nvc0: delete compute object on screen destruction
[mesa.git] / src / gallium / drivers / 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/nouveau_vp3_video.h"
31
32 #include "nvc0_context.h"
33 #include "nvc0_screen.h"
34
35 #include "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_COMBINED_SAMPLERS:
75 return 16 * 5;
76 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
77 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
78 return 15;
79 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
80 return (class_3d >= NVE4_3D_CLASS) ? 13 : 12;
81 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
82 return 2048;
83 case PIPE_CAP_MIN_TEXEL_OFFSET:
84 return -8;
85 case PIPE_CAP_MAX_TEXEL_OFFSET:
86 return 7;
87 case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
88 case PIPE_CAP_TEXTURE_SWIZZLE:
89 case PIPE_CAP_TEXTURE_SHADOW_MAP:
90 case PIPE_CAP_NPOT_TEXTURES:
91 case PIPE_CAP_ANISOTROPIC_FILTER:
92 case PIPE_CAP_SEAMLESS_CUBE_MAP:
93 case PIPE_CAP_CUBE_MAP_ARRAY:
94 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
95 case PIPE_CAP_TEXTURE_MULTISAMPLE:
96 return 1;
97 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
98 return 65536;
99 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
100 return (class_3d >= NVE4_3D_CLASS) ? 1 : 0;
101 case PIPE_CAP_TWO_SIDED_STENCIL:
102 case PIPE_CAP_DEPTH_CLIP_DISABLE:
103 case PIPE_CAP_POINT_SPRITE:
104 case PIPE_CAP_TGSI_TEXCOORD:
105 return 1;
106 case PIPE_CAP_SM3:
107 return 1;
108 case PIPE_CAP_GLSL_FEATURE_LEVEL:
109 return 150;
110 case PIPE_CAP_MAX_RENDER_TARGETS:
111 return 8;
112 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
113 return 1;
114 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
115 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
116 case PIPE_CAP_VERTEX_COLOR_CLAMPED:
117 return 1;
118 case PIPE_CAP_QUERY_TIMESTAMP:
119 case PIPE_CAP_QUERY_TIME_ELAPSED:
120 case PIPE_CAP_OCCLUSION_QUERY:
121 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
122 case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
123 return 1;
124 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
125 return 4;
126 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
127 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
128 return 128;
129 case PIPE_CAP_BLEND_EQUATION_SEPARATE:
130 case PIPE_CAP_INDEP_BLEND_ENABLE:
131 case PIPE_CAP_INDEP_BLEND_FUNC:
132 return 1;
133 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
134 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
135 return 1;
136 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
137 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
138 return 0;
139 case PIPE_CAP_SHADER_STENCIL_EXPORT:
140 return 0;
141 case PIPE_CAP_PRIMITIVE_RESTART:
142 case PIPE_CAP_TGSI_INSTANCEID:
143 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
144 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
145 case PIPE_CAP_CONDITIONAL_RENDER:
146 case PIPE_CAP_TEXTURE_BARRIER:
147 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
148 case PIPE_CAP_START_INSTANCE:
149 return 1;
150 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
151 return 0; /* state trackers will know better */
152 case PIPE_CAP_USER_CONSTANT_BUFFERS:
153 case PIPE_CAP_USER_INDEX_BUFFERS:
154 case PIPE_CAP_USER_VERTEX_BUFFERS:
155 return 1;
156 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
157 return 256;
158 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
159 return 1; /* 256 for binding as RT, but that's not possible in GL */
160 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
161 return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
162 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
163 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
164 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
165 return 0;
166 case PIPE_CAP_COMPUTE:
167 return (class_3d >= NVE4_3D_CLASS) ? 1 : 0;
168 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
169 return 1;
170 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
171 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
172 case PIPE_CAP_ENDIANNESS:
173 return PIPE_ENDIAN_LITTLE;
174 default:
175 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
176 return 0;
177 }
178 }
179
180 static int
181 nvc0_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
182 enum pipe_shader_cap param)
183 {
184 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
185
186 switch (shader) {
187 case PIPE_SHADER_VERTEX:
188 /*
189 case PIPE_SHADER_TESSELLATION_CONTROL:
190 case PIPE_SHADER_TESSELLATION_EVALUATION:
191 */
192 case PIPE_SHADER_GEOMETRY:
193 case PIPE_SHADER_FRAGMENT:
194 break;
195 case PIPE_SHADER_COMPUTE:
196 if (class_3d < NVE4_3D_CLASS)
197 return 0;
198 break;
199 default:
200 return 0;
201 }
202
203 switch (param) {
204 case PIPE_SHADER_CAP_PREFERRED_IR:
205 return PIPE_SHADER_IR_TGSI;
206 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
207 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
208 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
209 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
210 return 16384;
211 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
212 return 16;
213 case PIPE_SHADER_CAP_MAX_INPUTS:
214 if (shader == PIPE_SHADER_VERTEX)
215 return 32;
216 /* NOTE: These only count our slots for GENERIC varyings.
217 * The address space may be larger, but the actual hard limit seems to be
218 * less than what the address space layout permits, so don't add TEXCOORD,
219 * COLOR, etc. here.
220 */
221 if (shader == PIPE_SHADER_FRAGMENT)
222 return 0x1f0 / 16;
223 /* Actually this counts CLIPVERTEX, which occupies the last generic slot,
224 * and excludes 0x60 per-patch inputs.
225 */
226 return 0x200 / 16;
227 case PIPE_SHADER_CAP_MAX_CONSTS:
228 return 65536 / 16;
229 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
230 if (shader == PIPE_SHADER_COMPUTE && class_3d >= NVE4_3D_CLASS)
231 return NVE4_MAX_PIPE_CONSTBUFS_COMPUTE;
232 return NVC0_MAX_PIPE_CONSTBUFS;
233 case PIPE_SHADER_CAP_MAX_ADDRS:
234 return 1;
235 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
236 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
237 return shader != PIPE_SHADER_FRAGMENT;
238 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
239 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
240 return 1;
241 case PIPE_SHADER_CAP_MAX_PREDS:
242 return 0;
243 case PIPE_SHADER_CAP_MAX_TEMPS:
244 return NVC0_CAP_MAX_PROGRAM_TEMPS;
245 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
246 return 1;
247 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
248 return 0;
249 case PIPE_SHADER_CAP_SUBROUTINES:
250 return 1;
251 case PIPE_SHADER_CAP_INTEGERS:
252 return 1;
253 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
254 return 16; /* would be 32 in linked (OpenGL-style) mode */
255 /*
256 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLER_VIEWS:
257 return 32;
258 */
259 default:
260 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
261 return 0;
262 }
263 }
264
265 static float
266 nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
267 {
268 switch (param) {
269 case PIPE_CAPF_MAX_LINE_WIDTH:
270 case PIPE_CAPF_MAX_LINE_WIDTH_AA:
271 return 10.0f;
272 case PIPE_CAPF_MAX_POINT_WIDTH:
273 return 63.0f;
274 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
275 return 63.375f;
276 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
277 return 16.0f;
278 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
279 return 15.0f;
280 default:
281 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
282 return 0.0f;
283 }
284 }
285
286 static int
287 nvc0_screen_get_compute_param(struct pipe_screen *pscreen,
288 enum pipe_compute_cap param, void *data)
289 {
290 uint64_t *data64 = (uint64_t *)data;
291 const uint16_t obj_class = nvc0_screen(pscreen)->compute->oclass;
292
293 switch (param) {
294 case PIPE_COMPUTE_CAP_GRID_DIMENSION:
295 data64[0] = 3;
296 return 8;
297 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
298 data64[0] = (obj_class >= NVE4_COMPUTE_CLASS) ? 0x7fffffff : 65535;
299 data64[1] = 65535;
300 data64[2] = 65535;
301 return 24;
302 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
303 data64[0] = 1024;
304 data64[1] = 1024;
305 data64[2] = 64;
306 return 24;
307 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
308 data64[0] = 1024;
309 return 8;
310 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */
311 data64[0] = (uint64_t)1 << 40;
312 return 8;
313 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
314 data64[0] = 48 << 10;
315 return 8;
316 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
317 data64[0] = 512 << 10;
318 return 8;
319 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
320 data64[0] = 4096;
321 return 8;
322 default:
323 return 0;
324 }
325 }
326
327 static void
328 nvc0_screen_destroy(struct pipe_screen *pscreen)
329 {
330 struct nvc0_screen *screen = nvc0_screen(pscreen);
331
332 if (screen->base.fence.current) {
333 nouveau_fence_wait(screen->base.fence.current);
334 nouveau_fence_ref(NULL, &screen->base.fence.current);
335 }
336 if (screen->base.pushbuf)
337 screen->base.pushbuf->user_priv = NULL;
338
339 if (screen->blitter)
340 nvc0_blitter_destroy(screen);
341 if (screen->pm.prog) {
342 screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
343 nvc0_program_destroy(NULL, screen->pm.prog);
344 }
345
346 nouveau_bo_ref(NULL, &screen->text);
347 nouveau_bo_ref(NULL, &screen->uniform_bo);
348 nouveau_bo_ref(NULL, &screen->tls);
349 nouveau_bo_ref(NULL, &screen->txc);
350 nouveau_bo_ref(NULL, &screen->fence.bo);
351 nouveau_bo_ref(NULL, &screen->poly_cache);
352 nouveau_bo_ref(NULL, &screen->parm);
353
354 nouveau_heap_destroy(&screen->lib_code);
355 nouveau_heap_destroy(&screen->text_heap);
356
357 FREE(screen->tic.entries);
358
359 nouveau_mm_destroy(screen->mm_VRAM_fe0);
360
361 nouveau_object_del(&screen->eng3d);
362 nouveau_object_del(&screen->eng2d);
363 nouveau_object_del(&screen->m2mf);
364 nouveau_object_del(&screen->compute);
365
366 nouveau_screen_fini(&screen->base);
367
368 FREE(screen);
369 }
370
371 static int
372 nvc0_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos,
373 unsigned size, const uint32_t *data)
374 {
375 struct nouveau_pushbuf *push = screen->base.pushbuf;
376
377 size /= 4;
378
379 BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2);
380 PUSH_DATA (push, (m - 0x3800) / 8);
381 PUSH_DATA (push, pos);
382 BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1);
383 PUSH_DATA (push, pos);
384 PUSH_DATAp(push, data, size);
385
386 return pos + size;
387 }
388
389 static void
390 nvc0_magic_3d_init(struct nouveau_pushbuf *push, uint16_t obj_class)
391 {
392 BEGIN_NVC0(push, SUBC_3D(0x10cc), 1);
393 PUSH_DATA (push, 0xff);
394 BEGIN_NVC0(push, SUBC_3D(0x10e0), 2);
395 PUSH_DATA (push, 0xff);
396 PUSH_DATA (push, 0xff);
397 BEGIN_NVC0(push, SUBC_3D(0x10ec), 2);
398 PUSH_DATA (push, 0xff);
399 PUSH_DATA (push, 0xff);
400 BEGIN_NVC0(push, SUBC_3D(0x074c), 1);
401 PUSH_DATA (push, 0x3f);
402
403 BEGIN_NVC0(push, SUBC_3D(0x16a8), 1);
404 PUSH_DATA (push, (3 << 16) | 3);
405 BEGIN_NVC0(push, SUBC_3D(0x1794), 1);
406 PUSH_DATA (push, (2 << 16) | 2);
407 BEGIN_NVC0(push, SUBC_3D(0x0de8), 1);
408 PUSH_DATA (push, 1);
409
410 BEGIN_NVC0(push, SUBC_3D(0x12ac), 1);
411 PUSH_DATA (push, 0);
412 BEGIN_NVC0(push, SUBC_3D(0x0218), 1);
413 PUSH_DATA (push, 0x10);
414 BEGIN_NVC0(push, SUBC_3D(0x10fc), 1);
415 PUSH_DATA (push, 0x10);
416 BEGIN_NVC0(push, SUBC_3D(0x1290), 1);
417 PUSH_DATA (push, 0x10);
418 BEGIN_NVC0(push, SUBC_3D(0x12d8), 2);
419 PUSH_DATA (push, 0x10);
420 PUSH_DATA (push, 0x10);
421 BEGIN_NVC0(push, SUBC_3D(0x1140), 1);
422 PUSH_DATA (push, 0x10);
423 BEGIN_NVC0(push, SUBC_3D(0x1610), 1);
424 PUSH_DATA (push, 0xe);
425
426 BEGIN_NVC0(push, SUBC_3D(0x164c), 1);
427 PUSH_DATA (push, 1 << 12);
428 BEGIN_NVC0(push, SUBC_3D(0x030c), 1);
429 PUSH_DATA (push, 0);
430 BEGIN_NVC0(push, SUBC_3D(0x0300), 1);
431 PUSH_DATA (push, 3);
432
433 BEGIN_NVC0(push, SUBC_3D(0x02d0), 1);
434 PUSH_DATA (push, 0x3fffff);
435 BEGIN_NVC0(push, SUBC_3D(0x0fdc), 1);
436 PUSH_DATA (push, 1);
437 BEGIN_NVC0(push, SUBC_3D(0x19c0), 1);
438 PUSH_DATA (push, 1);
439 BEGIN_NVC0(push, SUBC_3D(0x075c), 1);
440 PUSH_DATA (push, 3);
441
442 if (obj_class >= NVE4_3D_CLASS) {
443 BEGIN_NVC0(push, SUBC_3D(0x07fc), 1);
444 PUSH_DATA (push, 1);
445 }
446
447 /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
448 * are supposed to do */
449 }
450
451 static void
452 nvc0_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
453 {
454 struct nvc0_screen *screen = nvc0_screen(pscreen);
455 struct nouveau_pushbuf *push = screen->base.pushbuf;
456
457 /* we need to do it after possible flush in MARK_RING */
458 *sequence = ++screen->base.fence.sequence;
459
460 BEGIN_NVC0(push, NVC0_3D(QUERY_ADDRESS_HIGH), 4);
461 PUSH_DATAh(push, screen->fence.bo->offset);
462 PUSH_DATA (push, screen->fence.bo->offset);
463 PUSH_DATA (push, *sequence);
464 PUSH_DATA (push, NVC0_3D_QUERY_GET_FENCE | NVC0_3D_QUERY_GET_SHORT |
465 (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT));
466 }
467
468 static u32
469 nvc0_screen_fence_update(struct pipe_screen *pscreen)
470 {
471 struct nvc0_screen *screen = nvc0_screen(pscreen);
472 return screen->fence.map[0];
473 }
474
475 static int
476 nvc0_screen_init_compute(struct nvc0_screen *screen)
477 {
478 screen->base.base.get_compute_param = nvc0_screen_get_compute_param;
479
480 switch (screen->base.device->chipset & 0xf0) {
481 case 0xc0:
482 case 0xd0:
483 /* Using COMPUTE has weird effects on 3D state, we need to
484 * investigate this further before enabling it by default.
485 */
486 if (debug_get_bool_option("NVC0_COMPUTE", FALSE))
487 return nvc0_screen_compute_setup(screen, screen->base.pushbuf);
488 return 0;
489 case 0xe0:
490 case 0xf0:
491 return nve4_screen_compute_setup(screen, screen->base.pushbuf);
492 default:
493 return -1;
494 }
495 }
496
497 boolean
498 nvc0_screen_resize_tls_area(struct nvc0_screen *screen,
499 uint32_t lpos, uint32_t lneg, uint32_t cstack)
500 {
501 struct nouveau_bo *bo = NULL;
502 int ret;
503 uint64_t size = (lpos + lneg) * 32 + cstack;
504
505 if (size >= (1 << 20)) {
506 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64"\n", size);
507 return FALSE;
508 }
509
510 size *= (screen->base.device->chipset >= 0xe0) ? 64 : 48; /* max warps */
511 size = align(size, 0x8000);
512 size *= screen->mp_count;
513
514 size = align(size, 1 << 17);
515
516 ret = nouveau_bo_new(screen->base.device, NOUVEAU_BO_VRAM, 1 << 17, size,
517 NULL, &bo);
518 if (ret) {
519 NOUVEAU_ERR("failed to allocate TLS area, size: 0x%"PRIx64"\n", size);
520 return FALSE;
521 }
522 nouveau_bo_ref(NULL, &screen->tls);
523 screen->tls = bo;
524 return TRUE;
525 }
526
527 #define FAIL_SCREEN_INIT(str, err) \
528 do { \
529 NOUVEAU_ERR(str, err); \
530 nvc0_screen_destroy(pscreen); \
531 return NULL; \
532 } while(0)
533
534 struct pipe_screen *
535 nvc0_screen_create(struct nouveau_device *dev)
536 {
537 struct nvc0_screen *screen;
538 struct pipe_screen *pscreen;
539 struct nouveau_object *chan;
540 struct nouveau_pushbuf *push;
541 uint64_t value;
542 uint32_t obj_class;
543 int ret;
544 unsigned i;
545 union nouveau_bo_config mm_config;
546
547 switch (dev->chipset & ~0xf) {
548 case 0xc0:
549 case 0xd0:
550 case 0xe0:
551 case 0xf0:
552 break;
553 default:
554 return NULL;
555 }
556
557 screen = CALLOC_STRUCT(nvc0_screen);
558 if (!screen)
559 return NULL;
560 pscreen = &screen->base.base;
561
562 ret = nouveau_screen_init(&screen->base, dev);
563 if (ret) {
564 nvc0_screen_destroy(pscreen);
565 return NULL;
566 }
567 chan = screen->base.channel;
568 push = screen->base.pushbuf;
569 push->user_priv = screen;
570 push->rsvd_kick = 5;
571
572 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
573 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
574 screen->base.sysmem_bindings |=
575 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
576
577 pscreen->destroy = nvc0_screen_destroy;
578 pscreen->context_create = nvc0_create;
579 pscreen->is_format_supported = nvc0_screen_is_format_supported;
580 pscreen->get_param = nvc0_screen_get_param;
581 pscreen->get_shader_param = nvc0_screen_get_shader_param;
582 pscreen->get_paramf = nvc0_screen_get_paramf;
583 pscreen->get_driver_query_info = nvc0_screen_get_driver_query_info;
584
585 nvc0_screen_init_resource_functions(pscreen);
586
587 screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
588 screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
589
590 ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096, NULL,
591 &screen->fence.bo);
592 if (ret)
593 goto fail;
594 nouveau_bo_map(screen->fence.bo, 0, NULL);
595 screen->fence.map = screen->fence.bo->map;
596 screen->base.fence.emit = nvc0_screen_fence_emit;
597 screen->base.fence.update = nvc0_screen_fence_update;
598
599 switch (dev->chipset & 0xf0) {
600 case 0xf0:
601 obj_class = NVF0_P2MF_CLASS;
602 break;
603 case 0xe0:
604 obj_class = NVE4_P2MF_CLASS;
605 break;
606 default:
607 obj_class = NVC0_M2MF_CLASS;
608 break;
609 }
610 ret = nouveau_object_new(chan, 0xbeef323f, obj_class, NULL, 0,
611 &screen->m2mf);
612 if (ret)
613 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
614
615 BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
616 PUSH_DATA (push, screen->m2mf->oclass);
617 if (screen->m2mf->oclass == NVE4_P2MF_CLASS) {
618 BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
619 PUSH_DATA (push, 0xa0b5);
620 }
621
622 ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
623 &screen->eng2d);
624 if (ret)
625 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
626
627 BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
628 PUSH_DATA (push, screen->eng2d->oclass);
629 BEGIN_NVC0(push, NVC0_2D(SINGLE_GPC), 1);
630 PUSH_DATA (push, 0);
631 BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
632 PUSH_DATA (push, NVC0_2D_OPERATION_SRCCOPY);
633 BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
634 PUSH_DATA (push, 0);
635 BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
636 PUSH_DATA (push, 0);
637 BEGIN_NVC0(push, SUBC_2D(0x0884), 1);
638 PUSH_DATA (push, 0x3f);
639 BEGIN_NVC0(push, SUBC_2D(0x0888), 1);
640 PUSH_DATA (push, 1);
641
642 BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
643 PUSH_DATAh(push, screen->fence.bo->offset + 16);
644 PUSH_DATA (push, screen->fence.bo->offset + 16);
645
646 switch (dev->chipset & 0xf0) {
647 case 0xf0:
648 obj_class = NVF0_3D_CLASS;
649 break;
650 case 0xe0:
651 obj_class = NVE4_3D_CLASS;
652 break;
653 case 0xd0:
654 case 0xc0:
655 default:
656 switch (dev->chipset) {
657 case 0xd9:
658 case 0xc8:
659 obj_class = NVC8_3D_CLASS;
660 break;
661 case 0xc1:
662 obj_class = NVC1_3D_CLASS;
663 break;
664 default:
665 obj_class = NVC0_3D_CLASS;
666 break;
667 }
668 break;
669 }
670 ret = nouveau_object_new(chan, 0xbeef003d, obj_class, NULL, 0,
671 &screen->eng3d);
672 if (ret)
673 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
674 screen->base.class_3d = obj_class;
675
676 BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
677 PUSH_DATA (push, screen->eng3d->oclass);
678
679 BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
680 PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
681
682 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", TRUE)) {
683 /* kill shaders after about 1 second (at 100 MHz) */
684 BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
685 PUSH_DATA (push, 0x17);
686 }
687
688 IMMED_NVC0(push, NVC0_3D(ZETA_COMP_ENABLE), dev->drm_version >= 0x01000101);
689 BEGIN_NVC0(push, NVC0_3D(RT_COMP_ENABLE(0)), 8);
690 for (i = 0; i < 8; ++i)
691 PUSH_DATA(push, dev->drm_version >= 0x01000101);
692
693 BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
694 PUSH_DATA (push, 1);
695
696 BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
697 PUSH_DATA (push, 0);
698 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
699 PUSH_DATA (push, 0);
700 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
701 PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
702 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
703 PUSH_DATA (push, 0);
704 BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
705 PUSH_DATA (push, 1);
706 BEGIN_NVC0(push, NVC0_3D(LINE_LAST_PIXEL), 1);
707 PUSH_DATA (push, 0);
708 BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
709 PUSH_DATA (push, 1);
710 BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
711 PUSH_DATA (push, 0);
712 if (screen->eng3d->oclass < NVE4_3D_CLASS) {
713 BEGIN_NVC0(push, NVC0_3D(TEX_MISC), 1);
714 PUSH_DATA (push, NVC0_3D_TEX_MISC_SEAMLESS_CUBE_MAP);
715 } else {
716 BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
717 PUSH_DATA (push, 15);
718 }
719 BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
720 PUSH_DATA (push, 8); /* 128 */
721 BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
722 PUSH_DATA (push, 1);
723 if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
724 BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
725 PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
726 }
727
728 nvc0_magic_3d_init(push, screen->eng3d->oclass);
729
730 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 20, NULL,
731 &screen->text);
732 if (ret)
733 goto fail;
734
735 /* XXX: getting a page fault at the end of the code buffer every few
736 * launches, don't use the last 256 bytes to work around them - prefetch ?
737 */
738 nouveau_heap_init(&screen->text_heap, 0, (1 << 20) - 0x100);
739
740 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 12, 6 << 16, NULL,
741 &screen->uniform_bo);
742 if (ret)
743 goto fail;
744
745 for (i = 0; i < 5; ++i) {
746 /* TIC and TSC entries for each unit (nve4+ only) */
747 /* auxiliary constants (6 user clip planes, base instance id) */
748 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
749 PUSH_DATA (push, 512);
750 PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (i << 9));
751 PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (i << 9));
752 BEGIN_NVC0(push, NVC0_3D(CB_BIND(i)), 1);
753 PUSH_DATA (push, (15 << 4) | 1);
754 if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
755 unsigned j;
756 BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
757 PUSH_DATA (push, 0);
758 for (j = 0; j < 8; ++j)
759 PUSH_DATA(push, j);
760 } else {
761 BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
762 PUSH_DATA (push, 0x54);
763 }
764 }
765 BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
766 PUSH_DATA (push, 0);
767
768 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
769 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
770 PUSH_DATA (push, 256);
771 PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
772 PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
773 BEGIN_1IC0(push, NVC0_3D(CB_POS), 5);
774 PUSH_DATA (push, 0);
775 PUSH_DATAf(push, 0.0f);
776 PUSH_DATAf(push, 0.0f);
777 PUSH_DATAf(push, 0.0f);
778 PUSH_DATAf(push, 0.0f);
779 BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
780 PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
781 PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
782
783 if (dev->drm_version >= 0x01000101) {
784 ret = nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
785 if (ret) {
786 NOUVEAU_ERR("NOUVEAU_GETPARAM_GRAPH_UNITS failed.\n");
787 goto fail;
788 }
789 } else {
790 if (dev->chipset >= 0xe0 && dev->chipset < 0xf0)
791 value = (8 << 8) | 4;
792 else
793 value = (16 << 8) | 4;
794 }
795 screen->mp_count = value >> 8;
796 screen->mp_count_compute = screen->mp_count;
797
798 nvc0_screen_resize_tls_area(screen, 128 * 16, 0, 0x200);
799
800 BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
801 PUSH_DATAh(push, screen->text->offset);
802 PUSH_DATA (push, screen->text->offset);
803 BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
804 PUSH_DATAh(push, screen->tls->offset);
805 PUSH_DATA (push, screen->tls->offset);
806 PUSH_DATA (push, screen->tls->size >> 32);
807 PUSH_DATA (push, screen->tls->size);
808 BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
809 PUSH_DATA (push, 0);
810 BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
811 PUSH_DATA (push, 0);
812
813 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 20, NULL,
814 &screen->poly_cache);
815 if (ret)
816 goto fail;
817
818 BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
819 PUSH_DATAh(push, screen->poly_cache->offset);
820 PUSH_DATA (push, screen->poly_cache->offset);
821 PUSH_DATA (push, 3);
822
823 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 17, NULL,
824 &screen->txc);
825 if (ret)
826 goto fail;
827
828 BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
829 PUSH_DATAh(push, screen->txc->offset);
830 PUSH_DATA (push, screen->txc->offset);
831 PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
832
833 BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
834 PUSH_DATAh(push, screen->txc->offset + 65536);
835 PUSH_DATA (push, screen->txc->offset + 65536);
836 PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
837
838 BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
839 PUSH_DATA (push, 0);
840 BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
841 PUSH_DATA (push, 0);
842 PUSH_DATA (push, 0);
843 BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
844 PUSH_DATA (push, 0x3f);
845
846 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
847 PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
848 BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
849 for (i = 0; i < 8 * 2; ++i)
850 PUSH_DATA(push, 0);
851 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
852 PUSH_DATA (push, 0);
853 BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
854 PUSH_DATA (push, 0);
855
856 /* neither scissors, viewport nor stencil mask should affect clears */
857 BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
858 PUSH_DATA (push, 0);
859
860 BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
861 PUSH_DATA (push, 1);
862 BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(0)), 2);
863 PUSH_DATAf(push, 0.0f);
864 PUSH_DATAf(push, 1.0f);
865 BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
866 PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
867
868 /* We use scissors instead of exact view volume clipping,
869 * so they're always enabled.
870 */
871 BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(0)), 3);
872 PUSH_DATA (push, 1);
873 PUSH_DATA (push, 8192 << 16);
874 PUSH_DATA (push, 8192 << 16);
875
876 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
877
878 i = 0;
879 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, nvc0_9097_per_instance_bf);
880 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, nvc0_9097_blend_enables);
881 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, nvc0_9097_vertex_array_select);
882 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, nvc0_9097_tep_select);
883 MK_MACRO(NVC0_3D_MACRO_GP_SELECT, nvc0_9097_gp_select);
884 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, nvc0_9097_poly_mode_front);
885 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, nvc0_9097_poly_mode_back);
886
887 BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
888 PUSH_DATA (push, 1);
889 BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
890 PUSH_DATA (push, 1);
891 BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
892 PUSH_DATA (push, 0x40);
893 BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
894 PUSH_DATA (push, 0);
895 BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
896 PUSH_DATA (push, 0x30);
897 BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
898 PUSH_DATA (push, 3);
899 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
900 PUSH_DATA (push, 0x20);
901 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
902 PUSH_DATA (push, 0x00);
903
904 BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
905 PUSH_DATA (push, 0);
906 BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
907 PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
908
909 IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
910
911 if (nvc0_screen_init_compute(screen))
912 goto fail;
913
914 PUSH_KICK (push);
915
916 screen->tic.entries = CALLOC(4096, sizeof(void *));
917 screen->tsc.entries = screen->tic.entries + 2048;
918
919 mm_config.nvc0.tile_mode = 0;
920 mm_config.nvc0.memtype = 0xfe0;
921 screen->mm_VRAM_fe0 = nouveau_mm_create(dev, NOUVEAU_BO_VRAM, &mm_config);
922
923 if (!nvc0_blitter_create(screen))
924 goto fail;
925
926 nouveau_fence_new(&screen->base, &screen->base.fence.current, FALSE);
927
928 return pscreen;
929
930 fail:
931 nvc0_screen_destroy(pscreen);
932 return NULL;
933 }
934
935 int
936 nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
937 {
938 int i = screen->tic.next;
939
940 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
941 i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
942
943 screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
944
945 if (screen->tic.entries[i])
946 nv50_tic_entry(screen->tic.entries[i])->id = -1;
947
948 screen->tic.entries[i] = entry;
949 return i;
950 }
951
952 int
953 nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
954 {
955 int i = screen->tsc.next;
956
957 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
958 i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
959
960 screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
961
962 if (screen->tsc.entries[i])
963 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
964
965 screen->tsc.entries[i] = entry;
966 return i;
967 }