trace: Dump result of create_stream_output_target
[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 "nvc0_context.h"
31 #include "nvc0_screen.h"
32
33 #include "nvc0_graph_macros.h"
34
35 #ifndef NOUVEAU_GETPARAM_GRAPH_UNITS
36 # define NOUVEAU_GETPARAM_GRAPH_UNITS 13
37 #endif
38
39 static boolean
40 nvc0_screen_is_format_supported(struct pipe_screen *pscreen,
41 enum pipe_format format,
42 enum pipe_texture_target target,
43 unsigned sample_count,
44 unsigned bindings)
45 {
46 if (sample_count > 8)
47 return FALSE;
48 if (!(0x117 & (1 << sample_count))) /* 0, 1, 2, 4 or 8 */
49 return FALSE;
50
51 if (!util_format_is_supported(format, bindings))
52 return FALSE;
53
54 if ((bindings & PIPE_BIND_SAMPLER_VIEW) && (target != PIPE_BUFFER))
55 if (util_format_get_blocksizebits(format) == 3 * 32)
56 return FALSE;
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
66 static boolean
67 nvc0_screen_video_supported(struct pipe_screen *screen,
68 enum pipe_format format,
69 enum pipe_video_profile profile)
70 {
71 if (profile != PIPE_VIDEO_PROFILE_UNKNOWN)
72 return format == PIPE_FORMAT_NV12;
73
74 return vl_video_buffer_is_format_supported(screen, format, profile);
75 }
76
77
78 static int
79 nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
80 {
81 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
82
83 switch (param) {
84 case PIPE_CAP_MAX_COMBINED_SAMPLERS:
85 return 16 * 5;
86 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
87 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
88 return 15;
89 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
90 return (class_3d >= NVE4_3D_CLASS) ? 13 : 12;
91 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
92 return 2048;
93 case PIPE_CAP_MIN_TEXEL_OFFSET:
94 return -8;
95 case PIPE_CAP_MAX_TEXEL_OFFSET:
96 return 7;
97 case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
98 case PIPE_CAP_TEXTURE_SWIZZLE:
99 case PIPE_CAP_TEXTURE_SHADOW_MAP:
100 case PIPE_CAP_NPOT_TEXTURES:
101 case PIPE_CAP_ANISOTROPIC_FILTER:
102 case PIPE_CAP_SEAMLESS_CUBE_MAP:
103 case PIPE_CAP_CUBE_MAP_ARRAY:
104 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
105 case PIPE_CAP_TEXTURE_MULTISAMPLE:
106 return 1;
107 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
108 return 65536;
109 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
110 return (class_3d >= NVE4_3D_CLASS) ? 1 : 0;
111 case PIPE_CAP_TWO_SIDED_STENCIL:
112 case PIPE_CAP_DEPTH_CLIP_DISABLE:
113 case PIPE_CAP_POINT_SPRITE:
114 case PIPE_CAP_TGSI_TEXCOORD:
115 return 1;
116 case PIPE_CAP_SM3:
117 return 1;
118 case PIPE_CAP_GLSL_FEATURE_LEVEL:
119 return 150;
120 case PIPE_CAP_MAX_RENDER_TARGETS:
121 return 8;
122 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
123 return 1;
124 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
125 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
126 case PIPE_CAP_VERTEX_COLOR_CLAMPED:
127 return 1;
128 case PIPE_CAP_QUERY_TIMESTAMP:
129 case PIPE_CAP_QUERY_TIME_ELAPSED:
130 case PIPE_CAP_OCCLUSION_QUERY:
131 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
132 case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
133 return 1;
134 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
135 return 4;
136 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
137 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
138 return 128;
139 case PIPE_CAP_BLEND_EQUATION_SEPARATE:
140 case PIPE_CAP_INDEP_BLEND_ENABLE:
141 case PIPE_CAP_INDEP_BLEND_FUNC:
142 return 1;
143 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
144 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
145 return 1;
146 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
147 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
148 return 0;
149 case PIPE_CAP_SHADER_STENCIL_EXPORT:
150 return 0;
151 case PIPE_CAP_PRIMITIVE_RESTART:
152 case PIPE_CAP_TGSI_INSTANCEID:
153 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
154 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
155 case PIPE_CAP_CONDITIONAL_RENDER:
156 case PIPE_CAP_TEXTURE_BARRIER:
157 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
158 case PIPE_CAP_START_INSTANCE:
159 return 1;
160 case PIPE_CAP_TGSI_CAN_COMPACT_VARYINGS:
161 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
162 return 0; /* state trackers will know better */
163 case PIPE_CAP_USER_CONSTANT_BUFFERS:
164 case PIPE_CAP_USER_INDEX_BUFFERS:
165 case PIPE_CAP_USER_VERTEX_BUFFERS:
166 return 1;
167 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
168 return 256;
169 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
170 return 1; /* 256 for binding as RT, but that's not possible in GL */
171 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
172 return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
173 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
174 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
175 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
176 return 0;
177 case PIPE_CAP_COMPUTE:
178 return (class_3d >= NVE4_3D_CLASS) ? 1 : 0;
179 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
180 return 1;
181 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
182 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
183 default:
184 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
185 return 0;
186 }
187 }
188
189 static int
190 nvc0_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
191 enum pipe_shader_cap param)
192 {
193 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
194
195 switch (shader) {
196 case PIPE_SHADER_VERTEX:
197 /*
198 case PIPE_SHADER_TESSELLATION_CONTROL:
199 case PIPE_SHADER_TESSELLATION_EVALUATION:
200 */
201 case PIPE_SHADER_GEOMETRY:
202 case PIPE_SHADER_FRAGMENT:
203 break;
204 case PIPE_SHADER_COMPUTE:
205 if (class_3d < NVE4_3D_CLASS)
206 return 0;
207 break;
208 default:
209 return 0;
210 }
211
212 switch (param) {
213 case PIPE_SHADER_CAP_PREFERRED_IR:
214 return PIPE_SHADER_IR_TGSI;
215 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
216 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
217 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
218 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
219 return 16384;
220 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
221 return 16;
222 case PIPE_SHADER_CAP_MAX_INPUTS:
223 if (shader == PIPE_SHADER_VERTEX)
224 return 32;
225 /* NOTE: These only count our slots for GENERIC varyings.
226 * The address space may be larger, but the actual hard limit seems to be
227 * less than what the address space layout permits, so don't add TEXCOORD,
228 * COLOR, etc. here.
229 */
230 if (shader == PIPE_SHADER_FRAGMENT)
231 return 0x1f0 / 16;
232 /* Actually this counts CLIPVERTEX, which occupies the last generic slot,
233 * and excludes 0x60 per-patch inputs.
234 */
235 return 0x200 / 16;
236 case PIPE_SHADER_CAP_MAX_CONSTS:
237 return 65536 / 16;
238 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
239 if (shader == PIPE_SHADER_COMPUTE && class_3d >= NVE4_3D_CLASS)
240 return NVE4_MAX_PIPE_CONSTBUFS_COMPUTE;
241 return NVC0_MAX_PIPE_CONSTBUFS;
242 case PIPE_SHADER_CAP_MAX_ADDRS:
243 return 1;
244 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
245 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
246 return shader != PIPE_SHADER_FRAGMENT;
247 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
248 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
249 return 1;
250 case PIPE_SHADER_CAP_MAX_PREDS:
251 return 0;
252 case PIPE_SHADER_CAP_MAX_TEMPS:
253 return NVC0_CAP_MAX_PROGRAM_TEMPS;
254 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
255 return 1;
256 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
257 return 0;
258 case PIPE_SHADER_CAP_SUBROUTINES:
259 return 1;
260 case PIPE_SHADER_CAP_INTEGERS:
261 return 1;
262 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
263 return 16; /* would be 32 in linked (OpenGL-style) mode */
264 /*
265 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLER_VIEWS:
266 return 32;
267 */
268 default:
269 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
270 return 0;
271 }
272 }
273
274 static float
275 nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
276 {
277 switch (param) {
278 case PIPE_CAPF_MAX_LINE_WIDTH:
279 case PIPE_CAPF_MAX_LINE_WIDTH_AA:
280 return 10.0f;
281 case PIPE_CAPF_MAX_POINT_WIDTH:
282 return 63.0f;
283 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
284 return 63.375f;
285 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
286 return 16.0f;
287 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
288 return 15.0f;
289 default:
290 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
291 return 0.0f;
292 }
293 }
294
295 static int
296 nvc0_screen_get_compute_param(struct pipe_screen *pscreen,
297 enum pipe_compute_cap param, void *data)
298 {
299 uint64_t *data64 = (uint64_t *)data;
300 const uint16_t obj_class = nvc0_screen(pscreen)->compute->oclass;
301
302 switch (param) {
303 case PIPE_COMPUTE_CAP_GRID_DIMENSION:
304 data64[0] = 3;
305 return 8;
306 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
307 data64[0] = (obj_class >= NVE4_COMPUTE_CLASS) ? 0x7fffffff : 65535;
308 data64[1] = 65535;
309 data64[2] = 65535;
310 return 24;
311 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
312 data64[0] = 1024;
313 data64[1] = 1024;
314 data64[2] = 64;
315 return 24;
316 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
317 data64[0] = 1024;
318 return 8;
319 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */
320 data64[0] = (uint64_t)1 << 40;
321 return 8;
322 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
323 data64[0] = 48 << 10;
324 return 8;
325 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
326 data64[0] = 512 << 10;
327 return 8;
328 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
329 data64[0] = 4096;
330 return 8;
331 default:
332 return 0;
333 }
334 }
335
336 static void
337 nvc0_screen_destroy(struct pipe_screen *pscreen)
338 {
339 struct nvc0_screen *screen = nvc0_screen(pscreen);
340
341 if (screen->base.fence.current) {
342 nouveau_fence_wait(screen->base.fence.current);
343 nouveau_fence_ref(NULL, &screen->base.fence.current);
344 }
345 if (screen->base.pushbuf)
346 screen->base.pushbuf->user_priv = NULL;
347
348 if (screen->blitter)
349 nvc0_blitter_destroy(screen);
350 if (screen->pm.prog) {
351 screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
352 nvc0_program_destroy(NULL, screen->pm.prog);
353 }
354
355 nouveau_bo_ref(NULL, &screen->text);
356 nouveau_bo_ref(NULL, &screen->uniform_bo);
357 nouveau_bo_ref(NULL, &screen->tls);
358 nouveau_bo_ref(NULL, &screen->txc);
359 nouveau_bo_ref(NULL, &screen->fence.bo);
360 nouveau_bo_ref(NULL, &screen->poly_cache);
361 nouveau_bo_ref(NULL, &screen->parm);
362
363 nouveau_heap_destroy(&screen->lib_code);
364 nouveau_heap_destroy(&screen->text_heap);
365
366 FREE(screen->tic.entries);
367
368 nouveau_mm_destroy(screen->mm_VRAM_fe0);
369
370 nouveau_object_del(&screen->eng3d);
371 nouveau_object_del(&screen->eng2d);
372 nouveau_object_del(&screen->m2mf);
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 & 0xf0) {
489 case 0xc0:
490 case 0xd0:
491 return 0;
492 case 0xe0:
493 case 0xf0:
494 return nve4_screen_compute_setup(screen, screen->base.pushbuf);
495 default:
496 return -1;
497 }
498 }
499
500 boolean
501 nvc0_screen_resize_tls_area(struct nvc0_screen *screen,
502 uint32_t lpos, uint32_t lneg, uint32_t cstack)
503 {
504 struct nouveau_bo *bo = NULL;
505 int ret;
506 uint64_t size = (lpos + lneg) * 32 + cstack;
507
508 if (size >= (1 << 20)) {
509 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64"\n", size);
510 return FALSE;
511 }
512
513 size *= (screen->base.device->chipset >= 0xe0) ? 64 : 48; /* max warps */
514 size *= screen->mp_count;
515
516 size = align(size, 1 << 17);
517
518 ret = nouveau_bo_new(screen->base.device, NOUVEAU_BO_VRAM, 1 << 17, size,
519 NULL, &bo);
520 if (ret) {
521 NOUVEAU_ERR("failed to allocate TLS area, size: 0x%"PRIx64"\n", size);
522 return FALSE;
523 }
524 nouveau_bo_ref(NULL, &screen->tls);
525 screen->tls = bo;
526 return TRUE;
527 }
528
529 #define FAIL_SCREEN_INIT(str, err) \
530 do { \
531 NOUVEAU_ERR(str, err); \
532 nvc0_screen_destroy(pscreen); \
533 return NULL; \
534 } while(0)
535
536 struct pipe_screen *
537 nvc0_screen_create(struct nouveau_device *dev)
538 {
539 struct nvc0_screen *screen;
540 struct pipe_screen *pscreen;
541 struct nouveau_object *chan;
542 struct nouveau_pushbuf *push;
543 uint64_t value;
544 uint32_t obj_class;
545 int ret;
546 unsigned i;
547 union nouveau_bo_config mm_config;
548
549 switch (dev->chipset & ~0xf) {
550 case 0xc0:
551 case 0xd0:
552 case 0xe0:
553 break;
554 default:
555 return NULL;
556 }
557
558 screen = CALLOC_STRUCT(nvc0_screen);
559 if (!screen)
560 return NULL;
561 pscreen = &screen->base.base;
562
563 ret = nouveau_screen_init(&screen->base, dev);
564 if (ret) {
565 nvc0_screen_destroy(pscreen);
566 return NULL;
567 }
568 chan = screen->base.channel;
569 push = screen->base.pushbuf;
570 push->user_priv = screen;
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 = nvc0_screen_get_video_param;
588 screen->base.base.is_video_format_supported = nvc0_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 0xe0:
601 obj_class = NVE4_P2MF_CLASS;
602 break;
603 default:
604 obj_class = NVC0_M2MF_CLASS;
605 break;
606 }
607 ret = nouveau_object_new(chan, 0xbeef323f, obj_class, NULL, 0,
608 &screen->m2mf);
609 if (ret)
610 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
611
612 BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
613 PUSH_DATA (push, screen->m2mf->oclass);
614 if (screen->m2mf->oclass == NVE4_P2MF_CLASS) {
615 BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
616 PUSH_DATA (push, 0xa0b5);
617 }
618
619 ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
620 &screen->eng2d);
621 if (ret)
622 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
623
624 BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
625 PUSH_DATA (push, screen->eng2d->oclass);
626 BEGIN_NVC0(push, NVC0_2D(SINGLE_GPC), 1);
627 PUSH_DATA (push, 0);
628 BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
629 PUSH_DATA (push, NVC0_2D_OPERATION_SRCCOPY);
630 BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
631 PUSH_DATA (push, 0);
632 BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
633 PUSH_DATA (push, 0);
634 BEGIN_NVC0(push, SUBC_2D(0x0884), 1);
635 PUSH_DATA (push, 0x3f);
636 BEGIN_NVC0(push, SUBC_2D(0x0888), 1);
637 PUSH_DATA (push, 1);
638
639 BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
640 PUSH_DATAh(push, screen->fence.bo->offset + 16);
641 PUSH_DATA (push, screen->fence.bo->offset + 16);
642
643 switch (dev->chipset & 0xf0) {
644 case 0xe0:
645 obj_class = NVE4_3D_CLASS;
646 break;
647 case 0xd0:
648 case 0xc0:
649 default:
650 switch (dev->chipset) {
651 case 0xd9:
652 case 0xc8:
653 obj_class = NVC8_3D_CLASS;
654 break;
655 case 0xc1:
656 obj_class = NVC1_3D_CLASS;
657 break;
658 default:
659 obj_class = NVC0_3D_CLASS;
660 break;
661 }
662 break;
663 }
664 ret = nouveau_object_new(chan, 0xbeef003d, obj_class, NULL, 0,
665 &screen->eng3d);
666 if (ret)
667 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
668 screen->base.class_3d = obj_class;
669
670 BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
671 PUSH_DATA (push, screen->eng3d->oclass);
672
673 BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
674 PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
675
676 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", TRUE)) {
677 /* kill shaders after about 1 second (at 100 MHz) */
678 BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
679 PUSH_DATA (push, 0x17);
680 }
681
682 IMMED_NVC0(push, NVC0_3D(ZETA_COMP_ENABLE), dev->drm_version >= 0x01000101);
683 BEGIN_NVC0(push, NVC0_3D(RT_COMP_ENABLE(0)), 8);
684 for (i = 0; i < 8; ++i)
685 PUSH_DATA(push, dev->drm_version >= 0x01000101);
686
687 BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
688 PUSH_DATA (push, 1);
689
690 BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
691 PUSH_DATA (push, 0);
692 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
693 PUSH_DATA (push, 0);
694 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
695 PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
696 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
697 PUSH_DATA (push, 0);
698 BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
699 PUSH_DATA (push, 1);
700 BEGIN_NVC0(push, NVC0_3D(LINE_LAST_PIXEL), 1);
701 PUSH_DATA (push, 0);
702 BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
703 PUSH_DATA (push, 1);
704 BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
705 PUSH_DATA (push, 0);
706 if (screen->eng3d->oclass < NVE4_3D_CLASS) {
707 BEGIN_NVC0(push, NVC0_3D(TEX_MISC), 1);
708 PUSH_DATA (push, NVC0_3D_TEX_MISC_SEAMLESS_CUBE_MAP);
709 } else {
710 BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
711 PUSH_DATA (push, 15);
712 }
713 BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
714 PUSH_DATA (push, 8); /* 128 */
715 BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
716 PUSH_DATA (push, 1);
717 if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
718 BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
719 PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
720 }
721
722 nvc0_magic_3d_init(push, screen->eng3d->oclass);
723
724 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 20, NULL,
725 &screen->text);
726 if (ret)
727 goto fail;
728
729 /* XXX: getting a page fault at the end of the code buffer every few
730 * launches, don't use the last 256 bytes to work around them - prefetch ?
731 */
732 nouveau_heap_init(&screen->text_heap, 0, (1 << 20) - 0x100);
733
734 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 12, 6 << 16, NULL,
735 &screen->uniform_bo);
736 if (ret)
737 goto fail;
738
739 for (i = 0; i < 5; ++i) {
740 /* TIC and TSC entries for each unit (nve4+ only) */
741 /* auxiliary constants (6 user clip planes, base instance id) */
742 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
743 PUSH_DATA (push, 512);
744 PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (i << 9));
745 PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (i << 9));
746 BEGIN_NVC0(push, NVC0_3D(CB_BIND(i)), 1);
747 PUSH_DATA (push, (15 << 4) | 1);
748 if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
749 unsigned j;
750 BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
751 PUSH_DATA (push, 0);
752 for (j = 0; j < 8; ++j)
753 PUSH_DATA(push, j);
754 } else {
755 BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
756 PUSH_DATA (push, 0x54);
757 }
758 }
759 BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
760 PUSH_DATA (push, 0);
761
762 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
763 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
764 PUSH_DATA (push, 256);
765 PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
766 PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
767 BEGIN_1IC0(push, NVC0_3D(CB_POS), 5);
768 PUSH_DATA (push, 0);
769 PUSH_DATAf(push, 0.0f);
770 PUSH_DATAf(push, 0.0f);
771 PUSH_DATAf(push, 0.0f);
772 PUSH_DATAf(push, 0.0f);
773 BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
774 PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
775 PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
776
777 if (dev->drm_version >= 0x01000101) {
778 ret = nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
779 if (ret) {
780 NOUVEAU_ERR("NOUVEAU_GETPARAM_GRAPH_UNITS failed.\n");
781 goto fail;
782 }
783 } else {
784 if (dev->chipset >= 0xe0 && dev->chipset < 0xf0)
785 value = (8 << 8) | 4;
786 else
787 value = (16 << 8) | 4;
788 }
789 screen->mp_count = value >> 8;
790 screen->mp_count_compute = screen->mp_count;
791
792 nvc0_screen_resize_tls_area(screen, 128 * 16, 0, 0x200);
793
794 BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
795 PUSH_DATAh(push, screen->text->offset);
796 PUSH_DATA (push, screen->text->offset);
797 BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
798 PUSH_DATAh(push, screen->tls->offset);
799 PUSH_DATA (push, screen->tls->offset);
800 PUSH_DATA (push, screen->tls->size >> 32);
801 PUSH_DATA (push, screen->tls->size);
802 BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
803 PUSH_DATA (push, 0);
804 BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
805 PUSH_DATA (push, 0);
806
807 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 20, NULL,
808 &screen->poly_cache);
809 if (ret)
810 goto fail;
811
812 BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
813 PUSH_DATAh(push, screen->poly_cache->offset);
814 PUSH_DATA (push, screen->poly_cache->offset);
815 PUSH_DATA (push, 3);
816
817 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 17, NULL,
818 &screen->txc);
819 if (ret)
820 goto fail;
821
822 BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
823 PUSH_DATAh(push, screen->txc->offset);
824 PUSH_DATA (push, screen->txc->offset);
825 PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
826
827 BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
828 PUSH_DATAh(push, screen->txc->offset + 65536);
829 PUSH_DATA (push, screen->txc->offset + 65536);
830 PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
831
832 BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
833 PUSH_DATA (push, 0);
834 BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
835 PUSH_DATA (push, 0);
836 PUSH_DATA (push, 0);
837 BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
838 PUSH_DATA (push, 0x3f);
839
840 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
841 PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
842 BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
843 for (i = 0; i < 8 * 2; ++i)
844 PUSH_DATA(push, 0);
845 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
846 PUSH_DATA (push, 0);
847 BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
848 PUSH_DATA (push, 0);
849
850 /* neither scissors, viewport nor stencil mask should affect clears */
851 BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
852 PUSH_DATA (push, 0);
853
854 BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
855 PUSH_DATA (push, 1);
856 BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(0)), 2);
857 PUSH_DATAf(push, 0.0f);
858 PUSH_DATAf(push, 1.0f);
859 BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
860 PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
861
862 /* We use scissors instead of exact view volume clipping,
863 * so they're always enabled.
864 */
865 BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(0)), 3);
866 PUSH_DATA (push, 1);
867 PUSH_DATA (push, 8192 << 16);
868 PUSH_DATA (push, 8192 << 16);
869
870 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
871
872 i = 0;
873 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, nvc0_9097_per_instance_bf);
874 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, nvc0_9097_blend_enables);
875 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, nvc0_9097_vertex_array_select);
876 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, nvc0_9097_tep_select);
877 MK_MACRO(NVC0_3D_MACRO_GP_SELECT, nvc0_9097_gp_select);
878 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, nvc0_9097_poly_mode_front);
879 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, nvc0_9097_poly_mode_back);
880
881 BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
882 PUSH_DATA (push, 1);
883 BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
884 PUSH_DATA (push, 1);
885 BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
886 PUSH_DATA (push, 0x40);
887 BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
888 PUSH_DATA (push, 0);
889 BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
890 PUSH_DATA (push, 0x30);
891 BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
892 PUSH_DATA (push, 3);
893 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
894 PUSH_DATA (push, 0x20);
895 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
896 PUSH_DATA (push, 0x00);
897
898 BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
899 PUSH_DATA (push, 0);
900 BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
901 PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
902
903 IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
904
905 if (nvc0_screen_init_compute(screen))
906 goto fail;
907
908 PUSH_KICK (push);
909
910 screen->tic.entries = CALLOC(4096, sizeof(void *));
911 screen->tsc.entries = screen->tic.entries + 2048;
912
913 mm_config.nvc0.tile_mode = 0;
914 mm_config.nvc0.memtype = 0xfe0;
915 screen->mm_VRAM_fe0 = nouveau_mm_create(dev, NOUVEAU_BO_VRAM, &mm_config);
916
917 if (!nvc0_blitter_create(screen))
918 goto fail;
919
920 nouveau_fence_new(&screen->base, &screen->base.fence.current, FALSE);
921
922 return pscreen;
923
924 fail:
925 nvc0_screen_destroy(pscreen);
926 return NULL;
927 }
928
929 int
930 nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
931 {
932 int i = screen->tic.next;
933
934 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
935 i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
936
937 screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
938
939 if (screen->tic.entries[i])
940 nv50_tic_entry(screen->tic.entries[i])->id = -1;
941
942 screen->tic.entries[i] = entry;
943 return i;
944 }
945
946 int
947 nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
948 {
949 int i = screen->tsc.next;
950
951 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
952 i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
953
954 screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
955
956 if (screen->tsc.entries[i])
957 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
958
959 screen->tsc.entries[i] = entry;
960 return i;
961 }