nouveau: create only 1 shared screen between vdpau and opengl
[mesa.git] / src / gallium / drivers / nouveau / nvc0 / nvc0_screen.c
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 OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE 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 "nouveau_vp3_video.h"
31
32 #include "nvc0/nvc0_context.h"
33 #include "nvc0/nvc0_screen.h"
34
35 #include "nvc0/nvc0_graph_macros.h"
36
37 #ifndef NOUVEAU_GETPARAM_GRAPH_UNITS
38 # define NOUVEAU_GETPARAM_GRAPH_UNITS 13
39 #endif
40
41 static boolean
42 nvc0_screen_is_format_supported(struct pipe_screen *pscreen,
43 enum pipe_format format,
44 enum pipe_texture_target target,
45 unsigned sample_count,
46 unsigned bindings)
47 {
48 if (sample_count > 8)
49 return FALSE;
50 if (!(0x117 & (1 << sample_count))) /* 0, 1, 2, 4 or 8 */
51 return FALSE;
52
53 if (!util_format_is_supported(format, bindings))
54 return FALSE;
55
56 if ((bindings & PIPE_BIND_SAMPLER_VIEW) && (target != PIPE_BUFFER))
57 if (util_format_get_blocksizebits(format) == 3 * 32)
58 return FALSE;
59
60 /* transfers & shared are always supported */
61 bindings &= ~(PIPE_BIND_TRANSFER_READ |
62 PIPE_BIND_TRANSFER_WRITE |
63 PIPE_BIND_SHARED);
64
65 return (nvc0_format_table[format].usage & bindings) == bindings;
66 }
67
68 static int
69 nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
70 {
71 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
72
73 switch (param) {
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 (class_3d >= NVE4_3D_CLASS) ? 13 : 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_MIXED_FRAMEBUFFER_SIZES:
90 case PIPE_CAP_ANISOTROPIC_FILTER:
91 case PIPE_CAP_SEAMLESS_CUBE_MAP:
92 case PIPE_CAP_CUBE_MAP_ARRAY:
93 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
94 case PIPE_CAP_TEXTURE_MULTISAMPLE:
95 return 1;
96 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
97 return 65536;
98 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
99 return (class_3d >= NVE4_3D_CLASS) ? 1 : 0;
100 case PIPE_CAP_TWO_SIDED_STENCIL:
101 case PIPE_CAP_DEPTH_CLIP_DISABLE:
102 case PIPE_CAP_POINT_SPRITE:
103 case PIPE_CAP_TGSI_TEXCOORD:
104 return 1;
105 case PIPE_CAP_SM3:
106 return 1;
107 case PIPE_CAP_GLSL_FEATURE_LEVEL:
108 return 330;
109 case PIPE_CAP_MAX_RENDER_TARGETS:
110 return 8;
111 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
112 return 1;
113 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
114 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
115 case PIPE_CAP_VERTEX_COLOR_CLAMPED:
116 return 1;
117 case PIPE_CAP_QUERY_TIMESTAMP:
118 case PIPE_CAP_QUERY_TIME_ELAPSED:
119 case PIPE_CAP_OCCLUSION_QUERY:
120 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
121 case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
122 return 1;
123 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
124 return 4;
125 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
126 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
127 return 128;
128 case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
129 case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
130 return 1024;
131 case PIPE_CAP_BLEND_EQUATION_SEPARATE:
132 case PIPE_CAP_INDEP_BLEND_ENABLE:
133 case PIPE_CAP_INDEP_BLEND_FUNC:
134 return 1;
135 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
136 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
137 return 1;
138 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
139 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
140 return 0;
141 case PIPE_CAP_SHADER_STENCIL_EXPORT:
142 return 0;
143 case PIPE_CAP_PRIMITIVE_RESTART:
144 case PIPE_CAP_TGSI_INSTANCEID:
145 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
146 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
147 case PIPE_CAP_CONDITIONAL_RENDER:
148 case PIPE_CAP_TEXTURE_BARRIER:
149 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
150 case PIPE_CAP_START_INSTANCE:
151 return 1;
152 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
153 return 0; /* state trackers will know better */
154 case PIPE_CAP_USER_CONSTANT_BUFFERS:
155 case PIPE_CAP_USER_INDEX_BUFFERS:
156 case PIPE_CAP_USER_VERTEX_BUFFERS:
157 return 1;
158 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
159 return 256;
160 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
161 return 1; /* 256 for binding as RT, but that's not possible in GL */
162 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
163 return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
164 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
165 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
166 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
167 return 0;
168 case PIPE_CAP_COMPUTE:
169 return (class_3d >= NVE4_3D_CLASS) ? 1 : 0;
170 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
171 return 1;
172 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
173 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
174 case PIPE_CAP_ENDIANNESS:
175 return PIPE_ENDIAN_LITTLE;
176 case PIPE_CAP_TGSI_VS_LAYER:
177 return 0;
178 case PIPE_CAP_MAX_VIEWPORTS:
179 return 1;
180 default:
181 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
182 return 0;
183 }
184 }
185
186 static int
187 nvc0_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
188 enum pipe_shader_cap param)
189 {
190 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
191
192 switch (shader) {
193 case PIPE_SHADER_VERTEX:
194 /*
195 case PIPE_SHADER_TESSELLATION_CONTROL:
196 case PIPE_SHADER_TESSELLATION_EVALUATION:
197 */
198 case PIPE_SHADER_GEOMETRY:
199 case PIPE_SHADER_FRAGMENT:
200 break;
201 case PIPE_SHADER_COMPUTE:
202 if (class_3d < NVE4_3D_CLASS)
203 return 0;
204 break;
205 default:
206 return 0;
207 }
208
209 switch (param) {
210 case PIPE_SHADER_CAP_PREFERRED_IR:
211 return PIPE_SHADER_IR_TGSI;
212 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
213 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
214 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
215 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
216 return 16384;
217 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
218 return 16;
219 case PIPE_SHADER_CAP_MAX_INPUTS:
220 if (shader == PIPE_SHADER_VERTEX)
221 return 32;
222 /* NOTE: These only count our slots for GENERIC varyings.
223 * The address space may be larger, but the actual hard limit seems to be
224 * less than what the address space layout permits, so don't add TEXCOORD,
225 * COLOR, etc. here.
226 */
227 if (shader == PIPE_SHADER_FRAGMENT)
228 return 0x1f0 / 16;
229 /* Actually this counts CLIPVERTEX, which occupies the last generic slot,
230 * and excludes 0x60 per-patch inputs.
231 */
232 return 0x200 / 16;
233 case PIPE_SHADER_CAP_MAX_CONSTS:
234 return 65536 / 16;
235 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
236 if (shader == PIPE_SHADER_COMPUTE && class_3d >= NVE4_3D_CLASS)
237 return NVE4_MAX_PIPE_CONSTBUFS_COMPUTE;
238 return NVC0_MAX_PIPE_CONSTBUFS;
239 case PIPE_SHADER_CAP_MAX_ADDRS:
240 return 1;
241 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
242 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
243 return shader != PIPE_SHADER_FRAGMENT;
244 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
245 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
246 return 1;
247 case PIPE_SHADER_CAP_MAX_PREDS:
248 return 0;
249 case PIPE_SHADER_CAP_MAX_TEMPS:
250 return NVC0_CAP_MAX_PROGRAM_TEMPS;
251 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
252 return 1;
253 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
254 return 0;
255 case PIPE_SHADER_CAP_SUBROUTINES:
256 return 1;
257 case PIPE_SHADER_CAP_INTEGERS:
258 return 1;
259 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
260 return 16; /* would be 32 in linked (OpenGL-style) mode */
261 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
262 return 16; /* XXX not sure if more are really safe */
263 default:
264 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
265 return 0;
266 }
267 }
268
269 static float
270 nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
271 {
272 switch (param) {
273 case PIPE_CAPF_MAX_LINE_WIDTH:
274 case PIPE_CAPF_MAX_LINE_WIDTH_AA:
275 return 10.0f;
276 case PIPE_CAPF_MAX_POINT_WIDTH:
277 return 63.0f;
278 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
279 return 63.375f;
280 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
281 return 16.0f;
282 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
283 return 15.0f;
284 default:
285 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
286 return 0.0f;
287 }
288 }
289
290 static int
291 nvc0_screen_get_compute_param(struct pipe_screen *pscreen,
292 enum pipe_compute_cap param, void *data)
293 {
294 uint64_t *data64 = (uint64_t *)data;
295 const uint16_t obj_class = nvc0_screen(pscreen)->compute->oclass;
296
297 switch (param) {
298 case PIPE_COMPUTE_CAP_GRID_DIMENSION:
299 data64[0] = 3;
300 return 8;
301 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
302 data64[0] = (obj_class >= NVE4_COMPUTE_CLASS) ? 0x7fffffff : 65535;
303 data64[1] = 65535;
304 data64[2] = 65535;
305 return 24;
306 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
307 data64[0] = 1024;
308 data64[1] = 1024;
309 data64[2] = 64;
310 return 24;
311 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
312 data64[0] = 1024;
313 return 8;
314 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */
315 data64[0] = (uint64_t)1 << 40;
316 return 8;
317 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
318 data64[0] = 48 << 10;
319 return 8;
320 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
321 data64[0] = 512 << 10;
322 return 8;
323 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
324 data64[0] = 4096;
325 return 8;
326 default:
327 return 0;
328 }
329 }
330
331 static void
332 nvc0_screen_destroy(struct pipe_screen *pscreen)
333 {
334 struct nvc0_screen *screen = nvc0_screen(pscreen);
335
336 if (!nouveau_drm_screen_unref(&screen->base))
337 return;
338
339 if (screen->base.fence.current) {
340 nouveau_fence_wait(screen->base.fence.current);
341 nouveau_fence_ref(NULL, &screen->base.fence.current);
342 }
343 if (screen->base.pushbuf)
344 screen->base.pushbuf->user_priv = NULL;
345
346 if (screen->blitter)
347 nvc0_blitter_destroy(screen);
348 if (screen->pm.prog) {
349 screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
350 nvc0_program_destroy(NULL, screen->pm.prog);
351 }
352
353 nouveau_bo_ref(NULL, &screen->text);
354 nouveau_bo_ref(NULL, &screen->uniform_bo);
355 nouveau_bo_ref(NULL, &screen->tls);
356 nouveau_bo_ref(NULL, &screen->txc);
357 nouveau_bo_ref(NULL, &screen->fence.bo);
358 nouveau_bo_ref(NULL, &screen->poly_cache);
359 nouveau_bo_ref(NULL, &screen->parm);
360
361 nouveau_heap_destroy(&screen->lib_code);
362 nouveau_heap_destroy(&screen->text_heap);
363
364 FREE(screen->tic.entries);
365
366 nouveau_mm_destroy(screen->mm_VRAM_fe0);
367
368 nouveau_object_del(&screen->eng3d);
369 nouveau_object_del(&screen->eng2d);
370 nouveau_object_del(&screen->m2mf);
371 nouveau_object_del(&screen->compute);
372 nouveau_object_del(&screen->nvsw);
373
374 nouveau_screen_fini(&screen->base);
375
376 FREE(screen);
377 }
378
379 static int
380 nvc0_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos,
381 unsigned size, const uint32_t *data)
382 {
383 struct nouveau_pushbuf *push = screen->base.pushbuf;
384
385 size /= 4;
386
387 BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2);
388 PUSH_DATA (push, (m - 0x3800) / 8);
389 PUSH_DATA (push, pos);
390 BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1);
391 PUSH_DATA (push, pos);
392 PUSH_DATAp(push, data, size);
393
394 return pos + size;
395 }
396
397 static void
398 nvc0_magic_3d_init(struct nouveau_pushbuf *push, uint16_t obj_class)
399 {
400 BEGIN_NVC0(push, SUBC_3D(0x10cc), 1);
401 PUSH_DATA (push, 0xff);
402 BEGIN_NVC0(push, SUBC_3D(0x10e0), 2);
403 PUSH_DATA (push, 0xff);
404 PUSH_DATA (push, 0xff);
405 BEGIN_NVC0(push, SUBC_3D(0x10ec), 2);
406 PUSH_DATA (push, 0xff);
407 PUSH_DATA (push, 0xff);
408 BEGIN_NVC0(push, SUBC_3D(0x074c), 1);
409 PUSH_DATA (push, 0x3f);
410
411 BEGIN_NVC0(push, SUBC_3D(0x16a8), 1);
412 PUSH_DATA (push, (3 << 16) | 3);
413 BEGIN_NVC0(push, SUBC_3D(0x1794), 1);
414 PUSH_DATA (push, (2 << 16) | 2);
415 BEGIN_NVC0(push, SUBC_3D(0x0de8), 1);
416 PUSH_DATA (push, 1);
417
418 BEGIN_NVC0(push, SUBC_3D(0x12ac), 1);
419 PUSH_DATA (push, 0);
420 BEGIN_NVC0(push, SUBC_3D(0x0218), 1);
421 PUSH_DATA (push, 0x10);
422 BEGIN_NVC0(push, SUBC_3D(0x10fc), 1);
423 PUSH_DATA (push, 0x10);
424 BEGIN_NVC0(push, SUBC_3D(0x1290), 1);
425 PUSH_DATA (push, 0x10);
426 BEGIN_NVC0(push, SUBC_3D(0x12d8), 2);
427 PUSH_DATA (push, 0x10);
428 PUSH_DATA (push, 0x10);
429 BEGIN_NVC0(push, SUBC_3D(0x1140), 1);
430 PUSH_DATA (push, 0x10);
431 BEGIN_NVC0(push, SUBC_3D(0x1610), 1);
432 PUSH_DATA (push, 0xe);
433
434 BEGIN_NVC0(push, SUBC_3D(0x164c), 1);
435 PUSH_DATA (push, 1 << 12);
436 BEGIN_NVC0(push, SUBC_3D(0x030c), 1);
437 PUSH_DATA (push, 0);
438 BEGIN_NVC0(push, SUBC_3D(0x0300), 1);
439 PUSH_DATA (push, 3);
440
441 BEGIN_NVC0(push, SUBC_3D(0x02d0), 1);
442 PUSH_DATA (push, 0x3fffff);
443 BEGIN_NVC0(push, SUBC_3D(0x0fdc), 1);
444 PUSH_DATA (push, 1);
445 BEGIN_NVC0(push, SUBC_3D(0x19c0), 1);
446 PUSH_DATA (push, 1);
447 BEGIN_NVC0(push, SUBC_3D(0x075c), 1);
448 PUSH_DATA (push, 3);
449
450 if (obj_class >= NVE4_3D_CLASS) {
451 BEGIN_NVC0(push, SUBC_3D(0x07fc), 1);
452 PUSH_DATA (push, 1);
453 }
454
455 /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
456 * are supposed to do */
457 }
458
459 static void
460 nvc0_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
461 {
462 struct nvc0_screen *screen = nvc0_screen(pscreen);
463 struct nouveau_pushbuf *push = screen->base.pushbuf;
464
465 /* we need to do it after possible flush in MARK_RING */
466 *sequence = ++screen->base.fence.sequence;
467
468 BEGIN_NVC0(push, NVC0_3D(QUERY_ADDRESS_HIGH), 4);
469 PUSH_DATAh(push, screen->fence.bo->offset);
470 PUSH_DATA (push, screen->fence.bo->offset);
471 PUSH_DATA (push, *sequence);
472 PUSH_DATA (push, NVC0_3D_QUERY_GET_FENCE | NVC0_3D_QUERY_GET_SHORT |
473 (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT));
474 }
475
476 static u32
477 nvc0_screen_fence_update(struct pipe_screen *pscreen)
478 {
479 struct nvc0_screen *screen = nvc0_screen(pscreen);
480 return screen->fence.map[0];
481 }
482
483 static int
484 nvc0_screen_init_compute(struct nvc0_screen *screen)
485 {
486 screen->base.base.get_compute_param = nvc0_screen_get_compute_param;
487
488 switch (screen->base.device->chipset & ~0xf) {
489 case 0xc0:
490 case 0xd0:
491 /* Using COMPUTE has weird effects on 3D state, we need to
492 * investigate this further before enabling it by default.
493 */
494 if (debug_get_bool_option("NVC0_COMPUTE", FALSE))
495 return nvc0_screen_compute_setup(screen, screen->base.pushbuf);
496 return 0;
497 case 0xe0:
498 case 0xf0:
499 case 0x100:
500 return nve4_screen_compute_setup(screen, screen->base.pushbuf);
501 default:
502 return -1;
503 }
504 }
505
506 boolean
507 nvc0_screen_resize_tls_area(struct nvc0_screen *screen,
508 uint32_t lpos, uint32_t lneg, uint32_t cstack)
509 {
510 struct nouveau_bo *bo = NULL;
511 int ret;
512 uint64_t size = (lpos + lneg) * 32 + cstack;
513
514 if (size >= (1 << 20)) {
515 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64"\n", size);
516 return FALSE;
517 }
518
519 size *= (screen->base.device->chipset >= 0xe0) ? 64 : 48; /* max warps */
520 size = align(size, 0x8000);
521 size *= screen->mp_count;
522
523 size = align(size, 1 << 17);
524
525 ret = nouveau_bo_new(screen->base.device, NOUVEAU_BO_VRAM, 1 << 17, size,
526 NULL, &bo);
527 if (ret) {
528 NOUVEAU_ERR("failed to allocate TLS area, size: 0x%"PRIx64"\n", size);
529 return FALSE;
530 }
531 nouveau_bo_ref(NULL, &screen->tls);
532 screen->tls = bo;
533 return TRUE;
534 }
535
536 #define FAIL_SCREEN_INIT(str, err) \
537 do { \
538 NOUVEAU_ERR(str, err); \
539 nvc0_screen_destroy(pscreen); \
540 return NULL; \
541 } while(0)
542
543 struct pipe_screen *
544 nvc0_screen_create(struct nouveau_device *dev)
545 {
546 struct nvc0_screen *screen;
547 struct pipe_screen *pscreen;
548 struct nouveau_object *chan;
549 struct nouveau_pushbuf *push;
550 uint64_t value;
551 uint32_t obj_class;
552 int ret;
553 unsigned i;
554 union nouveau_bo_config mm_config;
555
556 switch (dev->chipset & ~0xf) {
557 case 0xc0:
558 case 0xd0:
559 case 0xe0:
560 case 0xf0:
561 case 0x100:
562 break;
563 default:
564 return NULL;
565 }
566
567 screen = CALLOC_STRUCT(nvc0_screen);
568 if (!screen)
569 return NULL;
570 pscreen = &screen->base.base;
571
572 ret = nouveau_screen_init(&screen->base, dev);
573 if (ret) {
574 nvc0_screen_destroy(pscreen);
575 return NULL;
576 }
577 chan = screen->base.channel;
578 push = screen->base.pushbuf;
579 push->user_priv = screen;
580 push->rsvd_kick = 5;
581
582 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
583 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
584 screen->base.sysmem_bindings |=
585 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
586
587 pscreen->destroy = nvc0_screen_destroy;
588 pscreen->context_create = nvc0_create;
589 pscreen->is_format_supported = nvc0_screen_is_format_supported;
590 pscreen->get_param = nvc0_screen_get_param;
591 pscreen->get_shader_param = nvc0_screen_get_shader_param;
592 pscreen->get_paramf = nvc0_screen_get_paramf;
593 pscreen->get_driver_query_info = nvc0_screen_get_driver_query_info;
594
595 nvc0_screen_init_resource_functions(pscreen);
596
597 screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
598 screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
599
600 ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096, NULL,
601 &screen->fence.bo);
602 if (ret)
603 goto fail;
604 nouveau_bo_map(screen->fence.bo, 0, NULL);
605 screen->fence.map = screen->fence.bo->map;
606 screen->base.fence.emit = nvc0_screen_fence_emit;
607 screen->base.fence.update = nvc0_screen_fence_update;
608
609
610 ret = nouveau_object_new(chan,
611 (dev->chipset < 0xe0) ? 0x1f906e : 0x906e, 0x906e,
612 NULL, 0, &screen->nvsw);
613 if (ret)
614 FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret);
615
616
617 switch (dev->chipset & ~0xf) {
618 case 0x100:
619 case 0xf0:
620 obj_class = NVF0_P2MF_CLASS;
621 break;
622 case 0xe0:
623 obj_class = NVE4_P2MF_CLASS;
624 break;
625 default:
626 obj_class = NVC0_M2MF_CLASS;
627 break;
628 }
629 ret = nouveau_object_new(chan, 0xbeef323f, obj_class, NULL, 0,
630 &screen->m2mf);
631 if (ret)
632 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
633
634 BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
635 PUSH_DATA (push, screen->m2mf->oclass);
636 if (screen->m2mf->oclass == NVE4_P2MF_CLASS) {
637 BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
638 PUSH_DATA (push, 0xa0b5);
639 }
640
641 ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
642 &screen->eng2d);
643 if (ret)
644 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
645
646 BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
647 PUSH_DATA (push, screen->eng2d->oclass);
648 BEGIN_NVC0(push, NVC0_2D(SINGLE_GPC), 1);
649 PUSH_DATA (push, 0);
650 BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
651 PUSH_DATA (push, NVC0_2D_OPERATION_SRCCOPY);
652 BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
653 PUSH_DATA (push, 0);
654 BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
655 PUSH_DATA (push, 0);
656 BEGIN_NVC0(push, SUBC_2D(0x0884), 1);
657 PUSH_DATA (push, 0x3f);
658 BEGIN_NVC0(push, SUBC_2D(0x0888), 1);
659 PUSH_DATA (push, 1);
660
661 BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
662 PUSH_DATAh(push, screen->fence.bo->offset + 16);
663 PUSH_DATA (push, screen->fence.bo->offset + 16);
664
665 switch (dev->chipset & ~0xf) {
666 case 0x100:
667 case 0xf0:
668 obj_class = NVF0_3D_CLASS;
669 break;
670 case 0xe0:
671 obj_class = NVE4_3D_CLASS;
672 break;
673 case 0xd0:
674 obj_class = NVC8_3D_CLASS;
675 break;
676 case 0xc0:
677 default:
678 switch (dev->chipset) {
679 case 0xc8:
680 obj_class = NVC8_3D_CLASS;
681 break;
682 case 0xc1:
683 obj_class = NVC1_3D_CLASS;
684 break;
685 default:
686 obj_class = NVC0_3D_CLASS;
687 break;
688 }
689 break;
690 }
691 ret = nouveau_object_new(chan, 0xbeef003d, obj_class, NULL, 0,
692 &screen->eng3d);
693 if (ret)
694 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
695 screen->base.class_3d = obj_class;
696
697 BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
698 PUSH_DATA (push, screen->eng3d->oclass);
699
700 BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
701 PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
702
703 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", TRUE)) {
704 /* kill shaders after about 1 second (at 100 MHz) */
705 BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
706 PUSH_DATA (push, 0x17);
707 }
708
709 IMMED_NVC0(push, NVC0_3D(ZETA_COMP_ENABLE), dev->drm_version >= 0x01000101);
710 BEGIN_NVC0(push, NVC0_3D(RT_COMP_ENABLE(0)), 8);
711 for (i = 0; i < 8; ++i)
712 PUSH_DATA(push, dev->drm_version >= 0x01000101);
713
714 BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
715 PUSH_DATA (push, 1);
716
717 BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
718 PUSH_DATA (push, 0);
719 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
720 PUSH_DATA (push, 0);
721 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
722 PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
723 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
724 PUSH_DATA (push, 0);
725 BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
726 PUSH_DATA (push, 1);
727 BEGIN_NVC0(push, NVC0_3D(LINE_LAST_PIXEL), 1);
728 PUSH_DATA (push, 0);
729 BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
730 PUSH_DATA (push, 1);
731 BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
732 PUSH_DATA (push, 0);
733 if (screen->eng3d->oclass < NVE4_3D_CLASS) {
734 BEGIN_NVC0(push, NVC0_3D(TEX_MISC), 1);
735 PUSH_DATA (push, NVC0_3D_TEX_MISC_SEAMLESS_CUBE_MAP);
736 } else {
737 BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
738 PUSH_DATA (push, 15);
739 }
740 BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
741 PUSH_DATA (push, 8); /* 128 */
742 BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
743 PUSH_DATA (push, 1);
744 if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
745 BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
746 PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
747 }
748
749 nvc0_magic_3d_init(push, screen->eng3d->oclass);
750
751 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 20, NULL,
752 &screen->text);
753 if (ret)
754 goto fail;
755
756 /* XXX: getting a page fault at the end of the code buffer every few
757 * launches, don't use the last 256 bytes to work around them - prefetch ?
758 */
759 nouveau_heap_init(&screen->text_heap, 0, (1 << 20) - 0x100);
760
761 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 12, 6 << 16, NULL,
762 &screen->uniform_bo);
763 if (ret)
764 goto fail;
765
766 PUSH_REFN (push, screen->uniform_bo, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
767
768 for (i = 0; i < 5; ++i) {
769 /* TIC and TSC entries for each unit (nve4+ only) */
770 /* auxiliary constants (6 user clip planes, base instance id) */
771 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
772 PUSH_DATA (push, 512);
773 PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (i << 9));
774 PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (i << 9));
775 BEGIN_NVC0(push, NVC0_3D(CB_BIND(i)), 1);
776 PUSH_DATA (push, (15 << 4) | 1);
777 if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
778 unsigned j;
779 BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
780 PUSH_DATA (push, 0);
781 for (j = 0; j < 8; ++j)
782 PUSH_DATA(push, j);
783 } else {
784 BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
785 PUSH_DATA (push, 0x54);
786 }
787 }
788 BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
789 PUSH_DATA (push, 0);
790
791 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
792 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
793 PUSH_DATA (push, 256);
794 PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
795 PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
796 BEGIN_1IC0(push, NVC0_3D(CB_POS), 5);
797 PUSH_DATA (push, 0);
798 PUSH_DATAf(push, 0.0f);
799 PUSH_DATAf(push, 0.0f);
800 PUSH_DATAf(push, 0.0f);
801 PUSH_DATAf(push, 0.0f);
802 BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
803 PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
804 PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
805
806 if (dev->drm_version >= 0x01000101) {
807 ret = nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
808 if (ret) {
809 NOUVEAU_ERR("NOUVEAU_GETPARAM_GRAPH_UNITS failed.\n");
810 goto fail;
811 }
812 } else {
813 if (dev->chipset >= 0xe0 && dev->chipset < 0xf0)
814 value = (8 << 8) | 4;
815 else
816 value = (16 << 8) | 4;
817 }
818 screen->mp_count = value >> 8;
819 screen->mp_count_compute = screen->mp_count;
820
821 nvc0_screen_resize_tls_area(screen, 128 * 16, 0, 0x200);
822
823 BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
824 PUSH_DATAh(push, screen->text->offset);
825 PUSH_DATA (push, screen->text->offset);
826 BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
827 PUSH_DATAh(push, screen->tls->offset);
828 PUSH_DATA (push, screen->tls->offset);
829 PUSH_DATA (push, screen->tls->size >> 32);
830 PUSH_DATA (push, screen->tls->size);
831 BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
832 PUSH_DATA (push, 0);
833 BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
834 PUSH_DATA (push, 0);
835
836 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 20, NULL,
837 &screen->poly_cache);
838 if (ret)
839 goto fail;
840
841 BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
842 PUSH_DATAh(push, screen->poly_cache->offset);
843 PUSH_DATA (push, screen->poly_cache->offset);
844 PUSH_DATA (push, 3);
845
846 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 17, NULL,
847 &screen->txc);
848 if (ret)
849 goto fail;
850
851 BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
852 PUSH_DATAh(push, screen->txc->offset);
853 PUSH_DATA (push, screen->txc->offset);
854 PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
855
856 BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
857 PUSH_DATAh(push, screen->txc->offset + 65536);
858 PUSH_DATA (push, screen->txc->offset + 65536);
859 PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
860
861 BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
862 PUSH_DATA (push, 0);
863 BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
864 PUSH_DATA (push, 0);
865 PUSH_DATA (push, 0);
866 BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
867 PUSH_DATA (push, 0x3f);
868
869 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
870 PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
871 BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
872 for (i = 0; i < 8 * 2; ++i)
873 PUSH_DATA(push, 0);
874 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
875 PUSH_DATA (push, 0);
876 BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
877 PUSH_DATA (push, 0);
878
879 /* neither scissors, viewport nor stencil mask should affect clears */
880 BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
881 PUSH_DATA (push, 0);
882
883 BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
884 PUSH_DATA (push, 1);
885 BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(0)), 2);
886 PUSH_DATAf(push, 0.0f);
887 PUSH_DATAf(push, 1.0f);
888 BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
889 PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
890
891 /* We use scissors instead of exact view volume clipping,
892 * so they're always enabled.
893 */
894 BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(0)), 3);
895 PUSH_DATA (push, 1);
896 PUSH_DATA (push, 8192 << 16);
897 PUSH_DATA (push, 8192 << 16);
898
899 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
900
901 i = 0;
902 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, nvc0_9097_per_instance_bf);
903 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, nvc0_9097_blend_enables);
904 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, nvc0_9097_vertex_array_select);
905 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, nvc0_9097_tep_select);
906 MK_MACRO(NVC0_3D_MACRO_GP_SELECT, nvc0_9097_gp_select);
907 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, nvc0_9097_poly_mode_front);
908 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, nvc0_9097_poly_mode_back);
909
910 BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
911 PUSH_DATA (push, 1);
912 BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
913 PUSH_DATA (push, 1);
914 BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
915 PUSH_DATA (push, 0x40);
916 BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
917 PUSH_DATA (push, 0);
918 BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
919 PUSH_DATA (push, 0x30);
920 BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
921 PUSH_DATA (push, 3);
922 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
923 PUSH_DATA (push, 0x20);
924 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
925 PUSH_DATA (push, 0x00);
926
927 BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
928 PUSH_DATA (push, 0);
929 BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
930 PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
931
932 IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
933
934 if (nvc0_screen_init_compute(screen))
935 goto fail;
936
937 PUSH_KICK (push);
938
939 screen->tic.entries = CALLOC(4096, sizeof(void *));
940 screen->tsc.entries = screen->tic.entries + 2048;
941
942 mm_config.nvc0.tile_mode = 0;
943 mm_config.nvc0.memtype = 0xfe0;
944 screen->mm_VRAM_fe0 = nouveau_mm_create(dev, NOUVEAU_BO_VRAM, &mm_config);
945
946 if (!nvc0_blitter_create(screen))
947 goto fail;
948
949 nouveau_fence_new(&screen->base, &screen->base.fence.current, FALSE);
950
951 return pscreen;
952
953 fail:
954 nvc0_screen_destroy(pscreen);
955 return NULL;
956 }
957
958 int
959 nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
960 {
961 int i = screen->tic.next;
962
963 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
964 i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
965
966 screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
967
968 if (screen->tic.entries[i])
969 nv50_tic_entry(screen->tic.entries[i])->id = -1;
970
971 screen->tic.entries[i] = entry;
972 return i;
973 }
974
975 int
976 nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
977 {
978 int i = screen->tsc.next;
979
980 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
981 i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
982
983 screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
984
985 if (screen->tsc.entries[i])
986 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
987
988 screen->tsc.entries[i] = entry;
989 return i;
990 }