projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
nvc0: add bindless image support for kepler
[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 case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
311 case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
312 case PIPE_CAP_CONTEXT_PRIORITY_MASK:
313 return 0;
314
315 case PIPE_CAP_VENDOR_ID:
316 return 0x10de;
317 case PIPE_CAP_DEVICE_ID: {
318 uint64_t device_id;
319 if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
320 NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
321 return -1;
322 }
323 return device_id;
324 }
325 case PIPE_CAP_ACCELERATED:
326 return 1;
327 case PIPE_CAP_VIDEO_MEMORY:
328 return dev->vram_size >> 20;
329 case PIPE_CAP_UMA:
330 return 0;
331 }
332
333 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
334 return 0;
335 }
336
337 static int
338 nvc0_screen_get_shader_param(struct pipe_screen *pscreen,
339 enum pipe_shader_type shader,
340 enum pipe_shader_cap param)
341 {
342 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
343
344 switch (shader) {
345 case PIPE_SHADER_VERTEX:
346 case PIPE_SHADER_GEOMETRY:
347 case PIPE_SHADER_FRAGMENT:
348 case PIPE_SHADER_COMPUTE:
349 case PIPE_SHADER_TESS_CTRL:
350 case PIPE_SHADER_TESS_EVAL:
351 break;
352 default:
353 return 0;
354 }
355
356 switch (param) {
357 case PIPE_SHADER_CAP_PREFERRED_IR:
358 return PIPE_SHADER_IR_TGSI;
359 case PIPE_SHADER_CAP_SUPPORTED_IRS:
360 return 1 << PIPE_SHADER_IR_TGSI;
361 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
362 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
363 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
364 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
365 return 16384;
366 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
367 return 16;
368 case PIPE_SHADER_CAP_MAX_INPUTS:
369 if (shader == PIPE_SHADER_VERTEX)
370 return 32;
371 /* NOTE: These only count our slots for GENERIC varyings.
372 * The address space may be larger, but the actual hard limit seems to be
373 * less than what the address space layout permits, so don't add TEXCOORD,
374 * COLOR, etc. here.
375 */
376 if (shader == PIPE_SHADER_FRAGMENT)
377 return 0x1f0 / 16;
378 /* Actually this counts CLIPVERTEX, which occupies the last generic slot,
379 * and excludes 0x60 per-patch inputs.
380 */
381 return 0x200 / 16;
382 case PIPE_SHADER_CAP_MAX_OUTPUTS:
383 return 32;
384 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
385 return 65536;
386 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
387 return NVC0_MAX_PIPE_CONSTBUFS;
388 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
389 return shader != PIPE_SHADER_FRAGMENT;
390 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
391 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
392 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
393 return 1;
394 case PIPE_SHADER_CAP_MAX_TEMPS:
395 return NVC0_CAP_MAX_PROGRAM_TEMPS;
396 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
397 return 1;
398 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
399 return 1;
400 case PIPE_SHADER_CAP_SUBROUTINES:
401 return 1;
402 case PIPE_SHADER_CAP_INTEGERS:
403 return 1;
404 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
405 return 1;
406 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
407 return 1;
408 case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
409 return 1;
410 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
411 case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
412 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
413 case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
414 case PIPE_SHADER_CAP_INT64_ATOMICS:
415 case PIPE_SHADER_CAP_FP16:
416 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
417 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
418 return 0;
419 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
420 return NVC0_MAX_BUFFERS;
421 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
422 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
423 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
424 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
425 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
426 return 32;
427 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
428 if (class_3d >= NVE4_3D_CLASS)
429 return NVC0_MAX_IMAGES;
430 if (shader == PIPE_SHADER_FRAGMENT || shader == PIPE_SHADER_COMPUTE)
431 return NVC0_MAX_IMAGES;
432 return 0;
433 default:
434 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
435 return 0;
436 }
437 }
438
439 static float
440 nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
441 {
442 switch (param) {
443 case PIPE_CAPF_MAX_LINE_WIDTH:
444 case PIPE_CAPF_MAX_LINE_WIDTH_AA:
445 return 10.0f;
446 case PIPE_CAPF_MAX_POINT_WIDTH:
447 return 63.0f;
448 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
449 return 63.375f;
450 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
451 return 16.0f;
452 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
453 return 15.0f;
454 case PIPE_CAPF_GUARD_BAND_LEFT:
455 case PIPE_CAPF_GUARD_BAND_TOP:
456 return 0.0f;
457 case PIPE_CAPF_GUARD_BAND_RIGHT:
458 case PIPE_CAPF_GUARD_BAND_BOTTOM:
459 return 0.0f; /* that or infinity */
460 }
461
462 NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
463 return 0.0f;
464 }
465
466 static int
467 nvc0_screen_get_compute_param(struct pipe_screen *pscreen,
468 enum pipe_shader_ir ir_type,
469 enum pipe_compute_cap param, void *data)
470 {
471 struct nvc0_screen *screen = nvc0_screen(pscreen);
472 const uint16_t obj_class = screen->compute->oclass;
473
474 #define RET(x) do { \
475 if (data) \
476 memcpy(data, x, sizeof(x)); \
477 return sizeof(x); \
478 } while (0)
479
480 switch (param) {
481 case PIPE_COMPUTE_CAP_GRID_DIMENSION:
482 RET((uint64_t []) { 3 });
483 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
484 if (obj_class >= NVE4_COMPUTE_CLASS) {
485 RET(((uint64_t []) { 0x7fffffff, 65535, 65535 }));
486 } else {
487 RET(((uint64_t []) { 65535, 65535, 65535 }));
488 }
489 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
490 RET(((uint64_t []) { 1024, 1024, 64 }));
491 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
492 RET((uint64_t []) { 1024 });
493 case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
494 if (obj_class >= NVE4_COMPUTE_CLASS) {
495 RET((uint64_t []) { 1024 });
496 } else {
497 RET((uint64_t []) { 512 });
498 }
499 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */
500 RET((uint64_t []) { 1ULL << 40 });
501 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
502 switch (obj_class) {
503 case GM200_COMPUTE_CLASS:
504 RET((uint64_t []) { 96 << 10 });
505 break;
506 case GM107_COMPUTE_CLASS:
507 RET((uint64_t []) { 64 << 10 });
508 break;
509 default:
510 RET((uint64_t []) { 48 << 10 });
511 break;
512 }
513 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
514 RET((uint64_t []) { 512 << 10 });
515 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
516 RET((uint64_t []) { 4096 });
517 case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
518 RET((uint32_t []) { 32 });
519 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
520 RET((uint64_t []) { 1ULL << 40 });
521 case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
522 RET((uint32_t []) { 0 });
523 case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
524 RET((uint32_t []) { screen->mp_count_compute });
525 case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
526 RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
527 case PIPE_COMPUTE_CAP_ADDRESS_BITS:
528 RET((uint32_t []) { 64 });
529 default:
530 return 0;
531 }
532
533 #undef RET
534 }
535
536 static void
537 nvc0_screen_destroy(struct pipe_screen *pscreen)
538 {
539 struct nvc0_screen *screen = nvc0_screen(pscreen);
540
541 if (!nouveau_drm_screen_unref(&screen->base))
542 return;
543
544 if (screen->base.fence.current) {
545 struct nouveau_fence *current = NULL;
546
547 /* nouveau_fence_wait will create a new current fence, so wait on the
548 * _current_ one, and remove both.
549 */
550 nouveau_fence_ref(screen->base.fence.current, &current);
551 nouveau_fence_wait(current, NULL);
552 nouveau_fence_ref(NULL, &current);
553 nouveau_fence_ref(NULL, &screen->base.fence.current);
554 }
555 if (screen->base.pushbuf)
556 screen->base.pushbuf->user_priv = NULL;
557
558 if (screen->blitter)
559 nvc0_blitter_destroy(screen);
560 if (screen->pm.prog) {
561 screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
562 nvc0_program_destroy(NULL, screen->pm.prog);
563 FREE(screen->pm.prog);
564 }
565
566 nouveau_bo_ref(NULL, &screen->text);
567 nouveau_bo_ref(NULL, &screen->uniform_bo);
568 nouveau_bo_ref(NULL, &screen->tls);
569 nouveau_bo_ref(NULL, &screen->txc);
570 nouveau_bo_ref(NULL, &screen->fence.bo);
571 nouveau_bo_ref(NULL, &screen->poly_cache);
572
573 nouveau_heap_destroy(&screen->lib_code);
574 nouveau_heap_destroy(&screen->text_heap);
575
576 FREE(screen->default_tsc);
577 FREE(screen->tic.entries);
578
579 nouveau_object_del(&screen->eng3d);
580 nouveau_object_del(&screen->eng2d);
581 nouveau_object_del(&screen->m2mf);
582 nouveau_object_del(&screen->compute);
583 nouveau_object_del(&screen->nvsw);
584
585 nouveau_screen_fini(&screen->base);
586
587 FREE(screen);
588 }
589
590 static int
591 nvc0_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos,
592 unsigned size, const uint32_t *data)
593 {
594 struct nouveau_pushbuf *push = screen->base.pushbuf;
595
596 size /= 4;
597
598 assert((pos + size) <= 0x800);
599
600 BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2);
601 PUSH_DATA (push, (m - 0x3800) / 8);
602 PUSH_DATA (push, pos);
603 BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1);
604 PUSH_DATA (push, pos);
605 PUSH_DATAp(push, data, size);
606
607 return pos + size;
608 }
609
610 static void
611 nvc0_magic_3d_init(struct nouveau_pushbuf *push, uint16_t obj_class)
612 {
613 BEGIN_NVC0(push, SUBC_3D(0x10cc), 1);
614 PUSH_DATA (push, 0xff);
615 BEGIN_NVC0(push, SUBC_3D(0x10e0), 2);
616 PUSH_DATA (push, 0xff);
617 PUSH_DATA (push, 0xff);
618 BEGIN_NVC0(push, SUBC_3D(0x10ec), 2);
619 PUSH_DATA (push, 0xff);
620 PUSH_DATA (push, 0xff);
621 BEGIN_NVC0(push, SUBC_3D(0x074c), 1);
622 PUSH_DATA (push, 0x3f);
623
624 BEGIN_NVC0(push, SUBC_3D(0x16a8), 1);
625 PUSH_DATA (push, (3 << 16) | 3);
626 BEGIN_NVC0(push, SUBC_3D(0x1794), 1);
627 PUSH_DATA (push, (2 << 16) | 2);
628
629 if (obj_class < GM107_3D_CLASS) {
630 BEGIN_NVC0(push, SUBC_3D(0x12ac), 1);
631 PUSH_DATA (push, 0);
632 }
633 BEGIN_NVC0(push, SUBC_3D(0x0218), 1);
634 PUSH_DATA (push, 0x10);
635 BEGIN_NVC0(push, SUBC_3D(0x10fc), 1);
636 PUSH_DATA (push, 0x10);
637 BEGIN_NVC0(push, SUBC_3D(0x1290), 1);
638 PUSH_DATA (push, 0x10);
639 BEGIN_NVC0(push, SUBC_3D(0x12d8), 2);
640 PUSH_DATA (push, 0x10);
641 PUSH_DATA (push, 0x10);
642 BEGIN_NVC0(push, SUBC_3D(0x1140), 1);
643 PUSH_DATA (push, 0x10);
644 BEGIN_NVC0(push, SUBC_3D(0x1610), 1);
645 PUSH_DATA (push, 0xe);
646
647 BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_GEN_MODE), 1);
648 PUSH_DATA (push, NVC0_3D_VERTEX_ID_GEN_MODE_DRAW_ARRAYS_ADD_START);
649 BEGIN_NVC0(push, SUBC_3D(0x030c), 1);
650 PUSH_DATA (push, 0);
651 BEGIN_NVC0(push, SUBC_3D(0x0300), 1);
652 PUSH_DATA (push, 3);
653
654 BEGIN_NVC0(push, SUBC_3D(0x02d0), 1);
655 PUSH_DATA (push, 0x3fffff);
656 BEGIN_NVC0(push, SUBC_3D(0x0fdc), 1);
657 PUSH_DATA (push, 1);
658 BEGIN_NVC0(push, SUBC_3D(0x19c0), 1);
659 PUSH_DATA (push, 1);
660
661 if (obj_class < GM107_3D_CLASS) {
662 BEGIN_NVC0(push, SUBC_3D(0x075c), 1);
663 PUSH_DATA (push, 3);
664
665 if (obj_class >= NVE4_3D_CLASS) {
666 BEGIN_NVC0(push, SUBC_3D(0x07fc), 1);
667 PUSH_DATA (push, 1);
668 }
669 }
670
671 /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
672 * are supposed to do */
673 }
674
675 static void
676 nvc0_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
677 {
678 struct nvc0_screen *screen = nvc0_screen(pscreen);
679 struct nouveau_pushbuf *push = screen->base.pushbuf;
680
681 /* we need to do it after possible flush in MARK_RING */
682 *sequence = ++screen->base.fence.sequence;
683
684 assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
685 PUSH_DATA (push, NVC0_FIFO_PKHDR_SQ(NVC0_3D(QUERY_ADDRESS_HIGH), 4));
686 PUSH_DATAh(push, screen->fence.bo->offset);
687 PUSH_DATA (push, screen->fence.bo->offset);
688 PUSH_DATA (push, *sequence);
689 PUSH_DATA (push, NVC0_3D_QUERY_GET_FENCE | NVC0_3D_QUERY_GET_SHORT |
690 (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT));
691 }
692
693 static u32
694 nvc0_screen_fence_update(struct pipe_screen *pscreen)
695 {
696 struct nvc0_screen *screen = nvc0_screen(pscreen);
697 return screen->fence.map[0];
698 }
699
700 static int
701 nvc0_screen_init_compute(struct nvc0_screen *screen)
702 {
703 screen->base.base.get_compute_param = nvc0_screen_get_compute_param;
704
705 switch (screen->base.device->chipset & ~0xf) {
706 case 0xc0:
707 case 0xd0:
708 return nvc0_screen_compute_setup(screen, screen->base.pushbuf);
709 case 0xe0:
710 case 0xf0:
711 case 0x100:
712 case 0x110:
713 case 0x120:
714 case 0x130:
715 return nve4_screen_compute_setup(screen, screen->base.pushbuf);
716 default:
717 return -1;
718 }
719 }
720
721 static int
722 nvc0_screen_resize_tls_area(struct nvc0_screen *screen,
723 uint32_t lpos, uint32_t lneg, uint32_t cstack)
724 {
725 struct nouveau_bo *bo = NULL;
726 int ret;
727 uint64_t size = (lpos + lneg) * 32 + cstack;
728
729 if (size >= (1 << 20)) {
730 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64"\n", size);
731 return -1;
732 }
733
734 size *= (screen->base.device->chipset >= 0xe0) ? 64 : 48; /* max warps */
735 size = align(size, 0x8000);
736 size *= screen->mp_count;
737
738 size = align(size, 1 << 17);
739
740 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base), 1 << 17, size,
741 NULL, &bo);
742 if (ret)
743 return ret;
744
745 /* Make sure that the pushbuf has acquired a reference to the old tls
746 * segment, as it may have commands that will reference it.
747 */
748 if (screen->tls)
749 PUSH_REFN(screen->base.pushbuf, screen->tls,
750 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RDWR);
751 nouveau_bo_ref(NULL, &screen->tls);
752 screen->tls = bo;
753 return 0;
754 }
755
756 int
757 nvc0_screen_resize_text_area(struct nvc0_screen *screen, uint64_t size)
758 {
759 struct nouveau_pushbuf *push = screen->base.pushbuf;
760 struct nouveau_bo *bo;
761 int ret;
762
763 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base),
764 1 << 17, size, NULL, &bo);
765 if (ret)
766 return ret;
767
768 /* Make sure that the pushbuf has acquired a reference to the old text
769 * segment, as it may have commands that will reference it.
770 */
771 if (screen->text)
772 PUSH_REFN(push, screen->text,
773 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RD);
774 nouveau_bo_ref(NULL, &screen->text);
775 screen->text = bo;
776
777 nouveau_heap_destroy(&screen->lib_code);
778 nouveau_heap_destroy(&screen->text_heap);
779
780 /* XXX: getting a page fault at the end of the code buffer every few
781 * launches, don't use the last 256 bytes to work around them - prefetch ?
782 */
783 nouveau_heap_init(&screen->text_heap, 0, size - 0x100);
784
785 /* update the code segment setup */
786 BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
787 PUSH_DATAh(push, screen->text->offset);
788 PUSH_DATA (push, screen->text->offset);
789 if (screen->compute) {
790 BEGIN_NVC0(push, NVC0_CP(CODE_ADDRESS_HIGH), 2);
791 PUSH_DATAh(push, screen->text->offset);
792 PUSH_DATA (push, screen->text->offset);
793 }
794
795 return 0;
796 }
797
798 #define FAIL_SCREEN_INIT(str, err) \
799 do { \
800 NOUVEAU_ERR(str, err); \
801 goto fail; \
802 } while(0)
803
804 struct nouveau_screen *
805 nvc0_screen_create(struct nouveau_device *dev)
806 {
807 struct nvc0_screen *screen;
808 struct pipe_screen *pscreen;
809 struct nouveau_object *chan;
810 struct nouveau_pushbuf *push;
811 uint64_t value;
812 uint32_t obj_class;
813 uint32_t flags;
814 int ret;
815 unsigned i;
816
817 switch (dev->chipset & ~0xf) {
818 case 0xc0:
819 case 0xd0:
820 case 0xe0:
821 case 0xf0:
822 case 0x100:
823 case 0x110:
824 case 0x120:
825 case 0x130:
826 break;
827 default:
828 return NULL;
829 }
830
831 screen = CALLOC_STRUCT(nvc0_screen);
832 if (!screen)
833 return NULL;
834 pscreen = &screen->base.base;
835 pscreen->destroy = nvc0_screen_destroy;
836
837 ret = nouveau_screen_init(&screen->base, dev);
838 if (ret)
839 FAIL_SCREEN_INIT("Base screen init failed: %d\n", ret);
840 chan = screen->base.channel;
841 push = screen->base.pushbuf;
842 push->user_priv = screen;
843 push->rsvd_kick = 5;
844
845 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
846 PIPE_BIND_SHADER_BUFFER |
847 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
848 PIPE_BIND_COMMAND_ARGS_BUFFER | PIPE_BIND_QUERY_BUFFER;
849 screen->base.sysmem_bindings |=
850 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
851
852 if (screen->base.vram_domain & NOUVEAU_BO_GART) {
853 screen->base.sysmem_bindings |= screen->base.vidmem_bindings;
854 screen->base.vidmem_bindings = 0;
855 }
856
857 pscreen->context_create = nvc0_create;
858 pscreen->is_format_supported = nvc0_screen_is_format_supported;
859 pscreen->get_param = nvc0_screen_get_param;
860 pscreen->get_shader_param = nvc0_screen_get_shader_param;
861 pscreen->get_paramf = nvc0_screen_get_paramf;
862 pscreen->get_driver_query_info = nvc0_screen_get_driver_query_info;
863 pscreen->get_driver_query_group_info = nvc0_screen_get_driver_query_group_info;
864
865 nvc0_screen_init_resource_functions(pscreen);
866
867 screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
868 screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
869
870 flags = NOUVEAU_BO_GART | NOUVEAU_BO_MAP;
871 if (screen->base.drm->version >= 0x01000202)
872 flags |= NOUVEAU_BO_COHERENT;
873
874 ret = nouveau_bo_new(dev, flags, 0, 4096, NULL, &screen->fence.bo);
875 if (ret)
876 FAIL_SCREEN_INIT("Error allocating fence BO: %d\n", ret);
877 nouveau_bo_map(screen->fence.bo, 0, NULL);
878 screen->fence.map = screen->fence.bo->map;
879 screen->base.fence.emit = nvc0_screen_fence_emit;
880 screen->base.fence.update = nvc0_screen_fence_update;
881
882
883 ret = nouveau_object_new(chan, (dev->chipset < 0xe0) ? 0x1f906e : 0x906e,
884 NVIF_CLASS_SW_GF100, NULL, 0, &screen->nvsw);
885 if (ret)
886 FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret);
887
888 BEGIN_NVC0(push, SUBC_SW(NV01_SUBCHAN_OBJECT), 1);
889 PUSH_DATA (push, screen->nvsw->handle);
890
891 switch (dev->chipset & ~0xf) {
892 case 0x130:
893 case 0x120:
894 case 0x110:
895 case 0x100:
896 case 0xf0:
897 obj_class = NVF0_P2MF_CLASS;
898 break;
899 case 0xe0:
900 obj_class = NVE4_P2MF_CLASS;
901 break;
902 default:
903 obj_class = NVC0_M2MF_CLASS;
904 break;
905 }
906 ret = nouveau_object_new(chan, 0xbeef323f, obj_class, NULL, 0,
907 &screen->m2mf);
908 if (ret)
909 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
910
911 BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
912 PUSH_DATA (push, screen->m2mf->oclass);
913 if (screen->m2mf->oclass == NVE4_P2MF_CLASS) {
914 BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
915 PUSH_DATA (push, 0xa0b5);
916 }
917
918 ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
919 &screen->eng2d);
920 if (ret)
921 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
922
923 BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
924 PUSH_DATA (push, screen->eng2d->oclass);
925 BEGIN_NVC0(push, SUBC_2D(NVC0_2D_SINGLE_GPC), 1);
926 PUSH_DATA (push, 0);
927 BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
928 PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
929 BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
930 PUSH_DATA (push, 0);
931 BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
932 PUSH_DATA (push, 0);
933 BEGIN_NVC0(push, SUBC_2D(0x0884), 1);
934 PUSH_DATA (push, 0x3f);
935 BEGIN_NVC0(push, SUBC_2D(0x0888), 1);
936 PUSH_DATA (push, 1);
937 BEGIN_NVC0(push, NVC0_2D(COND_MODE), 1);
938 PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
939
940 BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
941 PUSH_DATAh(push, screen->fence.bo->offset + 16);
942 PUSH_DATA (push, screen->fence.bo->offset + 16);
943
944 switch (dev->chipset & ~0xf) {
945 case 0x130:
946 switch (dev->chipset) {
947 case 0x130:
948 case 0x13b:
949 obj_class = GP100_3D_CLASS;
950 break;
951 default:
952 obj_class = GP102_3D_CLASS;
953 break;
954 }
955 break;
956 case 0x120:
957 obj_class = GM200_3D_CLASS;
958 break;
959 case 0x110:
960 obj_class = GM107_3D_CLASS;
961 break;
962 case 0x100:
963 case 0xf0:
964 obj_class = NVF0_3D_CLASS;
965 break;
966 case 0xe0:
967 switch (dev->chipset) {
968 case 0xea:
969 obj_class = NVEA_3D_CLASS;
970 break;
971 default:
972 obj_class = NVE4_3D_CLASS;
973 break;
974 }
975 break;
976 case 0xd0:
977 obj_class = NVC8_3D_CLASS;
978 break;
979 case 0xc0:
980 default:
981 switch (dev->chipset) {
982 case 0xc8:
983 obj_class = NVC8_3D_CLASS;
984 break;
985 case 0xc1:
986 obj_class = NVC1_3D_CLASS;
987 break;
988 default:
989 obj_class = NVC0_3D_CLASS;
990 break;
991 }
992 break;
993 }
994 ret = nouveau_object_new(chan, 0xbeef003d, obj_class, NULL, 0,
995 &screen->eng3d);
996 if (ret)
997 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
998 screen->base.class_3d = obj_class;
999
1000 BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
1001 PUSH_DATA (push, screen->eng3d->oclass);
1002
1003 BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
1004 PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
1005
1006 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
1007 /* kill shaders after about 1 second (at 100 MHz) */
1008 BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
1009 PUSH_DATA (push, 0x17);
1010 }
1011
1012 IMMED_NVC0(push, NVC0_3D(ZETA_COMP_ENABLE),
1013 screen->base.drm->version >= 0x01000101);
1014 BEGIN_NVC0(push, NVC0_3D(RT_COMP_ENABLE(0)), 8);
1015 for (i = 0; i < 8; ++i)
1016 PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
1017
1018 BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
1019 PUSH_DATA (push, 1);
1020
1021 BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
1022 PUSH_DATA (push, 0);
1023 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
1024 PUSH_DATA (push, 0);
1025 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
1026 PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
1027 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
1028 PUSH_DATA (push, 0);
1029 BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
1030 PUSH_DATA (push, 1);
1031 BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
1032 PUSH_DATA (push, 1);
1033 BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
1034 PUSH_DATA (push, 1);
1035 BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
1036 PUSH_DATA (push, 0);
1037 BEGIN_NVC0(push, NVC0_3D(SHADE_MODEL), 1);
1038 PUSH_DATA (push, NVC0_3D_SHADE_MODEL_SMOOTH);
1039 if (screen->eng3d->oclass < NVE4_3D_CLASS) {
1040 IMMED_NVC0(push, NVC0_3D(TEX_MISC), 0);
1041 } else {
1042 BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
1043 PUSH_DATA (push, 15);
1044 }
1045 BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
1046 PUSH_DATA (push, 8); /* 128 */
1047 BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
1048 PUSH_DATA (push, 1);
1049 if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
1050 BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
1051 PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
1052 }
1053
1054 nvc0_magic_3d_init(push, screen->eng3d->oclass);
1055
1056 ret = nvc0_screen_resize_text_area(screen, 1 << 19);
1057 if (ret)
1058 FAIL_SCREEN_INIT("Error allocating TEXT area: %d\n", ret);
1059
1060 /* 6 user uniform areas, 6 driver areas, and 1 for the runout */
1061 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 12, 13 << 16, NULL,
1062 &screen->uniform_bo);
1063 if (ret)
1064 FAIL_SCREEN_INIT("Error allocating uniform BO: %d\n", ret);
1065
1066 PUSH_REFN (push, screen->uniform_bo, NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_WR);
1067
1068 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
1069 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1070 PUSH_DATA (push, 256);
1071 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1072 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1073 BEGIN_1IC0(push, NVC0_3D(CB_POS), 5);
1074 PUSH_DATA (push, 0);
1075 PUSH_DATAf(push, 0.0f);
1076 PUSH_DATAf(push, 0.0f);
1077 PUSH_DATAf(push, 0.0f);
1078 PUSH_DATAf(push, 0.0f);
1079 BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
1080 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1081 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1082
1083 if (screen->base.drm->version >= 0x01000101) {
1084 ret = nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
1085 if (ret)
1086 FAIL_SCREEN_INIT("NOUVEAU_GETPARAM_GRAPH_UNITS failed: %d\n", ret);
1087 } else {
1088 if (dev->chipset >= 0xe0 && dev->chipset < 0xf0)
1089 value = (8 << 8) | 4;
1090 else
1091 value = (16 << 8) | 4;
1092 }
1093 screen->gpc_count = value & 0x000000ff;
1094 screen->mp_count = value >> 8;
1095 screen->mp_count_compute = screen->mp_count;
1096
1097 ret = nvc0_screen_resize_tls_area(screen, 128 * 16, 0, 0x200);
1098 if (ret)
1099 FAIL_SCREEN_INIT("Error allocating TLS area: %d\n", ret);
1100
1101 BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
1102 PUSH_DATAh(push, screen->tls->offset);
1103 PUSH_DATA (push, screen->tls->offset);
1104 PUSH_DATA (push, screen->tls->size >> 32);
1105 PUSH_DATA (push, screen->tls->size);
1106 BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
1107 PUSH_DATA (push, 0);
1108 /* Reduce likelihood of collision with real buffers by placing the hole at
1109 * the top of the 4G area. This will have to be dealt with for real
1110 * eventually by blocking off that area from the VM.
1111 */
1112 BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
1113 PUSH_DATA (push, 0xff << 24);
1114
1115 if (screen->eng3d->oclass < GM107_3D_CLASS) {
1116 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 20, NULL,
1117 &screen->poly_cache);
1118 if (ret)
1119 FAIL_SCREEN_INIT("Error allocating poly cache BO: %d\n", ret);
1120
1121 BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
1122 PUSH_DATAh(push, screen->poly_cache->offset);
1123 PUSH_DATA (push, screen->poly_cache->offset);
1124 PUSH_DATA (push, 3);
1125 }
1126
1127 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 17, NULL,
1128 &screen->txc);
1129 if (ret)
1130 FAIL_SCREEN_INIT("Error allocating txc BO: %d\n", ret);
1131
1132 BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
1133 PUSH_DATAh(push, screen->txc->offset);
1134 PUSH_DATA (push, screen->txc->offset);
1135 PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
1136 if (screen->eng3d->oclass >= GM107_3D_CLASS) {
1137 screen->tic.maxwell = true;
1138 if (screen->eng3d->oclass == GM107_3D_CLASS) {
1139 screen->tic.maxwell =
1140 debug_get_bool_option("NOUVEAU_MAXWELL_TIC", true);
1141 IMMED_NVC0(push, SUBC_3D(0x0f10), screen->tic.maxwell);
1142 }
1143 }
1144
1145 BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
1146 PUSH_DATAh(push, screen->txc->offset + 65536);
1147 PUSH_DATA (push, screen->txc->offset + 65536);
1148 PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
1149
1150 BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
1151 PUSH_DATA (push, 0);
1152 BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
1153 PUSH_DATA (push, 0);
1154 PUSH_DATA (push, 0);
1155 BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
1156 PUSH_DATA (push, 0x3f);
1157
1158 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
1159 PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
1160 BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
1161 for (i = 0; i < 8 * 2; ++i)
1162 PUSH_DATA(push, 0);
1163 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
1164 PUSH_DATA (push, 0);
1165 BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
1166 PUSH_DATA (push, 0);
1167
1168 /* neither scissors, viewport nor stencil mask should affect clears */
1169 BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
1170 PUSH_DATA (push, 0);
1171
1172 BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
1173 PUSH_DATA (push, 1);
1174 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1175 BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(i)), 2);
1176 PUSH_DATAf(push, 0.0f);
1177 PUSH_DATAf(push, 1.0f);
1178 }
1179 BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
1180 PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
1181
1182 /* We use scissors instead of exact view volume clipping,
1183 * so they're always enabled.
1184 */
1185 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1186 BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(i)), 3);
1187 PUSH_DATA (push, 1);
1188 PUSH_DATA (push, 8192 << 16);
1189 PUSH_DATA (push, 8192 << 16);
1190 }
1191
1192 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
1193
1194 i = 0;
1195 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, mme9097_per_instance_bf);
1196 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, mme9097_blend_enables);
1197 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, mme9097_vertex_array_select);
1198 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, mme9097_tep_select);
1199 MK_MACRO(NVC0_3D_MACRO_GP_SELECT, mme9097_gp_select);
1200 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, mme9097_poly_mode_front);
1201 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, mme9097_poly_mode_back);
1202 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT, mme9097_draw_arrays_indirect);
1203 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT, mme9097_draw_elts_indirect);
1204 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT, mme9097_draw_arrays_indirect_count);
1205 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT, mme9097_draw_elts_indirect_count);
1206 MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE, mme9097_query_buffer_write);
1207 MK_MACRO(NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT, mme90c0_launch_grid_indirect);
1208
1209 BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
1210 PUSH_DATA (push, 1);
1211 BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
1212 PUSH_DATA (push, 1);
1213 BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
1214 PUSH_DATA (push, 0x40);
1215 BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
1216 PUSH_DATA (push, 0);
1217 BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
1218 PUSH_DATA (push, 0x30);
1219 BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
1220 PUSH_DATA (push, 3);
1221 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
1222 PUSH_DATA (push, 0x20);
1223 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
1224 PUSH_DATA (push, 0x00);
1225 screen->save_state.patch_vertices = 3;
1226
1227 BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
1228 PUSH_DATA (push, 0);
1229 BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
1230 PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
1231
1232 IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
1233
1234 if (nvc0_screen_init_compute(screen))
1235 goto fail;
1236
1237 /* XXX: Compute and 3D are somehow aliased on Fermi. */
1238 for (i = 0; i < 5; ++i) {
1239 /* TIC and TSC entries for each unit (nve4+ only) */
1240 /* auxiliary constants (6 user clip planes, base instance id) */
1241 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1242 PUSH_DATA (push, NVC0_CB_AUX_SIZE);
1243 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(i));
1244 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(i));
1245 BEGIN_NVC0(push, NVC0_3D(CB_BIND(i)), 1);
1246 PUSH_DATA (push, (15 << 4) | 1);
1247 if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
1248 unsigned j;
1249 BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
1250 PUSH_DATA (push, NVC0_CB_AUX_UNK_INFO);
1251 for (j = 0; j < 8; ++j)
1252 PUSH_DATA(push, j);
1253 } else {
1254 BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
1255 PUSH_DATA (push, 0x54);
1256 }
1257
1258 /* MS sample coordinate offsets: these do not work with _ALT modes ! */
1259 BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 2 * 8);
1260 PUSH_DATA (push, NVC0_CB_AUX_MS_INFO);
1261 PUSH_DATA (push, 0); /* 0 */
1262 PUSH_DATA (push, 0);
1263 PUSH_DATA (push, 1); /* 1 */
1264 PUSH_DATA (push, 0);
1265 PUSH_DATA (push, 0); /* 2 */
1266 PUSH_DATA (push, 1);
1267 PUSH_DATA (push, 1); /* 3 */
1268 PUSH_DATA (push, 1);
1269 PUSH_DATA (push, 2); /* 4 */
1270 PUSH_DATA (push, 0);
1271 PUSH_DATA (push, 3); /* 5 */
1272 PUSH_DATA (push, 0);
1273 PUSH_DATA (push, 2); /* 6 */
1274 PUSH_DATA (push, 1);
1275 PUSH_DATA (push, 3); /* 7 */
1276 PUSH_DATA (push, 1);
1277 }
1278 BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
1279 PUSH_DATA (push, 0);
1280
1281 PUSH_KICK (push);
1282
1283 screen->tic.entries = CALLOC(
1284 NVC0_TIC_MAX_ENTRIES + NVC0_TSC_MAX_ENTRIES + NVE4_IMG_MAX_HANDLES,
1285 sizeof(void *));
1286 screen->tsc.entries = screen->tic.entries + NVC0_TIC_MAX_ENTRIES;
1287 screen->img.entries = (void *)(screen->tsc.entries + NVC0_TSC_MAX_ENTRIES);
1288
1289 if (!nvc0_blitter_create(screen))
1290 goto fail;
1291
1292 screen->default_tsc = CALLOC_STRUCT(nv50_tsc_entry);
1293 screen->default_tsc->tsc[0] = G80_TSC_0_SRGB_CONVERSION;
1294
1295 nouveau_fence_new(&screen->base, &screen->base.fence.current);
1296
1297 return &screen->base;
1298
1299 fail:
1300 screen->base.base.context_create = NULL;
1301 return &screen->base;
1302 }
1303
1304 int
1305 nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
1306 {
1307 int i = screen->tic.next;
1308
1309 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1310 i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1311
1312 screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1313
1314 if (screen->tic.entries[i])
1315 nv50_tic_entry(screen->tic.entries[i])->id = -1;
1316
1317 screen->tic.entries[i] = entry;
1318 return i;
1319 }
1320
1321 int
1322 nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
1323 {
1324 int i = screen->tsc.next;
1325
1326 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1327 i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1328
1329 screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1330
1331 if (screen->tsc.entries[i])
1332 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1333
1334 screen->tsc.entries[i] = entry;
1335 return i;
1336 }