nvc0: create the SW object
[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_BLEND_EQUATION_SEPARATE:
129 case PIPE_CAP_INDEP_BLEND_ENABLE:
130 case PIPE_CAP_INDEP_BLEND_FUNC:
131 return 1;
132 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
133 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
134 return 1;
135 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
136 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
137 return 0;
138 case PIPE_CAP_SHADER_STENCIL_EXPORT:
139 return 0;
140 case PIPE_CAP_PRIMITIVE_RESTART:
141 case PIPE_CAP_TGSI_INSTANCEID:
142 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
143 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
144 case PIPE_CAP_CONDITIONAL_RENDER:
145 case PIPE_CAP_TEXTURE_BARRIER:
146 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
147 case PIPE_CAP_START_INSTANCE:
148 return 1;
149 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
150 return 0; /* state trackers will know better */
151 case PIPE_CAP_USER_CONSTANT_BUFFERS:
152 case PIPE_CAP_USER_INDEX_BUFFERS:
153 case PIPE_CAP_USER_VERTEX_BUFFERS:
154 return 1;
155 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
156 return 256;
157 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
158 return 1; /* 256 for binding as RT, but that's not possible in GL */
159 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
160 return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
161 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
162 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
163 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
164 return 0;
165 case PIPE_CAP_COMPUTE:
166 return (class_3d >= NVE4_3D_CLASS) ? 1 : 0;
167 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
168 return 1;
169 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
170 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
171 case PIPE_CAP_ENDIANNESS:
172 return PIPE_ENDIAN_LITTLE;
173 case PIPE_CAP_TGSI_VS_LAYER:
174 return 0;
175 default:
176 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
177 return 0;
178 }
179 }
180
181 static int
182 nvc0_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
183 enum pipe_shader_cap param)
184 {
185 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
186
187 switch (shader) {
188 case PIPE_SHADER_VERTEX:
189 /*
190 case PIPE_SHADER_TESSELLATION_CONTROL:
191 case PIPE_SHADER_TESSELLATION_EVALUATION:
192 */
193 case PIPE_SHADER_GEOMETRY:
194 case PIPE_SHADER_FRAGMENT:
195 break;
196 case PIPE_SHADER_COMPUTE:
197 if (class_3d < NVE4_3D_CLASS)
198 return 0;
199 break;
200 default:
201 return 0;
202 }
203
204 switch (param) {
205 case PIPE_SHADER_CAP_PREFERRED_IR:
206 return PIPE_SHADER_IR_TGSI;
207 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
208 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
209 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
210 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
211 return 16384;
212 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
213 return 16;
214 case PIPE_SHADER_CAP_MAX_INPUTS:
215 if (shader == PIPE_SHADER_VERTEX)
216 return 32;
217 /* NOTE: These only count our slots for GENERIC varyings.
218 * The address space may be larger, but the actual hard limit seems to be
219 * less than what the address space layout permits, so don't add TEXCOORD,
220 * COLOR, etc. here.
221 */
222 if (shader == PIPE_SHADER_FRAGMENT)
223 return 0x1f0 / 16;
224 /* Actually this counts CLIPVERTEX, which occupies the last generic slot,
225 * and excludes 0x60 per-patch inputs.
226 */
227 return 0x200 / 16;
228 case PIPE_SHADER_CAP_MAX_CONSTS:
229 return 65536 / 16;
230 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
231 if (shader == PIPE_SHADER_COMPUTE && class_3d >= NVE4_3D_CLASS)
232 return NVE4_MAX_PIPE_CONSTBUFS_COMPUTE;
233 return NVC0_MAX_PIPE_CONSTBUFS;
234 case PIPE_SHADER_CAP_MAX_ADDRS:
235 return 1;
236 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
237 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
238 return shader != PIPE_SHADER_FRAGMENT;
239 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
240 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
241 return 1;
242 case PIPE_SHADER_CAP_MAX_PREDS:
243 return 0;
244 case PIPE_SHADER_CAP_MAX_TEMPS:
245 return NVC0_CAP_MAX_PROGRAM_TEMPS;
246 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
247 return 1;
248 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
249 return 0;
250 case PIPE_SHADER_CAP_SUBROUTINES:
251 return 1;
252 case PIPE_SHADER_CAP_INTEGERS:
253 return 1;
254 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
255 return 16; /* would be 32 in linked (OpenGL-style) mode */
256 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
257 return 16; /* XXX not sure if more are really safe */
258 default:
259 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
260 return 0;
261 }
262 }
263
264 static float
265 nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
266 {
267 switch (param) {
268 case PIPE_CAPF_MAX_LINE_WIDTH:
269 case PIPE_CAPF_MAX_LINE_WIDTH_AA:
270 return 10.0f;
271 case PIPE_CAPF_MAX_POINT_WIDTH:
272 return 63.0f;
273 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
274 return 63.375f;
275 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
276 return 16.0f;
277 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
278 return 15.0f;
279 default:
280 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
281 return 0.0f;
282 }
283 }
284
285 static int
286 nvc0_screen_get_compute_param(struct pipe_screen *pscreen,
287 enum pipe_compute_cap param, void *data)
288 {
289 uint64_t *data64 = (uint64_t *)data;
290 const uint16_t obj_class = nvc0_screen(pscreen)->compute->oclass;
291
292 switch (param) {
293 case PIPE_COMPUTE_CAP_GRID_DIMENSION:
294 data64[0] = 3;
295 return 8;
296 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
297 data64[0] = (obj_class >= NVE4_COMPUTE_CLASS) ? 0x7fffffff : 65535;
298 data64[1] = 65535;
299 data64[2] = 65535;
300 return 24;
301 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
302 data64[0] = 1024;
303 data64[1] = 1024;
304 data64[2] = 64;
305 return 24;
306 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
307 data64[0] = 1024;
308 return 8;
309 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */
310 data64[0] = (uint64_t)1 << 40;
311 return 8;
312 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
313 data64[0] = 48 << 10;
314 return 8;
315 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
316 data64[0] = 512 << 10;
317 return 8;
318 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
319 data64[0] = 4096;
320 return 8;
321 default:
322 return 0;
323 }
324 }
325
326 static void
327 nvc0_screen_destroy(struct pipe_screen *pscreen)
328 {
329 struct nvc0_screen *screen = nvc0_screen(pscreen);
330
331 if (screen->base.fence.current) {
332 nouveau_fence_wait(screen->base.fence.current);
333 nouveau_fence_ref(NULL, &screen->base.fence.current);
334 }
335 if (screen->base.pushbuf)
336 screen->base.pushbuf->user_priv = NULL;
337
338 if (screen->blitter)
339 nvc0_blitter_destroy(screen);
340 if (screen->pm.prog) {
341 screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
342 nvc0_program_destroy(NULL, screen->pm.prog);
343 }
344
345 nouveau_bo_ref(NULL, &screen->text);
346 nouveau_bo_ref(NULL, &screen->uniform_bo);
347 nouveau_bo_ref(NULL, &screen->tls);
348 nouveau_bo_ref(NULL, &screen->txc);
349 nouveau_bo_ref(NULL, &screen->fence.bo);
350 nouveau_bo_ref(NULL, &screen->poly_cache);
351 nouveau_bo_ref(NULL, &screen->parm);
352
353 nouveau_heap_destroy(&screen->lib_code);
354 nouveau_heap_destroy(&screen->text_heap);
355
356 FREE(screen->tic.entries);
357
358 nouveau_mm_destroy(screen->mm_VRAM_fe0);
359
360 nouveau_object_del(&screen->eng3d);
361 nouveau_object_del(&screen->eng2d);
362 nouveau_object_del(&screen->m2mf);
363 nouveau_object_del(&screen->compute);
364 nouveau_object_del(&screen->nvsw);
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 & ~0xf) {
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 case 0x100:
492 return nve4_screen_compute_setup(screen, screen->base.pushbuf);
493 default:
494 return -1;
495 }
496 }
497
498 boolean
499 nvc0_screen_resize_tls_area(struct nvc0_screen *screen,
500 uint32_t lpos, uint32_t lneg, uint32_t cstack)
501 {
502 struct nouveau_bo *bo = NULL;
503 int ret;
504 uint64_t size = (lpos + lneg) * 32 + cstack;
505
506 if (size >= (1 << 20)) {
507 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64"\n", size);
508 return FALSE;
509 }
510
511 size *= (screen->base.device->chipset >= 0xe0) ? 64 : 48; /* max warps */
512 size = align(size, 0x8000);
513 size *= screen->mp_count;
514
515 size = align(size, 1 << 17);
516
517 ret = nouveau_bo_new(screen->base.device, NOUVEAU_BO_VRAM, 1 << 17, size,
518 NULL, &bo);
519 if (ret) {
520 NOUVEAU_ERR("failed to allocate TLS area, size: 0x%"PRIx64"\n", size);
521 return FALSE;
522 }
523 nouveau_bo_ref(NULL, &screen->tls);
524 screen->tls = bo;
525 return TRUE;
526 }
527
528 #define FAIL_SCREEN_INIT(str, err) \
529 do { \
530 NOUVEAU_ERR(str, err); \
531 nvc0_screen_destroy(pscreen); \
532 return NULL; \
533 } while(0)
534
535 struct pipe_screen *
536 nvc0_screen_create(struct nouveau_device *dev)
537 {
538 struct nvc0_screen *screen;
539 struct pipe_screen *pscreen;
540 struct nouveau_object *chan;
541 struct nouveau_pushbuf *push;
542 uint64_t value;
543 uint32_t obj_class;
544 int ret;
545 unsigned i;
546 union nouveau_bo_config mm_config;
547
548 switch (dev->chipset & ~0xf) {
549 case 0xc0:
550 case 0xd0:
551 case 0xe0:
552 case 0xf0:
553 case 0x100:
554 break;
555 default:
556 return NULL;
557 }
558
559 screen = CALLOC_STRUCT(nvc0_screen);
560 if (!screen)
561 return NULL;
562 pscreen = &screen->base.base;
563
564 ret = nouveau_screen_init(&screen->base, dev);
565 if (ret) {
566 nvc0_screen_destroy(pscreen);
567 return NULL;
568 }
569 chan = screen->base.channel;
570 push = screen->base.pushbuf;
571 push->user_priv = screen;
572 push->rsvd_kick = 5;
573
574 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
575 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
576 screen->base.sysmem_bindings |=
577 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
578
579 pscreen->destroy = nvc0_screen_destroy;
580 pscreen->context_create = nvc0_create;
581 pscreen->is_format_supported = nvc0_screen_is_format_supported;
582 pscreen->get_param = nvc0_screen_get_param;
583 pscreen->get_shader_param = nvc0_screen_get_shader_param;
584 pscreen->get_paramf = nvc0_screen_get_paramf;
585 pscreen->get_driver_query_info = nvc0_screen_get_driver_query_info;
586
587 nvc0_screen_init_resource_functions(pscreen);
588
589 screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
590 screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
591
592 ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096, NULL,
593 &screen->fence.bo);
594 if (ret)
595 goto fail;
596 nouveau_bo_map(screen->fence.bo, 0, NULL);
597 screen->fence.map = screen->fence.bo->map;
598 screen->base.fence.emit = nvc0_screen_fence_emit;
599 screen->base.fence.update = nvc0_screen_fence_update;
600
601
602 ret = nouveau_object_new(chan,
603 (dev->chipset < 0xe0) ? 0x1f906e : 0x906e, 0x906e,
604 NULL, 0, &screen->nvsw);
605 if (ret)
606 FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret);
607
608
609 switch (dev->chipset & ~0xf) {
610 case 0x100:
611 case 0xf0:
612 obj_class = NVF0_P2MF_CLASS;
613 break;
614 case 0xe0:
615 obj_class = NVE4_P2MF_CLASS;
616 break;
617 default:
618 obj_class = NVC0_M2MF_CLASS;
619 break;
620 }
621 ret = nouveau_object_new(chan, 0xbeef323f, obj_class, NULL, 0,
622 &screen->m2mf);
623 if (ret)
624 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
625
626 BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
627 PUSH_DATA (push, screen->m2mf->oclass);
628 if (screen->m2mf->oclass == NVE4_P2MF_CLASS) {
629 BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
630 PUSH_DATA (push, 0xa0b5);
631 }
632
633 ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
634 &screen->eng2d);
635 if (ret)
636 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
637
638 BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
639 PUSH_DATA (push, screen->eng2d->oclass);
640 BEGIN_NVC0(push, NVC0_2D(SINGLE_GPC), 1);
641 PUSH_DATA (push, 0);
642 BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
643 PUSH_DATA (push, NVC0_2D_OPERATION_SRCCOPY);
644 BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
645 PUSH_DATA (push, 0);
646 BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
647 PUSH_DATA (push, 0);
648 BEGIN_NVC0(push, SUBC_2D(0x0884), 1);
649 PUSH_DATA (push, 0x3f);
650 BEGIN_NVC0(push, SUBC_2D(0x0888), 1);
651 PUSH_DATA (push, 1);
652
653 BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
654 PUSH_DATAh(push, screen->fence.bo->offset + 16);
655 PUSH_DATA (push, screen->fence.bo->offset + 16);
656
657 switch (dev->chipset & ~0xf) {
658 case 0x100:
659 case 0xf0:
660 obj_class = NVF0_3D_CLASS;
661 break;
662 case 0xe0:
663 obj_class = NVE4_3D_CLASS;
664 break;
665 case 0xd0:
666 obj_class = NVC8_3D_CLASS;
667 break;
668 case 0xc0:
669 default:
670 switch (dev->chipset) {
671 case 0xc8:
672 obj_class = NVC8_3D_CLASS;
673 break;
674 case 0xc1:
675 obj_class = NVC1_3D_CLASS;
676 break;
677 default:
678 obj_class = NVC0_3D_CLASS;
679 break;
680 }
681 break;
682 }
683 ret = nouveau_object_new(chan, 0xbeef003d, obj_class, NULL, 0,
684 &screen->eng3d);
685 if (ret)
686 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
687 screen->base.class_3d = obj_class;
688
689 BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
690 PUSH_DATA (push, screen->eng3d->oclass);
691
692 BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
693 PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
694
695 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", TRUE)) {
696 /* kill shaders after about 1 second (at 100 MHz) */
697 BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
698 PUSH_DATA (push, 0x17);
699 }
700
701 IMMED_NVC0(push, NVC0_3D(ZETA_COMP_ENABLE), dev->drm_version >= 0x01000101);
702 BEGIN_NVC0(push, NVC0_3D(RT_COMP_ENABLE(0)), 8);
703 for (i = 0; i < 8; ++i)
704 PUSH_DATA(push, dev->drm_version >= 0x01000101);
705
706 BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
707 PUSH_DATA (push, 1);
708
709 BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
710 PUSH_DATA (push, 0);
711 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
712 PUSH_DATA (push, 0);
713 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
714 PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
715 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
716 PUSH_DATA (push, 0);
717 BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
718 PUSH_DATA (push, 1);
719 BEGIN_NVC0(push, NVC0_3D(LINE_LAST_PIXEL), 1);
720 PUSH_DATA (push, 0);
721 BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
722 PUSH_DATA (push, 1);
723 BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
724 PUSH_DATA (push, 0);
725 if (screen->eng3d->oclass < NVE4_3D_CLASS) {
726 BEGIN_NVC0(push, NVC0_3D(TEX_MISC), 1);
727 PUSH_DATA (push, NVC0_3D_TEX_MISC_SEAMLESS_CUBE_MAP);
728 } else {
729 BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
730 PUSH_DATA (push, 15);
731 }
732 BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
733 PUSH_DATA (push, 8); /* 128 */
734 BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
735 PUSH_DATA (push, 1);
736 if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
737 BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
738 PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
739 }
740
741 nvc0_magic_3d_init(push, screen->eng3d->oclass);
742
743 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 20, NULL,
744 &screen->text);
745 if (ret)
746 goto fail;
747
748 /* XXX: getting a page fault at the end of the code buffer every few
749 * launches, don't use the last 256 bytes to work around them - prefetch ?
750 */
751 nouveau_heap_init(&screen->text_heap, 0, (1 << 20) - 0x100);
752
753 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 12, 6 << 16, NULL,
754 &screen->uniform_bo);
755 if (ret)
756 goto fail;
757
758 PUSH_REFN (push, screen->uniform_bo, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
759
760 for (i = 0; i < 5; ++i) {
761 /* TIC and TSC entries for each unit (nve4+ only) */
762 /* auxiliary constants (6 user clip planes, base instance id) */
763 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
764 PUSH_DATA (push, 512);
765 PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (i << 9));
766 PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (i << 9));
767 BEGIN_NVC0(push, NVC0_3D(CB_BIND(i)), 1);
768 PUSH_DATA (push, (15 << 4) | 1);
769 if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
770 unsigned j;
771 BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
772 PUSH_DATA (push, 0);
773 for (j = 0; j < 8; ++j)
774 PUSH_DATA(push, j);
775 } else {
776 BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
777 PUSH_DATA (push, 0x54);
778 }
779 }
780 BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
781 PUSH_DATA (push, 0);
782
783 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
784 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
785 PUSH_DATA (push, 256);
786 PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
787 PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
788 BEGIN_1IC0(push, NVC0_3D(CB_POS), 5);
789 PUSH_DATA (push, 0);
790 PUSH_DATAf(push, 0.0f);
791 PUSH_DATAf(push, 0.0f);
792 PUSH_DATAf(push, 0.0f);
793 PUSH_DATAf(push, 0.0f);
794 BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
795 PUSH_DATAh(push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
796 PUSH_DATA (push, screen->uniform_bo->offset + (5 << 16) + (6 << 9));
797
798 if (dev->drm_version >= 0x01000101) {
799 ret = nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
800 if (ret) {
801 NOUVEAU_ERR("NOUVEAU_GETPARAM_GRAPH_UNITS failed.\n");
802 goto fail;
803 }
804 } else {
805 if (dev->chipset >= 0xe0 && dev->chipset < 0xf0)
806 value = (8 << 8) | 4;
807 else
808 value = (16 << 8) | 4;
809 }
810 screen->mp_count = value >> 8;
811 screen->mp_count_compute = screen->mp_count;
812
813 nvc0_screen_resize_tls_area(screen, 128 * 16, 0, 0x200);
814
815 BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
816 PUSH_DATAh(push, screen->text->offset);
817 PUSH_DATA (push, screen->text->offset);
818 BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
819 PUSH_DATAh(push, screen->tls->offset);
820 PUSH_DATA (push, screen->tls->offset);
821 PUSH_DATA (push, screen->tls->size >> 32);
822 PUSH_DATA (push, screen->tls->size);
823 BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
824 PUSH_DATA (push, 0);
825 BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
826 PUSH_DATA (push, 0);
827
828 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 20, NULL,
829 &screen->poly_cache);
830 if (ret)
831 goto fail;
832
833 BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
834 PUSH_DATAh(push, screen->poly_cache->offset);
835 PUSH_DATA (push, screen->poly_cache->offset);
836 PUSH_DATA (push, 3);
837
838 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 17, 1 << 17, NULL,
839 &screen->txc);
840 if (ret)
841 goto fail;
842
843 BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
844 PUSH_DATAh(push, screen->txc->offset);
845 PUSH_DATA (push, screen->txc->offset);
846 PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
847
848 BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
849 PUSH_DATAh(push, screen->txc->offset + 65536);
850 PUSH_DATA (push, screen->txc->offset + 65536);
851 PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
852
853 BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
854 PUSH_DATA (push, 0);
855 BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
856 PUSH_DATA (push, 0);
857 PUSH_DATA (push, 0);
858 BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
859 PUSH_DATA (push, 0x3f);
860
861 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
862 PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
863 BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
864 for (i = 0; i < 8 * 2; ++i)
865 PUSH_DATA(push, 0);
866 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
867 PUSH_DATA (push, 0);
868 BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
869 PUSH_DATA (push, 0);
870
871 /* neither scissors, viewport nor stencil mask should affect clears */
872 BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
873 PUSH_DATA (push, 0);
874
875 BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
876 PUSH_DATA (push, 1);
877 BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(0)), 2);
878 PUSH_DATAf(push, 0.0f);
879 PUSH_DATAf(push, 1.0f);
880 BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
881 PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
882
883 /* We use scissors instead of exact view volume clipping,
884 * so they're always enabled.
885 */
886 BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(0)), 3);
887 PUSH_DATA (push, 1);
888 PUSH_DATA (push, 8192 << 16);
889 PUSH_DATA (push, 8192 << 16);
890
891 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
892
893 i = 0;
894 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, nvc0_9097_per_instance_bf);
895 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, nvc0_9097_blend_enables);
896 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, nvc0_9097_vertex_array_select);
897 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, nvc0_9097_tep_select);
898 MK_MACRO(NVC0_3D_MACRO_GP_SELECT, nvc0_9097_gp_select);
899 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, nvc0_9097_poly_mode_front);
900 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, nvc0_9097_poly_mode_back);
901
902 BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
903 PUSH_DATA (push, 1);
904 BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
905 PUSH_DATA (push, 1);
906 BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
907 PUSH_DATA (push, 0x40);
908 BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
909 PUSH_DATA (push, 0);
910 BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
911 PUSH_DATA (push, 0x30);
912 BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
913 PUSH_DATA (push, 3);
914 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
915 PUSH_DATA (push, 0x20);
916 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
917 PUSH_DATA (push, 0x00);
918
919 BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
920 PUSH_DATA (push, 0);
921 BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
922 PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
923
924 IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
925
926 if (nvc0_screen_init_compute(screen))
927 goto fail;
928
929 PUSH_KICK (push);
930
931 screen->tic.entries = CALLOC(4096, sizeof(void *));
932 screen->tsc.entries = screen->tic.entries + 2048;
933
934 mm_config.nvc0.tile_mode = 0;
935 mm_config.nvc0.memtype = 0xfe0;
936 screen->mm_VRAM_fe0 = nouveau_mm_create(dev, NOUVEAU_BO_VRAM, &mm_config);
937
938 if (!nvc0_blitter_create(screen))
939 goto fail;
940
941 nouveau_fence_new(&screen->base, &screen->base.fence.current, FALSE);
942
943 return pscreen;
944
945 fail:
946 nvc0_screen_destroy(pscreen);
947 return NULL;
948 }
949
950 int
951 nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
952 {
953 int i = screen->tic.next;
954
955 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
956 i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
957
958 screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
959
960 if (screen->tic.entries[i])
961 nv50_tic_entry(screen->tic.entries[i])->id = -1;
962
963 screen->tic.entries[i] = entry;
964 return i;
965 }
966
967 int
968 nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
969 {
970 int i = screen->tsc.next;
971
972 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
973 i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
974
975 screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
976
977 if (screen->tsc.entries[i])
978 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
979
980 screen->tsc.entries[i] = entry;
981 return i;
982 }