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