nvc0_screen.c revision ef7bb281292c17b762b57779306e874704c87328
1/*
2 * Copyright 2010 Christoph Bumiller
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17 * THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
18 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
19 * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
20 * SOFTWARE.
21 */
22
23#include "util/u_format.h"
24#include "util/u_format_s3tc.h"
25#include "pipe/p_screen.h"
26
27#include "vl/vl_decoder.h"
28#include "vl/vl_video_buffer.h"
29
30#include "nvc0_context.h"
31#include "nvc0_screen.h"
32
33#include "nvc0_graph_macros.h"
34
35static boolean
36nvc0_screen_is_format_supported(struct pipe_screen *pscreen,
37                                enum pipe_format format,
38                                enum pipe_texture_target target,
39                                unsigned sample_count,
40                                unsigned bindings)
41{
42   if (!(0x117 & (1 << sample_count))) /* 0, 1, 2, 4 or 8 */
43      return FALSE;
44
45   if (!util_format_is_supported(format, bindings))
46      return FALSE;
47
48   switch (format) {
49   case PIPE_FORMAT_R8G8B8A8_UNORM:
50   case PIPE_FORMAT_R8G8B8X8_UNORM:
51      /* HACK: GL requires equal formats for MS resolve and window is BGRA */
52      if (bindings & PIPE_BIND_RENDER_TARGET)
53         return FALSE;
54   default:
55      break;
56   }
57
58   /* transfers & shared are always supported */
59   bindings &= ~(PIPE_BIND_TRANSFER_READ |
60                 PIPE_BIND_TRANSFER_WRITE |
61                 PIPE_BIND_SHARED);
62
63   return (nvc0_format_table[format].usage & bindings) == bindings;
64}
65
66static int
67nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
68{
69   const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
70
71   switch (param) {
72   case PIPE_CAP_MAX_COMBINED_SAMPLERS:
73      return 16 * PIPE_SHADER_TYPES; /* NOTE: should not count COMPUTE */
74   case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
75   case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
76      return 15;
77   case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
78      return 12;
79   case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
80      return 2048;
81   case PIPE_CAP_MIN_TEXEL_OFFSET:
82      return -8;
83   case PIPE_CAP_MAX_TEXEL_OFFSET:
84      return 7;
85   case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
86   case PIPE_CAP_TEXTURE_SWIZZLE:
87   case PIPE_CAP_TEXTURE_SHADOW_MAP:
88   case PIPE_CAP_NPOT_TEXTURES:
89   case PIPE_CAP_ANISOTROPIC_FILTER:
90   case PIPE_CAP_SEAMLESS_CUBE_MAP:
91      return 1;
92   case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
93      return (class_3d >= NVE4_3D_CLASS) ? 1 : 0;
94   case PIPE_CAP_TWO_SIDED_STENCIL:
95   case PIPE_CAP_DEPTH_CLIP_DISABLE:
96   case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE:
97   case PIPE_CAP_POINT_SPRITE:
98      return 1;
99   case PIPE_CAP_SM3:
100      return 1;
101   case PIPE_CAP_GLSL_FEATURE_LEVEL:
102      return 150;
103   case PIPE_CAP_MAX_RENDER_TARGETS:
104      return 8;
105   case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
106      return 1;
107   case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
108   case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
109   case PIPE_CAP_VERTEX_COLOR_CLAMPED:
110      return 1;
111   case PIPE_CAP_TIMER_QUERY:
112   case PIPE_CAP_OCCLUSION_QUERY:
113   case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
114      return 1;
115   case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
116      return 4;
117   case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
118   case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
119      return 128;
120   case PIPE_CAP_BLEND_EQUATION_SEPARATE:
121   case PIPE_CAP_INDEP_BLEND_ENABLE:
122   case PIPE_CAP_INDEP_BLEND_FUNC:
123      return 1;
124   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
125   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
126      return 1;
127   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
128   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
129      return 0;
130   case PIPE_CAP_SHADER_STENCIL_EXPORT:
131      return 0;
132   case PIPE_CAP_PRIMITIVE_RESTART:
133   case PIPE_CAP_TGSI_INSTANCEID:
134   case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
135   case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
136   case PIPE_CAP_CONDITIONAL_RENDER:
137   case PIPE_CAP_TEXTURE_BARRIER:
138   case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
139      return 1;
140   case PIPE_CAP_TGSI_CAN_COMPACT_VARYINGS:
141   case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
142   case PIPE_CAP_USER_VERTEX_BUFFERS:
143      return 0; /* state trackers will know better */
144   case PIPE_CAP_USER_CONSTANT_BUFFERS:
145   case PIPE_CAP_USER_INDEX_BUFFERS:
146      return 1;
147   case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
148      return 256;
149   case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
150   case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
151   case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
152      return 0;
153   default:
154      NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
155      return 0;
156   }
157}
158
159static int
160nvc0_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
161                             enum pipe_shader_cap param)
162{
163   switch (shader) {
164   case PIPE_SHADER_VERTEX:
165      /*
166   case PIPE_SHADER_TESSELLATION_CONTROL:
167   case PIPE_SHADER_TESSELLATION_EVALUATION:
168      */
169   case PIPE_SHADER_GEOMETRY:
170   case PIPE_SHADER_FRAGMENT:
171      break;
172   default:
173      return 0;
174   }
175
176   switch (param) {
177   case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
178   case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
179   case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
180   case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
181      return 16384;
182   case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
183      return 16;
184   case PIPE_SHADER_CAP_MAX_INPUTS:
185      if (shader == PIPE_SHADER_VERTEX)
186         return 32;
187      if (shader == PIPE_SHADER_FRAGMENT)
188         return (0x200 + 0x20 + 0x80) / 16; /* generic + colors + TexCoords */
189      return (0x200 + 0x40 + 0x80) / 16; /* without 0x60 for per-patch inputs */
190   case PIPE_SHADER_CAP_MAX_CONSTS:
191      return 65536 / 16;
192   case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
193      return NVC0_MAX_PIPE_CONSTBUFS;
194   case PIPE_SHADER_CAP_MAX_ADDRS:
195      return 1;
196   case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
197   case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
198      return shader != PIPE_SHADER_FRAGMENT;
199   case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
200   case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
201      return 1;
202   case PIPE_SHADER_CAP_MAX_PREDS:
203      return 0;
204   case PIPE_SHADER_CAP_MAX_TEMPS:
205      return NVC0_CAP_MAX_PROGRAM_TEMPS;
206   case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
207      return 1;
208   case PIPE_SHADER_CAP_SUBROUTINES:
209      return 1; /* but inlining everything, we need function declarations */
210   case PIPE_SHADER_CAP_INTEGERS:
211      return 1;
212   case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
213      return 16; /* would be 32 in linked (OpenGL-style) mode */
214      /*
215   case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLER_VIEWS:
216      return 32;
217      */
218   default:
219      NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
220      return 0;
221   }
222}
223
224static float
225nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
226{
227   switch (param) {
228   case PIPE_CAPF_MAX_LINE_WIDTH:
229   case PIPE_CAPF_MAX_LINE_WIDTH_AA:
230      return 10.0f;
231   case PIPE_CAPF_MAX_POINT_WIDTH:
232      return 63.0f;
233   case PIPE_CAPF_MAX_POINT_WIDTH_AA:
234      return 63.375f;
235   case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
236      return 16.0f;
237   case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
238      return 15.0f;
239   default:
240      NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
241      return 0.0f;
242   }
243}
244
245static void
246nvc0_screen_destroy(struct pipe_screen *pscreen)
247{
248   struct nvc0_screen *screen = nvc0_screen(pscreen);
249
250   if (screen->base.fence.current) {
251      nouveau_fence_wait(screen->base.fence.current);
252      nouveau_fence_ref(NULL, &screen->base.fence.current);
253   }
254   if (screen->base.pushbuf)
255      screen->base.pushbuf->user_priv = NULL;
256
257   if (screen->blitctx)
258      FREE(screen->blitctx);
259
260   nouveau_bo_ref(NULL, &screen->text);
261   nouveau_bo_ref(NULL, &screen->uniform_bo);
262   nouveau_bo_ref(NULL, &screen->tls);
263   nouveau_bo_ref(NULL, &screen->txc);
264   nouveau_bo_ref(NULL, &screen->fence.bo);
265   nouveau_bo_ref(NULL, &screen->poly_cache);
266
267   nouveau_heap_destroy(&screen->lib_code);
268   nouveau_heap_destroy(&screen->text_heap);
269
270   if (screen->tic.entries)
271      FREE(screen->tic.entries);
272
273   nouveau_mm_destroy(screen->mm_VRAM_fe0);
274
275   nouveau_object_del(&screen->eng3d);
276   nouveau_object_del(&screen->eng2d);
277   nouveau_object_del(&screen->m2mf);
278
279   nouveau_screen_fini(&screen->base);
280
281   FREE(screen);
282}
283
284static int
285nvc0_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos,
286                     unsigned size, const uint32_t *data)
287{
288   struct nouveau_pushbuf *push = screen->base.pushbuf;
289
290   size /= 4;
291
292   BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2);
293   PUSH_DATA (push, (m - 0x3800) / 8);
294   PUSH_DATA (push, pos);
295   BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1);
296   PUSH_DATA (push, pos);
297   PUSH_DATAp(push, data, size);
298
299   return pos + size;
300}
301
302static void
303nvc0_magic_3d_init(struct nouveau_pushbuf *push, uint16_t obj_class)
304{
305   BEGIN_NVC0(push, SUBC_3D(0x10cc), 1);
306   PUSH_DATA (push, 0xff);
307   BEGIN_NVC0(push, SUBC_3D(0x10e0), 2);
308   PUSH_DATA (push, 0xff);
309   PUSH_DATA (push, 0xff);
310   BEGIN_NVC0(push, SUBC_3D(0x10ec), 2);
311   PUSH_DATA (push, 0xff);
312   PUSH_DATA (push, 0xff);
313   BEGIN_NVC0(push, SUBC_3D(0x074c), 1);
314   PUSH_DATA (push, 0x3f);
315
316   BEGIN_NVC0(push, SUBC_3D(0x16a8), 1);
317   PUSH_DATA (push, (3 << 16) | 3);
318   BEGIN_NVC0(push, SUBC_3D(0x1794), 1);
319   PUSH_DATA (push, (2 << 16) | 2);
320   BEGIN_NVC0(push, SUBC_3D(0x0de8), 1);
321   PUSH_DATA (push, 1);
322
323   BEGIN_NVC0(push, SUBC_3D(0x12ac), 1);
324   PUSH_DATA (push, 0);
325   BEGIN_NVC0(push, SUBC_3D(0x0218), 1);
326   PUSH_DATA (push, 0x10);
327   BEGIN_NVC0(push, SUBC_3D(0x10fc), 1);
328   PUSH_DATA (push, 0x10);
329   BEGIN_NVC0(push, SUBC_3D(0x1290), 1);
330   PUSH_DATA (push, 0x10);
331   BEGIN_NVC0(push, SUBC_3D(0x12d8), 2);
332   PUSH_DATA (push, 0x10);
333   PUSH_DATA (push, 0x10);
334   BEGIN_NVC0(push, SUBC_3D(0x1140), 1);
335   PUSH_DATA (push, 0x10);
336   BEGIN_NVC0(push, SUBC_3D(0x1610), 1);
337   PUSH_DATA (push, 0xe);
338
339   BEGIN_NVC0(push, SUBC_3D(0x164c), 1);
340   PUSH_DATA (push, 1 << 12);
341   BEGIN_NVC0(push, SUBC_3D(0x030c), 1);
342   PUSH_DATA (push, 0);
343   BEGIN_NVC0(push, SUBC_3D(0x0300), 1);
344   PUSH_DATA (push, 3);
345
346   BEGIN_NVC0(push, SUBC_3D(0x02d0), 1);
347   PUSH_DATA (push, 0x3fffff);
348   BEGIN_NVC0(push, SUBC_3D(0x0fdc), 1);
349   PUSH_DATA (push, 1);
350   BEGIN_NVC0(push, SUBC_3D(0x19c0), 1);
351   PUSH_DATA (push, 1);
352   BEGIN_NVC0(push, SUBC_3D(0x075c), 1);
353   PUSH_DATA (push, 3);
354
355   if (obj_class >= NVE4_3D_CLASS) {
356      BEGIN_NVC0(push, SUBC_3D(0x07fc), 1);
357      PUSH_DATA (push, 1);
358   }
359
360   /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
361    * are supposed to do */
362}
363
364static void
365nvc0_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
366{
367   struct nvc0_screen *screen = nvc0_screen(pscreen);
368   struct nouveau_pushbuf *push = screen->base.pushbuf;
369
370   /* we need to do it after possible flush in MARK_RING */
371   *sequence = ++screen->base.fence.sequence;
372
373   BEGIN_NVC0(push, NVC0_3D(QUERY_ADDRESS_HIGH), 4);
374   PUSH_DATAh(push, screen->fence.bo->offset);
375   PUSH_DATA (push, screen->fence.bo->offset);
376   PUSH_DATA (push, *sequence);
377   PUSH_DATA (push, NVC0_3D_QUERY_GET_FENCE | NVC0_3D_QUERY_GET_SHORT |
378              (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT));
379}
380
381static u32
382nvc0_screen_fence_update(struct pipe_screen *pscreen)
383{
384   struct nvc0_screen *screen = nvc0_screen(pscreen);
385   return screen->fence.map[0];
386}
387
388#define FAIL_SCREEN_INIT(str, err)                    \
389   do {                                               \
390      NOUVEAU_ERR(str, err);                          \
391      nvc0_screen_destroy(pscreen);                   \
392      return NULL;                                    \
393   } while(0)
394
395struct pipe_screen *
396nvc0_screen_create(struct nouveau_device *dev)
397{
398   struct nvc0_screen *screen;
399   struct pipe_screen *pscreen;
400   struct nouveau_object *chan;
401   struct nouveau_pushbuf *push;
402   uint32_t obj_class;
403   int ret;
404   unsigned i;
405   union nouveau_bo_config mm_config;
406
407   switch (dev->chipset & ~0xf) {
408   case 0xc0:
409   case 0xd0:
410   case 0xe0:
411      break;
412   default:
413      return NULL;
414   }
415
416   screen = CALLOC_STRUCT(nvc0_screen);
417   if (!screen)
418      return NULL;
419   pscreen = &screen->base.base;
420
421   screen->base.sysmem_bindings = PIPE_BIND_CONSTANT_BUFFER;
422
423   ret = nouveau_screen_init(&screen->base, dev);
424   if (ret) {
425      nvc0_screen_destroy(pscreen);
426      return NULL;
427   }
428   chan = screen->base.channel;
429   push = screen->base.pushbuf;
430   push->user_priv = screen;
431
432   pscreen->destroy = nvc0_screen_destroy;
433   pscreen->context_create = nvc0_create;
434   pscreen->is_format_supported = nvc0_screen_is_format_supported;
435   pscreen->get_param = nvc0_screen_get_param;
436   pscreen->get_shader_param = nvc0_screen_get_shader_param;
437   pscreen->get_paramf = nvc0_screen_get_paramf;
438
439   nvc0_screen_init_resource_functions(pscreen);
440
441   nouveau_screen_init_vdec(&screen->base);
442
443   ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096, NULL,
444                        &screen->fence.bo);
445   if (ret)
446      goto fail;
447   nouveau_bo_map(screen->fence.bo, 0, NULL);
448   screen->fence.map = screen->fence.bo->map;
449   screen->base.fence.emit = nvc0_screen_fence_emit;
450   screen->base.fence.update = nvc0_screen_fence_update;
451
452   switch (dev->chipset & 0xf0) {
453   case 0xe0:
454      obj_class = NVE4_P2MF_CLASS;
455      break;
456   default:
457      obj_class = NVC0_M2MF_CLASS;
458      break;
459   }
460   ret = nouveau_object_new(chan, 0xbeef323f, obj_class, NULL, 0,
461                            &screen->m2mf);
462   if (ret)
463      FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
464
465   BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
466   PUSH_DATA (push, screen->m2mf->oclass);
467   if (screen->m2mf->oclass == NVE4_P2MF_CLASS) {
468      BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
469      PUSH_DATA (push, 0xa0b5);
470   }
471
472   ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
473                            &screen->eng2d);
474   if (ret)
475      FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
476
477   BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
478   PUSH_DATA (push, screen->eng2d->oclass);
479   BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
480   PUSH_DATA (push, NVC0_2D_OPERATION_SRCCOPY);
481   BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
482   PUSH_DATA (push, 0);
483   BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
484   PUSH_DATA (push, 0);
485   BEGIN_NVC0(push, SUBC_2D(0x0884), 1);
486   PUSH_DATA (push, 0x3f);
487   BEGIN_NVC0(push, SUBC_2D(0x0888), 1);
488   PUSH_DATA (push, 1);
489
490   BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
491   PUSH_DATAh(push, screen->fence.bo->offset + 16);
492   PUSH_DATA (push, screen->fence.bo->offset + 16);
493
494   switch (dev->chipset & 0xf0) {
495   case 0xe0:
496      obj_class = NVE4_3D_CLASS;
497      break;
498   case 0xd0:
499   case 0xc0:
500   default:
501      switch (dev->chipset) {
502      case 0xd9:
503      case 0xc8:
504         obj_class = NVC8_3D_CLASS;
505         break;
506      case 0xc1:
507         obj_class = NVC1_3D_CLASS;
508         break;
509      default:
510         obj_class = NVC0_3D_CLASS;
511         break;
512      }
513      break;
514   }
515   ret = nouveau_object_new(chan, 0xbeef003d, obj_class, NULL, 0,
516                            &screen->eng3d);
517   if (ret)
518      FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
519   screen->base.class_3d = obj_class;
520
521   BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
522   PUSH_DATA (push, screen->eng3d->oclass);
523
524   BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
525   PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
526
527   if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", TRUE)) {
528      /* kill shaders after about 1 second (at 100 MHz) */
529      BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
530      PUSH_DATA (push, 0x17);
531   }
532
533   BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
534   PUSH_DATA (push, 1);
535
536   BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
537   PUSH_DATA (push, 0);
538   BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
539   PUSH_DATA (push, 0);
540   BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
541   PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
542   BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
543   PUSH_DATA (push, 0);
544   BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
545   PUSH_DATA (push, 1);
546   BEGIN_NVC0(push, NVC0_3D(LINE_LAST_PIXEL), 1);
547   PUSH_DATA (push, 0);
548   BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
549   PUSH_DATA (push, 1);
550   BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
551   PUSH_DATA (push, 0);
552   if (screen->eng3d->oclass < NVE4_3D_CLASS) {
553      BEGIN_NVC0(push, NVC0_3D(TEX_MISC), 1);
554      PUSH_DATA (push, NVC0_3D_TEX_MISC_SEAMLESS_CUBE_MAP);
555   } else {
556      BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
557      PUSH_DATA (push, 15);
558   }
559   BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
560   PUSH_DATA (push, 8); /* 128 */
561   BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
562   PUSH_DATA (push, 1);
563   if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
564      BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
565      PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
566   }
567
568   nvc0_magic_3d_init(push, screen->eng3d->oclass);
569
570   ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 20, NULL,
571                        &screen->text);
572   if (ret)
573      goto fail;
574
575   /* XXX: getting a page fault at the end of the code buffer every few
576    *  launches, don't use the last 256 bytes to work around them - prefetch ?
577    */
578   nouveau_heap_init(&screen->text_heap, 0, (1 << 20) - 0x100);
579
580   ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 12, 6 << 16, NULL,
581                        &screen->uniform_bo);
582   if (ret)
583      goto fail;
584
585   for (i = 0; i < 5; ++i) {
586      /* TIC and TSC entries for each unit (nve4+ only) */
587      /* auxiliary constants (6 user clip planes, base instance id */
588      BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
589      PUSH_DATA (push, 512);
590      PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (i << 9));
591      PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (i << 9));
592      BEGIN_NVC0(push, NVC0_3D(CB_BIND(i)), 1);
593      PUSH_DATA (push, (15 << 4) | 1);
594      if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
595         unsigned j;
596         BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
597         PUSH_DATA (push, 0);
598         for (j = 0; j < 8; ++j)
599            PUSH_DATA(push, j);
600      } else {
601         BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
602         PUSH_DATA (push, 0x54);
603      }
604   }
605   BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
606   PUSH_DATA (push, 0);
607
608   /* max MPs * max warps per MP (TODO: ask kernel) */
609   if (screen->eng3d->oclass >= NVE4_3D_CLASS)
610      screen->tls_size = 8 * 64;
611   else
612      screen->tls_size = 16 * 48;
613   screen->tls_size *= NVC0_CAP_MAX_PROGRAM_TEMPS * 16;
614   screen->tls_size = align(screen->tls_size, 1 << 17);
615
616   ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17,
617                        screen->tls_size, NULL, &screen->tls);
618   if (ret)
619      goto fail;
620
621   BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
622   PUSH_DATAh(push, screen->text->offset);
623   PUSH_DATA (push, screen->text->offset);
624   BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
625   PUSH_DATAh(push, screen->tls->offset);
626   PUSH_DATA (push, screen->tls->offset);
627   PUSH_DATA (push, screen->tls_size >> 32);
628   PUSH_DATA (push, screen->tls_size);
629   BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
630   PUSH_DATA (push, 0);
631   BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
632   PUSH_DATA (push, 0);
633
634   ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 20, NULL,
635                        &screen->poly_cache);
636   if (ret)
637      goto fail;
638
639   BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
640   PUSH_DATAh(push, screen->poly_cache->offset);
641   PUSH_DATA (push, screen->poly_cache->offset);
642   PUSH_DATA (push, 3);
643
644   ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 17, NULL,
645                        &screen->txc);
646   if (ret)
647      goto fail;
648
649   BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
650   PUSH_DATAh(push, screen->txc->offset);
651   PUSH_DATA (push, screen->txc->offset);
652   PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
653
654   BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
655   PUSH_DATAh(push, screen->txc->offset + 65536);
656   PUSH_DATA (push, screen->txc->offset + 65536);
657   PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
658
659   BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
660   PUSH_DATA (push, 0);
661   BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
662   PUSH_DATA (push, 0);
663   PUSH_DATA (push, 0);
664   BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
665   PUSH_DATA (push, 0x3f);
666
667   BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
668   PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
669   BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
670   for (i = 0; i < 8 * 2; ++i)
671      PUSH_DATA(push, 0);
672   BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
673   PUSH_DATA (push, 0);
674   BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
675   PUSH_DATA (push, 0);
676
677   /* neither scissors, viewport nor stencil mask should affect clears */
678   BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
679   PUSH_DATA (push, 0);
680
681   BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
682   PUSH_DATA (push, 1);
683   BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(0)), 2);
684   PUSH_DATAf(push, 0.0f);
685   PUSH_DATAf(push, 1.0f);
686   BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
687   PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
688
689   /* We use scissors instead of exact view volume clipping,
690    * so they're always enabled.
691    */
692   BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(0)), 3);
693   PUSH_DATA (push, 1);
694   PUSH_DATA (push, 8192 << 16);
695   PUSH_DATA (push, 8192 << 16);
696
697#define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
698
699   i = 0;
700   MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, nvc0_9097_per_instance_bf);
701   MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, nvc0_9097_blend_enables);
702   MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, nvc0_9097_vertex_array_select);
703   MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, nvc0_9097_tep_select);
704   MK_MACRO(NVC0_3D_MACRO_GP_SELECT, nvc0_9097_gp_select);
705   MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, nvc0_9097_poly_mode_front);
706   MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, nvc0_9097_poly_mode_back);
707
708   BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
709   PUSH_DATA (push, 1);
710   BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
711   PUSH_DATA (push, 1);
712   BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
713   PUSH_DATA (push, 0x40);
714   BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
715   PUSH_DATA (push, 0);
716   BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
717   PUSH_DATA (push, 0x30);
718   BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
719   PUSH_DATA (push, 3);
720   BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
721   PUSH_DATA (push, 0x20);
722   BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
723   PUSH_DATA (push, 0x00);
724
725   BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
726   PUSH_DATA (push, 0);
727   BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
728   PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
729
730   IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
731
732   BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
733   PUSH_DATA (push, 0xab);
734   PUSH_DATA (push, 0x00000000);
735
736   PUSH_KICK (push);
737
738   screen->tic.entries = CALLOC(4096, sizeof(void *));
739   screen->tsc.entries = screen->tic.entries + 2048;
740
741   mm_config.nvc0.tile_mode = 0;
742   mm_config.nvc0.memtype = 0xfe0;
743   screen->mm_VRAM_fe0 = nouveau_mm_create(dev, NOUVEAU_BO_VRAM, &mm_config);
744
745   if (!nvc0_blitctx_create(screen))
746      goto fail;
747
748   nouveau_fence_new(&screen->base, &screen->base.fence.current, FALSE);
749
750   return pscreen;
751
752fail:
753   nvc0_screen_destroy(pscreen);
754   return NULL;
755}
756
757int
758nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
759{
760   int i = screen->tic.next;
761
762   while (screen->tic.lock[i / 32] & (1 << (i % 32)))
763      i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
764
765   screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
766
767   if (screen->tic.entries[i])
768      nv50_tic_entry(screen->tic.entries[i])->id = -1;
769
770   screen->tic.entries[i] = entry;
771   return i;
772}
773
774int
775nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
776{
777   int i = screen->tsc.next;
778
779   while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
780      i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
781
782   screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
783
784   if (screen->tsc.entries[i])
785      nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
786
787   screen->tsc.entries[i] = entry;
788   return i;
789}
790