gallium: remove PIPE_CAP_MAX_COMBINED_SAMPLERS
[mesa.git] / src / gallium / drivers / nouveau / nv50 / nv50_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 "nv50/nv50_context.h"
28 #include "nv50/nv50_screen.h"
29
30 #include "nouveau_vp3_video.h"
31
32 #include "nv_object.xml.h"
33 #include <errno.h>
34
35 #ifndef NOUVEAU_GETPARAM_GRAPH_UNITS
36 # define NOUVEAU_GETPARAM_GRAPH_UNITS 13
37 #endif
38
39 /* affected by LOCAL_WARPS_LOG_ALLOC / LOCAL_WARPS_NO_CLAMP */
40 #define LOCAL_WARPS_ALLOC 32
41 /* affected by STACK_WARPS_LOG_ALLOC / STACK_WARPS_NO_CLAMP */
42 #define STACK_WARPS_ALLOC 32
43
44 #define THREADS_IN_WARP 32
45
46 #define ONE_TEMP_SIZE (4/*vector*/ * sizeof(float))
47
48 static boolean
49 nv50_screen_is_format_supported(struct pipe_screen *pscreen,
50 enum pipe_format format,
51 enum pipe_texture_target target,
52 unsigned sample_count,
53 unsigned bindings)
54 {
55 if (sample_count > 8)
56 return FALSE;
57 if (!(0x117 & (1 << sample_count))) /* 0, 1, 2, 4 or 8 */
58 return FALSE;
59 if (sample_count == 8 && util_format_get_blocksizebits(format) >= 128)
60 return FALSE;
61
62 if (!util_format_is_supported(format, bindings))
63 return FALSE;
64
65 switch (format) {
66 case PIPE_FORMAT_Z16_UNORM:
67 if (nv50_screen(pscreen)->tesla->oclass < NVA0_3D_CLASS)
68 return FALSE;
69 break;
70 default:
71 break;
72 }
73
74 /* transfers & shared are always supported */
75 bindings &= ~(PIPE_BIND_TRANSFER_READ |
76 PIPE_BIND_TRANSFER_WRITE |
77 PIPE_BIND_SHARED);
78
79 return (nv50_format_table[format].usage & bindings) == bindings;
80 }
81
82 static int
83 nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
84 {
85 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
86
87 switch (param) {
88 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
89 return 14;
90 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
91 return 12;
92 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
93 return 14;
94 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
95 return 512;
96 case PIPE_CAP_MIN_TEXEL_OFFSET:
97 return -8;
98 case PIPE_CAP_MAX_TEXEL_OFFSET:
99 return 7;
100 case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
101 case PIPE_CAP_TEXTURE_SWIZZLE:
102 case PIPE_CAP_TEXTURE_SHADOW_MAP:
103 case PIPE_CAP_NPOT_TEXTURES:
104 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
105 case PIPE_CAP_ANISOTROPIC_FILTER:
106 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
107 return 1;
108 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
109 return 65536;
110 case PIPE_CAP_SEAMLESS_CUBE_MAP:
111 return 1; /* nv50_screen(pscreen)->tesla->oclass >= NVA0_3D_CLASS; */
112 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
113 return 0;
114 case PIPE_CAP_CUBE_MAP_ARRAY:
115 return 0;
116 /*
117 return nv50_screen(pscreen)->tesla->oclass >= NVA3_3D_CLASS;
118 */
119 case PIPE_CAP_TWO_SIDED_STENCIL:
120 case PIPE_CAP_DEPTH_CLIP_DISABLE:
121 case PIPE_CAP_POINT_SPRITE:
122 return 1;
123 case PIPE_CAP_SM3:
124 return 1;
125 case PIPE_CAP_GLSL_FEATURE_LEVEL:
126 return 330;
127 case PIPE_CAP_MAX_RENDER_TARGETS:
128 return 8;
129 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
130 return 1;
131 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
132 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
133 case PIPE_CAP_VERTEX_COLOR_CLAMPED:
134 return 1;
135 case PIPE_CAP_QUERY_TIMESTAMP:
136 case PIPE_CAP_QUERY_TIME_ELAPSED:
137 case PIPE_CAP_OCCLUSION_QUERY:
138 return 1;
139 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
140 return 4;
141 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
142 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
143 return 64;
144 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
145 return (class_3d >= NVA0_3D_CLASS) ? 1 : 0;
146 case PIPE_CAP_BLEND_EQUATION_SEPARATE:
147 case PIPE_CAP_INDEP_BLEND_ENABLE:
148 return 1;
149 case PIPE_CAP_INDEP_BLEND_FUNC:
150 return nv50_screen(pscreen)->tesla->oclass >= NVA3_3D_CLASS;
151 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
152 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
153 return 1;
154 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
155 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
156 return 0;
157 case PIPE_CAP_SHADER_STENCIL_EXPORT:
158 return 0;
159 case PIPE_CAP_PRIMITIVE_RESTART:
160 case PIPE_CAP_TGSI_INSTANCEID:
161 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
162 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
163 case PIPE_CAP_CONDITIONAL_RENDER:
164 case PIPE_CAP_TEXTURE_BARRIER:
165 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
166 case PIPE_CAP_START_INSTANCE:
167 return 1;
168 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
169 return 0; /* state trackers will know better */
170 case PIPE_CAP_USER_CONSTANT_BUFFERS:
171 case PIPE_CAP_USER_INDEX_BUFFERS:
172 case PIPE_CAP_USER_VERTEX_BUFFERS:
173 return 1;
174 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
175 return 256;
176 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
177 return 1; /* 256 for binding as RT, but that's not possible in GL */
178 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
179 return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
180 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
181 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
182 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
183 case PIPE_CAP_TGSI_TEXCOORD:
184 return 0;
185 case PIPE_CAP_TEXTURE_MULTISAMPLE:
186 return 1;
187 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
188 return 1;
189 case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
190 return 0;
191 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
192 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
193 case PIPE_CAP_ENDIANNESS:
194 return PIPE_ENDIAN_LITTLE;
195 case PIPE_CAP_TGSI_VS_LAYER:
196 return 0;
197 default:
198 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
199 return 0;
200 }
201 }
202
203 static int
204 nv50_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
205 enum pipe_shader_cap param)
206 {
207 switch (shader) {
208 case PIPE_SHADER_VERTEX:
209 case PIPE_SHADER_GEOMETRY:
210 case PIPE_SHADER_FRAGMENT:
211 break;
212 default:
213 return 0;
214 }
215
216 switch (param) {
217 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
218 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
219 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
220 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
221 return 16384;
222 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
223 return 4;
224 case PIPE_SHADER_CAP_MAX_INPUTS:
225 if (shader == PIPE_SHADER_VERTEX)
226 return 32;
227 return 15;
228 case PIPE_SHADER_CAP_MAX_CONSTS:
229 return 65536 / 16;
230 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
231 return NV50_MAX_PIPE_CONSTBUFS;
232 case PIPE_SHADER_CAP_MAX_ADDRS:
233 return 1;
234 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
235 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
236 return shader != PIPE_SHADER_FRAGMENT;
237 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
238 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
239 return 1;
240 case PIPE_SHADER_CAP_MAX_PREDS:
241 return 0;
242 case PIPE_SHADER_CAP_MAX_TEMPS:
243 return nv50_screen(pscreen)->max_tls_space / ONE_TEMP_SIZE;
244 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
245 return 1;
246 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
247 return 0;
248 case PIPE_SHADER_CAP_SUBROUTINES:
249 return 0; /* please inline, or provide function declarations */
250 case PIPE_SHADER_CAP_INTEGERS:
251 return 1;
252 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
253 /* The chip could handle more sampler views than samplers */
254 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
255 return MIN2(32, PIPE_MAX_SAMPLERS);
256 default:
257 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
258 return 0;
259 }
260 }
261
262 static float
263 nv50_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
264 {
265 switch (param) {
266 case PIPE_CAPF_MAX_LINE_WIDTH:
267 case PIPE_CAPF_MAX_LINE_WIDTH_AA:
268 return 10.0f;
269 case PIPE_CAPF_MAX_POINT_WIDTH:
270 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
271 return 64.0f;
272 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
273 return 16.0f;
274 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
275 return 4.0f;
276 default:
277 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
278 return 0.0f;
279 }
280 }
281
282 static void
283 nv50_screen_destroy(struct pipe_screen *pscreen)
284 {
285 struct nv50_screen *screen = nv50_screen(pscreen);
286
287 if (screen->base.fence.current) {
288 nouveau_fence_wait(screen->base.fence.current);
289 nouveau_fence_ref (NULL, &screen->base.fence.current);
290 }
291 if (screen->base.pushbuf)
292 screen->base.pushbuf->user_priv = NULL;
293
294 if (screen->blitter)
295 nv50_blitter_destroy(screen);
296
297 nouveau_bo_ref(NULL, &screen->code);
298 nouveau_bo_ref(NULL, &screen->tls_bo);
299 nouveau_bo_ref(NULL, &screen->stack_bo);
300 nouveau_bo_ref(NULL, &screen->txc);
301 nouveau_bo_ref(NULL, &screen->uniforms);
302 nouveau_bo_ref(NULL, &screen->fence.bo);
303
304 nouveau_heap_destroy(&screen->vp_code_heap);
305 nouveau_heap_destroy(&screen->gp_code_heap);
306 nouveau_heap_destroy(&screen->fp_code_heap);
307
308 FREE(screen->tic.entries);
309
310 nouveau_object_del(&screen->tesla);
311 nouveau_object_del(&screen->eng2d);
312 nouveau_object_del(&screen->m2mf);
313 nouveau_object_del(&screen->sync);
314
315 nouveau_screen_fini(&screen->base);
316
317 FREE(screen);
318 }
319
320 static void
321 nv50_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
322 {
323 struct nv50_screen *screen = nv50_screen(pscreen);
324 struct nouveau_pushbuf *push = screen->base.pushbuf;
325
326 /* we need to do it after possible flush in MARK_RING */
327 *sequence = ++screen->base.fence.sequence;
328
329 PUSH_DATA (push, NV50_FIFO_PKHDR(NV50_3D(QUERY_ADDRESS_HIGH), 4));
330 PUSH_DATAh(push, screen->fence.bo->offset);
331 PUSH_DATA (push, screen->fence.bo->offset);
332 PUSH_DATA (push, *sequence);
333 PUSH_DATA (push, NV50_3D_QUERY_GET_MODE_WRITE_UNK0 |
334 NV50_3D_QUERY_GET_UNK4 |
335 NV50_3D_QUERY_GET_UNIT_CROP |
336 NV50_3D_QUERY_GET_TYPE_QUERY |
337 NV50_3D_QUERY_GET_QUERY_SELECT_ZERO |
338 NV50_3D_QUERY_GET_SHORT);
339 }
340
341 static u32
342 nv50_screen_fence_update(struct pipe_screen *pscreen)
343 {
344 return nv50_screen(pscreen)->fence.map[0];
345 }
346
347 static void
348 nv50_screen_init_hwctx(struct nv50_screen *screen)
349 {
350 struct nouveau_pushbuf *push = screen->base.pushbuf;
351 struct nv04_fifo *fifo;
352 unsigned i;
353
354 fifo = (struct nv04_fifo *)screen->base.channel->data;
355
356 BEGIN_NV04(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
357 PUSH_DATA (push, screen->m2mf->handle);
358 BEGIN_NV04(push, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY), 3);
359 PUSH_DATA (push, screen->sync->handle);
360 PUSH_DATA (push, fifo->vram);
361 PUSH_DATA (push, fifo->vram);
362
363 BEGIN_NV04(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
364 PUSH_DATA (push, screen->eng2d->handle);
365 BEGIN_NV04(push, NV50_2D(DMA_NOTIFY), 4);
366 PUSH_DATA (push, screen->sync->handle);
367 PUSH_DATA (push, fifo->vram);
368 PUSH_DATA (push, fifo->vram);
369 PUSH_DATA (push, fifo->vram);
370 BEGIN_NV04(push, NV50_2D(OPERATION), 1);
371 PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
372 BEGIN_NV04(push, NV50_2D(CLIP_ENABLE), 1);
373 PUSH_DATA (push, 0);
374 BEGIN_NV04(push, NV50_2D(COLOR_KEY_ENABLE), 1);
375 PUSH_DATA (push, 0);
376 BEGIN_NV04(push, SUBC_2D(0x0888), 1);
377 PUSH_DATA (push, 1);
378
379 BEGIN_NV04(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
380 PUSH_DATA (push, screen->tesla->handle);
381
382 BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
383 PUSH_DATA (push, NV50_3D_COND_MODE_ALWAYS);
384
385 BEGIN_NV04(push, NV50_3D(DMA_NOTIFY), 1);
386 PUSH_DATA (push, screen->sync->handle);
387 BEGIN_NV04(push, NV50_3D(DMA_ZETA), 11);
388 for (i = 0; i < 11; ++i)
389 PUSH_DATA(push, fifo->vram);
390 BEGIN_NV04(push, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN);
391 for (i = 0; i < NV50_3D_DMA_COLOR__LEN; ++i)
392 PUSH_DATA(push, fifo->vram);
393
394 BEGIN_NV04(push, NV50_3D(REG_MODE), 1);
395 PUSH_DATA (push, NV50_3D_REG_MODE_STRIPED);
396 BEGIN_NV04(push, NV50_3D(UNK1400_LANES), 1);
397 PUSH_DATA (push, 0xf);
398
399 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", TRUE)) {
400 BEGIN_NV04(push, NV50_3D(WATCHDOG_TIMER), 1);
401 PUSH_DATA (push, 0x18);
402 }
403
404 BEGIN_NV04(push, NV50_3D(RT_CONTROL), 1);
405 PUSH_DATA (push, 1);
406
407 BEGIN_NV04(push, NV50_3D(CSAA_ENABLE), 1);
408 PUSH_DATA (push, 0);
409 BEGIN_NV04(push, NV50_3D(MULTISAMPLE_ENABLE), 1);
410 PUSH_DATA (push, 0);
411 BEGIN_NV04(push, NV50_3D(MULTISAMPLE_MODE), 1);
412 PUSH_DATA (push, NV50_3D_MULTISAMPLE_MODE_MS1);
413 BEGIN_NV04(push, NV50_3D(MULTISAMPLE_CTRL), 1);
414 PUSH_DATA (push, 0);
415 BEGIN_NV04(push, NV50_3D(LINE_LAST_PIXEL), 1);
416 PUSH_DATA (push, 0);
417 BEGIN_NV04(push, NV50_3D(BLEND_SEPARATE_ALPHA), 1);
418 PUSH_DATA (push, 1);
419
420 if (screen->tesla->oclass >= NVA0_3D_CLASS) {
421 BEGIN_NV04(push, SUBC_3D(NVA0_3D_TEX_MISC), 1);
422 PUSH_DATA (push, NVA0_3D_TEX_MISC_SEAMLESS_CUBE_MAP);
423 }
424
425 BEGIN_NV04(push, NV50_3D(SCREEN_Y_CONTROL), 1);
426 PUSH_DATA (push, 0);
427 BEGIN_NV04(push, NV50_3D(WINDOW_OFFSET_X), 2);
428 PUSH_DATA (push, 0);
429 PUSH_DATA (push, 0);
430 BEGIN_NV04(push, NV50_3D(ZCULL_REGION), 1);
431 PUSH_DATA (push, 0x3f);
432
433 BEGIN_NV04(push, NV50_3D(VP_ADDRESS_HIGH), 2);
434 PUSH_DATAh(push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
435 PUSH_DATA (push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
436
437 BEGIN_NV04(push, NV50_3D(FP_ADDRESS_HIGH), 2);
438 PUSH_DATAh(push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
439 PUSH_DATA (push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
440
441 BEGIN_NV04(push, NV50_3D(GP_ADDRESS_HIGH), 2);
442 PUSH_DATAh(push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
443 PUSH_DATA (push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
444
445 BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
446 PUSH_DATAh(push, screen->tls_bo->offset);
447 PUSH_DATA (push, screen->tls_bo->offset);
448 PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
449
450 BEGIN_NV04(push, NV50_3D(STACK_ADDRESS_HIGH), 3);
451 PUSH_DATAh(push, screen->stack_bo->offset);
452 PUSH_DATA (push, screen->stack_bo->offset);
453 PUSH_DATA (push, 4);
454
455 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
456 PUSH_DATAh(push, screen->uniforms->offset + (0 << 16));
457 PUSH_DATA (push, screen->uniforms->offset + (0 << 16));
458 PUSH_DATA (push, (NV50_CB_PVP << 16) | 0x0000);
459
460 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
461 PUSH_DATAh(push, screen->uniforms->offset + (1 << 16));
462 PUSH_DATA (push, screen->uniforms->offset + (1 << 16));
463 PUSH_DATA (push, (NV50_CB_PGP << 16) | 0x0000);
464
465 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
466 PUSH_DATAh(push, screen->uniforms->offset + (2 << 16));
467 PUSH_DATA (push, screen->uniforms->offset + (2 << 16));
468 PUSH_DATA (push, (NV50_CB_PFP << 16) | 0x0000);
469
470 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
471 PUSH_DATAh(push, screen->uniforms->offset + (3 << 16));
472 PUSH_DATA (push, screen->uniforms->offset + (3 << 16));
473 PUSH_DATA (push, (NV50_CB_AUX << 16) | (NV50_CB_AUX_SIZE & 0xffff));
474
475 BEGIN_NI04(push, NV50_3D(SET_PROGRAM_CB), 3);
476 PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf01);
477 PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf21);
478 PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf31);
479
480 /* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
481 BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
482 PUSH_DATA (push, (NV50_CB_AUX_RUNOUT_OFFSET << (8 - 2)) | NV50_CB_AUX);
483 BEGIN_NI04(push, NV50_3D(CB_DATA(0)), 4);
484 PUSH_DATAf(push, 0.0f);
485 PUSH_DATAf(push, 0.0f);
486 PUSH_DATAf(push, 0.0f);
487 PUSH_DATAf(push, 0.0f);
488 BEGIN_NV04(push, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
489 PUSH_DATAh(push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
490 PUSH_DATA (push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
491
492 nv50_upload_ms_info(push);
493
494 /* max TIC (bits 4:8) & TSC bindings, per program type */
495 for (i = 0; i < 3; ++i) {
496 BEGIN_NV04(push, NV50_3D(TEX_LIMITS(i)), 1);
497 PUSH_DATA (push, 0x54);
498 }
499
500 BEGIN_NV04(push, NV50_3D(TIC_ADDRESS_HIGH), 3);
501 PUSH_DATAh(push, screen->txc->offset);
502 PUSH_DATA (push, screen->txc->offset);
503 PUSH_DATA (push, NV50_TIC_MAX_ENTRIES - 1);
504
505 BEGIN_NV04(push, NV50_3D(TSC_ADDRESS_HIGH), 3);
506 PUSH_DATAh(push, screen->txc->offset + 65536);
507 PUSH_DATA (push, screen->txc->offset + 65536);
508 PUSH_DATA (push, NV50_TSC_MAX_ENTRIES - 1);
509
510 BEGIN_NV04(push, NV50_3D(LINKED_TSC), 1);
511 PUSH_DATA (push, 0);
512
513 BEGIN_NV04(push, NV50_3D(CLIP_RECTS_EN), 1);
514 PUSH_DATA (push, 0);
515 BEGIN_NV04(push, NV50_3D(CLIP_RECTS_MODE), 1);
516 PUSH_DATA (push, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY);
517 BEGIN_NV04(push, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
518 for (i = 0; i < 8 * 2; ++i)
519 PUSH_DATA(push, 0);
520 BEGIN_NV04(push, NV50_3D(CLIPID_ENABLE), 1);
521 PUSH_DATA (push, 0);
522
523 BEGIN_NV04(push, NV50_3D(VIEWPORT_TRANSFORM_EN), 1);
524 PUSH_DATA (push, 1);
525 BEGIN_NV04(push, NV50_3D(DEPTH_RANGE_NEAR(0)), 2);
526 PUSH_DATAf(push, 0.0f);
527 PUSH_DATAf(push, 1.0f);
528
529 BEGIN_NV04(push, NV50_3D(VIEW_VOLUME_CLIP_CTRL), 1);
530 #ifdef NV50_SCISSORS_CLIPPING
531 PUSH_DATA (push, 0x0000);
532 #else
533 PUSH_DATA (push, 0x1080);
534 #endif
535
536 BEGIN_NV04(push, NV50_3D(CLEAR_FLAGS), 1);
537 PUSH_DATA (push, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT);
538
539 /* We use scissors instead of exact view volume clipping,
540 * so they're always enabled.
541 */
542 BEGIN_NV04(push, NV50_3D(SCISSOR_ENABLE(0)), 3);
543 PUSH_DATA (push, 1);
544 PUSH_DATA (push, 8192 << 16);
545 PUSH_DATA (push, 8192 << 16);
546
547 BEGIN_NV04(push, NV50_3D(RASTERIZE_ENABLE), 1);
548 PUSH_DATA (push, 1);
549 BEGIN_NV04(push, NV50_3D(POINT_RASTER_RULES), 1);
550 PUSH_DATA (push, NV50_3D_POINT_RASTER_RULES_OGL);
551 BEGIN_NV04(push, NV50_3D(FRAG_COLOR_CLAMP_EN), 1);
552 PUSH_DATA (push, 0x11111111);
553 BEGIN_NV04(push, NV50_3D(EDGEFLAG), 1);
554 PUSH_DATA (push, 1);
555
556 PUSH_KICK (push);
557 }
558
559 static int nv50_tls_alloc(struct nv50_screen *screen, unsigned tls_space,
560 uint64_t *tls_size)
561 {
562 struct nouveau_device *dev = screen->base.device;
563 int ret;
564
565 screen->cur_tls_space = util_next_power_of_two(tls_space / ONE_TEMP_SIZE) *
566 ONE_TEMP_SIZE;
567 if (nouveau_mesa_debug)
568 debug_printf("allocating space for %u temps\n",
569 util_next_power_of_two(tls_space / ONE_TEMP_SIZE));
570 *tls_size = screen->cur_tls_space * util_next_power_of_two(screen->TPs) *
571 screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP;
572
573 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
574 *tls_size, NULL, &screen->tls_bo);
575 if (ret) {
576 NOUVEAU_ERR("Failed to allocate local bo: %d\n", ret);
577 return ret;
578 }
579
580 return 0;
581 }
582
583 int nv50_tls_realloc(struct nv50_screen *screen, unsigned tls_space)
584 {
585 struct nouveau_pushbuf *push = screen->base.pushbuf;
586 int ret;
587 uint64_t tls_size;
588
589 if (tls_space < screen->cur_tls_space)
590 return 0;
591 if (tls_space > screen->max_tls_space) {
592 /* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
593 * LOCAL_WARPS_NO_CLAMP) */
594 NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
595 (unsigned)(tls_space / ONE_TEMP_SIZE),
596 (unsigned)(screen->max_tls_space / ONE_TEMP_SIZE));
597 return -ENOMEM;
598 }
599
600 nouveau_bo_ref(NULL, &screen->tls_bo);
601 ret = nv50_tls_alloc(screen, tls_space, &tls_size);
602 if (ret)
603 return ret;
604
605 BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
606 PUSH_DATAh(push, screen->tls_bo->offset);
607 PUSH_DATA (push, screen->tls_bo->offset);
608 PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
609
610 return 1;
611 }
612
613 struct pipe_screen *
614 nv50_screen_create(struct nouveau_device *dev)
615 {
616 struct nv50_screen *screen;
617 struct pipe_screen *pscreen;
618 struct nouveau_object *chan;
619 uint64_t value;
620 uint32_t tesla_class;
621 unsigned stack_size;
622 int ret;
623
624 screen = CALLOC_STRUCT(nv50_screen);
625 if (!screen)
626 return NULL;
627 pscreen = &screen->base.base;
628
629 ret = nouveau_screen_init(&screen->base, dev);
630 if (ret) {
631 NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret);
632 goto fail;
633 }
634
635 /* TODO: Prevent FIFO prefetch before transfer of index buffers and
636 * admit them to VRAM.
637 */
638 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
639 PIPE_BIND_VERTEX_BUFFER;
640 screen->base.sysmem_bindings |=
641 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
642
643 screen->base.pushbuf->user_priv = screen;
644 screen->base.pushbuf->rsvd_kick = 5;
645
646 chan = screen->base.channel;
647
648 pscreen->destroy = nv50_screen_destroy;
649 pscreen->context_create = nv50_create;
650 pscreen->is_format_supported = nv50_screen_is_format_supported;
651 pscreen->get_param = nv50_screen_get_param;
652 pscreen->get_shader_param = nv50_screen_get_shader_param;
653 pscreen->get_paramf = nv50_screen_get_paramf;
654
655 nv50_screen_init_resource_functions(pscreen);
656
657 if (screen->base.device->chipset < 0x84 ||
658 debug_get_bool_option("NOUVEAU_PMPEG", FALSE)) {
659 /* PMPEG */
660 nouveau_screen_init_vdec(&screen->base);
661 } else if (screen->base.device->chipset < 0x98 ||
662 screen->base.device->chipset == 0xa0) {
663 /* VP2 */
664 screen->base.base.get_video_param = nv84_screen_get_video_param;
665 screen->base.base.is_video_format_supported = nv84_screen_video_supported;
666 } else {
667 /* VP3/4 */
668 screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
669 screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
670 }
671
672 ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096,
673 NULL, &screen->fence.bo);
674 if (ret) {
675 NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret);
676 goto fail;
677 }
678
679 nouveau_bo_map(screen->fence.bo, 0, NULL);
680 screen->fence.map = screen->fence.bo->map;
681 screen->base.fence.emit = nv50_screen_fence_emit;
682 screen->base.fence.update = nv50_screen_fence_update;
683
684 ret = nouveau_object_new(chan, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS,
685 &(struct nv04_notify){ .length = 32 },
686 sizeof(struct nv04_notify), &screen->sync);
687 if (ret) {
688 NOUVEAU_ERR("Failed to allocate notifier: %d\n", ret);
689 goto fail;
690 }
691
692 ret = nouveau_object_new(chan, 0xbeef5039, NV50_M2MF_CLASS,
693 NULL, 0, &screen->m2mf);
694 if (ret) {
695 NOUVEAU_ERR("Failed to allocate PGRAPH context for M2MF: %d\n", ret);
696 goto fail;
697 }
698
699 ret = nouveau_object_new(chan, 0xbeef502d, NV50_2D_CLASS,
700 NULL, 0, &screen->eng2d);
701 if (ret) {
702 NOUVEAU_ERR("Failed to allocate PGRAPH context for 2D: %d\n", ret);
703 goto fail;
704 }
705
706 switch (dev->chipset & 0xf0) {
707 case 0x50:
708 tesla_class = NV50_3D_CLASS;
709 break;
710 case 0x80:
711 case 0x90:
712 tesla_class = NV84_3D_CLASS;
713 break;
714 case 0xa0:
715 switch (dev->chipset) {
716 case 0xa0:
717 case 0xaa:
718 case 0xac:
719 tesla_class = NVA0_3D_CLASS;
720 break;
721 case 0xaf:
722 tesla_class = NVAF_3D_CLASS;
723 break;
724 default:
725 tesla_class = NVA3_3D_CLASS;
726 break;
727 }
728 break;
729 default:
730 NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", dev->chipset);
731 goto fail;
732 }
733 screen->base.class_3d = tesla_class;
734
735 ret = nouveau_object_new(chan, 0xbeef5097, tesla_class,
736 NULL, 0, &screen->tesla);
737 if (ret) {
738 NOUVEAU_ERR("Failed to allocate PGRAPH context for 3D: %d\n", ret);
739 goto fail;
740 }
741
742 /* This over-allocates by a whole code BO. The GP, which would execute at
743 * the end of the last page, would trigger faults. The going theory is that
744 * it prefetches up to a certain amount. This avoids dmesg spam.
745 */
746 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
747 4 << NV50_CODE_BO_SIZE_LOG2, NULL, &screen->code);
748 if (ret) {
749 NOUVEAU_ERR("Failed to allocate code bo: %d\n", ret);
750 goto fail;
751 }
752
753 nouveau_heap_init(&screen->vp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
754 nouveau_heap_init(&screen->gp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
755 nouveau_heap_init(&screen->fp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
756
757 nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
758
759 screen->TPs = util_bitcount(value & 0xffff);
760 screen->MPsInTP = util_bitcount((value >> 24) & 0xf);
761
762 stack_size = util_next_power_of_two(screen->TPs) * screen->MPsInTP *
763 STACK_WARPS_ALLOC * 64 * 8;
764
765 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, stack_size, NULL,
766 &screen->stack_bo);
767 if (ret) {
768 NOUVEAU_ERR("Failed to allocate stack bo: %d\n", ret);
769 goto fail;
770 }
771
772 uint64_t size_of_one_temp = util_next_power_of_two(screen->TPs) *
773 screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP *
774 ONE_TEMP_SIZE;
775 screen->max_tls_space = dev->vram_size / size_of_one_temp * ONE_TEMP_SIZE;
776 screen->max_tls_space /= 2; /* half of vram */
777
778 /* hw can address max 64 KiB */
779 screen->max_tls_space = MIN2(screen->max_tls_space, 64 << 10);
780
781 uint64_t tls_size;
782 unsigned tls_space = 4/*temps*/ * ONE_TEMP_SIZE;
783 ret = nv50_tls_alloc(screen, tls_space, &tls_size);
784 if (ret)
785 goto fail;
786
787 if (nouveau_mesa_debug)
788 debug_printf("TPs = %u, MPsInTP = %u, VRAM = %"PRIu64" MiB, tls_size = %"PRIu64" KiB\n",
789 screen->TPs, screen->MPsInTP, dev->vram_size >> 20, tls_size >> 10);
790
791 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 4 << 16, NULL,
792 &screen->uniforms);
793 if (ret) {
794 NOUVEAU_ERR("Failed to allocate uniforms bo: %d\n", ret);
795 goto fail;
796 }
797
798 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 3 << 16, NULL,
799 &screen->txc);
800 if (ret) {
801 NOUVEAU_ERR("Failed to allocate TIC/TSC bo: %d\n", ret);
802 goto fail;
803 }
804
805 screen->tic.entries = CALLOC(4096, sizeof(void *));
806 screen->tsc.entries = screen->tic.entries + 2048;
807
808 if (!nv50_blitter_create(screen))
809 goto fail;
810
811 nv50_screen_init_hwctx(screen);
812
813 nouveau_fence_new(&screen->base, &screen->base.fence.current, FALSE);
814
815 return pscreen;
816
817 fail:
818 nv50_screen_destroy(pscreen);
819 return NULL;
820 }
821
822 int
823 nv50_screen_tic_alloc(struct nv50_screen *screen, void *entry)
824 {
825 int i = screen->tic.next;
826
827 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
828 i = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
829
830 screen->tic.next = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
831
832 if (screen->tic.entries[i])
833 nv50_tic_entry(screen->tic.entries[i])->id = -1;
834
835 screen->tic.entries[i] = entry;
836 return i;
837 }
838
839 int
840 nv50_screen_tsc_alloc(struct nv50_screen *screen, void *entry)
841 {
842 int i = screen->tsc.next;
843
844 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
845 i = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
846
847 screen->tsc.next = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
848
849 if (screen->tsc.entries[i])
850 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
851
852 screen->tsc.entries[i] = entry;
853 return i;
854 }