by
52       * hardware.
53       */
54      return SVGA3D_TEX_ADDRESS_CLAMP;
55   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
56      return SVGA3D_TEX_ADDRESS_BORDER;
57   case PIPE_TEX_WRAP_MIRROR_REPEAT:
58      return SVGA3D_TEX_ADDRESS_MIRROR;
59   case PIPE_TEX_WRAP_MIRROR_CLAMP:
60   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
61   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
62      return SVGA3D_TEX_ADDRESS_MIRRORONCE;
63   default:
64      assert(0);
65      return SVGA3D_TEX_ADDRESS_WRAP;
66   }
67}
68
69
70static inline unsigned
71translate_img_filter(unsigned filter)
72{
73   switch (filter) {
74   case PIPE_TEX_FILTER_NEAREST:
75      return SVGA3D_TEX_FILTER_NEAREST;
76   case PIPE_TEX_FILTER_LINEAR:
77      return SVGA3D_TEX_FILTER_LINEAR;
78   default:
79      assert(0);
80      return SVGA3D_TEX_FILTER_NEAREST;
81   }
82}
83
84
85static inline unsigned
86translate_mip_filter(unsigned filter)
87{
88   switch (filter) {
89   case PIPE_TEX_MIPFILTER_NONE:
90      return SVGA3D_TEX_FILTER_NONE;
91   case PIPE_TEX_MIPFILTER_NEAREST:
92      return SVGA3D_TEX_FILTER_NEAREST;
93   case PIPE_TEX_MIPFILTER_LINEAR:
94      return SVGA3D_TEX_FILTER_LINEAR;
95   default:
96      assert(0);
97      return SVGA3D_TEX_FILTER_NONE;
98   }
99}
100
101
102static uint8
103translate_comparison_func(unsigned func)
104{
105   switch (func) {
106   case PIPE_FUNC_NEVER:
107      return SVGA3D_COMPARISON_NEVER;
108   case PIPE_FUNC_LESS:
109      return SVGA3D_COMPARISON_LESS;
110   case PIPE_FUNC_EQUAL:
111      return SVGA3D_COMPARISON_EQUAL;
112   case PIPE_FUNC_LEQUAL:
113      return SVGA3D_COMPARISON_LESS_EQUAL;
114   case PIPE_FUNC_GREATER:
115      return SVGA3D_COMPARISON_GREATER;
116   case PIPE_FUNC_NOTEQUAL:
117      return SVGA3D_COMPARISON_NOT_EQUAL;
118   case PIPE_FUNC_GEQUAL:
119      return SVGA3D_COMPARISON_GREATER_EQUAL;
120   case PIPE_FUNC_ALWAYS:
121      return SVGA3D_COMPARISON_ALWAYS;
122   default:
123      assert(!"Invalid comparison function");
124      return SVGA3D_COMPARISON_ALWAYS;
125   }
126}
127
128
129/**
130 * Translate filtering state to vgpu10 format.
131 */
132static SVGA3dFilter
133translate_filter_mode(unsigned img_filter,
134                      unsigned min_filter,
135                      unsigned mag_filter,
136                      boolean anisotropic,
137                      boolean compare)
138{
139   SVGA3dFilter mode = 0;
140
141   if (img_filter == PIPE_TEX_FILTER_LINEAR)
142      mode |= SVGA3D_FILTER_MIP_LINEAR;
143   if (min_filter == PIPE_TEX_FILTER_LINEAR)
144      mode |= SVGA3D_FILTER_MIN_LINEAR;
145   if (mag_filter == PIPE_TEX_FILTER_LINEAR)
146      mode |= SVGA3D_FILTER_MAG_LINEAR;
147   if (anisotropic)
148      mode |= SVGA3D_FILTER_ANISOTROPIC;
149   if (compare)
150      mode |= SVGA3D_FILTER_COMPARE;
151
152   return mode;
153}
154
155
156/**
157 * Define a vgpu10 sampler state.
158 */
159static void
160define_sampler_state_object(struct svga_context *svga,
161                            struct svga_sampler_state *ss,
162                            const struct pipe_sampler_state *ps)
163{
164   uint8_t max_aniso = (uint8_t) 255; /* XXX fix me */
165   boolean anisotropic;
166   uint8 compare_func;
167   SVGA3dFilter filter;
168   SVGA3dRGBAFloat bcolor;
169   float min_lod, max_lod;
170
171   assert(svga_have_vgpu10(svga));
172
173   anisotropic = ss->aniso_level > 1.0f;
174
175   filter = translate_filter_mode(ps->min_mip_filter,
176                                  ps->min_img_filter,
177                                  ps->mag_img_filter,
178                                  anisotropic,
179                                  ss->compare_mode);
180
181   compare_func = translate_comparison_func(ss->compare_func);
182
183   COPY_4V(bcolor.value, ps->border_color.f);
184
185   assert(ps->min_lod <= ps->max_lod);
186
187   if (ps->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) {
188      /* just use the base level image */
189      min_lod = max_lod = 0.0f;
190   }
191   else {
192      min_lod = ps->min_lod;
193      max_lod = ps->max_lod;
194   }
195
196   /* If shadow comparisons are enabled, create two sampler states: one
197    * with the given shadow compare mode, another with shadow comparison off.
198    * We need the later because in some cases, we have to do the shadow
199    * compare in the shader.  So, we don't want to do it twice.
200    */
201   STATIC_ASSERT(PIPE_TEX_COMPARE_NONE == 0);
202   STATIC_ASSERT(PIPE_TEX_COMPARE_R_TO_TEXTURE == 1);
203   ss->id[1] = SVGA3D_INVALID_ID;
204
205   unsigned i;
206   for (i = 0; i <= ss->compare_mode; i++) {
207      ss->id[i] = util_bitmask_add(svga->sampler_object_id_bm);
208
209      SVGA_RETRY(svga, SVGA3D_vgpu10_DefineSamplerState
210                 (svga->swc,
211                  ss->id[i],
212                  filter,
213                  ss->addressu,
214                  ss->addressv,
215                  ss->addressw,
216                  ss->lod_bias, /* float */
217                  max_aniso,
218                  compare_func,
219                  bcolor,
220                  min_lod,       /* float */
221                  max_lod));      /* float */
222
223      /* turn off the shadow compare option for second iteration */
224      filter &= ~SVGA3D_FILTER_COMPARE;
225   }
226}
227
228
229static void *
230svga_create_sampler_state(struct pipe_context *pipe,
231                          const struct pipe_sampler_state *sampler)
232{
233   struct svga_context *svga = svga_context(pipe);
234   struct svga_sampler_state *cso = CALLOC_STRUCT( svga_sampler_state );
235
236   if (!cso)
237      return NULL;
238
239   cso->mipfilter = translate_mip_filter(sampler->min_mip_filter);
240   cso->magfilter = translate_img_filter( sampler->mag_img_filter );
241   cso->minfilter = translate_img_filter( sampler->min_img_filter );
242   cso->aniso_level = MAX2( sampler->max_anisotropy, 1 );
243   if (sampler->max_anisotropy)
244      cso->magfilter = cso->minfilter = SVGA3D_TEX_FILTER_ANISOTROPIC;
245   cso->lod_bias = sampler->lod_bias;
246   cso->addressu = translate_wrap_mode(sampler->wrap_s);
247   cso->addressv = translate_wrap_mode(sampler->wrap_t);
248   cso->addressw = translate_wrap_mode(sampler->wrap_r);
249   cso->normalized_coords = sampler->normalized_coords;
250   cso->compare_mode = sampler->compare_mode;
251   cso->compare_func = sampler->compare_func;
252
253   {
254      uint32 r = float_to_ubyte(sampler->border_color.f[0]);
255      uint32 g = float_to_ubyte(sampler->border_color.f[1]);
256      uint32 b = float_to_ubyte(sampler->border_color.f[2]);
257      uint32 a = float_to_ubyte(sampler->border_color.f[3]);
258
259      cso->bordercolor = (a << 24) | (r << 16) | (g << 8) | b;
260   }
261
262   /* No SVGA3D support for:
263    *    - min/max LOD clamping
264    */
265   cso->min_lod = 0;
266   cso->view_min_lod = MAX2((int) (sampler->min_lod + 0.5), 0);
267   cso->view_max_lod = MAX2((int) (sampler->max_lod + 0.5), 0);
268
269   /* Use min_mipmap */
270   if (svga->debug.use_min_mipmap) {
271      if (cso->view_min_lod == cso->view_max_lod) {
272         cso->min_lod = cso->view_min_lod;
273         cso->view_min_lod = 0;
274         cso->view_max_lod = 1000; /* Just a high number */
275         cso->mipfilter = SVGA3D_TEX_FILTER_NONE;
276      }
277   }
278
279   if (svga_have_vgpu10(svga)) {
280      define_sampler_state_object(svga, cso, sampler);
281   }
282
283   SVGA_DBG(DEBUG_SAMPLERS,
284            "New sampler: min %u, view(min %u, max %u) lod, mipfilter %s\n",
285            cso->min_lod, cso->view_min_lod, cso->view_max_lod,
286            cso->mipfilter == SVGA3D_TEX_FILTER_NONE ? "SVGA3D_TEX_FILTER_NONE" : "SOMETHING");
287
288   svga->hud.num_sampler_objects++;
289   SVGA_STATS_COUNT_INC(svga_screen(svga->pipe.screen)->sws,
290                        SVGA_STATS_COUNT_SAMPLER);
291
292   return cso;
293}
294
295
296static void
297svga_bind_sampler_states(struct pipe_context *pipe,
298                         enum pipe_shader_type shader,
299                         unsigned start,
300                         unsigned num,
301                         void **samplers)
302{
303   struct svga_context *svga = svga_context(pipe);
304   unsigned i;
305   boolean any_change = FALSE;
306
307   assert(shader < PIPE_SHADER_TYPES);
308   assert(start + num <= PIPE_MAX_SAMPLERS);
309
310   /* Pre-VGPU10 only supports FS textures */
311   if (!svga_have_vgpu10(svga) && shader != PIPE_SHADER_FRAGMENT)
312      return;
313
314   for (i = 0; i < num; i++) {
315      if (svga->curr.sampler[shader][start + i] != samplers[i])
316         any_change = TRUE;
317      svga->curr.sampler[shader][start + i] = samplers[i];
318   }
319
320   if (!any_change) {
321      return;
322   }
323
324   /* find highest non-null sampler[] entry */
325   {
326      unsigned j = MAX2(svga->curr.num_samplers[shader], start + num);
327      while (j > 0 && svga->curr.sampler[shader][j - 1] == NULL)
328         j--;
329      svga->curr.num_samplers[shader] = j;
330   }
331
332   svga->dirty |= SVGA_NEW_SAMPLER;
333}
334
335
336static void
337svga_delete_sampler_state(struct pipe_context *pipe, void *sampler)
338{
339   struct svga_sampler_state *ss = (struct svga_sampler_state *) sampler;
340   struct svga_context *svga = svga_context(pipe);
341
342   if (svga_have_vgpu10(svga)) {
343      unsigned i;
344      for (i = 0; i < ARRAY_SIZE(ss->id); i++) {
345         if (ss->id[i] != SVGA3D_INVALID_ID) {
346            svga_hwtnl_flush_retry(svga);
347
348            SVGA_RETRY(svga, SVGA3D_vgpu10_DestroySamplerState(svga->swc,
349                                                               ss->id[i]));
350            util_bitmask_clear(svga->sampler_object_id_bm, ss->id[i]);
351         }
352      }
353   }
354
355   FREE(sampler);
356   svga->hud.num_sampler_objects--;
357}
358
359
360static struct pipe_sampler_view *
361svga_create_sampler_view(struct pipe_context *pipe,
362                         struct pipe_resource *texture,
363                         const struct pipe_sampler_view *templ)
364{
365   struct svga_context *svga = svga_context(pipe);
366   struct svga_pipe_sampler_view *sv = CALLOC_STRUCT(svga_pipe_sampler_view);
367
368   if (!sv) {
369      return NULL;
370   }
371
372   sv->base = *templ;
373   sv->base.reference.count = 1;
374   sv->base.texture = NULL;
375   pipe_resource_reference(&sv->base.texture, texture);
376
377   sv->base.context = pipe;
378   sv->id = SVGA3D_INVALID_ID;
379
380   svga->hud.num_samplerview_objects++;
381   SVGA_STATS_COUNT_INC(svga_screen(svga->pipe.screen)->sws,
382                        SVGA_STATS_COUNT_SAMPLERVIEW);
383
384   return &sv->base;
385}
386
387
388static void
389svga_sampler_view_destroy(struct pipe_context *pipe,
390                          struct pipe_sampler_view *view)
391{
392   struct svga_context *svga = svga_context(pipe);
393   struct svga_pipe_sampler_view *sv = svga_pipe_sampler_view(view);
394
395   if (svga_have_vgpu10(svga) && sv->id != SVGA3D_INVALID_ID) {
396      assert(view->context == pipe);
397
398      svga_hwtnl_flush_retry(svga);
399
400      SVGA_RETRY(svga, SVGA3D_vgpu10_DestroyShaderResourceView(svga->swc,
401                                                               sv->id));
402      util_bitmask_clear(svga->sampler_view_id_bm, sv->id);
403   }
404
405   pipe_resource_reference(&sv->base.texture, NULL);
406
407   FREE(sv);
408   svga->hud.num_samplerview_objects--;
409}
410
411
412static void
413svga_set_sampler_views(struct pipe_context *pipe,
414                       enum pipe_shader_type shader,
415                       unsigned start,
416                       unsigned num,
417                       unsigned unbind_num_trailing_slots,
418                       bool take_ownership,
419                       struct pipe_sampler_view **views)
420{
421   struct svga_context *svga = svga_context(pipe);
422   unsigned flag_1d = 0;
423   unsigned flag_srgb = 0;
424   uint i;
425   boolean any_change = FALSE;
426
427   assert(shader < PIPE_SHADER_TYPES);
428   assert(start + num <= ARRAY_SIZE(svga->curr.sampler_views[shader]));
429
430   /* Pre-VGPU10 only supports FS textures */
431   if (!svga_have_vgpu10(svga) && shader != PIPE_SHADER_FRAGMENT) {
432      for (unsigned i = 0; i < num; i++) {
433         struct pipe_sampler_view *view = views[i];
434         pipe_sampler_view_reference(&view, NULL);
435      }
436      return;
437   }
438
439   SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_SETSAMPLERVIEWS);
440
441   /* This bit of code works around a quirk in the CSO module.
442    * If start=num=0 it means all sampler views should be released.
443    * Note that the CSO module treats sampler views for fragment shaders
444    * differently than other shader types.
445    */
446   if (start == 0 && num == 0 && svga->curr.num_sampler_views[shader] > 0) {
447      for (i = 0; i < svga->curr.num_sampler_views[shader]; i++) {
448         pipe_sampler_view_reference(&svga->curr.sampler_views[shader][i],
449                                     NULL);
450      }
451      any_change = TRUE;
452   }
453
454   for (i = 0; i < num; i++) {
455      enum pipe_texture_target target;
456
457      any_change |= svga->curr.sampler_views[shader][start + i] != views[i];
458
459      if (take_ownership) {
460         pipe_sampler_view_reference(&svga->curr.sampler_views[shader][start + i],
461               NULL);
462         svga->curr.sampler_views[shader][start + i] = views[i];
463      } else if (svga->curr.sampler_views[shader][start + i] != views[i]) {
464         pipe_sampler_view_reference(&svga->curr.sampler_views[shader][start + i],
465                                     views[i]);
466      }
467
468      if (!views[i])
469         continue;
470
471      if (util_format_is_srgb(views[i]->format))
472         flag_srgb |= 1 << (start + i);
473
474      target = views[i]->target;
475      if (target == PIPE_TEXTURE_1D) {
476         flag_1d |= 1 << (start + i);
477      } else if (target == PIPE_TEXTURE_RECT) {
478         /* If the size of the bound texture changes, we need to emit new
479          * const buffer values.
480          */
481         svga->dirty |= SVGA_NEW_TEXTURE_CONSTS;
482      } else if (target == PIPE_BUFFER) {
483         /* If the size of the bound buffer changes, we need to emit new
484          * const buffer values.
485          */
486         svga->dirty |= SVGA_NEW_TEXTURE_CONSTS;
487      }
488   }
489
490   for (; i < num + unbind_num_trailing_slots; i++) {
491      if (svga->curr.sampler_views[shader][start + i]) {
492         pipe_sampler_view_reference(&svga->curr.sampler_views[shader][start + i],
493                                     NULL);
494         any_change = TRUE;
495      }
496   }
497
498   if (!any_change) {
499      goto done;
500   }
501
502   /* find highest non-null sampler_views[] entry */
503   {
504      unsigned j = MAX2(svga->curr.num_sampler_views[shader], start + num);
505      while (j > 0 && svga->curr.sampler_views[shader][j - 1] == NULL)
506         j--;
507      svga->curr.num_sampler_views[shader] = j;
508   }
509
510   svga->dirty |= SVGA_NEW_TEXTURE_BINDING;
511
512   if (flag_srgb != svga->curr.tex_flags.flag_srgb ||
513       flag_1d != svga->curr.tex_flags.flag_1d) {
514      svga->dirty |= SVGA_NEW_TEXTURE_FLAGS;
515      svga->curr.tex_flags.flag_1d = flag_1d;
516      svga->curr.tex_flags.flag_srgb = flag_srgb;
517   }
518
519   /* Check if any of the sampler view resources collide with the framebuffer
520    * color buffers or depth stencil resource. If so, set the NEW_FRAME_BUFFER
521    * dirty bit so that emit_framebuffer can be invoked to create backed view
522    * for the conflicted surface view.
523    */
524   if (svga_check_sampler_framebuffer_resource_collision(svga, shader)) {
525      svga->dirty |= SVGA_NEW_FRAME_BUFFER;
526   }
527
528done:
529   SVGA_STATS_TIME_POP(svga_sws(svga));
530}
531
532/**
533 * Clean up sampler, sampler view state at context destruction time
534 */
535void
536svga_cleanup_sampler_state(struct svga_context *svga)
537{
538   enum pipe_shader_type shader;
539
540   for (shader = 0; shader <= PIPE_SHADER_COMPUTE; shader++) {
541      unsigned i;
542
543      for (i = 0; i < svga->state.hw_draw.num_sampler_views[shader]; i++) {
544         pipe_sampler_view_reference(&svga->state.hw_draw.sampler_views[shader][i],
545                                     NULL);
546      }
547   }
548
549   /* free polygon stipple state */
550   if (svga->polygon_stipple.sampler) {
551      svga->pipe.delete_sampler_state(&svga->pipe, svga->polygon_stipple.sampler);
552   }
553
554   if (svga->polygon_stipple.sampler_view) {
555      svga->pipe.sampler_view_destroy(&svga->pipe,
556                                      &svga->polygon_stipple.sampler_view->base);
557   }
558   pipe_resource_reference(&svga->polygon_stipple.texture, NULL);
559}
560
561void
562svga_init_sampler_functions( struct svga_context *svga )
563{
564   svga->pipe.create_sampler_state = svga_create_sampler_state;
565   svga->pipe.bind_sampler_states = svga_bind_sampler_states;
566   svga->pipe.delete_sampler_state = svga_delete_sampler_state;
567   svga->pipe.set_sampler_views = svga_set_sampler_views;
568   svga->pipe.create_sampler_view = svga_create_sampler_view;
569   svga->pipe.sampler_view_destroy = svga_sampler_view_destroy;
570}
571