nv50: add support for texelFetch'ing MS textures, ARB_texture_multisample
[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_COMBINED_SAMPLERS:
89 return 64;
90 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
91 return 14;
92 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
93 return 12;
94 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
95 return 14;
96 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
97 return 512;
98 case PIPE_CAP_MIN_TEXEL_OFFSET:
99 return -8;
100 case PIPE_CAP_MAX_TEXEL_OFFSET:
101 return 7;
102 case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
103 case PIPE_CAP_TEXTURE_SWIZZLE:
104 case PIPE_CAP_TEXTURE_SHADOW_MAP:
105 case PIPE_CAP_NPOT_TEXTURES:
106 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
107 case PIPE_CAP_ANISOTROPIC_FILTER:
108 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
109 return 1;
110 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
111 return 65536;
112 case PIPE_CAP_SEAMLESS_CUBE_MAP:
113 return nv50_screen(pscreen)->tesla->oclass >= NVA0_3D_CLASS;
114 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
115 return 0;
116 case PIPE_CAP_CUBE_MAP_ARRAY:
117 return 0;
118 /*
119 return nv50_screen(pscreen)->tesla->oclass >= NVA3_3D_CLASS;
120 */
121 case PIPE_CAP_TWO_SIDED_STENCIL:
122 case PIPE_CAP_DEPTH_CLIP_DISABLE:
123 case PIPE_CAP_POINT_SPRITE:
124 return 1;
125 case PIPE_CAP_SM3:
126 return 1;
127 case PIPE_CAP_GLSL_FEATURE_LEVEL:
128 return 140;
129 case PIPE_CAP_MAX_RENDER_TARGETS:
130 return 8;
131 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
132 return 1;
133 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
134 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
135 case PIPE_CAP_VERTEX_COLOR_CLAMPED:
136 return 1;
137 case PIPE_CAP_QUERY_TIMESTAMP:
138 case PIPE_CAP_QUERY_TIME_ELAPSED:
139 case PIPE_CAP_OCCLUSION_QUERY:
140 return 1;
141 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
142 return 4;
143 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
144 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
145 return 64;
146 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
147 return (class_3d >= NVA0_3D_CLASS) ? 1 : 0;
148 case PIPE_CAP_BLEND_EQUATION_SEPARATE:
149 case PIPE_CAP_INDEP_BLEND_ENABLE:
150 return 1;
151 case PIPE_CAP_INDEP_BLEND_FUNC:
152 return nv50_screen(pscreen)->tesla->oclass >= NVA3_3D_CLASS;
153 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
154 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
155 return 1;
156 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
157 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
158 return 0;
159 case PIPE_CAP_SHADER_STENCIL_EXPORT:
160 return 0;
161 case PIPE_CAP_PRIMITIVE_RESTART:
162 case PIPE_CAP_TGSI_INSTANCEID:
163 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
164 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
165 case PIPE_CAP_CONDITIONAL_RENDER:
166 case PIPE_CAP_TEXTURE_BARRIER:
167 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
168 case PIPE_CAP_START_INSTANCE:
169 return 1;
170 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
171 return 0; /* state trackers will know better */
172 case PIPE_CAP_USER_CONSTANT_BUFFERS:
173 case PIPE_CAP_USER_INDEX_BUFFERS:
174 case PIPE_CAP_USER_VERTEX_BUFFERS:
175 return 1;
176 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
177 return 256;
178 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
179 return 1; /* 256 for binding as RT, but that's not possible in GL */
180 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
181 return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
182 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
183 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
184 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
185 case PIPE_CAP_TGSI_TEXCOORD:
186 return 0;
187 case PIPE_CAP_TEXTURE_MULTISAMPLE:
188 return 1;
189 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
190 return 1;
191 case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
192 return 0;
193 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
194 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
195 case PIPE_CAP_ENDIANNESS:
196 return PIPE_ENDIAN_LITTLE;
197 case PIPE_CAP_TGSI_VS_LAYER:
198 return 0;
199 default:
200 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
201 return 0;
202 }
203 }
204
205 static int
206 nv50_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
207 enum pipe_shader_cap param)
208 {
209 switch (shader) {
210 case PIPE_SHADER_VERTEX:
211 case PIPE_SHADER_GEOMETRY:
212 case PIPE_SHADER_FRAGMENT:
213 break;
214 default:
215 return 0;
216 }
217
218 switch (param) {
219 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
220 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
221 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
222 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
223 return 16384;
224 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
225 return 4;
226 case PIPE_SHADER_CAP_MAX_INPUTS:
227 if (shader == PIPE_SHADER_VERTEX)
228 return 32;
229 return 15;
230 case PIPE_SHADER_CAP_MAX_CONSTS:
231 return 65536 / 16;
232 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
233 return NV50_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 nv50_screen(pscreen)->max_tls_space / ONE_TEMP_SIZE;
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 0; /* please inline, or provide function declarations */
252 case PIPE_SHADER_CAP_INTEGERS:
253 return 1;
254 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
255 /* The chip could handle more sampler views than samplers */
256 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
257 return MIN2(32, PIPE_MAX_SAMPLERS);
258 default:
259 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
260 return 0;
261 }
262 }
263
264 static float
265 nv50_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 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
273 return 64.0f;
274 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
275 return 16.0f;
276 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
277 return 4.0f;
278 default:
279 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
280 return 0.0f;
281 }
282 }
283
284 static void
285 nv50_screen_destroy(struct pipe_screen *pscreen)
286 {
287 struct nv50_screen *screen = nv50_screen(pscreen);
288
289 if (screen->base.fence.current) {
290 nouveau_fence_wait(screen->base.fence.current);
291 nouveau_fence_ref (NULL, &screen->base.fence.current);
292 }
293 if (screen->base.pushbuf)
294 screen->base.pushbuf->user_priv = NULL;
295
296 if (screen->blitter)
297 nv50_blitter_destroy(screen);
298
299 nouveau_bo_ref(NULL, &screen->code);
300 nouveau_bo_ref(NULL, &screen->tls_bo);
301 nouveau_bo_ref(NULL, &screen->stack_bo);
302 nouveau_bo_ref(NULL, &screen->txc);
303 nouveau_bo_ref(NULL, &screen->uniforms);
304 nouveau_bo_ref(NULL, &screen->fence.bo);
305
306 nouveau_heap_destroy(&screen->vp_code_heap);
307 nouveau_heap_destroy(&screen->gp_code_heap);
308 nouveau_heap_destroy(&screen->fp_code_heap);
309
310 FREE(screen->tic.entries);
311
312 nouveau_object_del(&screen->tesla);
313 nouveau_object_del(&screen->eng2d);
314 nouveau_object_del(&screen->m2mf);
315 nouveau_object_del(&screen->sync);
316
317 nouveau_screen_fini(&screen->base);
318
319 FREE(screen);
320 }
321
322 static void
323 nv50_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
324 {
325 struct nv50_screen *screen = nv50_screen(pscreen);
326 struct nouveau_pushbuf *push = screen->base.pushbuf;
327
328 /* we need to do it after possible flush in MARK_RING */
329 *sequence = ++screen->base.fence.sequence;
330
331 PUSH_DATA (push, NV50_FIFO_PKHDR(NV50_3D(QUERY_ADDRESS_HIGH), 4));
332 PUSH_DATAh(push, screen->fence.bo->offset);
333 PUSH_DATA (push, screen->fence.bo->offset);
334 PUSH_DATA (push, *sequence);
335 PUSH_DATA (push, NV50_3D_QUERY_GET_MODE_WRITE_UNK0 |
336 NV50_3D_QUERY_GET_UNK4 |
337 NV50_3D_QUERY_GET_UNIT_CROP |
338 NV50_3D_QUERY_GET_TYPE_QUERY |
339 NV50_3D_QUERY_GET_QUERY_SELECT_ZERO |
340 NV50_3D_QUERY_GET_SHORT);
341 }
342
343 static u32
344 nv50_screen_fence_update(struct pipe_screen *pscreen)
345 {
346 return nv50_screen(pscreen)->fence.map[0];
347 }
348
349 static void
350 nv50_screen_init_hwctx(struct nv50_screen *screen)
351 {
352 struct nouveau_pushbuf *push = screen->base.pushbuf;
353 struct nv04_fifo *fifo;
354 unsigned i;
355
356 fifo = (struct nv04_fifo *)screen->base.channel->data;
357
358 BEGIN_NV04(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
359 PUSH_DATA (push, screen->m2mf->handle);
360 BEGIN_NV04(push, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY), 3);
361 PUSH_DATA (push, screen->sync->handle);
362 PUSH_DATA (push, fifo->vram);
363 PUSH_DATA (push, fifo->vram);
364
365 BEGIN_NV04(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
366 PUSH_DATA (push, screen->eng2d->handle);
367 BEGIN_NV04(push, NV50_2D(DMA_NOTIFY), 4);
368 PUSH_DATA (push, screen->sync->handle);
369 PUSH_DATA (push, fifo->vram);
370 PUSH_DATA (push, fifo->vram);
371 PUSH_DATA (push, fifo->vram);
372 BEGIN_NV04(push, NV50_2D(OPERATION), 1);
373 PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
374 BEGIN_NV04(push, NV50_2D(CLIP_ENABLE), 1);
375 PUSH_DATA (push, 0);
376 BEGIN_NV04(push, NV50_2D(COLOR_KEY_ENABLE), 1);
377 PUSH_DATA (push, 0);
378 BEGIN_NV04(push, SUBC_2D(0x0888), 1);
379 PUSH_DATA (push, 1);
380
381 BEGIN_NV04(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
382 PUSH_DATA (push, screen->tesla->handle);
383
384 BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
385 PUSH_DATA (push, NV50_3D_COND_MODE_ALWAYS);
386
387 BEGIN_NV04(push, NV50_3D(DMA_NOTIFY), 1);
388 PUSH_DATA (push, screen->sync->handle);
389 BEGIN_NV04(push, NV50_3D(DMA_ZETA), 11);
390 for (i = 0; i < 11; ++i)
391 PUSH_DATA(push, fifo->vram);
392 BEGIN_NV04(push, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN);
393 for (i = 0; i < NV50_3D_DMA_COLOR__LEN; ++i)
394 PUSH_DATA(push, fifo->vram);
395
396 BEGIN_NV04(push, NV50_3D(REG_MODE), 1);
397 PUSH_DATA (push, NV50_3D_REG_MODE_STRIPED);
398 BEGIN_NV04(push, NV50_3D(UNK1400_LANES), 1);
399 PUSH_DATA (push, 0xf);
400
401 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", TRUE)) {
402 BEGIN_NV04(push, NV50_3D(WATCHDOG_TIMER), 1);
403 PUSH_DATA (push, 0x18);
404 }
405
406 BEGIN_NV04(push, NV50_3D(RT_CONTROL), 1);
407 PUSH_DATA (push, 1);
408
409 BEGIN_NV04(push, NV50_3D(CSAA_ENABLE), 1);
410 PUSH_DATA (push, 0);
411 BEGIN_NV04(push, NV50_3D(MULTISAMPLE_ENABLE), 1);
412 PUSH_DATA (push, 0);
413 BEGIN_NV04(push, NV50_3D(MULTISAMPLE_MODE), 1);
414 PUSH_DATA (push, NV50_3D_MULTISAMPLE_MODE_MS1);
415 BEGIN_NV04(push, NV50_3D(MULTISAMPLE_CTRL), 1);
416 PUSH_DATA (push, 0);
417 BEGIN_NV04(push, NV50_3D(LINE_LAST_PIXEL), 1);
418 PUSH_DATA (push, 0);
419 BEGIN_NV04(push, NV50_3D(BLEND_SEPARATE_ALPHA), 1);
420 PUSH_DATA (push, 1);
421
422 if (screen->tesla->oclass >= NVA0_3D_CLASS) {
423 BEGIN_NV04(push, SUBC_3D(NVA0_3D_TEX_MISC), 1);
424 PUSH_DATA (push, NVA0_3D_TEX_MISC_SEAMLESS_CUBE_MAP);
425 }
426
427 BEGIN_NV04(push, NV50_3D(SCREEN_Y_CONTROL), 1);
428 PUSH_DATA (push, 0);
429 BEGIN_NV04(push, NV50_3D(WINDOW_OFFSET_X), 2);
430 PUSH_DATA (push, 0);
431 PUSH_DATA (push, 0);
432 BEGIN_NV04(push, NV50_3D(ZCULL_REGION), 1);
433 PUSH_DATA (push, 0x3f);
434
435 BEGIN_NV04(push, NV50_3D(VP_ADDRESS_HIGH), 2);
436 PUSH_DATAh(push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
437 PUSH_DATA (push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
438
439 BEGIN_NV04(push, NV50_3D(FP_ADDRESS_HIGH), 2);
440 PUSH_DATAh(push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
441 PUSH_DATA (push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
442
443 BEGIN_NV04(push, NV50_3D(GP_ADDRESS_HIGH), 2);
444 PUSH_DATAh(push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
445 PUSH_DATA (push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
446
447 BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
448 PUSH_DATAh(push, screen->tls_bo->offset);
449 PUSH_DATA (push, screen->tls_bo->offset);
450 PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
451
452 BEGIN_NV04(push, NV50_3D(STACK_ADDRESS_HIGH), 3);
453 PUSH_DATAh(push, screen->stack_bo->offset);
454 PUSH_DATA (push, screen->stack_bo->offset);
455 PUSH_DATA (push, 4);
456
457 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
458 PUSH_DATAh(push, screen->uniforms->offset + (0 << 16));
459 PUSH_DATA (push, screen->uniforms->offset + (0 << 16));
460 PUSH_DATA (push, (NV50_CB_PVP << 16) | 0x0000);
461
462 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
463 PUSH_DATAh(push, screen->uniforms->offset + (1 << 16));
464 PUSH_DATA (push, screen->uniforms->offset + (1 << 16));
465 PUSH_DATA (push, (NV50_CB_PGP << 16) | 0x0000);
466
467 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
468 PUSH_DATAh(push, screen->uniforms->offset + (2 << 16));
469 PUSH_DATA (push, screen->uniforms->offset + (2 << 16));
470 PUSH_DATA (push, (NV50_CB_PFP << 16) | 0x0000);
471
472 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
473 PUSH_DATAh(push, screen->uniforms->offset + (3 << 16));
474 PUSH_DATA (push, screen->uniforms->offset + (3 << 16));
475 PUSH_DATA (push, (NV50_CB_AUX << 16) | (NV50_CB_AUX_SIZE & 0xffff));
476
477 BEGIN_NI04(push, NV50_3D(SET_PROGRAM_CB), 3);
478 PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf01);
479 PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf21);
480 PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf31);
481
482 /* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
483 BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
484 PUSH_DATA (push, (NV50_CB_AUX_RUNOUT_OFFSET << (8 - 2)) | NV50_CB_AUX);
485 BEGIN_NI04(push, NV50_3D(CB_DATA(0)), 4);
486 PUSH_DATAf(push, 0.0f);
487 PUSH_DATAf(push, 0.0f);
488 PUSH_DATAf(push, 0.0f);
489 PUSH_DATAf(push, 0.0f);
490 BEGIN_NV04(push, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
491 PUSH_DATAh(push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
492 PUSH_DATA (push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
493
494 nv50_upload_ms_info(push);
495
496 /* max TIC (bits 4:8) & TSC bindings, per program type */
497 for (i = 0; i < 3; ++i) {
498 BEGIN_NV04(push, NV50_3D(TEX_LIMITS(i)), 1);
499 PUSH_DATA (push, 0x54);
500 }
501
502 BEGIN_NV04(push, NV50_3D(TIC_ADDRESS_HIGH), 3);
503 PUSH_DATAh(push, screen->txc->offset);
504 PUSH_DATA (push, screen->txc->offset);
505 PUSH_DATA (push, NV50_TIC_MAX_ENTRIES - 1);
506
507 BEGIN_NV04(push, NV50_3D(TSC_ADDRESS_HIGH), 3);
508 PUSH_DATAh(push, screen->txc->offset + 65536);
509 PUSH_DATA (push, screen->txc->offset + 65536);
510 PUSH_DATA (push, NV50_TSC_MAX_ENTRIES - 1);
511
512 BEGIN_NV04(push, NV50_3D(LINKED_TSC), 1);
513 PUSH_DATA (push, 0);
514
515 BEGIN_NV04(push, NV50_3D(CLIP_RECTS_EN), 1);
516 PUSH_DATA (push, 0);
517 BEGIN_NV04(push, NV50_3D(CLIP_RECTS_MODE), 1);
518 PUSH_DATA (push, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY);
519 BEGIN_NV04(push, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
520 for (i = 0; i < 8 * 2; ++i)
521 PUSH_DATA(push, 0);
522 BEGIN_NV04(push, NV50_3D(CLIPID_ENABLE), 1);
523 PUSH_DATA (push, 0);
524
525 BEGIN_NV04(push, NV50_3D(VIEWPORT_TRANSFORM_EN), 1);
526 PUSH_DATA (push, 1);
527 BEGIN_NV04(push, NV50_3D(DEPTH_RANGE_NEAR(0)), 2);
528 PUSH_DATAf(push, 0.0f);
529 PUSH_DATAf(push, 1.0f);
530
531 BEGIN_NV04(push, NV50_3D(VIEW_VOLUME_CLIP_CTRL), 1);
532 #ifdef NV50_SCISSORS_CLIPPING
533 PUSH_DATA (push, 0x0000);
534 #else
535 PUSH_DATA (push, 0x1080);
536 #endif
537
538 BEGIN_NV04(push, NV50_3D(CLEAR_FLAGS), 1);
539 PUSH_DATA (push, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT);
540
541 /* We use scissors instead of exact view volume clipping,
542 * so they're always enabled.
543 */
544 BEGIN_NV04(push, NV50_3D(SCISSOR_ENABLE(0)), 3);
545 PUSH_DATA (push, 1);
546 PUSH_DATA (push, 8192 << 16);
547 PUSH_DATA (push, 8192 << 16);
548
549 BEGIN_NV04(push, NV50_3D(RASTERIZE_ENABLE), 1);
550 PUSH_DATA (push, 1);
551 BEGIN_NV04(push, NV50_3D(POINT_RASTER_RULES), 1);
552 PUSH_DATA (push, NV50_3D_POINT_RASTER_RULES_OGL);
553 BEGIN_NV04(push, NV50_3D(FRAG_COLOR_CLAMP_EN), 1);
554 PUSH_DATA (push, 0x11111111);
555 BEGIN_NV04(push, NV50_3D(EDGEFLAG), 1);
556 PUSH_DATA (push, 1);
557
558 PUSH_KICK (push);
559 }
560
561 static int nv50_tls_alloc(struct nv50_screen *screen, unsigned tls_space,
562 uint64_t *tls_size)
563 {
564 struct nouveau_device *dev = screen->base.device;
565 int ret;
566
567 screen->cur_tls_space = util_next_power_of_two(tls_space / ONE_TEMP_SIZE) *
568 ONE_TEMP_SIZE;
569 if (nouveau_mesa_debug)
570 debug_printf("allocating space for %u temps\n",
571 util_next_power_of_two(tls_space / ONE_TEMP_SIZE));
572 *tls_size = screen->cur_tls_space * util_next_power_of_two(screen->TPs) *
573 screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP;
574
575 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
576 *tls_size, NULL, &screen->tls_bo);
577 if (ret) {
578 NOUVEAU_ERR("Failed to allocate local bo: %d\n", ret);
579 return ret;
580 }
581
582 return 0;
583 }
584
585 int nv50_tls_realloc(struct nv50_screen *screen, unsigned tls_space)
586 {
587 struct nouveau_pushbuf *push = screen->base.pushbuf;
588 int ret;
589 uint64_t tls_size;
590
591 if (tls_space < screen->cur_tls_space)
592 return 0;
593 if (tls_space > screen->max_tls_space) {
594 /* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
595 * LOCAL_WARPS_NO_CLAMP) */
596 NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
597 (unsigned)(tls_space / ONE_TEMP_SIZE),
598 (unsigned)(screen->max_tls_space / ONE_TEMP_SIZE));
599 return -ENOMEM;
600 }
601
602 nouveau_bo_ref(NULL, &screen->tls_bo);
603 ret = nv50_tls_alloc(screen, tls_space, &tls_size);
604 if (ret)
605 return ret;
606
607 BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
608 PUSH_DATAh(push, screen->tls_bo->offset);
609 PUSH_DATA (push, screen->tls_bo->offset);
610 PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
611
612 return 1;
613 }
614
615 struct pipe_screen *
616 nv50_screen_create(struct nouveau_device *dev)
617 {
618 struct nv50_screen *screen;
619 struct pipe_screen *pscreen;
620 struct nouveau_object *chan;
621 uint64_t value;
622 uint32_t tesla_class;
623 unsigned stack_size;
624 int ret;
625
626 screen = CALLOC_STRUCT(nv50_screen);
627 if (!screen)
628 return NULL;
629 pscreen = &screen->base.base;
630
631 ret = nouveau_screen_init(&screen->base, dev);
632 if (ret) {
633 NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret);
634 goto fail;
635 }
636
637 /* TODO: Prevent FIFO prefetch before transfer of index buffers and
638 * admit them to VRAM.
639 */
640 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
641 PIPE_BIND_VERTEX_BUFFER;
642 screen->base.sysmem_bindings |=
643 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
644
645 screen->base.pushbuf->user_priv = screen;
646 screen->base.pushbuf->rsvd_kick = 5;
647
648 chan = screen->base.channel;
649
650 pscreen->destroy = nv50_screen_destroy;
651 pscreen->context_create = nv50_create;
652 pscreen->is_format_supported = nv50_screen_is_format_supported;
653 pscreen->get_param = nv50_screen_get_param;
654 pscreen->get_shader_param = nv50_screen_get_shader_param;
655 pscreen->get_paramf = nv50_screen_get_paramf;
656
657 nv50_screen_init_resource_functions(pscreen);
658
659 if (screen->base.device->chipset < 0x84 ||
660 debug_get_bool_option("NOUVEAU_PMPEG", FALSE)) {
661 /* PMPEG */
662 nouveau_screen_init_vdec(&screen->base);
663 } else if (screen->base.device->chipset < 0x98 ||
664 screen->base.device->chipset == 0xa0) {
665 /* VP2 */
666 screen->base.base.get_video_param = nv84_screen_get_video_param;
667 screen->base.base.is_video_format_supported = nv84_screen_video_supported;
668 } else {
669 /* VP3/4 */
670 screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
671 screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
672 }
673
674 ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096,
675 NULL, &screen->fence.bo);
676 if (ret) {
677 NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret);
678 goto fail;
679 }
680
681 nouveau_bo_map(screen->fence.bo, 0, NULL);
682 screen->fence.map = screen->fence.bo->map;
683 screen->base.fence.emit = nv50_screen_fence_emit;
684 screen->base.fence.update = nv50_screen_fence_update;
685
686 ret = nouveau_object_new(chan, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS,
687 &(struct nv04_notify){ .length = 32 },
688 sizeof(struct nv04_notify), &screen->sync);
689 if (ret) {
690 NOUVEAU_ERR("Failed to allocate notifier: %d\n", ret);
691 goto fail;
692 }
693
694 ret = nouveau_object_new(chan, 0xbeef5039, NV50_M2MF_CLASS,
695 NULL, 0, &screen->m2mf);
696 if (ret) {
697 NOUVEAU_ERR("Failed to allocate PGRAPH context for M2MF: %d\n", ret);
698 goto fail;
699 }
700
701 ret = nouveau_object_new(chan, 0xbeef502d, NV50_2D_CLASS,
702 NULL, 0, &screen->eng2d);
703 if (ret) {
704 NOUVEAU_ERR("Failed to allocate PGRAPH context for 2D: %d\n", ret);
705 goto fail;
706 }
707
708 switch (dev->chipset & 0xf0) {
709 case 0x50:
710 tesla_class = NV50_3D_CLASS;
711 break;
712 case 0x80:
713 case 0x90:
714 tesla_class = NV84_3D_CLASS;
715 break;
716 case 0xa0:
717 switch (dev->chipset) {
718 case 0xa0:
719 case 0xaa:
720 case 0xac:
721 tesla_class = NVA0_3D_CLASS;
722 break;
723 case 0xaf:
724 tesla_class = NVAF_3D_CLASS;
725 break;
726 default:
727 tesla_class = NVA3_3D_CLASS;
728 break;
729 }
730 break;
731 default:
732 NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", dev->chipset);
733 goto fail;
734 }
735 screen->base.class_3d = tesla_class;
736
737 ret = nouveau_object_new(chan, 0xbeef5097, tesla_class,
738 NULL, 0, &screen->tesla);
739 if (ret) {
740 NOUVEAU_ERR("Failed to allocate PGRAPH context for 3D: %d\n", ret);
741 goto fail;
742 }
743
744 /* This over-allocates by a whole code BO. The GP, which would execute at
745 * the end of the last page, would trigger faults. The going theory is that
746 * it prefetches up to a certain amount. This avoids dmesg spam.
747 */
748 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
749 4 << NV50_CODE_BO_SIZE_LOG2, NULL, &screen->code);
750 if (ret) {
751 NOUVEAU_ERR("Failed to allocate code bo: %d\n", ret);
752 goto fail;
753 }
754
755 nouveau_heap_init(&screen->vp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
756 nouveau_heap_init(&screen->gp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
757 nouveau_heap_init(&screen->fp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
758
759 nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
760
761 screen->TPs = util_bitcount(value & 0xffff);
762 screen->MPsInTP = util_bitcount((value >> 24) & 0xf);
763
764 stack_size = util_next_power_of_two(screen->TPs) * screen->MPsInTP *
765 STACK_WARPS_ALLOC * 64 * 8;
766
767 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, stack_size, NULL,
768 &screen->stack_bo);
769 if (ret) {
770 NOUVEAU_ERR("Failed to allocate stack bo: %d\n", ret);
771 goto fail;
772 }
773
774 uint64_t size_of_one_temp = util_next_power_of_two(screen->TPs) *
775 screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP *
776 ONE_TEMP_SIZE;
777 screen->max_tls_space = dev->vram_size / size_of_one_temp * ONE_TEMP_SIZE;
778 screen->max_tls_space /= 2; /* half of vram */
779
780 /* hw can address max 64 KiB */
781 screen->max_tls_space = MIN2(screen->max_tls_space, 64 << 10);
782
783 uint64_t tls_size;
784 unsigned tls_space = 4/*temps*/ * ONE_TEMP_SIZE;
785 ret = nv50_tls_alloc(screen, tls_space, &tls_size);
786 if (ret)
787 goto fail;
788
789 if (nouveau_mesa_debug)
790 debug_printf("TPs = %u, MPsInTP = %u, VRAM = %"PRIu64" MiB, tls_size = %"PRIu64" KiB\n",
791 screen->TPs, screen->MPsInTP, dev->vram_size >> 20, tls_size >> 10);
792
793 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 4 << 16, NULL,
794 &screen->uniforms);
795 if (ret) {
796 NOUVEAU_ERR("Failed to allocate uniforms bo: %d\n", ret);
797 goto fail;
798 }
799
800 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 3 << 16, NULL,
801 &screen->txc);
802 if (ret) {
803 NOUVEAU_ERR("Failed to allocate TIC/TSC bo: %d\n", ret);
804 goto fail;
805 }
806
807 screen->tic.entries = CALLOC(4096, sizeof(void *));
808 screen->tsc.entries = screen->tic.entries + 2048;
809
810 if (!nv50_blitter_create(screen))
811 goto fail;
812
813 nv50_screen_init_hwctx(screen);
814
815 nouveau_fence_new(&screen->base, &screen->base.fence.current, FALSE);
816
817 return pscreen;
818
819 fail:
820 nv50_screen_destroy(pscreen);
821 return NULL;
822 }
823
824 int
825 nv50_screen_tic_alloc(struct nv50_screen *screen, void *entry)
826 {
827 int i = screen->tic.next;
828
829 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
830 i = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
831
832 screen->tic.next = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
833
834 if (screen->tic.entries[i])
835 nv50_tic_entry(screen->tic.entries[i])->id = -1;
836
837 screen->tic.entries[i] = entry;
838 return i;
839 }
840
841 int
842 nv50_screen_tsc_alloc(struct nv50_screen *screen, void *entry)
843 {
844 int i = screen->tsc.next;
845
846 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
847 i = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
848
849 screen->tsc.next = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
850
851 if (screen->tsc.entries[i])
852 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
853
854 screen->tsc.entries[i] = entry;
855 return i;
856 }