nv50,nvc0: add ARB_clear_texture support
[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 <errno.h>
24 #include <xf86drm.h>
25 #include <nouveau_drm.h>
26 #include "util/u_format.h"
27 #include "util/u_format_s3tc.h"
28 #include "pipe/p_screen.h"
29
30 #include "nv50/nv50_context.h"
31 #include "nv50/nv50_screen.h"
32
33 #include "nouveau_vp3_video.h"
34
35 #include "nv_object.xml.h"
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 struct nouveau_device *dev = nouveau_screen(pscreen)->device;
85
86 switch (param) {
87 /* non-boolean caps */
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_TEXTURE_GATHER_OFFSET:
97 case PIPE_CAP_MIN_TEXEL_OFFSET:
98 return -8;
99 case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
100 case PIPE_CAP_MAX_TEXEL_OFFSET:
101 return 7;
102 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
103 return 128 * 1024 * 1024;
104 case PIPE_CAP_GLSL_FEATURE_LEVEL:
105 return 330;
106 case PIPE_CAP_MAX_RENDER_TARGETS:
107 return 8;
108 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
109 return 1;
110 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
111 return 4;
112 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
113 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
114 return 64;
115 case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
116 case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
117 return 1024;
118 case PIPE_CAP_MAX_VERTEX_STREAMS:
119 return 1;
120 case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
121 return 2048;
122 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
123 return 256;
124 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
125 return 1; /* 256 for binding as RT, but that's not possible in GL */
126 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
127 return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
128 case PIPE_CAP_MAX_VIEWPORTS:
129 return NV50_MAX_VIEWPORTS;
130 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
131 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
132 case PIPE_CAP_ENDIANNESS:
133 return PIPE_ENDIAN_LITTLE;
134 case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
135 return (class_3d >= NVA3_3D_CLASS) ? 4 : 0;
136
137 /* supported caps */
138 case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
139 case PIPE_CAP_TEXTURE_SWIZZLE:
140 case PIPE_CAP_TEXTURE_SHADOW_MAP:
141 case PIPE_CAP_NPOT_TEXTURES:
142 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
143 case PIPE_CAP_ANISOTROPIC_FILTER:
144 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
145 case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
146 case PIPE_CAP_TWO_SIDED_STENCIL:
147 case PIPE_CAP_DEPTH_CLIP_DISABLE:
148 case PIPE_CAP_POINT_SPRITE:
149 case PIPE_CAP_SM3:
150 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
151 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
152 case PIPE_CAP_VERTEX_COLOR_CLAMPED:
153 case PIPE_CAP_QUERY_TIMESTAMP:
154 case PIPE_CAP_QUERY_TIME_ELAPSED:
155 case PIPE_CAP_OCCLUSION_QUERY:
156 case PIPE_CAP_BLEND_EQUATION_SEPARATE:
157 case PIPE_CAP_INDEP_BLEND_ENABLE:
158 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
159 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
160 case PIPE_CAP_PRIMITIVE_RESTART:
161 case PIPE_CAP_TGSI_INSTANCEID:
162 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
163 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
164 case PIPE_CAP_CONDITIONAL_RENDER:
165 case PIPE_CAP_TEXTURE_BARRIER:
166 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
167 case PIPE_CAP_START_INSTANCE:
168 case PIPE_CAP_USER_CONSTANT_BUFFERS:
169 case PIPE_CAP_USER_INDEX_BUFFERS:
170 case PIPE_CAP_USER_VERTEX_BUFFERS:
171 case PIPE_CAP_TEXTURE_MULTISAMPLE:
172 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
173 case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
174 case PIPE_CAP_SAMPLER_VIEW_TARGET:
175 case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
176 case PIPE_CAP_CLIP_HALFZ:
177 case PIPE_CAP_POLYGON_OFFSET_CLAMP:
178 case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
179 case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
180 case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
181 case PIPE_CAP_DEPTH_BOUNDS_TEST:
182 case PIPE_CAP_TGSI_TXQS:
183 case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
184 case PIPE_CAP_SHAREABLE_SHADERS:
185 case PIPE_CAP_CLEAR_TEXTURE:
186 return 1;
187 case PIPE_CAP_SEAMLESS_CUBE_MAP:
188 return 1; /* class_3d >= NVA0_3D_CLASS; */
189 /* supported on nva0+ */
190 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
191 return class_3d >= NVA0_3D_CLASS;
192 /* supported on nva3+ */
193 case PIPE_CAP_CUBE_MAP_ARRAY:
194 case PIPE_CAP_INDEP_BLEND_FUNC:
195 case PIPE_CAP_TEXTURE_QUERY_LOD:
196 case PIPE_CAP_SAMPLE_SHADING:
197 case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
198 return class_3d >= NVA3_3D_CLASS;
199
200 /* unsupported caps */
201 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
202 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
203 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
204 case PIPE_CAP_SHADER_STENCIL_EXPORT:
205 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
206 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
207 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
208 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
209 case PIPE_CAP_TGSI_TEXCOORD:
210 case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
211 case PIPE_CAP_TEXTURE_GATHER_SM5:
212 case PIPE_CAP_FAKE_SW_MSAA:
213 case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
214 case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
215 case PIPE_CAP_COMPUTE:
216 case PIPE_CAP_DRAW_INDIRECT:
217 case PIPE_CAP_VERTEXID_NOBASE:
218 case PIPE_CAP_MULTISAMPLE_Z_RESOLVE: /* potentially supported on some hw */
219 case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
220 case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
221 case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
222 return 0;
223
224 case PIPE_CAP_VENDOR_ID:
225 return 0x10de;
226 case PIPE_CAP_DEVICE_ID: {
227 uint64_t device_id;
228 if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
229 NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
230 return -1;
231 }
232 return device_id;
233 }
234 case PIPE_CAP_ACCELERATED:
235 return 1;
236 case PIPE_CAP_VIDEO_MEMORY:
237 return dev->vram_size >> 20;
238 case PIPE_CAP_UMA:
239 return 0;
240 }
241
242 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
243 return 0;
244 }
245
246 static int
247 nv50_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
248 enum pipe_shader_cap param)
249 {
250 switch (shader) {
251 case PIPE_SHADER_VERTEX:
252 case PIPE_SHADER_GEOMETRY:
253 case PIPE_SHADER_FRAGMENT:
254 break;
255 default:
256 return 0;
257 }
258
259 switch (param) {
260 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
261 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
262 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
263 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
264 return 16384;
265 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
266 return 4;
267 case PIPE_SHADER_CAP_MAX_INPUTS:
268 if (shader == PIPE_SHADER_VERTEX)
269 return 32;
270 return 15;
271 case PIPE_SHADER_CAP_MAX_OUTPUTS:
272 return 16;
273 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
274 return 65536;
275 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
276 return NV50_MAX_PIPE_CONSTBUFS;
277 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
278 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
279 return shader != PIPE_SHADER_FRAGMENT;
280 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
281 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
282 return 1;
283 case PIPE_SHADER_CAP_MAX_PREDS:
284 return 0;
285 case PIPE_SHADER_CAP_MAX_TEMPS:
286 return nv50_screen(pscreen)->max_tls_space / ONE_TEMP_SIZE;
287 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
288 return 1;
289 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
290 return 0;
291 case PIPE_SHADER_CAP_SUBROUTINES:
292 return 0; /* please inline, or provide function declarations */
293 case PIPE_SHADER_CAP_INTEGERS:
294 return 1;
295 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
296 /* The chip could handle more sampler views than samplers */
297 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
298 return MIN2(16, PIPE_MAX_SAMPLERS);
299 case PIPE_SHADER_CAP_DOUBLES:
300 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
301 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
302 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
303 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
304 return 0;
305 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
306 return 32;
307 default:
308 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
309 return 0;
310 }
311 }
312
313 static float
314 nv50_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
315 {
316 switch (param) {
317 case PIPE_CAPF_MAX_LINE_WIDTH:
318 case PIPE_CAPF_MAX_LINE_WIDTH_AA:
319 return 10.0f;
320 case PIPE_CAPF_MAX_POINT_WIDTH:
321 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
322 return 64.0f;
323 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
324 return 16.0f;
325 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
326 return 4.0f;
327 case PIPE_CAPF_GUARD_BAND_LEFT:
328 case PIPE_CAPF_GUARD_BAND_TOP:
329 return 0.0f;
330 case PIPE_CAPF_GUARD_BAND_RIGHT:
331 case PIPE_CAPF_GUARD_BAND_BOTTOM:
332 return 0.0f; /* that or infinity */
333 }
334
335 NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
336 return 0.0f;
337 }
338
339 static void
340 nv50_screen_destroy(struct pipe_screen *pscreen)
341 {
342 struct nv50_screen *screen = nv50_screen(pscreen);
343
344 if (!nouveau_drm_screen_unref(&screen->base))
345 return;
346
347 if (screen->base.fence.current) {
348 struct nouveau_fence *current = NULL;
349
350 /* nouveau_fence_wait will create a new current fence, so wait on the
351 * _current_ one, and remove both.
352 */
353 nouveau_fence_ref(screen->base.fence.current, &current);
354 nouveau_fence_wait(current, NULL);
355 nouveau_fence_ref(NULL, &current);
356 nouveau_fence_ref(NULL, &screen->base.fence.current);
357 }
358 if (screen->base.pushbuf)
359 screen->base.pushbuf->user_priv = NULL;
360
361 if (screen->blitter)
362 nv50_blitter_destroy(screen);
363
364 nouveau_bo_ref(NULL, &screen->code);
365 nouveau_bo_ref(NULL, &screen->tls_bo);
366 nouveau_bo_ref(NULL, &screen->stack_bo);
367 nouveau_bo_ref(NULL, &screen->txc);
368 nouveau_bo_ref(NULL, &screen->uniforms);
369 nouveau_bo_ref(NULL, &screen->fence.bo);
370
371 nouveau_heap_destroy(&screen->vp_code_heap);
372 nouveau_heap_destroy(&screen->gp_code_heap);
373 nouveau_heap_destroy(&screen->fp_code_heap);
374
375 FREE(screen->tic.entries);
376
377 nouveau_object_del(&screen->tesla);
378 nouveau_object_del(&screen->eng2d);
379 nouveau_object_del(&screen->m2mf);
380 nouveau_object_del(&screen->sync);
381
382 nouveau_screen_fini(&screen->base);
383
384 FREE(screen);
385 }
386
387 static void
388 nv50_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
389 {
390 struct nv50_screen *screen = nv50_screen(pscreen);
391 struct nouveau_pushbuf *push = screen->base.pushbuf;
392
393 /* we need to do it after possible flush in MARK_RING */
394 *sequence = ++screen->base.fence.sequence;
395
396 assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
397 PUSH_DATA (push, NV50_FIFO_PKHDR(NV50_3D(QUERY_ADDRESS_HIGH), 4));
398 PUSH_DATAh(push, screen->fence.bo->offset);
399 PUSH_DATA (push, screen->fence.bo->offset);
400 PUSH_DATA (push, *sequence);
401 PUSH_DATA (push, NV50_3D_QUERY_GET_MODE_WRITE_UNK0 |
402 NV50_3D_QUERY_GET_UNK4 |
403 NV50_3D_QUERY_GET_UNIT_CROP |
404 NV50_3D_QUERY_GET_TYPE_QUERY |
405 NV50_3D_QUERY_GET_QUERY_SELECT_ZERO |
406 NV50_3D_QUERY_GET_SHORT);
407 }
408
409 static u32
410 nv50_screen_fence_update(struct pipe_screen *pscreen)
411 {
412 return nv50_screen(pscreen)->fence.map[0];
413 }
414
415 static void
416 nv50_screen_init_hwctx(struct nv50_screen *screen)
417 {
418 struct nouveau_pushbuf *push = screen->base.pushbuf;
419 struct nv04_fifo *fifo;
420 unsigned i;
421
422 fifo = (struct nv04_fifo *)screen->base.channel->data;
423
424 BEGIN_NV04(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
425 PUSH_DATA (push, screen->m2mf->handle);
426 BEGIN_NV04(push, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY), 3);
427 PUSH_DATA (push, screen->sync->handle);
428 PUSH_DATA (push, fifo->vram);
429 PUSH_DATA (push, fifo->vram);
430
431 BEGIN_NV04(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
432 PUSH_DATA (push, screen->eng2d->handle);
433 BEGIN_NV04(push, NV50_2D(DMA_NOTIFY), 4);
434 PUSH_DATA (push, screen->sync->handle);
435 PUSH_DATA (push, fifo->vram);
436 PUSH_DATA (push, fifo->vram);
437 PUSH_DATA (push, fifo->vram);
438 BEGIN_NV04(push, NV50_2D(OPERATION), 1);
439 PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
440 BEGIN_NV04(push, NV50_2D(CLIP_ENABLE), 1);
441 PUSH_DATA (push, 0);
442 BEGIN_NV04(push, NV50_2D(COLOR_KEY_ENABLE), 1);
443 PUSH_DATA (push, 0);
444 BEGIN_NV04(push, SUBC_2D(0x0888), 1);
445 PUSH_DATA (push, 1);
446 BEGIN_NV04(push, NV50_2D(COND_MODE), 1);
447 PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
448
449 BEGIN_NV04(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
450 PUSH_DATA (push, screen->tesla->handle);
451
452 BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
453 PUSH_DATA (push, NV50_3D_COND_MODE_ALWAYS);
454
455 BEGIN_NV04(push, NV50_3D(DMA_NOTIFY), 1);
456 PUSH_DATA (push, screen->sync->handle);
457 BEGIN_NV04(push, NV50_3D(DMA_ZETA), 11);
458 for (i = 0; i < 11; ++i)
459 PUSH_DATA(push, fifo->vram);
460 BEGIN_NV04(push, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN);
461 for (i = 0; i < NV50_3D_DMA_COLOR__LEN; ++i)
462 PUSH_DATA(push, fifo->vram);
463
464 BEGIN_NV04(push, NV50_3D(REG_MODE), 1);
465 PUSH_DATA (push, NV50_3D_REG_MODE_STRIPED);
466 BEGIN_NV04(push, NV50_3D(UNK1400_LANES), 1);
467 PUSH_DATA (push, 0xf);
468
469 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
470 BEGIN_NV04(push, NV50_3D(WATCHDOG_TIMER), 1);
471 PUSH_DATA (push, 0x18);
472 }
473
474 BEGIN_NV04(push, NV50_3D(ZETA_COMP_ENABLE), 1);
475 PUSH_DATA(push, screen->base.device->drm_version >= 0x01000101);
476
477 BEGIN_NV04(push, NV50_3D(RT_COMP_ENABLE(0)), 8);
478 for (i = 0; i < 8; ++i)
479 PUSH_DATA(push, screen->base.device->drm_version >= 0x01000101);
480
481 BEGIN_NV04(push, NV50_3D(RT_CONTROL), 1);
482 PUSH_DATA (push, 1);
483
484 BEGIN_NV04(push, NV50_3D(CSAA_ENABLE), 1);
485 PUSH_DATA (push, 0);
486 BEGIN_NV04(push, NV50_3D(MULTISAMPLE_ENABLE), 1);
487 PUSH_DATA (push, 0);
488 BEGIN_NV04(push, NV50_3D(MULTISAMPLE_MODE), 1);
489 PUSH_DATA (push, NV50_3D_MULTISAMPLE_MODE_MS1);
490 BEGIN_NV04(push, NV50_3D(MULTISAMPLE_CTRL), 1);
491 PUSH_DATA (push, 0);
492 BEGIN_NV04(push, NV50_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
493 PUSH_DATA (push, 1);
494 BEGIN_NV04(push, NV50_3D(BLEND_SEPARATE_ALPHA), 1);
495 PUSH_DATA (push, 1);
496
497 if (screen->tesla->oclass >= NVA0_3D_CLASS) {
498 BEGIN_NV04(push, SUBC_3D(NVA0_3D_TEX_MISC), 1);
499 PUSH_DATA (push, NVA0_3D_TEX_MISC_SEAMLESS_CUBE_MAP);
500 }
501
502 BEGIN_NV04(push, NV50_3D(SCREEN_Y_CONTROL), 1);
503 PUSH_DATA (push, 0);
504 BEGIN_NV04(push, NV50_3D(WINDOW_OFFSET_X), 2);
505 PUSH_DATA (push, 0);
506 PUSH_DATA (push, 0);
507 BEGIN_NV04(push, NV50_3D(ZCULL_REGION), 1);
508 PUSH_DATA (push, 0x3f);
509
510 BEGIN_NV04(push, NV50_3D(VP_ADDRESS_HIGH), 2);
511 PUSH_DATAh(push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
512 PUSH_DATA (push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
513
514 BEGIN_NV04(push, NV50_3D(FP_ADDRESS_HIGH), 2);
515 PUSH_DATAh(push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
516 PUSH_DATA (push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
517
518 BEGIN_NV04(push, NV50_3D(GP_ADDRESS_HIGH), 2);
519 PUSH_DATAh(push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
520 PUSH_DATA (push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
521
522 BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
523 PUSH_DATAh(push, screen->tls_bo->offset);
524 PUSH_DATA (push, screen->tls_bo->offset);
525 PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
526
527 BEGIN_NV04(push, NV50_3D(STACK_ADDRESS_HIGH), 3);
528 PUSH_DATAh(push, screen->stack_bo->offset);
529 PUSH_DATA (push, screen->stack_bo->offset);
530 PUSH_DATA (push, 4);
531
532 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
533 PUSH_DATAh(push, screen->uniforms->offset + (0 << 16));
534 PUSH_DATA (push, screen->uniforms->offset + (0 << 16));
535 PUSH_DATA (push, (NV50_CB_PVP << 16) | 0x0000);
536
537 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
538 PUSH_DATAh(push, screen->uniforms->offset + (1 << 16));
539 PUSH_DATA (push, screen->uniforms->offset + (1 << 16));
540 PUSH_DATA (push, (NV50_CB_PGP << 16) | 0x0000);
541
542 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
543 PUSH_DATAh(push, screen->uniforms->offset + (2 << 16));
544 PUSH_DATA (push, screen->uniforms->offset + (2 << 16));
545 PUSH_DATA (push, (NV50_CB_PFP << 16) | 0x0000);
546
547 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
548 PUSH_DATAh(push, screen->uniforms->offset + (3 << 16));
549 PUSH_DATA (push, screen->uniforms->offset + (3 << 16));
550 PUSH_DATA (push, (NV50_CB_AUX << 16) | (NV50_CB_AUX_SIZE & 0xffff));
551
552 BEGIN_NI04(push, NV50_3D(SET_PROGRAM_CB), 3);
553 PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf01);
554 PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf21);
555 PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf31);
556
557 /* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
558 BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
559 PUSH_DATA (push, (NV50_CB_AUX_RUNOUT_OFFSET << (8 - 2)) | NV50_CB_AUX);
560 BEGIN_NI04(push, NV50_3D(CB_DATA(0)), 4);
561 PUSH_DATAf(push, 0.0f);
562 PUSH_DATAf(push, 0.0f);
563 PUSH_DATAf(push, 0.0f);
564 PUSH_DATAf(push, 0.0f);
565 BEGIN_NV04(push, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
566 PUSH_DATAh(push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
567 PUSH_DATA (push, screen->uniforms->offset + (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
568
569 nv50_upload_ms_info(push);
570
571 /* max TIC (bits 4:8) & TSC bindings, per program type */
572 for (i = 0; i < 3; ++i) {
573 BEGIN_NV04(push, NV50_3D(TEX_LIMITS(i)), 1);
574 PUSH_DATA (push, 0x54);
575 }
576
577 BEGIN_NV04(push, NV50_3D(TIC_ADDRESS_HIGH), 3);
578 PUSH_DATAh(push, screen->txc->offset);
579 PUSH_DATA (push, screen->txc->offset);
580 PUSH_DATA (push, NV50_TIC_MAX_ENTRIES - 1);
581
582 BEGIN_NV04(push, NV50_3D(TSC_ADDRESS_HIGH), 3);
583 PUSH_DATAh(push, screen->txc->offset + 65536);
584 PUSH_DATA (push, screen->txc->offset + 65536);
585 PUSH_DATA (push, NV50_TSC_MAX_ENTRIES - 1);
586
587 BEGIN_NV04(push, NV50_3D(LINKED_TSC), 1);
588 PUSH_DATA (push, 0);
589
590 BEGIN_NV04(push, NV50_3D(CLIP_RECTS_EN), 1);
591 PUSH_DATA (push, 0);
592 BEGIN_NV04(push, NV50_3D(CLIP_RECTS_MODE), 1);
593 PUSH_DATA (push, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY);
594 BEGIN_NV04(push, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
595 for (i = 0; i < 8 * 2; ++i)
596 PUSH_DATA(push, 0);
597 BEGIN_NV04(push, NV50_3D(CLIPID_ENABLE), 1);
598 PUSH_DATA (push, 0);
599
600 BEGIN_NV04(push, NV50_3D(VIEWPORT_TRANSFORM_EN), 1);
601 PUSH_DATA (push, 1);
602 for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
603 BEGIN_NV04(push, NV50_3D(DEPTH_RANGE_NEAR(i)), 2);
604 PUSH_DATAf(push, 0.0f);
605 PUSH_DATAf(push, 1.0f);
606 BEGIN_NV04(push, NV50_3D(VIEWPORT_HORIZ(i)), 2);
607 PUSH_DATA (push, 8192 << 16);
608 PUSH_DATA (push, 8192 << 16);
609 }
610
611 BEGIN_NV04(push, NV50_3D(VIEW_VOLUME_CLIP_CTRL), 1);
612 #ifdef NV50_SCISSORS_CLIPPING
613 PUSH_DATA (push, 0x0000);
614 #else
615 PUSH_DATA (push, 0x1080);
616 #endif
617
618 BEGIN_NV04(push, NV50_3D(CLEAR_FLAGS), 1);
619 PUSH_DATA (push, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT);
620
621 /* We use scissors instead of exact view volume clipping,
622 * so they're always enabled.
623 */
624 for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
625 BEGIN_NV04(push, NV50_3D(SCISSOR_ENABLE(i)), 3);
626 PUSH_DATA (push, 1);
627 PUSH_DATA (push, 8192 << 16);
628 PUSH_DATA (push, 8192 << 16);
629 }
630
631 BEGIN_NV04(push, NV50_3D(RASTERIZE_ENABLE), 1);
632 PUSH_DATA (push, 1);
633 BEGIN_NV04(push, NV50_3D(POINT_RASTER_RULES), 1);
634 PUSH_DATA (push, NV50_3D_POINT_RASTER_RULES_OGL);
635 BEGIN_NV04(push, NV50_3D(FRAG_COLOR_CLAMP_EN), 1);
636 PUSH_DATA (push, 0x11111111);
637 BEGIN_NV04(push, NV50_3D(EDGEFLAG), 1);
638 PUSH_DATA (push, 1);
639
640 BEGIN_NV04(push, NV50_3D(VB_ELEMENT_BASE), 1);
641 PUSH_DATA (push, 0);
642 if (screen->base.class_3d >= NV84_3D_CLASS) {
643 BEGIN_NV04(push, SUBC_3D(NV84_3D_VERTEX_ID_BASE), 1);
644 PUSH_DATA (push, 0);
645 }
646
647 PUSH_KICK (push);
648 }
649
650 static int nv50_tls_alloc(struct nv50_screen *screen, unsigned tls_space,
651 uint64_t *tls_size)
652 {
653 struct nouveau_device *dev = screen->base.device;
654 int ret;
655
656 screen->cur_tls_space = util_next_power_of_two(tls_space / ONE_TEMP_SIZE) *
657 ONE_TEMP_SIZE;
658 if (nouveau_mesa_debug)
659 debug_printf("allocating space for %u temps\n",
660 util_next_power_of_two(tls_space / ONE_TEMP_SIZE));
661 *tls_size = screen->cur_tls_space * util_next_power_of_two(screen->TPs) *
662 screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP;
663
664 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
665 *tls_size, NULL, &screen->tls_bo);
666 if (ret) {
667 NOUVEAU_ERR("Failed to allocate local bo: %d\n", ret);
668 return ret;
669 }
670
671 return 0;
672 }
673
674 int nv50_tls_realloc(struct nv50_screen *screen, unsigned tls_space)
675 {
676 struct nouveau_pushbuf *push = screen->base.pushbuf;
677 int ret;
678 uint64_t tls_size;
679
680 if (tls_space < screen->cur_tls_space)
681 return 0;
682 if (tls_space > screen->max_tls_space) {
683 /* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
684 * LOCAL_WARPS_NO_CLAMP) */
685 NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
686 (unsigned)(tls_space / ONE_TEMP_SIZE),
687 (unsigned)(screen->max_tls_space / ONE_TEMP_SIZE));
688 return -ENOMEM;
689 }
690
691 nouveau_bo_ref(NULL, &screen->tls_bo);
692 ret = nv50_tls_alloc(screen, tls_space, &tls_size);
693 if (ret)
694 return ret;
695
696 BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
697 PUSH_DATAh(push, screen->tls_bo->offset);
698 PUSH_DATA (push, screen->tls_bo->offset);
699 PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
700
701 return 1;
702 }
703
704 struct pipe_screen *
705 nv50_screen_create(struct nouveau_device *dev)
706 {
707 struct nv50_screen *screen;
708 struct pipe_screen *pscreen;
709 struct nouveau_object *chan;
710 uint64_t value;
711 uint32_t tesla_class;
712 unsigned stack_size;
713 int ret;
714
715 screen = CALLOC_STRUCT(nv50_screen);
716 if (!screen)
717 return NULL;
718 pscreen = &screen->base.base;
719
720 ret = nouveau_screen_init(&screen->base, dev);
721 if (ret) {
722 NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret);
723 goto fail;
724 }
725
726 /* TODO: Prevent FIFO prefetch before transfer of index buffers and
727 * admit them to VRAM.
728 */
729 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
730 PIPE_BIND_VERTEX_BUFFER;
731 screen->base.sysmem_bindings |=
732 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
733
734 screen->base.pushbuf->user_priv = screen;
735 screen->base.pushbuf->rsvd_kick = 5;
736
737 chan = screen->base.channel;
738
739 pscreen->destroy = nv50_screen_destroy;
740 pscreen->context_create = nv50_create;
741 pscreen->is_format_supported = nv50_screen_is_format_supported;
742 pscreen->get_param = nv50_screen_get_param;
743 pscreen->get_shader_param = nv50_screen_get_shader_param;
744 pscreen->get_paramf = nv50_screen_get_paramf;
745
746 nv50_screen_init_resource_functions(pscreen);
747
748 if (screen->base.device->chipset < 0x84 ||
749 debug_get_bool_option("NOUVEAU_PMPEG", false)) {
750 /* PMPEG */
751 nouveau_screen_init_vdec(&screen->base);
752 } else if (screen->base.device->chipset < 0x98 ||
753 screen->base.device->chipset == 0xa0) {
754 /* VP2 */
755 screen->base.base.get_video_param = nv84_screen_get_video_param;
756 screen->base.base.is_video_format_supported = nv84_screen_video_supported;
757 } else {
758 /* VP3/4 */
759 screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
760 screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
761 }
762
763 ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096,
764 NULL, &screen->fence.bo);
765 if (ret) {
766 NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret);
767 goto fail;
768 }
769
770 nouveau_bo_map(screen->fence.bo, 0, NULL);
771 screen->fence.map = screen->fence.bo->map;
772 screen->base.fence.emit = nv50_screen_fence_emit;
773 screen->base.fence.update = nv50_screen_fence_update;
774
775 ret = nouveau_object_new(chan, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS,
776 &(struct nv04_notify){ .length = 32 },
777 sizeof(struct nv04_notify), &screen->sync);
778 if (ret) {
779 NOUVEAU_ERR("Failed to allocate notifier: %d\n", ret);
780 goto fail;
781 }
782
783 ret = nouveau_object_new(chan, 0xbeef5039, NV50_M2MF_CLASS,
784 NULL, 0, &screen->m2mf);
785 if (ret) {
786 NOUVEAU_ERR("Failed to allocate PGRAPH context for M2MF: %d\n", ret);
787 goto fail;
788 }
789
790 ret = nouveau_object_new(chan, 0xbeef502d, NV50_2D_CLASS,
791 NULL, 0, &screen->eng2d);
792 if (ret) {
793 NOUVEAU_ERR("Failed to allocate PGRAPH context for 2D: %d\n", ret);
794 goto fail;
795 }
796
797 switch (dev->chipset & 0xf0) {
798 case 0x50:
799 tesla_class = NV50_3D_CLASS;
800 break;
801 case 0x80:
802 case 0x90:
803 tesla_class = NV84_3D_CLASS;
804 break;
805 case 0xa0:
806 switch (dev->chipset) {
807 case 0xa0:
808 case 0xaa:
809 case 0xac:
810 tesla_class = NVA0_3D_CLASS;
811 break;
812 case 0xaf:
813 tesla_class = NVAF_3D_CLASS;
814 break;
815 default:
816 tesla_class = NVA3_3D_CLASS;
817 break;
818 }
819 break;
820 default:
821 NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", dev->chipset);
822 goto fail;
823 }
824 screen->base.class_3d = tesla_class;
825
826 ret = nouveau_object_new(chan, 0xbeef5097, tesla_class,
827 NULL, 0, &screen->tesla);
828 if (ret) {
829 NOUVEAU_ERR("Failed to allocate PGRAPH context for 3D: %d\n", ret);
830 goto fail;
831 }
832
833 /* This over-allocates by a page. The GP, which would execute at the end of
834 * the last page, would trigger faults. The going theory is that it
835 * prefetches up to a certain amount.
836 */
837 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
838 (3 << NV50_CODE_BO_SIZE_LOG2) + 0x1000,
839 NULL, &screen->code);
840 if (ret) {
841 NOUVEAU_ERR("Failed to allocate code bo: %d\n", ret);
842 goto fail;
843 }
844
845 nouveau_heap_init(&screen->vp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
846 nouveau_heap_init(&screen->gp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
847 nouveau_heap_init(&screen->fp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
848
849 nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
850
851 screen->TPs = util_bitcount(value & 0xffff);
852 screen->MPsInTP = util_bitcount((value >> 24) & 0xf);
853
854 stack_size = util_next_power_of_two(screen->TPs) * screen->MPsInTP *
855 STACK_WARPS_ALLOC * 64 * 8;
856
857 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, stack_size, NULL,
858 &screen->stack_bo);
859 if (ret) {
860 NOUVEAU_ERR("Failed to allocate stack bo: %d\n", ret);
861 goto fail;
862 }
863
864 uint64_t size_of_one_temp = util_next_power_of_two(screen->TPs) *
865 screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP *
866 ONE_TEMP_SIZE;
867 screen->max_tls_space = dev->vram_size / size_of_one_temp * ONE_TEMP_SIZE;
868 screen->max_tls_space /= 2; /* half of vram */
869
870 /* hw can address max 64 KiB */
871 screen->max_tls_space = MIN2(screen->max_tls_space, 64 << 10);
872
873 uint64_t tls_size;
874 unsigned tls_space = 4/*temps*/ * ONE_TEMP_SIZE;
875 ret = nv50_tls_alloc(screen, tls_space, &tls_size);
876 if (ret)
877 goto fail;
878
879 if (nouveau_mesa_debug)
880 debug_printf("TPs = %u, MPsInTP = %u, VRAM = %"PRIu64" MiB, tls_size = %"PRIu64" KiB\n",
881 screen->TPs, screen->MPsInTP, dev->vram_size >> 20, tls_size >> 10);
882
883 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 4 << 16, NULL,
884 &screen->uniforms);
885 if (ret) {
886 NOUVEAU_ERR("Failed to allocate uniforms bo: %d\n", ret);
887 goto fail;
888 }
889
890 ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 3 << 16, NULL,
891 &screen->txc);
892 if (ret) {
893 NOUVEAU_ERR("Failed to allocate TIC/TSC bo: %d\n", ret);
894 goto fail;
895 }
896
897 screen->tic.entries = CALLOC(4096, sizeof(void *));
898 screen->tsc.entries = screen->tic.entries + 2048;
899
900 if (!nv50_blitter_create(screen))
901 goto fail;
902
903 nv50_screen_init_hwctx(screen);
904
905 nouveau_fence_new(&screen->base, &screen->base.fence.current, false);
906
907 return pscreen;
908
909 fail:
910 nv50_screen_destroy(pscreen);
911 return NULL;
912 }
913
914 int
915 nv50_screen_tic_alloc(struct nv50_screen *screen, void *entry)
916 {
917 int i = screen->tic.next;
918
919 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
920 i = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
921
922 screen->tic.next = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
923
924 if (screen->tic.entries[i])
925 nv50_tic_entry(screen->tic.entries[i])->id = -1;
926
927 screen->tic.entries[i] = entry;
928 return i;
929 }
930
931 int
932 nv50_screen_tsc_alloc(struct nv50_screen *screen, void *entry)
933 {
934 int i = screen->tsc.next;
935
936 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
937 i = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
938
939 screen->tsc.next = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
940
941 if (screen->tsc.entries[i])
942 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
943
944 screen->tsc.entries[i] = entry;
945 return i;
946 }