nv50,nvc0: add RGBX16/32_FLOAT formats
[mesa.git] / src / gallium / drivers / 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 BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
18 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
19 * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
20 * 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_context.h"
28 #include "nv50_screen.h"
29
30 #include "nouveau/nv_object.xml.h"
31 #include <errno.h>
32
33 #ifndef NOUVEAU_GETPARAM_GRAPH_UNITS
34 # define NOUVEAU_GETPARAM_GRAPH_UNITS 13
35 #endif
36
37 /* affected by LOCAL_WARPS_LOG_ALLOC / LOCAL_WARPS_NO_CLAMP */
38 #define LOCAL_WARPS_ALLOC 32
39 /* affected by STACK_WARPS_LOG_ALLOC / STACK_WARPS_NO_CLAMP */
40 #define STACK_WARPS_ALLOC 32
41
42 #define THREADS_IN_WARP 32
43
44 #define ONE_TEMP_SIZE (4/*vector*/ * sizeof(float))
45
46 static boolean
47 nv50_screen_is_format_supported(struct pipe_screen *pscreen,
48 enum pipe_format format,
49 enum pipe_texture_target target,
50 unsigned sample_count,
51 unsigned bindings)
52 {
53 if (sample_count > 8)
54 return FALSE;
55 if (!(0x117 & (1 << sample_count))) /* 0, 1, 2, 4 or 8 */
56 return FALSE;
57 if (sample_count == 8 && util_format_get_blocksizebits(format) >= 128)
58 return FALSE;
59
60 if (!util_format_is_supported(format, bindings))
61 return FALSE;
62
63 switch (format) {
64 case PIPE_FORMAT_Z16_UNORM:
65 if (nv50_screen(pscreen)->tesla->oclass < NVA0_3D_CLASS)
66 return FALSE;
67 break;
68 default:
69 break;
70 }
71
72 /* transfers & shared are always supported */
73 bindings &= ~(PIPE_BIND_TRANSFER_READ |
74 PIPE_BIND_TRANSFER_WRITE |
75 PIPE_BIND_SHARED);
76
77 return (nv50_format_table[format].usage & bindings) == bindings;
78 }
79
80 static int
81 nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
82 {
83 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
84
85 switch (param) {
86 case PIPE_CAP_MAX_COMBINED_SAMPLERS:
87 return 64;
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_ANISOTROPIC_FILTER:
105 case PIPE_CAP_SCALED_RESOLVE:
106 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
107 return 1;
108 case PIPE_CAP_SEAMLESS_CUBE_MAP:
109 return nv50_screen(pscreen)->tesla->oclass >= NVA0_3D_CLASS;
110 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
111 return 0;
112 case PIPE_CAP_CUBE_MAP_ARRAY:
113 return 0;
114 /*
115 return nv50_screen(pscreen)->tesla->oclass >= NVA3_3D_CLASS;
116 */
117 case PIPE_CAP_TWO_SIDED_STENCIL:
118 case PIPE_CAP_DEPTH_CLIP_DISABLE:
119 case PIPE_CAP_POINT_SPRITE:
120 return 1;
121 case PIPE_CAP_SM3:
122 return 1;
123 case PIPE_CAP_GLSL_FEATURE_LEVEL:
124 return 140;
125 case PIPE_CAP_MAX_RENDER_TARGETS:
126 return 8;
127 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
128 return 1;
129 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
130 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
131 case PIPE_CAP_VERTEX_COLOR_CLAMPED:
132 return 1;
133 case PIPE_CAP_QUERY_TIMESTAMP:
134 case PIPE_CAP_QUERY_TIME_ELAPSED:
135 case PIPE_CAP_OCCLUSION_QUERY:
136 return 1;
137 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
138 return 4;
139 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
140 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
141 return 64;
142 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
143 return (class_3d >= NVA0_3D_CLASS) ? 1 : 0;
144 case PIPE_CAP_BLEND_EQUATION_SEPARATE:
145 case PIPE_CAP_INDEP_BLEND_ENABLE:
146 return 1;
147 case PIPE_CAP_INDEP_BLEND_FUNC:
148 return nv50_screen(pscreen)->tesla->oclass >= NVA3_3D_CLASS;
149 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
150 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
151 return 1;
152 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
153 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
154 return 0;
155 case PIPE_CAP_SHADER_STENCIL_EXPORT:
156 return 0;
157 case PIPE_CAP_PRIMITIVE_RESTART:
158 case PIPE_CAP_TGSI_INSTANCEID:
159 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
160 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
161 case PIPE_CAP_CONDITIONAL_RENDER:
162 case PIPE_CAP_TEXTURE_BARRIER:
163 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
164 case PIPE_CAP_START_INSTANCE:
165 return 1;
166 case PIPE_CAP_TGSI_CAN_COMPACT_VARYINGS:
167 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
168 return 0; /* state trackers will know better */
169 case PIPE_CAP_USER_CONSTANT_BUFFERS:
170 case PIPE_CAP_USER_INDEX_BUFFERS:
171 case PIPE_CAP_USER_VERTEX_BUFFERS:
172 return 1;
173 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
174 return 256;
175 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
176 return 1; /* 256 for binding as RT, but that's not possible in GL */
177 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
178 return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
179 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
180 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
181 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
182 case PIPE_CAP_TGSI_TEXCOORD:
183 case PIPE_CAP_TEXTURE_MULTISAMPLE:
184 return 0;
185 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
186 return 1;
187 case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
188 return 0;
189 default:
190 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
191 return 0;
192 }
193 }
194
195 static int
196 nv50_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
197 enum pipe_shader_cap param)
198 {
199 switch (shader) {
200 case PIPE_SHADER_VERTEX:
201 case PIPE_SHADER_GEOMETRY:
202 case PIPE_SHADER_FRAGMENT:
203 break;
204 default:
205 return 0;
206 }
207
208 switch (param) {
209 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
210 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
211 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
212 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
213 return 16384;
214 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
215 return 4;
216 case PIPE_SHADER_CAP_MAX_INPUTS:
217 if (shader == PIPE_SHADER_VERTEX)
218 return 32;
219 return 0x300 / 16;
220 case PIPE_SHADER_CAP_MAX_CONSTS:
221 return 65536 / 16;
222 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
223 return NV50_MAX_PIPE_CONSTBUFS;
224 case PIPE_SHADER_CAP_MAX_ADDRS:
225 return 1;
226 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
227 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
228 return shader != PIPE_SHADER_FRAGMENT;
229 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
230 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
231 return 1;
232 case PIPE_SHADER_CAP_MAX_PREDS:
233 return 0;
234 case PIPE_SHADER_CAP_MAX_TEMPS:
235 return nv50_screen(pscreen)->max_tls_space / ONE_TEMP_SIZE;
236 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
237 return 1;
238 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
239 return 0;
240 case PIPE_SHADER_CAP_SUBROUTINES:
241 return 0; /* please inline, or provide function declarations */
242 case PIPE_SHADER_CAP_INTEGERS:
243 return 1;
244 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
245 return 32;
246 default:
247 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
248 return 0;
249 }
250 }
251
252 static float
253 nv50_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
254 {
255 switch (param) {
256 case PIPE_CAPF_MAX_LINE_WIDTH:
257 case PIPE_CAPF_MAX_LINE_WIDTH_AA:
258 return 10.0f;
259 case PIPE_CAPF_MAX_POINT_WIDTH:
260 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
261 return 64.0f;
262 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
263 return 16.0f;
264 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
265 return 4.0f;
266 default:
267 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
268 return 0.0f;
269 }
270 }
271
272 static void
273 nv50_screen_destroy(struct pipe_screen *pscreen)
274 {
275 struct nv50_screen *screen = nv50_screen(pscreen);
276
277 if (screen->base.fence.current) {
278 nouveau_fence_wait(screen->base.fence.current);
279 nouveau_fence_ref (NULL, &screen->base.fence.current);
280 }
281 if (screen->base.pushbuf)
282 screen->base.pushbuf->user_priv = NULL;
283
284 if (screen->blitter)
285 nv50_blitter_destroy(screen);
286
287 nouveau_bo_ref(NULL, &screen->code);
288 nouveau_bo_ref(NULL, &screen->tls_bo);
289 nouveau_bo_ref(NULL, &screen->stack_bo);
290 nouveau_bo_ref(NULL, &screen->txc);
291 nouveau_bo_ref(NULL, &screen->uniforms);
292 nouveau_bo_ref(NULL, &screen->fence.bo);
293
294 nouveau_heap_destroy(&screen->vp_code_heap);
295 nouveau_heap_destroy(&screen->gp_code_heap);
296 nouveau_heap_destroy(&screen->fp_code_heap);
297
298 FREE(screen->tic.entries);
299
300 nouveau_object_del(&screen->tesla);
301 nouveau_object_del(&screen->eng2d);
302 nouveau_object_del(&screen->m2mf);
303 nouveau_object_del(&screen->sync);
304
305 nouveau_screen_fini(&screen->base);
306
307 FREE(screen);
308 }
309
310 static void
311 nv50_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
312 {
313 struct nv50_screen *screen = nv50_screen(pscreen);
314 struct nouveau_pushbuf *push = screen->base.pushbuf;
315
316 /* we need to do it after possible flush in MARK_RING */
317 *sequence = ++screen->base.fence.sequence;
318
319 PUSH_DATA (push, NV50_FIFO_PKHDR(NV50_3D(QUERY_ADDRESS_HIGH), 4));
320 PUSH_DATAh(push, screen->fence.bo->offset);
321 PUSH_DATA (push, screen->fence.bo->offset);
322 PUSH_DATA (push, *sequence);
323 PUSH_DATA (push, NV50_3D_QUERY_GET_MODE_WRITE_UNK0 |
324 NV50_3D_QUERY_GET_UNK4 |
325 NV50_3D_QUERY_GET_UNIT_CROP |
326 NV50_3D_QUERY_GET_TYPE_QUERY |
327 NV50_3D_QUERY_GET_QUERY_SELECT_ZERO |
328 NV50_3D_QUERY_GET_SHORT);
329 }
330
331 static u32
332 nv50_screen_fence_update(struct pipe_screen *pscreen)
333 {
334 return nv50_screen(pscreen)->fence.map[0];
335 }
336
337 static void
338 nv50_screen_init_hwctx(struct nv50_screen *screen)
339 {
340 struct nouveau_pushbuf *push = screen->base.pushbuf;
341 struct nv04_fifo *fifo;
342 unsigned i;
343
344 fifo = (struct nv04_fifo *)screen->base.channel->data;
345
346 BEGIN_NV04(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
347 PUSH_DATA (push, screen->m2mf->handle);
348 BEGIN_NV04(push, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY), 3);
349 PUSH_DATA (push, screen->sync->handle);
350 PUSH_DATA (push, fifo->vram);
351 PUSH_DATA (push, fifo->vram);
352
353 BEGIN_NV04(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
354 PUSH_DATA (push, screen->eng2d->handle);
355 BEGIN_NV04(push, NV50_2D(DMA_NOTIFY), 4);
356 PUSH_DATA (push, screen->sync->handle);
357 PUSH_DATA (push, fifo->vram);
358 PUSH_DATA (push, fifo->vram);
359 PUSH_DATA (push, fifo->vram);
360 BEGIN_NV04(push, NV50_2D(OPERATION), 1);
361 PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
362 BEGIN_NV04(push, NV50_2D(CLIP_ENABLE), 1);
363 PUSH_DATA (push, 0);
364 BEGIN_NV04(push, NV50_2D(COLOR_KEY_ENABLE), 1);
365 PUSH_DATA (push, 0);
366 BEGIN_NV04(push, SUBC_2D(0x0888), 1);
367 PUSH_DATA (push, 1);
368
369 BEGIN_NV04(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
370 PUSH_DATA (push, screen->tesla->handle);
371
372 BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
373 PUSH_DATA (push, NV50_3D_COND_MODE_ALWAYS);
374
375 BEGIN_NV04(push, NV50_3D(DMA_NOTIFY), 1);
376 PUSH_DATA (push, screen->sync->handle);
377 BEGIN_NV04(push, NV50_3D(DMA_ZETA), 11);
378 for (i = 0; i < 11; ++i)
379 PUSH_DATA(push, fifo->vram);
380 BEGIN_NV04(push, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN);
381 for (i = 0; i < NV50_3D_DMA_COLOR__LEN; ++i)
382 PUSH_DATA(push, fifo->vram);
383
384 BEGIN_NV04(push, NV50_3D(REG_MODE), 1);
385 PUSH_DATA (push, NV50_3D_REG_MODE_STRIPED);
386 BEGIN_NV04(push, NV50_3D(UNK1400_LANES), 1);
387 PUSH_DATA (push, 0xf);
388
389 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", TRUE)) {
390 BEGIN_NV04(push, NV50_3D(WATCHDOG_TIMER), 1);
391 PUSH_DATA (push, 0x18);
392 }
393
394 BEGIN_NV04(push, NV50_3D(RT_CONTROL), 1);
395 PUSH_DATA (push, 1);
396
397 BEGIN_NV04(push, NV50_3D(CSAA_ENABLE), 1);
398 PUSH_DATA (push, 0);
399 BEGIN_NV04(push, NV50_3D(MULTISAMPLE_ENABLE), 1);
400 PUSH_DATA (push, 0);
401 BEGIN_NV04(push, NV50_3D(MULTISAMPLE_MODE), 1);
402 PUSH_DATA (push, NV50_3D_MULTISAMPLE_MODE_MS1);
403 BEGIN_NV04(push, NV50_3D(MULTISAMPLE_CTRL), 1);
404 PUSH_DATA (push, 0);
405 BEGIN_NV04(push, NV50_3D(LINE_LAST_PIXEL), 1);
406 PUSH_DATA (push, 0);
407 BEGIN_NV04(push, NV50_3D(BLEND_SEPARATE_ALPHA), 1);
408 PUSH_DATA (push, 1);
409
410 if (screen->tesla->oclass >= NVA0_3D_CLASS) {
411 BEGIN_NV04(push, SUBC_3D(NVA0_3D_TEX_MISC), 1);
412 PUSH_DATA (push, NVA0_3D_TEX_MISC_SEAMLESS_CUBE_MAP);
413 }
414
415 BEGIN_NV04(push, NV50_3D(SCREEN_Y_CONTROL), 1);
416 PUSH_DATA (push, 0);
417 BEGIN_NV04(push, NV50_3D(WINDOW_OFFSET_X), 2);
418 PUSH_DATA (push, 0);
419 PUSH_DATA (push, 0);
420 BEGIN_NV04(push, NV50_3D(ZCULL_REGION), 1);
421 PUSH_DATA (push, 0x3f);
422
423 BEGIN_NV04(push, NV50_3D(VP_ADDRESS_HIGH), 2);
424 PUSH_DATAh(push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
425 PUSH_DATA (push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
426
427 BEGIN_NV04(push, NV50_3D(FP_ADDRESS_HIGH), 2);
428 PUSH_DATAh(push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
429 PUSH_DATA (push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
430
431 BEGIN_NV04(push, NV50_3D(GP_ADDRESS_HIGH), 2);
432 PUSH_DATAh(push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
433 PUSH_DATA (push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
434
435 BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
436 PUSH_DATAh(push, screen->tls_bo->offset);
437 PUSH_DATA (push, screen->tls_bo->offset);
438 PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
439
440 BEGIN_NV04(push, NV50_3D(STACK_ADDRESS_HIGH), 3);
441 PUSH_DATAh(push, screen->stack_bo->offset);
442 PUSH_DATA (push, screen->stack_bo->offset);
443 PUSH_DATA (push, 4);
444
445 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
446 PUSH_DATAh(push, screen->uniforms->offset + (0 << 16));
447 PUSH_DATA (push, screen->uniforms->offset + (0 << 16));
448 PUSH_DATA (push, (NV50_CB_PVP << 16) | 0x0000);
449
450 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
451 PUSH_DATAh(push, screen->uniforms->offset + (1 << 16));
452 PUSH_DATA (push, screen->uniforms->offset + (1 << 16));
453 PUSH_DATA (push, (NV50_CB_PGP << 16) | 0x0000);
454
455 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
456 PUSH_DATAh(push, screen->uniforms->offset + (2 << 16));
457 PUSH_DATA (push, screen->uniforms->offset + (2 << 16));
458 PUSH_DATA (push, (NV50_CB_PFP << 16) | 0x0000);
459
460 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
461 PUSH_DATAh(push, screen->uniforms->offset + (3 << 16));
462 PUSH_DATA (push, screen->uniforms->offset + (3 << 16));
463 PUSH_DATA (push, (NV50_CB_AUX << 16) | 0x0200);
464
465 BEGIN_NI04(push, NV50_3D(SET_PROGRAM_CB), 3);
466 PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf01);
467 PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf21);
468 PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf31);
469
470 /* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
471 BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
472 PUSH_DATA (push, ((1 << 9) << 6) | NV50_CB_AUX);
473 BEGIN_NI04(push, NV50_3D(CB_DATA(0)), 4);
474 PUSH_DATAf(push, 0.0f);
475 PUSH_DATAf(push, 0.0f);
476 PUSH_DATAf(push, 0.0f);
477 PUSH_DATAf(push, 0.0f);
478 BEGIN_NV04(push, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
479 PUSH_DATAh(push, screen->uniforms->offset + (3 << 16) + (1 << 9));
480 PUSH_DATA (push, screen->uniforms->offset + (3 << 16) + (1 << 9));
481
482 /* max TIC (bits 4:8) & TSC bindings, per program type */
483 for (i = 0; i < 3; ++i) {
484 BEGIN_NV04(push, NV50_3D(TEX_LIMITS(i)), 1);
485 PUSH_DATA (push, 0x54);
486 }
487
488 BEGIN_NV04(push, NV50_3D(TIC_ADDRESS_HIGH), 3);
489 PUSH_DATAh(push, screen->txc->offset);
490 PUSH_DATA (push, screen->txc->offset);
491 PUSH_DATA (push, NV50_TIC_MAX_ENTRIES - 1);
492
493 BEGIN_NV04(push, NV50_3D(TSC_ADDRESS_HIGH), 3);
494 PUSH_DATAh(push, screen->txc->offset + 65536);
495 PUSH_DATA (push, screen->txc->offset + 65536);
496 PUSH_DATA (push, NV50_TSC_MAX_ENTRIES - 1);
497
498 BEGIN_NV04(push, NV50_3D(LINKED_TSC), 1);
499 PUSH_DATA (push, 0);
500
501 BEGIN_NV04(push, NV50_3D(CLIP_RECTS_EN), 1);
502 PUSH_DATA (push, 0);
503 BEGIN_NV04(push, NV50_3D(CLIP_RECTS_MODE), 1);
504 PUSH_DATA (push, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY);
505 BEGIN_NV04(push, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
506 for (i = 0; i < 8 * 2; ++i)
507 PUSH_DATA(push, 0);
508 BEGIN_NV04(push, NV50_3D(CLIPID_ENABLE), 1);
509 PUSH_DATA (push, 0);
510
511 BEGIN_NV04(push, NV50_3D(VIEWPORT_TRANSFORM_EN), 1);
512 PUSH_DATA (push, 1);
513 BEGIN_NV04(push, NV50_3D(DEPTH_RANGE_NEAR(0)), 2);
514 PUSH_DATAf(push, 0.0f);
515 PUSH_DATAf(push, 1.0f);
516
517 BEGIN_NV04(push, NV50_3D(VIEW_VOLUME_CLIP_CTRL), 1);
518 #ifdef NV50_SCISSORS_CLIPPING
519 PUSH_DATA (push, 0x0000);
520 #else
521 PUSH_DATA (push, 0x1080);
522 #endif
523
524 BEGIN_NV04(push, NV50_3D(CLEAR_FLAGS), 1);
525 PUSH_DATA (push, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT);
526
527 /* We use scissors instead of exact view volume clipping,
528 * so they're always enabled.
529 */
530 BEGIN_NV04(push, NV50_3D(SCISSOR_ENABLE(0)), 3);
531 PUSH_DATA (push, 1);
532 PUSH_DATA (push, 8192 << 16);
533 PUSH_DATA (push, 8192 << 16);
534
535 BEGIN_NV04(push, NV50_3D(RASTERIZE_ENABLE), 1);
536 PUSH_DATA (push, 1);
537 BEGIN_NV04(push, NV50_3D(POINT_RASTER_RULES), 1);
538 PUSH_DATA (push, NV50_3D_POINT_RASTER_RULES_OGL);
539 BEGIN_NV04(push, NV50_3D(FRAG_COLOR_CLAMP_EN), 1);
540 PUSH_DATA (push, 0x11111111);
541 BEGIN_NV04(push, NV50_3D(EDGEFLAG), 1);
542 PUSH_DATA (push, 1);
543
544 PUSH_KICK (push);
545 }
546
547 static int nv50_tls_alloc(struct nv50_screen *screen, unsigned tls_space,
548 uint64_t *tls_size)
549 {
550 struct nouveau_device *dev = screen->base.device;
551 int ret;
552
553 screen->cur_tls_space = util_next_power_of_two(tls_space / ONE_TEMP_SIZE) *
554 ONE_TEMP_SIZE;
555 if (nouveau_mesa_debug)
556 debug_printf("allocating space for %u temps\n",
557 util_next_power_of_two(tls_space / ONE_TEMP_SIZE));
558 *tls_size = screen->cur_tls_space * util_next_power_of_two(screen->TPs) *
559 screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP;
560
561 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
562 *tls_size, NULL, &screen->tls_bo);
563 if (ret) {
564 NOUVEAU_ERR("Failed to allocate local bo: %d\n", ret);
565 return ret;
566 }
567
568 return 0;
569 }
570
571 int nv50_tls_realloc(struct nv50_screen *screen, unsigned tls_space)
572 {
573 struct nouveau_pushbuf *push = screen->base.pushbuf;
574 int ret;
575 uint64_t tls_size;
576
577 if (tls_space < screen->cur_tls_space)
578 return 0;
579 if (tls_space > screen->max_tls_space) {
580 /* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
581 * LOCAL_WARPS_NO_CLAMP) */
582 NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
583 (unsigned)(tls_space / ONE_TEMP_SIZE),
584 (unsigned)(screen->max_tls_space / ONE_TEMP_SIZE));
585 return -ENOMEM;
586 }
587
588 nouveau_bo_ref(NULL, &screen->tls_bo);
589 ret = nv50_tls_alloc(screen, tls_space, &tls_size);
590 if (ret)
591 return ret;
592
593 BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
594 PUSH_DATAh(push, screen->tls_bo->offset);
595 PUSH_DATA (push, screen->tls_bo->offset);
596 PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
597
598 return 1;
599 }
600
601 struct pipe_screen *
602 nv50_screen_create(struct nouveau_device *dev)
603 {
604 struct nv50_screen *screen;
605 struct pipe_screen *pscreen;
606 struct nouveau_object *chan;
607 uint64_t value;
608 uint32_t tesla_class;
609 unsigned stack_size;
610 int ret;
611
612 screen = CALLOC_STRUCT(nv50_screen);
613 if (!screen)
614 return NULL;
615 pscreen = &screen->base.base;
616
617 ret = nouveau_screen_init(&screen->base, dev);
618 if (ret) {
619 NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret);
620 goto fail;
621 }
622
623 /* TODO: Prevent FIFO prefetch before transfer of index buffers and
624 * admit them to VRAM.
625 */
626 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
627 PIPE_BIND_VERTEX_BUFFER;
628 screen->base.sysmem_bindings |=
629 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
630
631 screen->base.pushbuf->user_priv = screen;
632 screen->base.pushbuf->rsvd_kick = 5;
633
634 chan = screen->base.channel;
635
636 pscreen->destroy = nv50_screen_destroy;
637 pscreen->context_create = nv50_create;
638 pscreen->is_format_supported = nv50_screen_is_format_supported;
639 pscreen->get_param = nv50_screen_get_param;
640 pscreen->get_shader_param = nv50_screen_get_shader_param;
641 pscreen->get_paramf = nv50_screen_get_paramf;
642
643 nv50_screen_init_resource_functions(pscreen);
644
645 nouveau_screen_init_vdec(&screen->base);
646
647 ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096,
648 NULL, &screen->fence.bo);
649 if (ret) {
650 NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret);
651 goto fail;
652 }
653
654 nouveau_bo_map(screen->fence.bo, 0, NULL);
655 screen->fence.map = screen->fence.bo->map;
656 screen->base.fence.emit = nv50_screen_fence_emit;
657 screen->base.fence.update = nv50_screen_fence_update;
658
659 ret = nouveau_object_new(chan, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS,
660 &(struct nv04_notify){ .length = 32 },
661 sizeof(struct nv04_notify), &screen->sync);
662 if (ret) {
663 NOUVEAU_ERR("Failed to allocate notifier: %d\n", ret);
664 goto fail;
665 }
666
667 ret = nouveau_object_new(chan, 0xbeef5039, NV50_M2MF_CLASS,
668 NULL, 0, &screen->m2mf);
669 if (ret) {
670 NOUVEAU_ERR("Failed to allocate PGRAPH context for M2MF: %d\n", ret);
671 goto fail;
672 }
673
674 ret = nouveau_object_new(chan, 0xbeef502d, NV50_2D_CLASS,
675 NULL, 0, &screen->eng2d);
676 if (ret) {
677 NOUVEAU_ERR("Failed to allocate PGRAPH context for 2D: %d\n", ret);
678 goto fail;
679 }
680
681 switch (dev->chipset & 0xf0) {
682 case 0x50:
683 tesla_class = NV50_3D_CLASS;
684 break;
685 case 0x80:
686 case 0x90:
687 tesla_class = NV84_3D_CLASS;
688 break;
689 case 0xa0:
690 switch (dev->chipset) {
691 case 0xa0:
692 case 0xaa:
693 case 0xac:
694 tesla_class = NVA0_3D_CLASS;
695 break;
696 case 0xaf:
697 tesla_class = NVAF_3D_CLASS;
698 break;
699 default:
700 tesla_class = NVA3_3D_CLASS;
701 break;
702 }
703 break;
704 default:
705 NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", dev->chipset);
706 goto fail;
707 }
708 screen->base.class_3d = tesla_class;
709
710 ret = nouveau_object_new(chan, 0xbeef5097, tesla_class,
711 NULL, 0, &screen->tesla);
712 if (ret) {
713 NOUVEAU_ERR("Failed to allocate PGRAPH context for 3D: %d\n", ret);
714 goto fail;
715 }
716
717 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
718 3 << NV50_CODE_BO_SIZE_LOG2, NULL, &screen->code);
719 if (ret) {
720 NOUVEAU_ERR("Failed to allocate code bo: %d\n", ret);
721 goto fail;
722 }
723
724 nouveau_heap_init(&screen->vp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
725 nouveau_heap_init(&screen->gp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
726 nouveau_heap_init(&screen->fp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
727
728 nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
729
730 screen->TPs = util_bitcount(value & 0xffff);
731 screen->MPsInTP = util_bitcount((value >> 24) & 0xf);
732
733 stack_size = util_next_power_of_two(screen->TPs) * screen->MPsInTP *
734 STACK_WARPS_ALLOC * 64 * 8;
735
736 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, stack_size, NULL,
737 &screen->stack_bo);
738 if (ret) {
739 NOUVEAU_ERR("Failed to allocate stack bo: %d\n", ret);
740 goto fail;
741 }
742
743 uint64_t size_of_one_temp = util_next_power_of_two(screen->TPs) *
744 screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP *
745 ONE_TEMP_SIZE;
746 screen->max_tls_space = dev->vram_size / size_of_one_temp * ONE_TEMP_SIZE;
747 screen->max_tls_space /= 2; /* half of vram */
748
749 /* hw can address max 64 KiB */
750 screen->max_tls_space = MIN2(screen->max_tls_space, 64 << 10);
751
752 uint64_t tls_size;
753 unsigned tls_space = 4/*temps*/ * ONE_TEMP_SIZE;
754 ret = nv50_tls_alloc(screen, tls_space, &tls_size);
755 if (ret)
756 goto fail;
757
758 if (nouveau_mesa_debug)
759 debug_printf("TPs = %u, MPsInTP = %u, VRAM = %"PRIu64" MiB, tls_size = %"PRIu64" KiB\n",
760 screen->TPs, screen->MPsInTP, dev->vram_size >> 20, tls_size >> 10);
761
762 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 4 << 16, NULL,
763 &screen->uniforms);
764 if (ret) {
765 NOUVEAU_ERR("Failed to allocate uniforms bo: %d\n", ret);
766 goto fail;
767 }
768
769 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 3 << 16, NULL,
770 &screen->txc);
771 if (ret) {
772 NOUVEAU_ERR("Failed to allocate TIC/TSC bo: %d\n", ret);
773 goto fail;
774 }
775
776 screen->tic.entries = CALLOC(4096, sizeof(void *));
777 screen->tsc.entries = screen->tic.entries + 2048;
778
779 if (!nv50_blitter_create(screen))
780 goto fail;
781
782 nv50_screen_init_hwctx(screen);
783
784 nouveau_fence_new(&screen->base, &screen->base.fence.current, FALSE);
785
786 return pscreen;
787
788 fail:
789 nv50_screen_destroy(pscreen);
790 return NULL;
791 }
792
793 int
794 nv50_screen_tic_alloc(struct nv50_screen *screen, void *entry)
795 {
796 int i = screen->tic.next;
797
798 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
799 i = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
800
801 screen->tic.next = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
802
803 if (screen->tic.entries[i])
804 nv50_tic_entry(screen->tic.entries[i])->id = -1;
805
806 screen->tic.entries[i] = entry;
807 return i;
808 }
809
810 int
811 nv50_screen_tsc_alloc(struct nv50_screen *screen, void *entry)
812 {
813 int i = screen->tsc.next;
814
815 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
816 i = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
817
818 screen->tsc.next = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
819
820 if (screen->tsc.entries[i])
821 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
822
823 screen->tsc.entries[i] = entry;
824 return i;
825 }