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
nvc0_screen_is_format_supported(struct pipe_screen * pscreen,enum pipe_format format,enum pipe_texture_target target,unsigned sample_count,unsigned bindings)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
nvc0_screen_get_param(struct pipe_screen * pscreen,enum pipe_cap param)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_NPOT_TEXTURES:
180 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
181 case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
182 case PIPE_CAP_ANISOTROPIC_FILTER:
183 case PIPE_CAP_SEAMLESS_CUBE_MAP:
184 case PIPE_CAP_CUBE_MAP_ARRAY:
185 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
186 case PIPE_CAP_TEXTURE_MULTISAMPLE:
187 case PIPE_CAP_DEPTH_CLIP_DISABLE:
188 case PIPE_CAP_POINT_SPRITE:
189 case PIPE_CAP_TGSI_TEXCOORD:
190 case PIPE_CAP_SM3:
191 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
192 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
193 case PIPE_CAP_VERTEX_COLOR_CLAMPED:
194 case PIPE_CAP_QUERY_TIMESTAMP:
195 case PIPE_CAP_QUERY_TIME_ELAPSED:
196 case PIPE_CAP_OCCLUSION_QUERY:
197 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
198 case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
199 case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
200 case PIPE_CAP_BLEND_EQUATION_SEPARATE:
201 case PIPE_CAP_INDEP_BLEND_ENABLE:
202 case PIPE_CAP_INDEP_BLEND_FUNC:
203 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
204 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
205 case PIPE_CAP_PRIMITIVE_RESTART:
206 case PIPE_CAP_TGSI_INSTANCEID:
207 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
208 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
209 case PIPE_CAP_CONDITIONAL_RENDER:
210 case PIPE_CAP_TEXTURE_BARRIER:
211 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
212 case PIPE_CAP_START_INSTANCE:
213 case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
214 case PIPE_CAP_DRAW_INDIRECT:
215 case PIPE_CAP_USER_VERTEX_BUFFERS:
216 case PIPE_CAP_TEXTURE_QUERY_LOD:
217 case PIPE_CAP_SAMPLE_SHADING:
218 case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
219 case PIPE_CAP_TEXTURE_GATHER_SM5:
220 case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
221 case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
222 case PIPE_CAP_SAMPLER_VIEW_TARGET:
223 case PIPE_CAP_CLIP_HALFZ:
224 case PIPE_CAP_POLYGON_OFFSET_CLAMP:
225 case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
226 case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
227 case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
228 case PIPE_CAP_DEPTH_BOUNDS_TEST:
229 case PIPE_CAP_TGSI_TXQS:
230 case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
231 case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
232 case PIPE_CAP_SHAREABLE_SHADERS:
233 case PIPE_CAP_CLEAR_TEXTURE:
234 case PIPE_CAP_DRAW_PARAMETERS:
235 case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
236 case PIPE_CAP_MULTI_DRAW_INDIRECT:
237 case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
238 case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
239 case PIPE_CAP_QUERY_BUFFER_OBJECT:
240 case PIPE_CAP_INVALIDATE_BUFFER:
241 case PIPE_CAP_STRING_MARKER:
242 case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
243 case PIPE_CAP_CULL_DISTANCE:
244 case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
245 case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
246 case PIPE_CAP_TGSI_VOTE:
247 case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
248 case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
249 case PIPE_CAP_TGSI_MUL_ZERO_WINS:
250 case PIPE_CAP_DOUBLES:
251 case PIPE_CAP_INT64:
252 case PIPE_CAP_TGSI_TEX_TXF_LZ:
253 case PIPE_CAP_TGSI_CLOCK:
254 case PIPE_CAP_COMPUTE:
255 case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
256 case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
257 return 1;
258 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
259 return nouveau_screen(pscreen)->vram_domain & NOUVEAU_BO_VRAM ? 1 : 0;
260 case PIPE_CAP_TGSI_FS_FBFETCH:
261 return class_3d >= NVE4_3D_CLASS; /* needs testing on fermi */
262 case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
263 case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
264 case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
265 case PIPE_CAP_POST_DEPTH_COVERAGE:
266 return class_3d >= GM200_3D_CLASS;
267 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
268 case PIPE_CAP_TGSI_BALLOT:
269 return class_3d >= NVE4_3D_CLASS;
270 case PIPE_CAP_BINDLESS_TEXTURE:
271 return class_3d >= NVE4_3D_CLASS && class_3d < GM107_3D_CLASS;
272
273 /* unsupported caps */
274 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
275 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
276 case PIPE_CAP_SHADER_STENCIL_EXPORT:
277 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
278 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
279 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
280 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
281 case PIPE_CAP_FAKE_SW_MSAA:
282 case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
283 case PIPE_CAP_VERTEXID_NOBASE:
284 case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
285 case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
286 case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
287 case PIPE_CAP_GENERATE_MIPMAP:
288 case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
289 case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
290 case PIPE_CAP_QUERY_MEMORY_INFO:
291 case PIPE_CAP_PCI_GROUP:
292 case PIPE_CAP_PCI_BUS:
293 case PIPE_CAP_PCI_DEVICE:
294 case PIPE_CAP_PCI_FUNCTION:
295 case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
296 case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
297 case PIPE_CAP_NATIVE_FENCE_FD:
298 case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
299 case PIPE_CAP_INT64_DIVMOD:
300 case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
301 case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
302 case PIPE_CAP_QUERY_SO_OVERFLOW:
303 case PIPE_CAP_MEMOBJ:
304 case PIPE_CAP_LOAD_CONSTBUF:
305 case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS:
306 case PIPE_CAP_TILE_RASTER_ORDER:
307 case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
308 case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
309 case PIPE_CAP_CONTEXT_PRIORITY_MASK:
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
nvc0_screen_get_shader_param(struct pipe_screen * pscreen,enum pipe_shader_type shader,enum pipe_shader_cap param)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 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
414 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
415 return 0;
416 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
417 return NVC0_MAX_BUFFERS;
418 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
419 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
420 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
421 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
422 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
423 return 32;
424 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
425 if (class_3d >= NVE4_3D_CLASS)
426 return NVC0_MAX_IMAGES;
427 if (shader == PIPE_SHADER_FRAGMENT || shader == PIPE_SHADER_COMPUTE)
428 return NVC0_MAX_IMAGES;
429 return 0;
430 default:
431 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
432 return 0;
433 }
434 }
435
436 static float
nvc0_screen_get_paramf(struct pipe_screen * pscreen,enum pipe_capf param)437 nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
438 {
439 switch (param) {
440 case PIPE_CAPF_MAX_LINE_WIDTH:
441 case PIPE_CAPF_MAX_LINE_WIDTH_AA:
442 return 10.0f;
443 case PIPE_CAPF_MAX_POINT_WIDTH:
444 return 63.0f;
445 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
446 return 63.375f;
447 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
448 return 16.0f;
449 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
450 return 15.0f;
451 case PIPE_CAPF_GUARD_BAND_LEFT:
452 case PIPE_CAPF_GUARD_BAND_TOP:
453 return 0.0f;
454 case PIPE_CAPF_GUARD_BAND_RIGHT:
455 case PIPE_CAPF_GUARD_BAND_BOTTOM:
456 return 0.0f; /* that or infinity */
457 }
458
459 NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
460 return 0.0f;
461 }
462
463 static int
nvc0_screen_get_compute_param(struct pipe_screen * pscreen,enum pipe_shader_ir ir_type,enum pipe_compute_cap param,void * data)464 nvc0_screen_get_compute_param(struct pipe_screen *pscreen,
465 enum pipe_shader_ir ir_type,
466 enum pipe_compute_cap param, void *data)
467 {
468 struct nvc0_screen *screen = nvc0_screen(pscreen);
469 const uint16_t obj_class = screen->compute->oclass;
470
471 #define RET(x) do { \
472 if (data) \
473 memcpy(data, x, sizeof(x)); \
474 return sizeof(x); \
475 } while (0)
476
477 switch (param) {
478 case PIPE_COMPUTE_CAP_GRID_DIMENSION:
479 RET((uint64_t []) { 3 });
480 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
481 if (obj_class >= NVE4_COMPUTE_CLASS) {
482 RET(((uint64_t []) { 0x7fffffff, 65535, 65535 }));
483 } else {
484 RET(((uint64_t []) { 65535, 65535, 65535 }));
485 }
486 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
487 RET(((uint64_t []) { 1024, 1024, 64 }));
488 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
489 RET((uint64_t []) { 1024 });
490 case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
491 if (obj_class >= NVE4_COMPUTE_CLASS) {
492 RET((uint64_t []) { 1024 });
493 } else {
494 RET((uint64_t []) { 512 });
495 }
496 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */
497 RET((uint64_t []) { 1ULL << 40 });
498 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
499 switch (obj_class) {
500 case GM200_COMPUTE_CLASS:
501 RET((uint64_t []) { 96 << 10 });
502 break;
503 case GM107_COMPUTE_CLASS:
504 RET((uint64_t []) { 64 << 10 });
505 break;
506 default:
507 RET((uint64_t []) { 48 << 10 });
508 break;
509 }
510 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
511 RET((uint64_t []) { 512 << 10 });
512 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
513 RET((uint64_t []) { 4096 });
514 case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
515 RET((uint32_t []) { 32 });
516 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
517 RET((uint64_t []) { 1ULL << 40 });
518 case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
519 RET((uint32_t []) { 0 });
520 case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
521 RET((uint32_t []) { screen->mp_count_compute });
522 case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
523 RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
524 case PIPE_COMPUTE_CAP_ADDRESS_BITS:
525 RET((uint32_t []) { 64 });
526 default:
527 return 0;
528 }
529
530 #undef RET
531 }
532
533 static void
nvc0_screen_destroy(struct pipe_screen * pscreen)534 nvc0_screen_destroy(struct pipe_screen *pscreen)
535 {
536 struct nvc0_screen *screen = nvc0_screen(pscreen);
537
538 if (!nouveau_drm_screen_unref(&screen->base))
539 return;
540
541 if (screen->base.fence.current) {
542 struct nouveau_fence *current = NULL;
543
544 /* nouveau_fence_wait will create a new current fence, so wait on the
545 * _current_ one, and remove both.
546 */
547 nouveau_fence_ref(screen->base.fence.current, ¤t);
548 nouveau_fence_wait(current, NULL);
549 nouveau_fence_ref(NULL, ¤t);
550 nouveau_fence_ref(NULL, &screen->base.fence.current);
551 }
552 if (screen->base.pushbuf)
553 screen->base.pushbuf->user_priv = NULL;
554
555 if (screen->blitter)
556 nvc0_blitter_destroy(screen);
557 if (screen->pm.prog) {
558 screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
559 nvc0_program_destroy(NULL, screen->pm.prog);
560 FREE(screen->pm.prog);
561 }
562
563 nouveau_bo_ref(NULL, &screen->text);
564 nouveau_bo_ref(NULL, &screen->uniform_bo);
565 nouveau_bo_ref(NULL, &screen->tls);
566 nouveau_bo_ref(NULL, &screen->txc);
567 nouveau_bo_ref(NULL, &screen->fence.bo);
568 nouveau_bo_ref(NULL, &screen->poly_cache);
569
570 nouveau_heap_destroy(&screen->lib_code);
571 nouveau_heap_destroy(&screen->text_heap);
572
573 FREE(screen->default_tsc);
574 FREE(screen->tic.entries);
575
576 nouveau_object_del(&screen->eng3d);
577 nouveau_object_del(&screen->eng2d);
578 nouveau_object_del(&screen->m2mf);
579 nouveau_object_del(&screen->compute);
580 nouveau_object_del(&screen->nvsw);
581
582 nouveau_screen_fini(&screen->base);
583
584 FREE(screen);
585 }
586
587 static int
nvc0_graph_set_macro(struct nvc0_screen * screen,uint32_t m,unsigned pos,unsigned size,const uint32_t * data)588 nvc0_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos,
589 unsigned size, const uint32_t *data)
590 {
591 struct nouveau_pushbuf *push = screen->base.pushbuf;
592
593 size /= 4;
594
595 assert((pos + size) <= 0x800);
596
597 BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2);
598 PUSH_DATA (push, (m - 0x3800) / 8);
599 PUSH_DATA (push, pos);
600 BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1);
601 PUSH_DATA (push, pos);
602 PUSH_DATAp(push, data, size);
603
604 return pos + size;
605 }
606
607 static void
nvc0_magic_3d_init(struct nouveau_pushbuf * push,uint16_t obj_class)608 nvc0_magic_3d_init(struct nouveau_pushbuf *push, uint16_t obj_class)
609 {
610 BEGIN_NVC0(push, SUBC_3D(0x10cc), 1);
611 PUSH_DATA (push, 0xff);
612 BEGIN_NVC0(push, SUBC_3D(0x10e0), 2);
613 PUSH_DATA (push, 0xff);
614 PUSH_DATA (push, 0xff);
615 BEGIN_NVC0(push, SUBC_3D(0x10ec), 2);
616 PUSH_DATA (push, 0xff);
617 PUSH_DATA (push, 0xff);
618 BEGIN_NVC0(push, SUBC_3D(0x074c), 1);
619 PUSH_DATA (push, 0x3f);
620
621 BEGIN_NVC0(push, SUBC_3D(0x16a8), 1);
622 PUSH_DATA (push, (3 << 16) | 3);
623 BEGIN_NVC0(push, SUBC_3D(0x1794), 1);
624 PUSH_DATA (push, (2 << 16) | 2);
625
626 if (obj_class < GM107_3D_CLASS) {
627 BEGIN_NVC0(push, SUBC_3D(0x12ac), 1);
628 PUSH_DATA (push, 0);
629 }
630 BEGIN_NVC0(push, SUBC_3D(0x0218), 1);
631 PUSH_DATA (push, 0x10);
632 BEGIN_NVC0(push, SUBC_3D(0x10fc), 1);
633 PUSH_DATA (push, 0x10);
634 BEGIN_NVC0(push, SUBC_3D(0x1290), 1);
635 PUSH_DATA (push, 0x10);
636 BEGIN_NVC0(push, SUBC_3D(0x12d8), 2);
637 PUSH_DATA (push, 0x10);
638 PUSH_DATA (push, 0x10);
639 BEGIN_NVC0(push, SUBC_3D(0x1140), 1);
640 PUSH_DATA (push, 0x10);
641 BEGIN_NVC0(push, SUBC_3D(0x1610), 1);
642 PUSH_DATA (push, 0xe);
643
644 BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_GEN_MODE), 1);
645 PUSH_DATA (push, NVC0_3D_VERTEX_ID_GEN_MODE_DRAW_ARRAYS_ADD_START);
646 BEGIN_NVC0(push, SUBC_3D(0x030c), 1);
647 PUSH_DATA (push, 0);
648 BEGIN_NVC0(push, SUBC_3D(0x0300), 1);
649 PUSH_DATA (push, 3);
650
651 BEGIN_NVC0(push, SUBC_3D(0x02d0), 1);
652 PUSH_DATA (push, 0x3fffff);
653 BEGIN_NVC0(push, SUBC_3D(0x0fdc), 1);
654 PUSH_DATA (push, 1);
655 BEGIN_NVC0(push, SUBC_3D(0x19c0), 1);
656 PUSH_DATA (push, 1);
657
658 if (obj_class < GM107_3D_CLASS) {
659 BEGIN_NVC0(push, SUBC_3D(0x075c), 1);
660 PUSH_DATA (push, 3);
661
662 if (obj_class >= NVE4_3D_CLASS) {
663 BEGIN_NVC0(push, SUBC_3D(0x07fc), 1);
664 PUSH_DATA (push, 1);
665 }
666 }
667
668 /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
669 * are supposed to do */
670 }
671
672 static void
nvc0_screen_fence_emit(struct pipe_screen * pscreen,u32 * sequence)673 nvc0_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
674 {
675 struct nvc0_screen *screen = nvc0_screen(pscreen);
676 struct nouveau_pushbuf *push = screen->base.pushbuf;
677
678 /* we need to do it after possible flush in MARK_RING */
679 *sequence = ++screen->base.fence.sequence;
680
681 assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
682 PUSH_DATA (push, NVC0_FIFO_PKHDR_SQ(NVC0_3D(QUERY_ADDRESS_HIGH), 4));
683 PUSH_DATAh(push, screen->fence.bo->offset);
684 PUSH_DATA (push, screen->fence.bo->offset);
685 PUSH_DATA (push, *sequence);
686 PUSH_DATA (push, NVC0_3D_QUERY_GET_FENCE | NVC0_3D_QUERY_GET_SHORT |
687 (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT));
688 }
689
690 static u32
nvc0_screen_fence_update(struct pipe_screen * pscreen)691 nvc0_screen_fence_update(struct pipe_screen *pscreen)
692 {
693 struct nvc0_screen *screen = nvc0_screen(pscreen);
694 return screen->fence.map[0];
695 }
696
697 static int
nvc0_screen_init_compute(struct nvc0_screen * screen)698 nvc0_screen_init_compute(struct nvc0_screen *screen)
699 {
700 screen->base.base.get_compute_param = nvc0_screen_get_compute_param;
701
702 switch (screen->base.device->chipset & ~0xf) {
703 case 0xc0:
704 case 0xd0:
705 return nvc0_screen_compute_setup(screen, screen->base.pushbuf);
706 case 0xe0:
707 case 0xf0:
708 case 0x100:
709 case 0x110:
710 case 0x120:
711 case 0x130:
712 return nve4_screen_compute_setup(screen, screen->base.pushbuf);
713 default:
714 return -1;
715 }
716 }
717
718 static int
nvc0_screen_resize_tls_area(struct nvc0_screen * screen,uint32_t lpos,uint32_t lneg,uint32_t cstack)719 nvc0_screen_resize_tls_area(struct nvc0_screen *screen,
720 uint32_t lpos, uint32_t lneg, uint32_t cstack)
721 {
722 struct nouveau_bo *bo = NULL;
723 int ret;
724 uint64_t size = (lpos + lneg) * 32 + cstack;
725
726 if (size >= (1 << 20)) {
727 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64"\n", size);
728 return -1;
729 }
730
731 size *= (screen->base.device->chipset >= 0xe0) ? 64 : 48; /* max warps */
732 size = align(size, 0x8000);
733 size *= screen->mp_count;
734
735 size = align(size, 1 << 17);
736
737 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base), 1 << 17, size,
738 NULL, &bo);
739 if (ret)
740 return ret;
741
742 /* Make sure that the pushbuf has acquired a reference to the old tls
743 * segment, as it may have commands that will reference it.
744 */
745 if (screen->tls)
746 PUSH_REFN(screen->base.pushbuf, screen->tls,
747 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RDWR);
748 nouveau_bo_ref(NULL, &screen->tls);
749 screen->tls = bo;
750 return 0;
751 }
752
753 int
nvc0_screen_resize_text_area(struct nvc0_screen * screen,uint64_t size)754 nvc0_screen_resize_text_area(struct nvc0_screen *screen, uint64_t size)
755 {
756 struct nouveau_pushbuf *push = screen->base.pushbuf;
757 struct nouveau_bo *bo;
758 int ret;
759
760 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base),
761 1 << 17, size, NULL, &bo);
762 if (ret)
763 return ret;
764
765 /* Make sure that the pushbuf has acquired a reference to the old text
766 * segment, as it may have commands that will reference it.
767 */
768 if (screen->text)
769 PUSH_REFN(push, screen->text,
770 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RD);
771 nouveau_bo_ref(NULL, &screen->text);
772 screen->text = bo;
773
774 nouveau_heap_destroy(&screen->lib_code);
775 nouveau_heap_destroy(&screen->text_heap);
776
777 /* XXX: getting a page fault at the end of the code buffer every few
778 * launches, don't use the last 256 bytes to work around them - prefetch ?
779 */
780 nouveau_heap_init(&screen->text_heap, 0, size - 0x100);
781
782 /* update the code segment setup */
783 BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
784 PUSH_DATAh(push, screen->text->offset);
785 PUSH_DATA (push, screen->text->offset);
786 if (screen->compute) {
787 BEGIN_NVC0(push, NVC0_CP(CODE_ADDRESS_HIGH), 2);
788 PUSH_DATAh(push, screen->text->offset);
789 PUSH_DATA (push, screen->text->offset);
790 }
791
792 return 0;
793 }
794
795 #define FAIL_SCREEN_INIT(str, err) \
796 do { \
797 NOUVEAU_ERR(str, err); \
798 goto fail; \
799 } while(0)
800
801 struct nouveau_screen *
nvc0_screen_create(struct nouveau_device * dev)802 nvc0_screen_create(struct nouveau_device *dev)
803 {
804 struct nvc0_screen *screen;
805 struct pipe_screen *pscreen;
806 struct nouveau_object *chan;
807 struct nouveau_pushbuf *push;
808 uint64_t value;
809 uint32_t obj_class;
810 uint32_t flags;
811 int ret;
812 unsigned i;
813
814 switch (dev->chipset & ~0xf) {
815 case 0xc0:
816 case 0xd0:
817 case 0xe0:
818 case 0xf0:
819 case 0x100:
820 case 0x110:
821 case 0x120:
822 case 0x130:
823 break;
824 default:
825 return NULL;
826 }
827
828 screen = CALLOC_STRUCT(nvc0_screen);
829 if (!screen)
830 return NULL;
831 pscreen = &screen->base.base;
832 pscreen->destroy = nvc0_screen_destroy;
833
834 ret = nouveau_screen_init(&screen->base, dev);
835 if (ret)
836 FAIL_SCREEN_INIT("Base screen init failed: %d\n", ret);
837 chan = screen->base.channel;
838 push = screen->base.pushbuf;
839 push->user_priv = screen;
840 push->rsvd_kick = 5;
841
842 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
843 PIPE_BIND_SHADER_BUFFER |
844 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
845 PIPE_BIND_COMMAND_ARGS_BUFFER | PIPE_BIND_QUERY_BUFFER;
846 screen->base.sysmem_bindings |=
847 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
848
849 if (screen->base.vram_domain & NOUVEAU_BO_GART) {
850 screen->base.sysmem_bindings |= screen->base.vidmem_bindings;
851 screen->base.vidmem_bindings = 0;
852 }
853
854 pscreen->context_create = nvc0_create;
855 pscreen->is_format_supported = nvc0_screen_is_format_supported;
856 pscreen->get_param = nvc0_screen_get_param;
857 pscreen->get_shader_param = nvc0_screen_get_shader_param;
858 pscreen->get_paramf = nvc0_screen_get_paramf;
859 pscreen->get_driver_query_info = nvc0_screen_get_driver_query_info;
860 pscreen->get_driver_query_group_info = nvc0_screen_get_driver_query_group_info;
861
862 nvc0_screen_init_resource_functions(pscreen);
863
864 screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
865 screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
866
867 flags = NOUVEAU_BO_GART | NOUVEAU_BO_MAP;
868 if (screen->base.drm->version >= 0x01000202)
869 flags |= NOUVEAU_BO_COHERENT;
870
871 ret = nouveau_bo_new(dev, flags, 0, 4096, NULL, &screen->fence.bo);
872 if (ret)
873 FAIL_SCREEN_INIT("Error allocating fence BO: %d\n", ret);
874 nouveau_bo_map(screen->fence.bo, 0, NULL);
875 screen->fence.map = screen->fence.bo->map;
876 screen->base.fence.emit = nvc0_screen_fence_emit;
877 screen->base.fence.update = nvc0_screen_fence_update;
878
879
880 ret = nouveau_object_new(chan, (dev->chipset < 0xe0) ? 0x1f906e : 0x906e,
881 NVIF_CLASS_SW_GF100, NULL, 0, &screen->nvsw);
882 if (ret)
883 FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret);
884
885 BEGIN_NVC0(push, SUBC_SW(NV01_SUBCHAN_OBJECT), 1);
886 PUSH_DATA (push, screen->nvsw->handle);
887
888 switch (dev->chipset & ~0xf) {
889 case 0x130:
890 case 0x120:
891 case 0x110:
892 case 0x100:
893 case 0xf0:
894 obj_class = NVF0_P2MF_CLASS;
895 break;
896 case 0xe0:
897 obj_class = NVE4_P2MF_CLASS;
898 break;
899 default:
900 obj_class = NVC0_M2MF_CLASS;
901 break;
902 }
903 ret = nouveau_object_new(chan, 0xbeef323f, obj_class, NULL, 0,
904 &screen->m2mf);
905 if (ret)
906 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
907
908 BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
909 PUSH_DATA (push, screen->m2mf->oclass);
910 if (screen->m2mf->oclass == NVE4_P2MF_CLASS) {
911 BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
912 PUSH_DATA (push, 0xa0b5);
913 }
914
915 ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
916 &screen->eng2d);
917 if (ret)
918 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
919
920 BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
921 PUSH_DATA (push, screen->eng2d->oclass);
922 BEGIN_NVC0(push, SUBC_2D(NVC0_2D_SINGLE_GPC), 1);
923 PUSH_DATA (push, 0);
924 BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
925 PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
926 BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
927 PUSH_DATA (push, 0);
928 BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
929 PUSH_DATA (push, 0);
930 BEGIN_NVC0(push, SUBC_2D(0x0884), 1);
931 PUSH_DATA (push, 0x3f);
932 BEGIN_NVC0(push, SUBC_2D(0x0888), 1);
933 PUSH_DATA (push, 1);
934 BEGIN_NVC0(push, NVC0_2D(COND_MODE), 1);
935 PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
936
937 BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
938 PUSH_DATAh(push, screen->fence.bo->offset + 16);
939 PUSH_DATA (push, screen->fence.bo->offset + 16);
940
941 switch (dev->chipset & ~0xf) {
942 case 0x130:
943 switch (dev->chipset) {
944 case 0x130:
945 case 0x13b:
946 obj_class = GP100_3D_CLASS;
947 break;
948 default:
949 obj_class = GP102_3D_CLASS;
950 break;
951 }
952 break;
953 case 0x120:
954 obj_class = GM200_3D_CLASS;
955 break;
956 case 0x110:
957 obj_class = GM107_3D_CLASS;
958 break;
959 case 0x100:
960 case 0xf0:
961 obj_class = NVF0_3D_CLASS;
962 break;
963 case 0xe0:
964 switch (dev->chipset) {
965 case 0xea:
966 obj_class = NVEA_3D_CLASS;
967 break;
968 default:
969 obj_class = NVE4_3D_CLASS;
970 break;
971 }
972 break;
973 case 0xd0:
974 obj_class = NVC8_3D_CLASS;
975 break;
976 case 0xc0:
977 default:
978 switch (dev->chipset) {
979 case 0xc8:
980 obj_class = NVC8_3D_CLASS;
981 break;
982 case 0xc1:
983 obj_class = NVC1_3D_CLASS;
984 break;
985 default:
986 obj_class = NVC0_3D_CLASS;
987 break;
988 }
989 break;
990 }
991 ret = nouveau_object_new(chan, 0xbeef003d, obj_class, NULL, 0,
992 &screen->eng3d);
993 if (ret)
994 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
995 screen->base.class_3d = obj_class;
996
997 BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
998 PUSH_DATA (push, screen->eng3d->oclass);
999
1000 BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
1001 PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
1002
1003 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
1004 /* kill shaders after about 1 second (at 100 MHz) */
1005 BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
1006 PUSH_DATA (push, 0x17);
1007 }
1008
1009 IMMED_NVC0(push, NVC0_3D(ZETA_COMP_ENABLE),
1010 screen->base.drm->version >= 0x01000101);
1011 BEGIN_NVC0(push, NVC0_3D(RT_COMP_ENABLE(0)), 8);
1012 for (i = 0; i < 8; ++i)
1013 PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
1014
1015 BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
1016 PUSH_DATA (push, 1);
1017
1018 BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
1019 PUSH_DATA (push, 0);
1020 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
1021 PUSH_DATA (push, 0);
1022 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
1023 PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
1024 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
1025 PUSH_DATA (push, 0);
1026 BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
1027 PUSH_DATA (push, 1);
1028 BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
1029 PUSH_DATA (push, 1);
1030 BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
1031 PUSH_DATA (push, 1);
1032 BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
1033 PUSH_DATA (push, 0);
1034 BEGIN_NVC0(push, NVC0_3D(SHADE_MODEL), 1);
1035 PUSH_DATA (push, NVC0_3D_SHADE_MODEL_SMOOTH);
1036 if (screen->eng3d->oclass < NVE4_3D_CLASS) {
1037 IMMED_NVC0(push, NVC0_3D(TEX_MISC), 0);
1038 } else {
1039 BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
1040 PUSH_DATA (push, 15);
1041 }
1042 BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
1043 PUSH_DATA (push, 8); /* 128 */
1044 BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
1045 PUSH_DATA (push, 1);
1046 if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
1047 BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
1048 PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
1049 }
1050
1051 nvc0_magic_3d_init(push, screen->eng3d->oclass);
1052
1053 ret = nvc0_screen_resize_text_area(screen, 1 << 19);
1054 if (ret)
1055 FAIL_SCREEN_INIT("Error allocating TEXT area: %d\n", ret);
1056
1057 /* 6 user uniform areas, 6 driver areas, and 1 for the runout */
1058 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 12, 13 << 16, NULL,
1059 &screen->uniform_bo);
1060 if (ret)
1061 FAIL_SCREEN_INIT("Error allocating uniform BO: %d\n", ret);
1062
1063 PUSH_REFN (push, screen->uniform_bo, NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_WR);
1064
1065 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
1066 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1067 PUSH_DATA (push, 256);
1068 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1069 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1070 BEGIN_1IC0(push, NVC0_3D(CB_POS), 5);
1071 PUSH_DATA (push, 0);
1072 PUSH_DATAf(push, 0.0f);
1073 PUSH_DATAf(push, 0.0f);
1074 PUSH_DATAf(push, 0.0f);
1075 PUSH_DATAf(push, 0.0f);
1076 BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
1077 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1078 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1079
1080 if (screen->base.drm->version >= 0x01000101) {
1081 ret = nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
1082 if (ret)
1083 FAIL_SCREEN_INIT("NOUVEAU_GETPARAM_GRAPH_UNITS failed: %d\n", ret);
1084 } else {
1085 if (dev->chipset >= 0xe0 && dev->chipset < 0xf0)
1086 value = (8 << 8) | 4;
1087 else
1088 value = (16 << 8) | 4;
1089 }
1090 screen->gpc_count = value & 0x000000ff;
1091 screen->mp_count = value >> 8;
1092 screen->mp_count_compute = screen->mp_count;
1093
1094 ret = nvc0_screen_resize_tls_area(screen, 128 * 16, 0, 0x200);
1095 if (ret)
1096 FAIL_SCREEN_INIT("Error allocating TLS area: %d\n", ret);
1097
1098 BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
1099 PUSH_DATAh(push, screen->tls->offset);
1100 PUSH_DATA (push, screen->tls->offset);
1101 PUSH_DATA (push, screen->tls->size >> 32);
1102 PUSH_DATA (push, screen->tls->size);
1103 BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
1104 PUSH_DATA (push, 0);
1105 /* Reduce likelihood of collision with real buffers by placing the hole at
1106 * the top of the 4G area. This will have to be dealt with for real
1107 * eventually by blocking off that area from the VM.
1108 */
1109 BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
1110 PUSH_DATA (push, 0xff << 24);
1111
1112 if (screen->eng3d->oclass < GM107_3D_CLASS) {
1113 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 20, NULL,
1114 &screen->poly_cache);
1115 if (ret)
1116 FAIL_SCREEN_INIT("Error allocating poly cache BO: %d\n", ret);
1117
1118 BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
1119 PUSH_DATAh(push, screen->poly_cache->offset);
1120 PUSH_DATA (push, screen->poly_cache->offset);
1121 PUSH_DATA (push, 3);
1122 }
1123
1124 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 17, NULL,
1125 &screen->txc);
1126 if (ret)
1127 FAIL_SCREEN_INIT("Error allocating txc BO: %d\n", ret);
1128
1129 BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
1130 PUSH_DATAh(push, screen->txc->offset);
1131 PUSH_DATA (push, screen->txc->offset);
1132 PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
1133 if (screen->eng3d->oclass >= GM107_3D_CLASS) {
1134 screen->tic.maxwell = true;
1135 if (screen->eng3d->oclass == GM107_3D_CLASS) {
1136 screen->tic.maxwell =
1137 debug_get_bool_option("NOUVEAU_MAXWELL_TIC", true);
1138 IMMED_NVC0(push, SUBC_3D(0x0f10), screen->tic.maxwell);
1139 }
1140 }
1141
1142 BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
1143 PUSH_DATAh(push, screen->txc->offset + 65536);
1144 PUSH_DATA (push, screen->txc->offset + 65536);
1145 PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
1146
1147 BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
1148 PUSH_DATA (push, 0);
1149 BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
1150 PUSH_DATA (push, 0);
1151 PUSH_DATA (push, 0);
1152 BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
1153 PUSH_DATA (push, 0x3f);
1154
1155 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
1156 PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
1157 BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
1158 for (i = 0; i < 8 * 2; ++i)
1159 PUSH_DATA(push, 0);
1160 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
1161 PUSH_DATA (push, 0);
1162 BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
1163 PUSH_DATA (push, 0);
1164
1165 /* neither scissors, viewport nor stencil mask should affect clears */
1166 BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
1167 PUSH_DATA (push, 0);
1168
1169 BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
1170 PUSH_DATA (push, 1);
1171 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1172 BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(i)), 2);
1173 PUSH_DATAf(push, 0.0f);
1174 PUSH_DATAf(push, 1.0f);
1175 }
1176 BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
1177 PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
1178
1179 /* We use scissors instead of exact view volume clipping,
1180 * so they're always enabled.
1181 */
1182 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1183 BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(i)), 3);
1184 PUSH_DATA (push, 1);
1185 PUSH_DATA (push, 8192 << 16);
1186 PUSH_DATA (push, 8192 << 16);
1187 }
1188
1189 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
1190
1191 i = 0;
1192 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, mme9097_per_instance_bf);
1193 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, mme9097_blend_enables);
1194 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, mme9097_vertex_array_select);
1195 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, mme9097_tep_select);
1196 MK_MACRO(NVC0_3D_MACRO_GP_SELECT, mme9097_gp_select);
1197 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, mme9097_poly_mode_front);
1198 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, mme9097_poly_mode_back);
1199 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT, mme9097_draw_arrays_indirect);
1200 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT, mme9097_draw_elts_indirect);
1201 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT, mme9097_draw_arrays_indirect_count);
1202 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT, mme9097_draw_elts_indirect_count);
1203 MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE, mme9097_query_buffer_write);
1204 MK_MACRO(NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT, mme90c0_launch_grid_indirect);
1205
1206 BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
1207 PUSH_DATA (push, 1);
1208 BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
1209 PUSH_DATA (push, 1);
1210 BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
1211 PUSH_DATA (push, 0x40);
1212 BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
1213 PUSH_DATA (push, 0);
1214 BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
1215 PUSH_DATA (push, 0x30);
1216 BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
1217 PUSH_DATA (push, 3);
1218 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
1219 PUSH_DATA (push, 0x20);
1220 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
1221 PUSH_DATA (push, 0x00);
1222 screen->save_state.patch_vertices = 3;
1223
1224 BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
1225 PUSH_DATA (push, 0);
1226 BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
1227 PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
1228
1229 IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
1230
1231 if (nvc0_screen_init_compute(screen))
1232 goto fail;
1233
1234 /* XXX: Compute and 3D are somehow aliased on Fermi. */
1235 for (i = 0; i < 5; ++i) {
1236 /* TIC and TSC entries for each unit (nve4+ only) */
1237 /* auxiliary constants (6 user clip planes, base instance id) */
1238 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1239 PUSH_DATA (push, NVC0_CB_AUX_SIZE);
1240 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(i));
1241 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(i));
1242 BEGIN_NVC0(push, NVC0_3D(CB_BIND(i)), 1);
1243 PUSH_DATA (push, (15 << 4) | 1);
1244 if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
1245 unsigned j;
1246 BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
1247 PUSH_DATA (push, NVC0_CB_AUX_UNK_INFO);
1248 for (j = 0; j < 8; ++j)
1249 PUSH_DATA(push, j);
1250 } else {
1251 BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
1252 PUSH_DATA (push, 0x54);
1253 }
1254
1255 /* MS sample coordinate offsets: these do not work with _ALT modes ! */
1256 BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 2 * 8);
1257 PUSH_DATA (push, NVC0_CB_AUX_MS_INFO);
1258 PUSH_DATA (push, 0); /* 0 */
1259 PUSH_DATA (push, 0);
1260 PUSH_DATA (push, 1); /* 1 */
1261 PUSH_DATA (push, 0);
1262 PUSH_DATA (push, 0); /* 2 */
1263 PUSH_DATA (push, 1);
1264 PUSH_DATA (push, 1); /* 3 */
1265 PUSH_DATA (push, 1);
1266 PUSH_DATA (push, 2); /* 4 */
1267 PUSH_DATA (push, 0);
1268 PUSH_DATA (push, 3); /* 5 */
1269 PUSH_DATA (push, 0);
1270 PUSH_DATA (push, 2); /* 6 */
1271 PUSH_DATA (push, 1);
1272 PUSH_DATA (push, 3); /* 7 */
1273 PUSH_DATA (push, 1);
1274 }
1275 BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
1276 PUSH_DATA (push, 0);
1277
1278 PUSH_KICK (push);
1279
1280 screen->tic.entries = CALLOC(
1281 NVC0_TIC_MAX_ENTRIES + NVC0_TSC_MAX_ENTRIES + NVE4_IMG_MAX_HANDLES,
1282 sizeof(void *));
1283 screen->tsc.entries = screen->tic.entries + NVC0_TIC_MAX_ENTRIES;
1284 screen->img.entries = (void *)(screen->tsc.entries + NVC0_TSC_MAX_ENTRIES);
1285
1286 if (!nvc0_blitter_create(screen))
1287 goto fail;
1288
1289 screen->default_tsc = CALLOC_STRUCT(nv50_tsc_entry);
1290 screen->default_tsc->tsc[0] = G80_TSC_0_SRGB_CONVERSION;
1291
1292 nouveau_fence_new(&screen->base, &screen->base.fence.current);
1293
1294 return &screen->base;
1295
1296 fail:
1297 screen->base.base.context_create = NULL;
1298 return &screen->base;
1299 }
1300
1301 int
nvc0_screen_tic_alloc(struct nvc0_screen * screen,void * entry)1302 nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
1303 {
1304 int i = screen->tic.next;
1305
1306 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1307 i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1308
1309 screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1310
1311 if (screen->tic.entries[i])
1312 nv50_tic_entry(screen->tic.entries[i])->id = -1;
1313
1314 screen->tic.entries[i] = entry;
1315 return i;
1316 }
1317
1318 int
nvc0_screen_tsc_alloc(struct nvc0_screen * screen,void * entry)1319 nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
1320 {
1321 int i = screen->tsc.next;
1322
1323 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1324 i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1325
1326 screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1327
1328 if (screen->tsc.entries[i])
1329 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1330
1331 screen->tsc.entries[i] = entry;
1332 return i;
1333 }
1334