nvc0_screen.c revision 587c221a29fe01e24c69952a7a9b0034f7652e4e
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      return 0; /* state trackers will know better */
143   case PIPE_CAP_USER_VERTEX_BUFFERS:
144      return 1;
145   case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
146   case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
147   case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
148      return 0;
149   default:
150      NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
151      return 0;
152   }
153}
154
155static int
156nvc0_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
157                             enum pipe_shader_cap param)
158{
159   switch (shader) {
160   case PIPE_SHADER_VERTEX:
161      /*
162   case PIPE_SHADER_TESSELLATION_CONTROL:
163   case PIPE_SHADER_TESSELLATION_EVALUATION:
164      */
165   case PIPE_SHADER_GEOMETRY:
166   case PIPE_SHADER_FRAGMENT:
167      break;
168   default:
169      return 0;
170   }
171
172   switch (param) {
173   case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
174   case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
175   case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
176   case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
177      return 16384;
178   case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
179      return 16;
180   case PIPE_SHADER_CAP_MAX_INPUTS:
181      if (shader == PIPE_SHADER_VERTEX)
182         return 32;
183      if (shader == PIPE_SHADER_FRAGMENT)
184         return (0x200 + 0x20 + 0x80) / 16; /* generic + colors + TexCoords */
185      return (0x200 + 0x40 + 0x80) / 16; /* without 0x60 for per-patch inputs */
186   case PIPE_SHADER_CAP_MAX_CONSTS:
187      return 65536 / 16;
188   case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
189      return 14;
190   case PIPE_SHADER_CAP_MAX_ADDRS:
191      return 1;
192   case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
193   case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
194      return shader != PIPE_SHADER_FRAGMENT;
195   case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
196   case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
197      return 1;
198   case PIPE_SHADER_CAP_MAX_PREDS:
199      return 0;
200   case PIPE_SHADER_CAP_MAX_TEMPS:
201      return NVC0_CAP_MAX_PROGRAM_TEMPS;
202   case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
203      return 1;
204   case PIPE_SHADER_CAP_SUBROUTINES:
205      return 1; /* but inlining everything, we need function declarations */
206   case PIPE_SHADER_CAP_INTEGERS:
207      return 1;
208   case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
209      return 16; /* would be 32 in linked (OpenGL-style) mode */
210      /*
211   case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLER_VIEWS:
212      return 32;
213      */
214   default:
215      NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
216      return 0;
217   }
218}
219
220static float
221nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
222{
223   switch (param) {
224   case PIPE_CAPF_MAX_LINE_WIDTH:
225   case PIPE_CAPF_MAX_LINE_WIDTH_AA:
226      return 10.0f;
227   case PIPE_CAPF_MAX_POINT_WIDTH:
228      return 63.0f;
229   case PIPE_CAPF_MAX_POINT_WIDTH_AA:
230      return 63.375f;
231   case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
232      return 16.0f;
233   case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
234      return 15.0f;
235   default:
236      NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
237      return 0.0f;
238   }
239}
240
241static void
242nvc0_screen_destroy(struct pipe_screen *pscreen)
243{
244   struct nvc0_screen *screen = nvc0_screen(pscreen);
245
246   if (screen->base.fence.current) {
247      nouveau_fence_wait(screen->base.fence.current);
248      nouveau_fence_ref(NULL, &screen->base.fence.current);
249   }
250   if (screen->base.pushbuf)
251      screen->base.pushbuf->user_priv = NULL;
252
253   if (screen->blitctx)
254      FREE(screen->blitctx);
255
256   nouveau_bo_ref(NULL, &screen->text);
257   nouveau_bo_ref(NULL, &screen->uniform_bo);
258   nouveau_bo_ref(NULL, &screen->tls);
259   nouveau_bo_ref(NULL, &screen->txc);
260   nouveau_bo_ref(NULL, &screen->fence.bo);
261   nouveau_bo_ref(NULL, &screen->poly_cache);
262
263   nouveau_heap_destroy(&screen->lib_code);
264   nouveau_heap_destroy(&screen->text_heap);
265
266   if (screen->tic.entries)
267      FREE(screen->tic.entries);
268
269   nouveau_mm_destroy(screen->mm_VRAM_fe0);
270
271   nouveau_object_del(&screen->eng3d);
272   nouveau_object_del(&screen->eng2d);
273   nouveau_object_del(&screen->m2mf);
274
275   nouveau_screen_fini(&screen->base);
276
277   FREE(screen);
278}
279
280static int
281nvc0_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos,
282                     unsigned size, const uint32_t *data)
283{
284   struct nouveau_pushbuf *push = screen->base.pushbuf;
285
286   size /= 4;
287
288   BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2);
289   PUSH_DATA (push, (m - 0x3800) / 8);
290   PUSH_DATA (push, pos);
291   BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1);
292   PUSH_DATA (push, pos);
293   PUSH_DATAp(push, data, size);
294
295   return pos + size;
296}
297
298static void
299nvc0_magic_3d_init(struct nouveau_pushbuf *push, uint16_t obj_class)
300{
301   BEGIN_NVC0(push, SUBC_3D(0x10cc), 1);
302   PUSH_DATA (push, 0xff);
303   BEGIN_NVC0(push, SUBC_3D(0x10e0), 2);
304   PUSH_DATA (push, 0xff);
305   PUSH_DATA (push, 0xff);
306   BEGIN_NVC0(push, SUBC_3D(0x10ec), 2);
307   PUSH_DATA (push, 0xff);
308   PUSH_DATA (push, 0xff);
309   BEGIN_NVC0(push, SUBC_3D(0x074c), 1);
310   PUSH_DATA (push, 0x3f);
311
312   BEGIN_NVC0(push, SUBC_3D(0x16a8), 1);
313   PUSH_DATA (push, (3 << 16) | 3);
314   BEGIN_NVC0(push, SUBC_3D(0x1794), 1);
315   PUSH_DATA (push, (2 << 16) | 2);
316   BEGIN_NVC0(push, SUBC_3D(0x0de8), 1);
317   PUSH_DATA (push, 1);
318
319   BEGIN_NVC0(push, SUBC_3D(0x12ac), 1);
320   PUSH_DATA (push, 0);
321   BEGIN_NVC0(push, SUBC_3D(0x0218), 1);
322   PUSH_DATA (push, 0x10);
323   BEGIN_NVC0(push, SUBC_3D(0x10fc), 1);
324   PUSH_DATA (push, 0x10);
325   BEGIN_NVC0(push, SUBC_3D(0x1290), 1);
326   PUSH_DATA (push, 0x10);
327   BEGIN_NVC0(push, SUBC_3D(0x12d8), 2);
328   PUSH_DATA (push, 0x10);
329   PUSH_DATA (push, 0x10);
330   BEGIN_NVC0(push, SUBC_3D(0x1140), 1);
331   PUSH_DATA (push, 0x10);
332   BEGIN_NVC0(push, SUBC_3D(0x1610), 1);
333   PUSH_DATA (push, 0xe);
334
335   BEGIN_NVC0(push, SUBC_3D(0x164c), 1);
336   PUSH_DATA (push, 1 << 12);
337   BEGIN_NVC0(push, SUBC_3D(0x030c), 1);
338   PUSH_DATA (push, 0);
339   BEGIN_NVC0(push, SUBC_3D(0x0300), 1);
340   PUSH_DATA (push, 3);
341
342   BEGIN_NVC0(push, SUBC_3D(0x02d0), 1);
343   PUSH_DATA (push, 0x3fffff);
344   BEGIN_NVC0(push, SUBC_3D(0x0fdc), 1);
345   PUSH_DATA (push, 1);
346   BEGIN_NVC0(push, SUBC_3D(0x19c0), 1);
347   PUSH_DATA (push, 1);
348   BEGIN_NVC0(push, SUBC_3D(0x075c), 1);
349   PUSH_DATA (push, 3);
350
351   if (obj_class >= NVE4_3D_CLASS) {
352      BEGIN_NVC0(push, SUBC_3D(0x07fc), 1);
353      PUSH_DATA (push, 1);
354   }
355
356   /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
357    * are supposed to do */
358}
359
360static void
361nvc0_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
362{
363   struct nvc0_screen *screen = nvc0_screen(pscreen);
364   struct nouveau_pushbuf *push = screen->base.pushbuf;
365
366   /* we need to do it after possible flush in MARK_RING */
367   *sequence = ++screen->base.fence.sequence;
368
369   BEGIN_NVC0(push, NVC0_3D(QUERY_ADDRESS_HIGH), 4);
370   PUSH_DATAh(push, screen->fence.bo->offset);
371   PUSH_DATA (push, screen->fence.bo->offset);
372   PUSH_DATA (push, *sequence);
373   PUSH_DATA (push, NVC0_3D_QUERY_GET_FENCE | NVC0_3D_QUERY_GET_SHORT |
374              (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT));
375}
376
377static u32
378nvc0_screen_fence_update(struct pipe_screen *pscreen)
379{
380   struct nvc0_screen *screen = nvc0_screen(pscreen);
381   return screen->fence.map[0];
382}
383
384#define FAIL_SCREEN_INIT(str, err)                    \
385   do {                                               \
386      NOUVEAU_ERR(str, err);                          \
387      nvc0_screen_destroy(pscreen);                   \
388      return NULL;                                    \
389   } while(0)
390
391struct pipe_screen *
392nvc0_screen_create(struct nouveau_device *dev)
393{
394   struct nvc0_screen *screen;
395   struct pipe_screen *pscreen;
396   struct nouveau_object *chan;
397   struct nouveau_pushbuf *push;
398   uint32_t obj_class;
399   int ret;
400   unsigned i;
401   union nouveau_bo_config mm_config;
402
403   switch (dev->chipset & ~0xf) {
404   case 0xc0:
405   case 0xd0:
406   case 0xe0:
407      break;
408   default:
409      return NULL;
410   }
411
412   screen = CALLOC_STRUCT(nvc0_screen);
413   if (!screen)
414      return NULL;
415   pscreen = &screen->base.base;
416
417   screen->base.sysmem_bindings = PIPE_BIND_CONSTANT_BUFFER;
418
419   ret = nouveau_screen_init(&screen->base, dev);
420   if (ret) {
421      nvc0_screen_destroy(pscreen);
422      return NULL;
423   }
424   chan = screen->base.channel;
425   push = screen->base.pushbuf;
426   push->user_priv = screen;
427
428   pscreen->destroy = nvc0_screen_destroy;
429   pscreen->context_create = nvc0_create;
430   pscreen->is_format_supported = nvc0_screen_is_format_supported;
431   pscreen->get_param = nvc0_screen_get_param;
432   pscreen->get_shader_param = nvc0_screen_get_shader_param;
433   pscreen->get_paramf = nvc0_screen_get_paramf;
434
435   nvc0_screen_init_resource_functions(pscreen);
436
437   nouveau_screen_init_vdec(&screen->base);
438
439   ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096, NULL,
440                        &screen->fence.bo);
441   if (ret)
442      goto fail;
443   nouveau_bo_map(screen->fence.bo, 0, NULL);
444   screen->fence.map = screen->fence.bo->map;
445   screen->base.fence.emit = nvc0_screen_fence_emit;
446   screen->base.fence.update = nvc0_screen_fence_update;
447
448   switch (dev->chipset & 0xf0) {
449   case 0xe0:
450      obj_class = NVE4_P2MF_CLASS;
451      break;
452   default:
453      obj_class = NVC0_M2MF_CLASS;
454      break;
455   }
456   ret = nouveau_object_new(chan, 0xbeef323f, obj_class, NULL, 0,
457                            &screen->m2mf);
458   if (ret)
459      FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
460
461   BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
462   PUSH_DATA (push, screen->m2mf->oclass);
463   if (screen->m2mf->oclass == NVE4_P2MF_CLASS) {
464      BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
465      PUSH_DATA (push, 0xa0b5);
466   }
467
468   ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
469                            &screen->eng2d);
470   if (ret)
471      FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
472
473   BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
474   PUSH_DATA (push, screen->eng2d->oclass);
475   BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
476   PUSH_DATA (push, NVC0_2D_OPERATION_SRCCOPY);
477   BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
478   PUSH_DATA (push, 0);
479   BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
480   PUSH_DATA (push, 0);
481   BEGIN_NVC0(push, SUBC_2D(0x0884), 1);
482   PUSH_DATA (push, 0x3f);
483   BEGIN_NVC0(push, SUBC_2D(0x0888), 1);
484   PUSH_DATA (push, 1);
485
486   BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
487   PUSH_DATAh(push, screen->fence.bo->offset + 16);
488   PUSH_DATA (push, screen->fence.bo->offset + 16);
489
490   switch (dev->chipset & 0xf0) {
491   case 0xe0:
492      obj_class = NVE4_3D_CLASS;
493      break;
494   case 0xd0:
495   case 0xc0:
496   default:
497      switch (dev->chipset) {
498      case 0xd9:
499      case 0xc8:
500         obj_class = NVC8_3D_CLASS;
501         break;
502      case 0xc1:
503         obj_class = NVC1_3D_CLASS;
504         break;
505      default:
506         obj_class = NVC0_3D_CLASS;
507         break;
508      }
509      break;
510   }
511   ret = nouveau_object_new(chan, 0xbeef003d, obj_class, NULL, 0,
512                            &screen->eng3d);
513   if (ret)
514      FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
515   screen->base.class_3d = obj_class;
516
517   BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
518   PUSH_DATA (push, screen->eng3d->oclass);
519
520   BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
521   PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
522
523   if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", TRUE)) {
524      /* kill shaders after about 1 second (at 100 MHz) */
525      BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
526      PUSH_DATA (push, 0x17);
527   }
528
529   BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
530   PUSH_DATA (push, 1);
531
532   BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
533   PUSH_DATA (push, 0);
534   BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
535   PUSH_DATA (push, 0);
536   BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
537   PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
538   BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
539   PUSH_DATA (push, 0);
540   BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
541   PUSH_DATA (push, 1);
542   BEGIN_NVC0(push, NVC0_3D(LINE_LAST_PIXEL), 1);
543   PUSH_DATA (push, 0);
544   BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
545   PUSH_DATA (push, 1);
546   BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
547   PUSH_DATA (push, 0);
548   if (screen->eng3d->oclass < NVE4_3D_CLASS) {
549      BEGIN_NVC0(push, NVC0_3D(TEX_MISC), 1);
550      PUSH_DATA (push, NVC0_3D_TEX_MISC_SEAMLESS_CUBE_MAP);
551   } else {
552      BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
553      PUSH_DATA (push, 15);
554   }
555   BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
556   PUSH_DATA (push, 8); /* 128 */
557   BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
558   PUSH_DATA (push, 1);
559   if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
560      BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
561      PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
562   }
563
564   nvc0_magic_3d_init(push, screen->eng3d->oclass);
565
566   ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 20, NULL,
567                        &screen->text);
568   if (ret)
569      goto fail;
570
571   /* XXX: getting a page fault at the end of the code buffer every few
572    *  launches, don't use the last 256 bytes to work around them - prefetch ?
573    */
574   nouveau_heap_init(&screen->text_heap, 0, (1 << 20) - 0x100);
575
576   ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 12, 6 << 16, NULL,
577                        &screen->uniform_bo);
578   if (ret)
579      goto fail;
580
581   for (i = 0; i < 5; ++i) {
582      /* TIC and TSC entries for each unit (nve4+ only) */
583      /* auxiliary constants (6 user clip planes, base instance id */
584      BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
585      PUSH_DATA (push, 512);
586      PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (i << 9));
587      PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (i << 9));
588      BEGIN_NVC0(push, NVC0_3D(CB_BIND(i)), 1);
589      PUSH_DATA (push, (15 << 4) | 1);
590      if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
591         unsigned j;
592         BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
593         PUSH_DATA (push, 0);
594         for (j = 0; j < 8; ++j)
595            PUSH_DATA(push, j);
596      } else {
597         BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
598         PUSH_DATA (push, 0x54);
599      }
600   }
601   BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
602   PUSH_DATA (push, 0);
603
604   /* max MPs * max warps per MP (TODO: ask kernel) */
605   if (screen->eng3d->oclass >= NVE4_3D_CLASS)
606      screen->tls_size = 8 * 64;
607   else
608      screen->tls_size = 16 * 48;
609   screen->tls_size *= NVC0_CAP_MAX_PROGRAM_TEMPS * 16;
610   screen->tls_size = align(screen->tls_size, 1 << 17);
611
612   ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17,
613                        screen->tls_size, NULL, &screen->tls);
614   if (ret)
615      goto fail;
616
617   BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
618   PUSH_DATAh(push, screen->text->offset);
619   PUSH_DATA (push, screen->text->offset);
620   BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
621   PUSH_DATAh(push, screen->tls->offset);
622   PUSH_DATA (push, screen->tls->offset);
623   PUSH_DATA (push, screen->tls_size >> 32);
624   PUSH_DATA (push, screen->tls_size);
625   BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
626   PUSH_DATA (push, 0);
627   BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
628   PUSH_DATA (push, 0);
629
630   ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 20, NULL,
631                        &screen->poly_cache);
632   if (ret)
633      goto fail;
634
635   BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
636   PUSH_DATAh(push, screen->poly_cache->offset);
637   PUSH_DATA (push, screen->poly_cache->offset);
638   PUSH_DATA (push, 3);
639
640   ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 17, NULL,
641                        &screen->txc);
642   if (ret)
643      goto fail;
644
645   BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
646   PUSH_DATAh(push, screen->txc->offset);
647   PUSH_DATA (push, screen->txc->offset);
648   PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
649
650   BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
651   PUSH_DATAh(push, screen->txc->offset + 65536);
652   PUSH_DATA (push, screen->txc->offset + 65536);
653   PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
654
655   BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
656   PUSH_DATA (push, 0);
657   BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
658   PUSH_DATA (push, 0);
659   PUSH_DATA (push, 0);
660   BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
661   PUSH_DATA (push, 0x3f);
662
663   BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
664   PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
665   BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
666   for (i = 0; i < 8 * 2; ++i)
667      PUSH_DATA(push, 0);
668   BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
669   PUSH_DATA (push, 0);
670   BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
671   PUSH_DATA (push, 0);
672
673   /* neither scissors, viewport nor stencil mask should affect clears */
674   BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
675   PUSH_DATA (push, 0);
676
677   BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
678   PUSH_DATA (push, 1);
679   BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(0)), 2);
680   PUSH_DATAf(push, 0.0f);
681   PUSH_DATAf(push, 1.0f);
682   BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
683   PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
684
685   /* We use scissors instead of exact view volume clipping,
686    * so they're always enabled.
687    */
688   BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(0)), 3);
689   PUSH_DATA (push, 1);
690   PUSH_DATA (push, 8192 << 16);
691   PUSH_DATA (push, 8192 << 16);
692
693#define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
694
695   i = 0;
696   MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, nvc0_9097_per_instance_bf);
697   MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, nvc0_9097_blend_enables);
698   MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, nvc0_9097_vertex_array_select);
699   MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, nvc0_9097_tep_select);
700   MK_MACRO(NVC0_3D_MACRO_GP_SELECT, nvc0_9097_gp_select);
701   MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, nvc0_9097_poly_mode_front);
702   MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, nvc0_9097_poly_mode_back);
703
704   BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
705   PUSH_DATA (push, 1);
706   BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
707   PUSH_DATA (push, 1);
708   BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
709   PUSH_DATA (push, 0x40);
710   BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
711   PUSH_DATA (push, 0);
712   BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
713   PUSH_DATA (push, 0x30);
714   BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
715   PUSH_DATA (push, 3);
716   BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
717   PUSH_DATA (push, 0x20);
718   BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
719   PUSH_DATA (push, 0x00);
720
721   BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
722   PUSH_DATA (push, 0);
723   BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
724   PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
725
726   IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
727
728   BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
729   PUSH_DATA (push, 0xab);
730   PUSH_DATA (push, 0x00000000);
731
732   PUSH_KICK (push);
733
734   screen->tic.entries = CALLOC(4096, sizeof(void *));
735   screen->tsc.entries = screen->tic.entries + 2048;
736
737   mm_config.nvc0.tile_mode = 0;
738   mm_config.nvc0.memtype = 0xfe0;
739   screen->mm_VRAM_fe0 = nouveau_mm_create(dev, NOUVEAU_BO_VRAM, &mm_config);
740
741   if (!nvc0_blitctx_create(screen))
742      goto fail;
743
744   nouveau_fence_new(&screen->base, &screen->base.fence.current, FALSE);
745
746   return pscreen;
747
748fail:
749   nvc0_screen_destroy(pscreen);
750   return NULL;
751}
752
753int
754nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
755{
756   int i = screen->tic.next;
757
758   while (screen->tic.lock[i / 32] & (1 << (i % 32)))
759      i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
760
761   screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
762
763   if (screen->tic.entries[i])
764      nv50_tic_entry(screen->tic.entries[i])->id = -1;
765
766   screen->tic.entries[i] = entry;
767   return i;
768}
769
770int
771nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
772{
773   int i = screen->tsc.next;
774
775   while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
776      i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
777
778   screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
779
780   if (screen->tsc.entries[i])
781      nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
782
783   screen->tsc.entries[i] = entry;
784   return i;
785}
786