projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
gallium: Create a new PIPE_CAP_TILE_RASTER_ORDER for vc4.
[mesa.git] / src / gallium / drivers / nouveau / nvc0 / nvc0_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 <xf86drm.h>
24 #include <nouveau_drm.h>
25 #include <nvif/class.h>
26 #include "util/u_format.h"
27 #include "util/u_format_s3tc.h"
28 #include "pipe/p_screen.h"
29
30 #include "nouveau_vp3_video.h"
31
32 #include "nvc0/nvc0_context.h"
33 #include "nvc0/nvc0_screen.h"
34
35 #include "nvc0/mme/com9097.mme.h"
36 #include "nvc0/mme/com90c0.mme.h"
37
38 #include "nv50/g80_texture.xml.h"
39
40 static boolean
41 nvc0_screen_is_format_supported(struct pipe_screen *pscreen,
42 enum pipe_format format,
43 enum pipe_texture_target target,
44 unsigned sample_count,
45 unsigned bindings)
46 {
47 const struct util_format_description *desc = util_format_description(format);
48
49 if (sample_count > 8)
50 return false;
51 if (!(0x117 & (1 << sample_count))) /* 0, 1, 2, 4 or 8 */
52 return false;
53
54 /* Short-circuit the rest of the logic -- this is used by the state tracker
55 * to determine valid MS levels in a no-attachments scenario.
56 */
57 if (format == PIPE_FORMAT_NONE && bindings & PIPE_BIND_RENDER_TARGET)
58 return true;
59
60 if (!util_format_is_supported(format, bindings))
61 return false;
62
63 if ((bindings & PIPE_BIND_SAMPLER_VIEW) && (target != PIPE_BUFFER))
64 if (util_format_get_blocksizebits(format) == 3 * 32)
65 return false;
66
67 if (bindings & PIPE_BIND_LINEAR)
68 if (util_format_is_depth_or_stencil(format) ||
69 (target != PIPE_TEXTURE_1D &&
70 target != PIPE_TEXTURE_2D &&
71 target != PIPE_TEXTURE_RECT) ||
72 sample_count > 1)
73 return false;
74
75 /* Restrict ETC2 and ASTC formats here. These are only supported on GK20A.
76 */
77 if ((desc->layout == UTIL_FORMAT_LAYOUT_ETC ||
78 desc->layout == UTIL_FORMAT_LAYOUT_ASTC) &&
79 /* The claim is that this should work on GM107 but it doesn't. Need to
80 * test further and figure out if it's a nouveau issue or a HW one.
81 nouveau_screen(pscreen)->class_3d < GM107_3D_CLASS &&
82 */
83 nouveau_screen(pscreen)->class_3d != NVEA_3D_CLASS)
84 return false;
85
86 /* shared is always supported */
87 bindings &= ~(PIPE_BIND_LINEAR |
88 PIPE_BIND_SHARED);
89
90 if (bindings & PIPE_BIND_SHADER_IMAGE) {
91 if (sample_count > 1 &&
92 nouveau_screen(pscreen)->class_3d >= GM107_3D_CLASS) {
93 /* MS images are currently unsupported on Maxwell because they have to
94 * be handled explicitly. */
95 return false;
96 }
97
98 if (format == PIPE_FORMAT_B8G8R8A8_UNORM &&
99 nouveau_screen(pscreen)->class_3d < NVE4_3D_CLASS) {
100 /* This should work on Fermi, but for currently unknown reasons it
101 * does not and results in breaking reads from pbos. */
102 return false;
103 }
104 }
105
106 return (( nvc0_format_table[format].usage |
107 nvc0_vertex_format[format].usage) & bindings) == bindings;
108 }
109
110 static int
111 nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
112 {
113 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
114 struct nouveau_device *dev = nouveau_screen(pscreen)->device;
115
116 switch (param) {
117 /* non-boolean caps */
118 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
119 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
120 return 15;
121 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
122 return 12;
123 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
124 return 2048;
125 case PIPE_CAP_MIN_TEXEL_OFFSET:
126 return -8;
127 case PIPE_CAP_MAX_TEXEL_OFFSET:
128 return 7;
129 case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
130 return -32;
131 case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
132 return 31;
133 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
134 return 128 * 1024 * 1024;
135 case PIPE_CAP_GLSL_FEATURE_LEVEL:
136 return 430;
137 case PIPE_CAP_MAX_RENDER_TARGETS:
138 return 8;
139 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
140 return 1;
141 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
142 return 4;
143 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
144 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
145 return 128;
146 case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
147 case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
148 return 1024;
149 case PIPE_CAP_MAX_VERTEX_STREAMS:
150 return 4;
151 case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
152 return 2048;
153 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
154 return 256;
155 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
156 if (class_3d < GM107_3D_CLASS)
157 return 256; /* IMAGE bindings require alignment to 256 */
158 return 16;
159 case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
160 return 16;
161 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
162 return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
163 case PIPE_CAP_MAX_VIEWPORTS:
164 return NVC0_MAX_VIEWPORTS;
165 case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
166 return 4;
167 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
168 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
169 case PIPE_CAP_ENDIANNESS:
170 return PIPE_ENDIAN_LITTLE;
171 case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
172 return 30;
173 case PIPE_CAP_MAX_WINDOW_RECTANGLES:
174 return NVC0_MAX_WINDOW_RECTANGLES;
175
176 /* supported caps */
177 case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
178 case PIPE_CAP_TEXTURE_SWIZZLE:
179 case PIPE_CAP_TEXTURE_SHADOW_MAP:
180 case PIPE_CAP_NPOT_TEXTURES:
181 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
182 case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
183 case PIPE_CAP_ANISOTROPIC_FILTER:
184 case PIPE_CAP_SEAMLESS_CUBE_MAP:
185 case PIPE_CAP_CUBE_MAP_ARRAY:
186 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
187 case PIPE_CAP_TEXTURE_MULTISAMPLE:
188 case PIPE_CAP_TWO_SIDED_STENCIL:
189 case PIPE_CAP_DEPTH_CLIP_DISABLE:
190 case PIPE_CAP_POINT_SPRITE:
191 case PIPE_CAP_TGSI_TEXCOORD:
192 case PIPE_CAP_SM3:
193 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
194 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
195 case PIPE_CAP_VERTEX_COLOR_CLAMPED:
196 case PIPE_CAP_QUERY_TIMESTAMP:
197 case PIPE_CAP_QUERY_TIME_ELAPSED:
198 case PIPE_CAP_OCCLUSION_QUERY:
199 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
200 case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
201 case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
202 case PIPE_CAP_BLEND_EQUATION_SEPARATE:
203 case PIPE_CAP_INDEP_BLEND_ENABLE:
204 case PIPE_CAP_INDEP_BLEND_FUNC:
205 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
206 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
207 case PIPE_CAP_PRIMITIVE_RESTART:
208 case PIPE_CAP_TGSI_INSTANCEID:
209 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
210 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
211 case PIPE_CAP_CONDITIONAL_RENDER:
212 case PIPE_CAP_TEXTURE_BARRIER:
213 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
214 case PIPE_CAP_START_INSTANCE:
215 case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
216 case PIPE_CAP_DRAW_INDIRECT:
217 case PIPE_CAP_USER_CONSTANT_BUFFERS:
218 case PIPE_CAP_USER_VERTEX_BUFFERS:
219 case PIPE_CAP_TEXTURE_QUERY_LOD:
220 case PIPE_CAP_SAMPLE_SHADING:
221 case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
222 case PIPE_CAP_TEXTURE_GATHER_SM5:
223 case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
224 case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
225 case PIPE_CAP_SAMPLER_VIEW_TARGET:
226 case PIPE_CAP_CLIP_HALFZ:
227 case PIPE_CAP_POLYGON_OFFSET_CLAMP:
228 case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
229 case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
230 case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
231 case PIPE_CAP_DEPTH_BOUNDS_TEST:
232 case PIPE_CAP_TGSI_TXQS:
233 case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
234 case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
235 case PIPE_CAP_SHAREABLE_SHADERS:
236 case PIPE_CAP_CLEAR_TEXTURE:
237 case PIPE_CAP_DRAW_PARAMETERS:
238 case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
239 case PIPE_CAP_MULTI_DRAW_INDIRECT:
240 case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
241 case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
242 case PIPE_CAP_QUERY_BUFFER_OBJECT:
243 case PIPE_CAP_INVALIDATE_BUFFER:
244 case PIPE_CAP_STRING_MARKER:
245 case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
246 case PIPE_CAP_CULL_DISTANCE:
247 case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
248 case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
249 case PIPE_CAP_TGSI_VOTE:
250 case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
251 case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
252 case PIPE_CAP_TGSI_MUL_ZERO_WINS:
253 case PIPE_CAP_DOUBLES:
254 case PIPE_CAP_INT64:
255 case PIPE_CAP_TGSI_TEX_TXF_LZ:
256 case PIPE_CAP_TGSI_CLOCK:
257 case PIPE_CAP_COMPUTE:
258 case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
259 case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
260 return 1;
261 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
262 return (class_3d >= NVE4_3D_CLASS) ? 1 : 0;
263 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
264 return nouveau_screen(pscreen)->vram_domain & NOUVEAU_BO_VRAM ? 1 : 0;
265 case PIPE_CAP_TGSI_FS_FBFETCH:
266 return class_3d >= NVE4_3D_CLASS; /* needs testing on fermi */
267 case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
268 case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
269 case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
270 case PIPE_CAP_POST_DEPTH_COVERAGE:
271 return class_3d >= GM200_3D_CLASS;
272 case PIPE_CAP_TGSI_BALLOT:
273 return class_3d >= NVE4_3D_CLASS;
274
275 /* unsupported caps */
276 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
277 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
278 case PIPE_CAP_SHADER_STENCIL_EXPORT:
279 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
280 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
281 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
282 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
283 case PIPE_CAP_FAKE_SW_MSAA:
284 case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
285 case PIPE_CAP_VERTEXID_NOBASE:
286 case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
287 case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
288 case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
289 case PIPE_CAP_GENERATE_MIPMAP:
290 case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
291 case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
292 case PIPE_CAP_QUERY_MEMORY_INFO:
293 case PIPE_CAP_PCI_GROUP:
294 case PIPE_CAP_PCI_BUS:
295 case PIPE_CAP_PCI_DEVICE:
296 case PIPE_CAP_PCI_FUNCTION:
297 case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
298 case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
299 case PIPE_CAP_NATIVE_FENCE_FD:
300 case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
301 case PIPE_CAP_INT64_DIVMOD:
302 case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
303 case PIPE_CAP_BINDLESS_TEXTURE:
304 case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
305 case PIPE_CAP_QUERY_SO_OVERFLOW:
306 case PIPE_CAP_MEMOBJ:
307 case PIPE_CAP_LOAD_CONSTBUF:
308 case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS:
309 case PIPE_CAP_TILE_RASTER_ORDER:
310 return 0;
311
312 case PIPE_CAP_VENDOR_ID:
313 return 0x10de;
314 case PIPE_CAP_DEVICE_ID: {
315 uint64_t device_id;
316 if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
317 NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
318 return -1;
319 }
320 return device_id;
321 }
322 case PIPE_CAP_ACCELERATED:
323 return 1;
324 case PIPE_CAP_VIDEO_MEMORY:
325 return dev->vram_size >> 20;
326 case PIPE_CAP_UMA:
327 return 0;
328 }
329
330 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
331 return 0;
332 }
333
334 static int
335 nvc0_screen_get_shader_param(struct pipe_screen *pscreen,
336 enum pipe_shader_type shader,
337 enum pipe_shader_cap param)
338 {
339 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
340
341 switch (shader) {
342 case PIPE_SHADER_VERTEX:
343 case PIPE_SHADER_GEOMETRY:
344 case PIPE_SHADER_FRAGMENT:
345 case PIPE_SHADER_COMPUTE:
346 case PIPE_SHADER_TESS_CTRL:
347 case PIPE_SHADER_TESS_EVAL:
348 break;
349 default:
350 return 0;
351 }
352
353 switch (param) {
354 case PIPE_SHADER_CAP_PREFERRED_IR:
355 return PIPE_SHADER_IR_TGSI;
356 case PIPE_SHADER_CAP_SUPPORTED_IRS:
357 return 1 << PIPE_SHADER_IR_TGSI;
358 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
359 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
360 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
361 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
362 return 16384;
363 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
364 return 16;
365 case PIPE_SHADER_CAP_MAX_INPUTS:
366 if (shader == PIPE_SHADER_VERTEX)
367 return 32;
368 /* NOTE: These only count our slots for GENERIC varyings.
369 * The address space may be larger, but the actual hard limit seems to be
370 * less than what the address space layout permits, so don't add TEXCOORD,
371 * COLOR, etc. here.
372 */
373 if (shader == PIPE_SHADER_FRAGMENT)
374 return 0x1f0 / 16;
375 /* Actually this counts CLIPVERTEX, which occupies the last generic slot,
376 * and excludes 0x60 per-patch inputs.
377 */
378 return 0x200 / 16;
379 case PIPE_SHADER_CAP_MAX_OUTPUTS:
380 return 32;
381 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
382 return 65536;
383 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
384 return NVC0_MAX_PIPE_CONSTBUFS;
385 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
386 return shader != PIPE_SHADER_FRAGMENT;
387 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
388 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
389 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
390 return 1;
391 case PIPE_SHADER_CAP_MAX_TEMPS:
392 return NVC0_CAP_MAX_PROGRAM_TEMPS;
393 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
394 return 1;
395 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
396 return 1;
397 case PIPE_SHADER_CAP_SUBROUTINES:
398 return 1;
399 case PIPE_SHADER_CAP_INTEGERS:
400 return 1;
401 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
402 return 1;
403 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
404 return 1;
405 case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
406 return 1;
407 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
408 case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
409 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
410 case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
411 case PIPE_SHADER_CAP_INT64_ATOMICS:
412 case PIPE_SHADER_CAP_FP16:
413 return 0;
414 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
415 return NVC0_MAX_BUFFERS;
416 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
417 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
418 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
419 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
420 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
421 return 32;
422 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
423 if (class_3d >= NVE4_3D_CLASS)
424 return NVC0_MAX_IMAGES;
425 if (shader == PIPE_SHADER_FRAGMENT || shader == PIPE_SHADER_COMPUTE)
426 return NVC0_MAX_IMAGES;
427 return 0;
428 default:
429 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
430 return 0;
431 }
432 }
433
434 static float
435 nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
436 {
437 switch (param) {
438 case PIPE_CAPF_MAX_LINE_WIDTH:
439 case PIPE_CAPF_MAX_LINE_WIDTH_AA:
440 return 10.0f;
441 case PIPE_CAPF_MAX_POINT_WIDTH:
442 return 63.0f;
443 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
444 return 63.375f;
445 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
446 return 16.0f;
447 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
448 return 15.0f;
449 case PIPE_CAPF_GUARD_BAND_LEFT:
450 case PIPE_CAPF_GUARD_BAND_TOP:
451 return 0.0f;
452 case PIPE_CAPF_GUARD_BAND_RIGHT:
453 case PIPE_CAPF_GUARD_BAND_BOTTOM:
454 return 0.0f; /* that or infinity */
455 }
456
457 NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
458 return 0.0f;
459 }
460
461 static int
462 nvc0_screen_get_compute_param(struct pipe_screen *pscreen,
463 enum pipe_shader_ir ir_type,
464 enum pipe_compute_cap param, void *data)
465 {
466 struct nvc0_screen *screen = nvc0_screen(pscreen);
467 const uint16_t obj_class = screen->compute->oclass;
468
469 #define RET(x) do { \
470 if (data) \
471 memcpy(data, x, sizeof(x)); \
472 return sizeof(x); \
473 } while (0)
474
475 switch (param) {
476 case PIPE_COMPUTE_CAP_GRID_DIMENSION:
477 RET((uint64_t []) { 3 });
478 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
479 if (obj_class >= NVE4_COMPUTE_CLASS) {
480 RET(((uint64_t []) { 0x7fffffff, 65535, 65535 }));
481 } else {
482 RET(((uint64_t []) { 65535, 65535, 65535 }));
483 }
484 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
485 RET(((uint64_t []) { 1024, 1024, 64 }));
486 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
487 RET((uint64_t []) { 1024 });
488 case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
489 if (obj_class >= NVE4_COMPUTE_CLASS) {
490 RET((uint64_t []) { 1024 });
491 } else {
492 RET((uint64_t []) { 512 });
493 }
494 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */
495 RET((uint64_t []) { 1ULL << 40 });
496 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
497 switch (obj_class) {
498 case GM200_COMPUTE_CLASS:
499 RET((uint64_t []) { 96 << 10 });
500 break;
501 case GM107_COMPUTE_CLASS:
502 RET((uint64_t []) { 64 << 10 });
503 break;
504 default:
505 RET((uint64_t []) { 48 << 10 });
506 break;
507 }
508 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
509 RET((uint64_t []) { 512 << 10 });
510 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
511 RET((uint64_t []) { 4096 });
512 case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
513 RET((uint32_t []) { 32 });
514 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
515 RET((uint64_t []) { 1ULL << 40 });
516 case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
517 RET((uint32_t []) { 0 });
518 case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
519 RET((uint32_t []) { screen->mp_count_compute });
520 case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
521 RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
522 case PIPE_COMPUTE_CAP_ADDRESS_BITS:
523 RET((uint32_t []) { 64 });
524 default:
525 return 0;
526 }
527
528 #undef RET
529 }
530
531 static void
532 nvc0_screen_destroy(struct pipe_screen *pscreen)
533 {
534 struct nvc0_screen *screen = nvc0_screen(pscreen);
535
536 if (!nouveau_drm_screen_unref(&screen->base))
537 return;
538
539 if (screen->base.fence.current) {
540 struct nouveau_fence *current = NULL;
541
542 /* nouveau_fence_wait will create a new current fence, so wait on the
543 * _current_ one, and remove both.
544 */
545 nouveau_fence_ref(screen->base.fence.current, &current);
546 nouveau_fence_wait(current, NULL);
547 nouveau_fence_ref(NULL, &current);
548 nouveau_fence_ref(NULL, &screen->base.fence.current);
549 }
550 if (screen->base.pushbuf)
551 screen->base.pushbuf->user_priv = NULL;
552
553 if (screen->blitter)
554 nvc0_blitter_destroy(screen);
555 if (screen->pm.prog) {
556 screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
557 nvc0_program_destroy(NULL, screen->pm.prog);
558 FREE(screen->pm.prog);
559 }
560
561 nouveau_bo_ref(NULL, &screen->text);
562 nouveau_bo_ref(NULL, &screen->uniform_bo);
563 nouveau_bo_ref(NULL, &screen->tls);
564 nouveau_bo_ref(NULL, &screen->txc);
565 nouveau_bo_ref(NULL, &screen->fence.bo);
566 nouveau_bo_ref(NULL, &screen->poly_cache);
567
568 nouveau_heap_destroy(&screen->lib_code);
569 nouveau_heap_destroy(&screen->text_heap);
570
571 FREE(screen->default_tsc);
572 FREE(screen->tic.entries);
573
574 nouveau_object_del(&screen->eng3d);
575 nouveau_object_del(&screen->eng2d);
576 nouveau_object_del(&screen->m2mf);
577 nouveau_object_del(&screen->compute);
578 nouveau_object_del(&screen->nvsw);
579
580 nouveau_screen_fini(&screen->base);
581
582 FREE(screen);
583 }
584
585 static int
586 nvc0_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos,
587 unsigned size, const uint32_t *data)
588 {
589 struct nouveau_pushbuf *push = screen->base.pushbuf;
590
591 size /= 4;
592
593 assert((pos + size) <= 0x800);
594
595 BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2);
596 PUSH_DATA (push, (m - 0x3800) / 8);
597 PUSH_DATA (push, pos);
598 BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1);
599 PUSH_DATA (push, pos);
600 PUSH_DATAp(push, data, size);
601
602 return pos + size;
603 }
604
605 static void
606 nvc0_magic_3d_init(struct nouveau_pushbuf *push, uint16_t obj_class)
607 {
608 BEGIN_NVC0(push, SUBC_3D(0x10cc), 1);
609 PUSH_DATA (push, 0xff);
610 BEGIN_NVC0(push, SUBC_3D(0x10e0), 2);
611 PUSH_DATA (push, 0xff);
612 PUSH_DATA (push, 0xff);
613 BEGIN_NVC0(push, SUBC_3D(0x10ec), 2);
614 PUSH_DATA (push, 0xff);
615 PUSH_DATA (push, 0xff);
616 BEGIN_NVC0(push, SUBC_3D(0x074c), 1);
617 PUSH_DATA (push, 0x3f);
618
619 BEGIN_NVC0(push, SUBC_3D(0x16a8), 1);
620 PUSH_DATA (push, (3 << 16) | 3);
621 BEGIN_NVC0(push, SUBC_3D(0x1794), 1);
622 PUSH_DATA (push, (2 << 16) | 2);
623
624 if (obj_class < GM107_3D_CLASS) {
625 BEGIN_NVC0(push, SUBC_3D(0x12ac), 1);
626 PUSH_DATA (push, 0);
627 }
628 BEGIN_NVC0(push, SUBC_3D(0x0218), 1);
629 PUSH_DATA (push, 0x10);
630 BEGIN_NVC0(push, SUBC_3D(0x10fc), 1);
631 PUSH_DATA (push, 0x10);
632 BEGIN_NVC0(push, SUBC_3D(0x1290), 1);
633 PUSH_DATA (push, 0x10);
634 BEGIN_NVC0(push, SUBC_3D(0x12d8), 2);
635 PUSH_DATA (push, 0x10);
636 PUSH_DATA (push, 0x10);
637 BEGIN_NVC0(push, SUBC_3D(0x1140), 1);
638 PUSH_DATA (push, 0x10);
639 BEGIN_NVC0(push, SUBC_3D(0x1610), 1);
640 PUSH_DATA (push, 0xe);
641
642 BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_GEN_MODE), 1);
643 PUSH_DATA (push, NVC0_3D_VERTEX_ID_GEN_MODE_DRAW_ARRAYS_ADD_START);
644 BEGIN_NVC0(push, SUBC_3D(0x030c), 1);
645 PUSH_DATA (push, 0);
646 BEGIN_NVC0(push, SUBC_3D(0x0300), 1);
647 PUSH_DATA (push, 3);
648
649 BEGIN_NVC0(push, SUBC_3D(0x02d0), 1);
650 PUSH_DATA (push, 0x3fffff);
651 BEGIN_NVC0(push, SUBC_3D(0x0fdc), 1);
652 PUSH_DATA (push, 1);
653 BEGIN_NVC0(push, SUBC_3D(0x19c0), 1);
654 PUSH_DATA (push, 1);
655
656 if (obj_class < GM107_3D_CLASS) {
657 BEGIN_NVC0(push, SUBC_3D(0x075c), 1);
658 PUSH_DATA (push, 3);
659
660 if (obj_class >= NVE4_3D_CLASS) {
661 BEGIN_NVC0(push, SUBC_3D(0x07fc), 1);
662 PUSH_DATA (push, 1);
663 }
664 }
665
666 /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
667 * are supposed to do */
668 }
669
670 static void
671 nvc0_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
672 {
673 struct nvc0_screen *screen = nvc0_screen(pscreen);
674 struct nouveau_pushbuf *push = screen->base.pushbuf;
675
676 /* we need to do it after possible flush in MARK_RING */
677 *sequence = ++screen->base.fence.sequence;
678
679 assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
680 PUSH_DATA (push, NVC0_FIFO_PKHDR_SQ(NVC0_3D(QUERY_ADDRESS_HIGH), 4));
681 PUSH_DATAh(push, screen->fence.bo->offset);
682 PUSH_DATA (push, screen->fence.bo->offset);
683 PUSH_DATA (push, *sequence);
684 PUSH_DATA (push, NVC0_3D_QUERY_GET_FENCE | NVC0_3D_QUERY_GET_SHORT |
685 (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT));
686 }
687
688 static u32
689 nvc0_screen_fence_update(struct pipe_screen *pscreen)
690 {
691 struct nvc0_screen *screen = nvc0_screen(pscreen);
692 return screen->fence.map[0];
693 }
694
695 static int
696 nvc0_screen_init_compute(struct nvc0_screen *screen)
697 {
698 screen->base.base.get_compute_param = nvc0_screen_get_compute_param;
699
700 switch (screen->base.device->chipset & ~0xf) {
701 case 0xc0:
702 case 0xd0:
703 return nvc0_screen_compute_setup(screen, screen->base.pushbuf);
704 case 0xe0:
705 case 0xf0:
706 case 0x100:
707 case 0x110:
708 case 0x120:
709 case 0x130:
710 return nve4_screen_compute_setup(screen, screen->base.pushbuf);
711 default:
712 return -1;
713 }
714 }
715
716 static int
717 nvc0_screen_resize_tls_area(struct nvc0_screen *screen,
718 uint32_t lpos, uint32_t lneg, uint32_t cstack)
719 {
720 struct nouveau_bo *bo = NULL;
721 int ret;
722 uint64_t size = (lpos + lneg) * 32 + cstack;
723
724 if (size >= (1 << 20)) {
725 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64"\n", size);
726 return -1;
727 }
728
729 size *= (screen->base.device->chipset >= 0xe0) ? 64 : 48; /* max warps */
730 size = align(size, 0x8000);
731 size *= screen->mp_count;
732
733 size = align(size, 1 << 17);
734
735 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base), 1 << 17, size,
736 NULL, &bo);
737 if (ret)
738 return ret;
739 nouveau_bo_ref(NULL, &screen->tls);
740 screen->tls = bo;
741 return 0;
742 }
743
744 int
745 nvc0_screen_resize_text_area(struct nvc0_screen *screen, uint64_t size)
746 {
747 struct nouveau_pushbuf *push = screen->base.pushbuf;
748 struct nouveau_bo *bo;
749 int ret;
750
751 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base),
752 1 << 17, size, NULL, &bo);
753 if (ret)
754 return ret;
755
756 nouveau_bo_ref(NULL, &screen->text);
757 screen->text = bo;
758
759 nouveau_heap_destroy(&screen->lib_code);
760 nouveau_heap_destroy(&screen->text_heap);
761
762 /* XXX: getting a page fault at the end of the code buffer every few
763 * launches, don't use the last 256 bytes to work around them - prefetch ?
764 */
765 nouveau_heap_init(&screen->text_heap, 0, size - 0x100);
766
767 /* update the code segment setup */
768 BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
769 PUSH_DATAh(push, screen->text->offset);
770 PUSH_DATA (push, screen->text->offset);
771 if (screen->compute) {
772 BEGIN_NVC0(push, NVC0_CP(CODE_ADDRESS_HIGH), 2);
773 PUSH_DATAh(push, screen->text->offset);
774 PUSH_DATA (push, screen->text->offset);
775 }
776
777 return 0;
778 }
779
780 #define FAIL_SCREEN_INIT(str, err) \
781 do { \
782 NOUVEAU_ERR(str, err); \
783 goto fail; \
784 } while(0)
785
786 struct nouveau_screen *
787 nvc0_screen_create(struct nouveau_device *dev)
788 {
789 struct nvc0_screen *screen;
790 struct pipe_screen *pscreen;
791 struct nouveau_object *chan;
792 struct nouveau_pushbuf *push;
793 uint64_t value;
794 uint32_t obj_class;
795 uint32_t flags;
796 int ret;
797 unsigned i;
798
799 switch (dev->chipset & ~0xf) {
800 case 0xc0:
801 case 0xd0:
802 case 0xe0:
803 case 0xf0:
804 case 0x100:
805 case 0x110:
806 case 0x120:
807 case 0x130:
808 break;
809 default:
810 return NULL;
811 }
812
813 screen = CALLOC_STRUCT(nvc0_screen);
814 if (!screen)
815 return NULL;
816 pscreen = &screen->base.base;
817 pscreen->destroy = nvc0_screen_destroy;
818
819 ret = nouveau_screen_init(&screen->base, dev);
820 if (ret)
821 FAIL_SCREEN_INIT("Base screen init failed: %d\n", ret);
822 chan = screen->base.channel;
823 push = screen->base.pushbuf;
824 push->user_priv = screen;
825 push->rsvd_kick = 5;
826
827 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
828 PIPE_BIND_SHADER_BUFFER |
829 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
830 PIPE_BIND_COMMAND_ARGS_BUFFER | PIPE_BIND_QUERY_BUFFER;
831 screen->base.sysmem_bindings |=
832 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
833
834 if (screen->base.vram_domain & NOUVEAU_BO_GART) {
835 screen->base.sysmem_bindings |= screen->base.vidmem_bindings;
836 screen->base.vidmem_bindings = 0;
837 }
838
839 pscreen->context_create = nvc0_create;
840 pscreen->is_format_supported = nvc0_screen_is_format_supported;
841 pscreen->get_param = nvc0_screen_get_param;
842 pscreen->get_shader_param = nvc0_screen_get_shader_param;
843 pscreen->get_paramf = nvc0_screen_get_paramf;
844 pscreen->get_driver_query_info = nvc0_screen_get_driver_query_info;
845 pscreen->get_driver_query_group_info = nvc0_screen_get_driver_query_group_info;
846
847 nvc0_screen_init_resource_functions(pscreen);
848
849 screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
850 screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
851
852 flags = NOUVEAU_BO_GART | NOUVEAU_BO_MAP;
853 if (screen->base.drm->version >= 0x01000202)
854 flags |= NOUVEAU_BO_COHERENT;
855
856 ret = nouveau_bo_new(dev, flags, 0, 4096, NULL, &screen->fence.bo);
857 if (ret)
858 FAIL_SCREEN_INIT("Error allocating fence BO: %d\n", ret);
859 nouveau_bo_map(screen->fence.bo, 0, NULL);
860 screen->fence.map = screen->fence.bo->map;
861 screen->base.fence.emit = nvc0_screen_fence_emit;
862 screen->base.fence.update = nvc0_screen_fence_update;
863
864
865 ret = nouveau_object_new(chan, (dev->chipset < 0xe0) ? 0x1f906e : 0x906e,
866 NVIF_CLASS_SW_GF100, NULL, 0, &screen->nvsw);
867 if (ret)
868 FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret);
869
870 BEGIN_NVC0(push, SUBC_SW(NV01_SUBCHAN_OBJECT), 1);
871 PUSH_DATA (push, screen->nvsw->handle);
872
873 switch (dev->chipset & ~0xf) {
874 case 0x130:
875 case 0x120:
876 case 0x110:
877 case 0x100:
878 case 0xf0:
879 obj_class = NVF0_P2MF_CLASS;
880 break;
881 case 0xe0:
882 obj_class = NVE4_P2MF_CLASS;
883 break;
884 default:
885 obj_class = NVC0_M2MF_CLASS;
886 break;
887 }
888 ret = nouveau_object_new(chan, 0xbeef323f, obj_class, NULL, 0,
889 &screen->m2mf);
890 if (ret)
891 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
892
893 BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
894 PUSH_DATA (push, screen->m2mf->oclass);
895 if (screen->m2mf->oclass == NVE4_P2MF_CLASS) {
896 BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
897 PUSH_DATA (push, 0xa0b5);
898 }
899
900 ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
901 &screen->eng2d);
902 if (ret)
903 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
904
905 BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
906 PUSH_DATA (push, screen->eng2d->oclass);
907 BEGIN_NVC0(push, SUBC_2D(NVC0_2D_SINGLE_GPC), 1);
908 PUSH_DATA (push, 0);
909 BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
910 PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
911 BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
912 PUSH_DATA (push, 0);
913 BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
914 PUSH_DATA (push, 0);
915 BEGIN_NVC0(push, SUBC_2D(0x0884), 1);
916 PUSH_DATA (push, 0x3f);
917 BEGIN_NVC0(push, SUBC_2D(0x0888), 1);
918 PUSH_DATA (push, 1);
919 BEGIN_NVC0(push, NVC0_2D(COND_MODE), 1);
920 PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
921
922 BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
923 PUSH_DATAh(push, screen->fence.bo->offset + 16);
924 PUSH_DATA (push, screen->fence.bo->offset + 16);
925
926 switch (dev->chipset & ~0xf) {
927 case 0x130:
928 switch (dev->chipset) {
929 case 0x130:
930 case 0x13b:
931 obj_class = GP100_3D_CLASS;
932 break;
933 default:
934 obj_class = GP102_3D_CLASS;
935 break;
936 }
937 break;
938 case 0x120:
939 obj_class = GM200_3D_CLASS;
940 break;
941 case 0x110:
942 obj_class = GM107_3D_CLASS;
943 break;
944 case 0x100:
945 case 0xf0:
946 obj_class = NVF0_3D_CLASS;
947 break;
948 case 0xe0:
949 switch (dev->chipset) {
950 case 0xea:
951 obj_class = NVEA_3D_CLASS;
952 break;
953 default:
954 obj_class = NVE4_3D_CLASS;
955 break;
956 }
957 break;
958 case 0xd0:
959 obj_class = NVC8_3D_CLASS;
960 break;
961 case 0xc0:
962 default:
963 switch (dev->chipset) {
964 case 0xc8:
965 obj_class = NVC8_3D_CLASS;
966 break;
967 case 0xc1:
968 obj_class = NVC1_3D_CLASS;
969 break;
970 default:
971 obj_class = NVC0_3D_CLASS;
972 break;
973 }
974 break;
975 }
976 ret = nouveau_object_new(chan, 0xbeef003d, obj_class, NULL, 0,
977 &screen->eng3d);
978 if (ret)
979 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
980 screen->base.class_3d = obj_class;
981
982 BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
983 PUSH_DATA (push, screen->eng3d->oclass);
984
985 BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
986 PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
987
988 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
989 /* kill shaders after about 1 second (at 100 MHz) */
990 BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
991 PUSH_DATA (push, 0x17);
992 }
993
994 IMMED_NVC0(push, NVC0_3D(ZETA_COMP_ENABLE),
995 screen->base.drm->version >= 0x01000101);
996 BEGIN_NVC0(push, NVC0_3D(RT_COMP_ENABLE(0)), 8);
997 for (i = 0; i < 8; ++i)
998 PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
999
1000 BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
1001 PUSH_DATA (push, 1);
1002
1003 BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
1004 PUSH_DATA (push, 0);
1005 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
1006 PUSH_DATA (push, 0);
1007 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
1008 PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
1009 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
1010 PUSH_DATA (push, 0);
1011 BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
1012 PUSH_DATA (push, 1);
1013 BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
1014 PUSH_DATA (push, 1);
1015 BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
1016 PUSH_DATA (push, 1);
1017 BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
1018 PUSH_DATA (push, 0);
1019 BEGIN_NVC0(push, NVC0_3D(SHADE_MODEL), 1);
1020 PUSH_DATA (push, NVC0_3D_SHADE_MODEL_SMOOTH);
1021 if (screen->eng3d->oclass < NVE4_3D_CLASS) {
1022 IMMED_NVC0(push, NVC0_3D(TEX_MISC), 0);
1023 } else {
1024 BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
1025 PUSH_DATA (push, 15);
1026 }
1027 BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
1028 PUSH_DATA (push, 8); /* 128 */
1029 BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
1030 PUSH_DATA (push, 1);
1031 if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
1032 BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
1033 PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
1034 }
1035
1036 nvc0_magic_3d_init(push, screen->eng3d->oclass);
1037
1038 ret = nvc0_screen_resize_text_area(screen, 1 << 19);
1039 if (ret)
1040 FAIL_SCREEN_INIT("Error allocating TEXT area: %d\n", ret);
1041
1042 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 12, 7 << 16, NULL,
1043 &screen->uniform_bo);
1044 if (ret)
1045 FAIL_SCREEN_INIT("Error allocating uniform BO: %d\n", ret);
1046
1047 PUSH_REFN (push, screen->uniform_bo, NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_WR);
1048
1049 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
1050 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1051 PUSH_DATA (push, 256);
1052 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1053 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1054 BEGIN_1IC0(push, NVC0_3D(CB_POS), 5);
1055 PUSH_DATA (push, 0);
1056 PUSH_DATAf(push, 0.0f);
1057 PUSH_DATAf(push, 0.0f);
1058 PUSH_DATAf(push, 0.0f);
1059 PUSH_DATAf(push, 0.0f);
1060 BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
1061 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1062 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1063
1064 if (screen->base.drm->version >= 0x01000101) {
1065 ret = nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
1066 if (ret)
1067 FAIL_SCREEN_INIT("NOUVEAU_GETPARAM_GRAPH_UNITS failed: %d\n", ret);
1068 } else {
1069 if (dev->chipset >= 0xe0 && dev->chipset < 0xf0)
1070 value = (8 << 8) | 4;
1071 else
1072 value = (16 << 8) | 4;
1073 }
1074 screen->gpc_count = value & 0x000000ff;
1075 screen->mp_count = value >> 8;
1076 screen->mp_count_compute = screen->mp_count;
1077
1078 ret = nvc0_screen_resize_tls_area(screen, 128 * 16, 0, 0x200);
1079 if (ret)
1080 FAIL_SCREEN_INIT("Error allocating TLS area: %d\n", ret);
1081
1082 BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
1083 PUSH_DATAh(push, screen->tls->offset);
1084 PUSH_DATA (push, screen->tls->offset);
1085 PUSH_DATA (push, screen->tls->size >> 32);
1086 PUSH_DATA (push, screen->tls->size);
1087 BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
1088 PUSH_DATA (push, 0);
1089 /* Reduce likelihood of collision with real buffers by placing the hole at
1090 * the top of the 4G area. This will have to be dealt with for real
1091 * eventually by blocking off that area from the VM.
1092 */
1093 BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
1094 PUSH_DATA (push, 0xff << 24);
1095
1096 if (screen->eng3d->oclass < GM107_3D_CLASS) {
1097 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 20, NULL,
1098 &screen->poly_cache);
1099 if (ret)
1100 FAIL_SCREEN_INIT("Error allocating poly cache BO: %d\n", ret);
1101
1102 BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
1103 PUSH_DATAh(push, screen->poly_cache->offset);
1104 PUSH_DATA (push, screen->poly_cache->offset);
1105 PUSH_DATA (push, 3);
1106 }
1107
1108 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 17, NULL,
1109 &screen->txc);
1110 if (ret)
1111 FAIL_SCREEN_INIT("Error allocating txc BO: %d\n", ret);
1112
1113 BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
1114 PUSH_DATAh(push, screen->txc->offset);
1115 PUSH_DATA (push, screen->txc->offset);
1116 PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
1117 if (screen->eng3d->oclass >= GM107_3D_CLASS) {
1118 screen->tic.maxwell = true;
1119 if (screen->eng3d->oclass == GM107_3D_CLASS) {
1120 screen->tic.maxwell =
1121 debug_get_bool_option("NOUVEAU_MAXWELL_TIC", true);
1122 IMMED_NVC0(push, SUBC_3D(0x0f10), screen->tic.maxwell);
1123 }
1124 }
1125
1126 BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
1127 PUSH_DATAh(push, screen->txc->offset + 65536);
1128 PUSH_DATA (push, screen->txc->offset + 65536);
1129 PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
1130
1131 BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
1132 PUSH_DATA (push, 0);
1133 BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
1134 PUSH_DATA (push, 0);
1135 PUSH_DATA (push, 0);
1136 BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
1137 PUSH_DATA (push, 0x3f);
1138
1139 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
1140 PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
1141 BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
1142 for (i = 0; i < 8 * 2; ++i)
1143 PUSH_DATA(push, 0);
1144 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
1145 PUSH_DATA (push, 0);
1146 BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
1147 PUSH_DATA (push, 0);
1148
1149 /* neither scissors, viewport nor stencil mask should affect clears */
1150 BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
1151 PUSH_DATA (push, 0);
1152
1153 BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
1154 PUSH_DATA (push, 1);
1155 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1156 BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(i)), 2);
1157 PUSH_DATAf(push, 0.0f);
1158 PUSH_DATAf(push, 1.0f);
1159 }
1160 BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
1161 PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
1162
1163 /* We use scissors instead of exact view volume clipping,
1164 * so they're always enabled.
1165 */
1166 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1167 BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(i)), 3);
1168 PUSH_DATA (push, 1);
1169 PUSH_DATA (push, 8192 << 16);
1170 PUSH_DATA (push, 8192 << 16);
1171 }
1172
1173 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
1174
1175 i = 0;
1176 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, mme9097_per_instance_bf);
1177 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, mme9097_blend_enables);
1178 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, mme9097_vertex_array_select);
1179 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, mme9097_tep_select);
1180 MK_MACRO(NVC0_3D_MACRO_GP_SELECT, mme9097_gp_select);
1181 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, mme9097_poly_mode_front);
1182 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, mme9097_poly_mode_back);
1183 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT, mme9097_draw_arrays_indirect);
1184 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT, mme9097_draw_elts_indirect);
1185 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT, mme9097_draw_arrays_indirect_count);
1186 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT, mme9097_draw_elts_indirect_count);
1187 MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE, mme9097_query_buffer_write);
1188 MK_MACRO(NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT, mme90c0_launch_grid_indirect);
1189
1190 BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
1191 PUSH_DATA (push, 1);
1192 BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
1193 PUSH_DATA (push, 1);
1194 BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
1195 PUSH_DATA (push, 0x40);
1196 BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
1197 PUSH_DATA (push, 0);
1198 BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
1199 PUSH_DATA (push, 0x30);
1200 BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
1201 PUSH_DATA (push, 3);
1202 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
1203 PUSH_DATA (push, 0x20);
1204 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
1205 PUSH_DATA (push, 0x00);
1206 screen->save_state.patch_vertices = 3;
1207
1208 BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
1209 PUSH_DATA (push, 0);
1210 BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
1211 PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
1212
1213 IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
1214
1215 if (nvc0_screen_init_compute(screen))
1216 goto fail;
1217
1218 /* XXX: Compute and 3D are somehow aliased on Fermi. */
1219 for (i = 0; i < 5; ++i) {
1220 /* TIC and TSC entries for each unit (nve4+ only) */
1221 /* auxiliary constants (6 user clip planes, base instance id) */
1222 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1223 PUSH_DATA (push, NVC0_CB_AUX_SIZE);
1224 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(i));
1225 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(i));
1226 BEGIN_NVC0(push, NVC0_3D(CB_BIND(i)), 1);
1227 PUSH_DATA (push, (15 << 4) | 1);
1228 if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
1229 unsigned j;
1230 BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
1231 PUSH_DATA (push, NVC0_CB_AUX_UNK_INFO);
1232 for (j = 0; j < 8; ++j)
1233 PUSH_DATA(push, j);
1234 } else {
1235 BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
1236 PUSH_DATA (push, 0x54);
1237 }
1238
1239 /* MS sample coordinate offsets: these do not work with _ALT modes ! */
1240 BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 2 * 8);
1241 PUSH_DATA (push, NVC0_CB_AUX_MS_INFO);
1242 PUSH_DATA (push, 0); /* 0 */
1243 PUSH_DATA (push, 0);
1244 PUSH_DATA (push, 1); /* 1 */
1245 PUSH_DATA (push, 0);
1246 PUSH_DATA (push, 0); /* 2 */
1247 PUSH_DATA (push, 1);
1248 PUSH_DATA (push, 1); /* 3 */
1249 PUSH_DATA (push, 1);
1250 PUSH_DATA (push, 2); /* 4 */
1251 PUSH_DATA (push, 0);
1252 PUSH_DATA (push, 3); /* 5 */
1253 PUSH_DATA (push, 0);
1254 PUSH_DATA (push, 2); /* 6 */
1255 PUSH_DATA (push, 1);
1256 PUSH_DATA (push, 3); /* 7 */
1257 PUSH_DATA (push, 1);
1258 }
1259 BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
1260 PUSH_DATA (push, 0);
1261
1262 PUSH_KICK (push);
1263
1264 screen->tic.entries = CALLOC(4096, sizeof(void *));
1265 screen->tsc.entries = screen->tic.entries + 2048;
1266
1267 if (!nvc0_blitter_create(screen))
1268 goto fail;
1269
1270 screen->default_tsc = CALLOC_STRUCT(nv50_tsc_entry);
1271 screen->default_tsc->tsc[0] = G80_TSC_0_SRGB_CONVERSION;
1272
1273 nouveau_fence_new(&screen->base, &screen->base.fence.current);
1274
1275 return &screen->base;
1276
1277 fail:
1278 screen->base.base.context_create = NULL;
1279 return &screen->base;
1280 }
1281
1282 int
1283 nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
1284 {
1285 int i = screen->tic.next;
1286
1287 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1288 i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1289
1290 screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1291
1292 if (screen->tic.entries[i])
1293 nv50_tic_entry(screen->tic.entries[i])->id = -1;
1294
1295 screen->tic.entries[i] = entry;
1296 return i;
1297 }
1298
1299 int
1300 nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
1301 {
1302 int i = screen->tsc.next;
1303
1304 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1305 i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1306
1307 screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1308
1309 if (screen->tsc.entries[i])
1310 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1311
1312 screen->tsc.entries[i] = entry;
1313 return i;
1314 }