nv50,nvc0: warn on not-explicitly-handled caps
[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 "util/u_screen.h"
29 #include "pipe/p_screen.h"
30
31 #include "nouveau_vp3_video.h"
32
33 #include "nvc0/nvc0_context.h"
34 #include "nvc0/nvc0_screen.h"
35
36 #include "nvc0/mme/com9097.mme.h"
37 #include "nvc0/mme/com90c0.mme.h"
38
39 #include "nv50/g80_texture.xml.h"
40
41 static boolean
42 nvc0_screen_is_format_supported(struct pipe_screen *pscreen,
43 enum pipe_format format,
44 enum pipe_texture_target target,
45 unsigned sample_count,
46 unsigned storage_sample_count,
47 unsigned bindings)
48 {
49 const struct util_format_description *desc = util_format_description(format);
50
51 if (sample_count > 8)
52 return false;
53 if (!(0x117 & (1 << sample_count))) /* 0, 1, 2, 4 or 8 */
54 return false;
55
56 if (MAX2(1, sample_count) != MAX2(1, storage_sample_count))
57 return false;
58
59 /* Short-circuit the rest of the logic -- this is used by the state tracker
60 * to determine valid MS levels in a no-attachments scenario.
61 */
62 if (format == PIPE_FORMAT_NONE && bindings & PIPE_BIND_RENDER_TARGET)
63 return true;
64
65 if ((bindings & PIPE_BIND_SAMPLER_VIEW) && (target != PIPE_BUFFER))
66 if (util_format_get_blocksizebits(format) == 3 * 32)
67 return false;
68
69 if (bindings & PIPE_BIND_LINEAR)
70 if (util_format_is_depth_or_stencil(format) ||
71 (target != PIPE_TEXTURE_1D &&
72 target != PIPE_TEXTURE_2D &&
73 target != PIPE_TEXTURE_RECT) ||
74 sample_count > 1)
75 return false;
76
77 /* Restrict ETC2 and ASTC formats here. These are only supported on GK20A.
78 */
79 if ((desc->layout == UTIL_FORMAT_LAYOUT_ETC ||
80 desc->layout == UTIL_FORMAT_LAYOUT_ASTC) &&
81 /* The claim is that this should work on GM107 but it doesn't. Need to
82 * test further and figure out if it's a nouveau issue or a HW one.
83 nouveau_screen(pscreen)->class_3d < GM107_3D_CLASS &&
84 */
85 nouveau_screen(pscreen)->class_3d != NVEA_3D_CLASS)
86 return false;
87
88 /* shared is always supported */
89 bindings &= ~(PIPE_BIND_LINEAR |
90 PIPE_BIND_SHARED);
91
92 if (bindings & PIPE_BIND_SHADER_IMAGE) {
93 if (format == PIPE_FORMAT_B8G8R8A8_UNORM &&
94 nouveau_screen(pscreen)->class_3d < NVE4_3D_CLASS) {
95 /* This should work on Fermi, but for currently unknown reasons it
96 * does not and results in breaking reads from pbos. */
97 return false;
98 }
99 }
100
101 return (( nvc0_format_table[format].usage |
102 nvc0_vertex_format[format].usage) & bindings) == bindings;
103 }
104
105 static int
106 nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
107 {
108 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
109 struct nouveau_device *dev = nouveau_screen(pscreen)->device;
110
111 switch (param) {
112 /* non-boolean caps */
113 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
114 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
115 return 15;
116 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
117 return 12;
118 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
119 return 2048;
120 case PIPE_CAP_MIN_TEXEL_OFFSET:
121 return -8;
122 case PIPE_CAP_MAX_TEXEL_OFFSET:
123 return 7;
124 case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
125 return -32;
126 case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
127 return 31;
128 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
129 return 128 * 1024 * 1024;
130 case PIPE_CAP_GLSL_FEATURE_LEVEL:
131 return 430;
132 case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY:
133 return 430;
134 case PIPE_CAP_MAX_RENDER_TARGETS:
135 return 8;
136 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
137 return 1;
138 case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
139 case PIPE_CAP_RASTERIZER_SUBPIXEL_BITS:
140 return 8;
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_GS_INVOCATIONS:
152 return 32;
153 case PIPE_CAP_MAX_SHADER_BUFFER_SIZE:
154 return 1 << 27;
155 case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
156 return 2048;
157 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
158 return 256;
159 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
160 if (class_3d < GM107_3D_CLASS)
161 return 256; /* IMAGE bindings require alignment to 256 */
162 return 16;
163 case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
164 return 16;
165 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
166 return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
167 case PIPE_CAP_MAX_VIEWPORTS:
168 return NVC0_MAX_VIEWPORTS;
169 case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
170 return 4;
171 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
172 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
173 case PIPE_CAP_ENDIANNESS:
174 return PIPE_ENDIAN_LITTLE;
175 case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
176 return 30;
177 case PIPE_CAP_MAX_WINDOW_RECTANGLES:
178 return NVC0_MAX_WINDOW_RECTANGLES;
179 case PIPE_CAP_MAX_CONSERVATIVE_RASTER_SUBPIXEL_PRECISION_BIAS:
180 return class_3d >= GM200_3D_CLASS ? 8 : 0;
181
182 /* supported caps */
183 case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
184 case PIPE_CAP_TEXTURE_MIRROR_CLAMP_TO_EDGE:
185 case PIPE_CAP_TEXTURE_SWIZZLE:
186 case PIPE_CAP_NPOT_TEXTURES:
187 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
188 case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
189 case PIPE_CAP_ANISOTROPIC_FILTER:
190 case PIPE_CAP_SEAMLESS_CUBE_MAP:
191 case PIPE_CAP_CUBE_MAP_ARRAY:
192 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
193 case PIPE_CAP_TEXTURE_MULTISAMPLE:
194 case PIPE_CAP_DEPTH_CLIP_DISABLE:
195 case PIPE_CAP_POINT_SPRITE:
196 case PIPE_CAP_TGSI_TEXCOORD:
197 case PIPE_CAP_SM3:
198 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
199 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
200 case PIPE_CAP_VERTEX_COLOR_CLAMPED:
201 case PIPE_CAP_QUERY_TIMESTAMP:
202 case PIPE_CAP_QUERY_TIME_ELAPSED:
203 case PIPE_CAP_OCCLUSION_QUERY:
204 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
205 case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
206 case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
207 case PIPE_CAP_BLEND_EQUATION_SEPARATE:
208 case PIPE_CAP_INDEP_BLEND_ENABLE:
209 case PIPE_CAP_INDEP_BLEND_FUNC:
210 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
211 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
212 case PIPE_CAP_PRIMITIVE_RESTART:
213 case PIPE_CAP_TGSI_INSTANCEID:
214 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
215 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
216 case PIPE_CAP_CONDITIONAL_RENDER:
217 case PIPE_CAP_TEXTURE_BARRIER:
218 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
219 case PIPE_CAP_START_INSTANCE:
220 case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
221 case PIPE_CAP_DRAW_INDIRECT:
222 case PIPE_CAP_USER_VERTEX_BUFFERS:
223 case PIPE_CAP_TEXTURE_QUERY_LOD:
224 case PIPE_CAP_SAMPLE_SHADING:
225 case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
226 case PIPE_CAP_TEXTURE_GATHER_SM5:
227 case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
228 case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
229 case PIPE_CAP_SAMPLER_VIEW_TARGET:
230 case PIPE_CAP_CLIP_HALFZ:
231 case PIPE_CAP_POLYGON_OFFSET_CLAMP:
232 case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
233 case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
234 case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
235 case PIPE_CAP_DEPTH_BOUNDS_TEST:
236 case PIPE_CAP_TGSI_TXQS:
237 case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
238 case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
239 case PIPE_CAP_SHAREABLE_SHADERS:
240 case PIPE_CAP_CLEAR_TEXTURE:
241 case PIPE_CAP_DRAW_PARAMETERS:
242 case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
243 case PIPE_CAP_MULTI_DRAW_INDIRECT:
244 case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
245 case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
246 case PIPE_CAP_QUERY_BUFFER_OBJECT:
247 case PIPE_CAP_INVALIDATE_BUFFER:
248 case PIPE_CAP_STRING_MARKER:
249 case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
250 case PIPE_CAP_CULL_DISTANCE:
251 case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
252 case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
253 case PIPE_CAP_TGSI_VOTE:
254 case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
255 case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
256 case PIPE_CAP_TGSI_MUL_ZERO_WINS:
257 case PIPE_CAP_DOUBLES:
258 case PIPE_CAP_INT64:
259 case PIPE_CAP_TGSI_TEX_TXF_LZ:
260 case PIPE_CAP_TGSI_CLOCK:
261 case PIPE_CAP_COMPUTE:
262 case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
263 case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
264 case PIPE_CAP_QUERY_SO_OVERFLOW:
265 return 1;
266 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
267 return nouveau_screen(pscreen)->vram_domain & NOUVEAU_BO_VRAM ? 1 : 0;
268 case PIPE_CAP_TGSI_FS_FBFETCH:
269 return class_3d >= NVE4_3D_CLASS; /* needs testing on fermi */
270 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
271 case PIPE_CAP_TGSI_BALLOT:
272 case PIPE_CAP_BINDLESS_TEXTURE:
273 return class_3d >= NVE4_3D_CLASS;
274 case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
275 case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
276 case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
277 case PIPE_CAP_POST_DEPTH_COVERAGE:
278 case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_TRIANGLES:
279 case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_POINTS_LINES:
280 case PIPE_CAP_CONSERVATIVE_RASTER_POST_DEPTH_COVERAGE:
281 case PIPE_CAP_PROGRAMMABLE_SAMPLE_LOCATIONS:
282 return class_3d >= GM200_3D_CLASS;
283 case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES:
284 return class_3d >= GP100_3D_CLASS;
285
286 /* unsupported caps */
287 case PIPE_CAP_DEPTH_CLIP_DISABLE_SEPARATE:
288 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
289 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
290 case PIPE_CAP_SHADER_STENCIL_EXPORT:
291 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
292 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
293 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
294 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
295 case PIPE_CAP_FAKE_SW_MSAA:
296 case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
297 case PIPE_CAP_VERTEXID_NOBASE:
298 case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
299 case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
300 case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
301 case PIPE_CAP_GENERATE_MIPMAP:
302 case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
303 case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
304 case PIPE_CAP_QUERY_MEMORY_INFO:
305 case PIPE_CAP_PCI_GROUP:
306 case PIPE_CAP_PCI_BUS:
307 case PIPE_CAP_PCI_DEVICE:
308 case PIPE_CAP_PCI_FUNCTION:
309 case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
310 case PIPE_CAP_NATIVE_FENCE_FD:
311 case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
312 case PIPE_CAP_INT64_DIVMOD:
313 case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
314 case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
315 case PIPE_CAP_MEMOBJ:
316 case PIPE_CAP_LOAD_CONSTBUF:
317 case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS:
318 case PIPE_CAP_TILE_RASTER_ORDER:
319 case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
320 case PIPE_CAP_FRAMEBUFFER_MSAA_CONSTRAINTS:
321 case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
322 case PIPE_CAP_CONTEXT_PRIORITY_MASK:
323 case PIPE_CAP_FENCE_SIGNAL:
324 case PIPE_CAP_CONSTBUF0_FLAGS:
325 case PIPE_CAP_PACKED_UNIFORMS:
326 case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_POINTS_LINES:
327 case PIPE_CAP_MAX_TEXTURE_UPLOAD_MEMORY_BUDGET:
328 case PIPE_CAP_MAX_COMBINED_SHADER_BUFFERS:
329 case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTERS:
330 case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTER_BUFFERS:
331 return 0;
332
333 case PIPE_CAP_VENDOR_ID:
334 return 0x10de;
335 case PIPE_CAP_DEVICE_ID: {
336 uint64_t device_id;
337 if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
338 NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
339 return -1;
340 }
341 return device_id;
342 }
343 case PIPE_CAP_ACCELERATED:
344 return 1;
345 case PIPE_CAP_VIDEO_MEMORY:
346 return dev->vram_size >> 20;
347 case PIPE_CAP_UMA:
348 return 0;
349 default:
350 debug_printf("%s: unhandled cap %d\n", __func__, param);
351 return u_pipe_screen_get_param_defaults(pscreen, param);
352 }
353 }
354
355 static int
356 nvc0_screen_get_shader_param(struct pipe_screen *pscreen,
357 enum pipe_shader_type shader,
358 enum pipe_shader_cap param)
359 {
360 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
361
362 switch (shader) {
363 case PIPE_SHADER_VERTEX:
364 case PIPE_SHADER_GEOMETRY:
365 case PIPE_SHADER_FRAGMENT:
366 case PIPE_SHADER_COMPUTE:
367 case PIPE_SHADER_TESS_CTRL:
368 case PIPE_SHADER_TESS_EVAL:
369 break;
370 default:
371 return 0;
372 }
373
374 switch (param) {
375 case PIPE_SHADER_CAP_PREFERRED_IR:
376 return PIPE_SHADER_IR_TGSI;
377 case PIPE_SHADER_CAP_SUPPORTED_IRS:
378 return 1 << PIPE_SHADER_IR_TGSI;
379 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
380 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
381 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
382 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
383 return 16384;
384 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
385 return 16;
386 case PIPE_SHADER_CAP_MAX_INPUTS:
387 if (shader == PIPE_SHADER_VERTEX)
388 return 32;
389 /* NOTE: These only count our slots for GENERIC varyings.
390 * The address space may be larger, but the actual hard limit seems to be
391 * less than what the address space layout permits, so don't add TEXCOORD,
392 * COLOR, etc. here.
393 */
394 if (shader == PIPE_SHADER_FRAGMENT)
395 return 0x1f0 / 16;
396 /* Actually this counts CLIPVERTEX, which occupies the last generic slot,
397 * and excludes 0x60 per-patch inputs.
398 */
399 return 0x200 / 16;
400 case PIPE_SHADER_CAP_MAX_OUTPUTS:
401 return 32;
402 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
403 return NVC0_MAX_CONSTBUF_SIZE;
404 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
405 return NVC0_MAX_PIPE_CONSTBUFS;
406 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
407 return shader != PIPE_SHADER_FRAGMENT;
408 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
409 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
410 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
411 return 1;
412 case PIPE_SHADER_CAP_MAX_TEMPS:
413 return NVC0_CAP_MAX_PROGRAM_TEMPS;
414 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
415 return 1;
416 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
417 return 1;
418 case PIPE_SHADER_CAP_SUBROUTINES:
419 return 1;
420 case PIPE_SHADER_CAP_INTEGERS:
421 return 1;
422 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
423 return 1;
424 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
425 return 1;
426 case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
427 return 1;
428 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
429 case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
430 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
431 case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
432 case PIPE_SHADER_CAP_INT64_ATOMICS:
433 case PIPE_SHADER_CAP_FP16:
434 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
435 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
436 return 0;
437 case PIPE_SHADER_CAP_SCALAR_ISA:
438 return 1;
439 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
440 return NVC0_MAX_BUFFERS;
441 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
442 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
443 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
444 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
445 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
446 return 32;
447 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
448 if (class_3d >= NVE4_3D_CLASS)
449 return NVC0_MAX_IMAGES;
450 if (shader == PIPE_SHADER_FRAGMENT || shader == PIPE_SHADER_COMPUTE)
451 return NVC0_MAX_IMAGES;
452 return 0;
453 default:
454 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
455 return 0;
456 }
457 }
458
459 static float
460 nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
461 {
462 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
463
464 switch (param) {
465 case PIPE_CAPF_MAX_LINE_WIDTH:
466 case PIPE_CAPF_MAX_LINE_WIDTH_AA:
467 return 10.0f;
468 case PIPE_CAPF_MAX_POINT_WIDTH:
469 return 63.0f;
470 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
471 return 63.375f;
472 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
473 return 16.0f;
474 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
475 return 15.0f;
476 case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE:
477 return 0.0f;
478 case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE:
479 return class_3d >= GM200_3D_CLASS ? 0.75f : 0.0f;
480 case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY:
481 return class_3d >= GM200_3D_CLASS ? 0.25f : 0.0f;
482 }
483
484 NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
485 return 0.0f;
486 }
487
488 static int
489 nvc0_screen_get_compute_param(struct pipe_screen *pscreen,
490 enum pipe_shader_ir ir_type,
491 enum pipe_compute_cap param, void *data)
492 {
493 struct nvc0_screen *screen = nvc0_screen(pscreen);
494 const uint16_t obj_class = screen->compute->oclass;
495
496 #define RET(x) do { \
497 if (data) \
498 memcpy(data, x, sizeof(x)); \
499 return sizeof(x); \
500 } while (0)
501
502 switch (param) {
503 case PIPE_COMPUTE_CAP_GRID_DIMENSION:
504 RET((uint64_t []) { 3 });
505 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
506 if (obj_class >= NVE4_COMPUTE_CLASS) {
507 RET(((uint64_t []) { 0x7fffffff, 65535, 65535 }));
508 } else {
509 RET(((uint64_t []) { 65535, 65535, 65535 }));
510 }
511 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
512 RET(((uint64_t []) { 1024, 1024, 64 }));
513 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
514 RET((uint64_t []) { 1024 });
515 case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
516 if (obj_class >= NVE4_COMPUTE_CLASS) {
517 RET((uint64_t []) { 1024 });
518 } else {
519 RET((uint64_t []) { 512 });
520 }
521 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */
522 RET((uint64_t []) { 1ULL << 40 });
523 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
524 switch (obj_class) {
525 case GM200_COMPUTE_CLASS:
526 RET((uint64_t []) { 96 << 10 });
527 break;
528 case GM107_COMPUTE_CLASS:
529 RET((uint64_t []) { 64 << 10 });
530 break;
531 default:
532 RET((uint64_t []) { 48 << 10 });
533 break;
534 }
535 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
536 RET((uint64_t []) { 512 << 10 });
537 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
538 RET((uint64_t []) { 4096 });
539 case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
540 RET((uint32_t []) { 32 });
541 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
542 RET((uint64_t []) { 1ULL << 40 });
543 case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
544 RET((uint32_t []) { 0 });
545 case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
546 RET((uint32_t []) { screen->mp_count_compute });
547 case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
548 RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
549 case PIPE_COMPUTE_CAP_ADDRESS_BITS:
550 RET((uint32_t []) { 64 });
551 default:
552 return 0;
553 }
554
555 #undef RET
556 }
557
558 static void
559 nvc0_screen_get_sample_pixel_grid(struct pipe_screen *pscreen,
560 unsigned sample_count,
561 unsigned *width, unsigned *height)
562 {
563 switch (sample_count) {
564 case 0:
565 case 1:
566 /* this could be 4x4, but the GL state tracker makes it difficult to
567 * create a 1x MSAA texture and smaller grids save CB space */
568 *width = 2;
569 *height = 4;
570 break;
571 case 2:
572 *width = 2;
573 *height = 4;
574 break;
575 case 4:
576 *width = 2;
577 *height = 2;
578 break;
579 case 8:
580 *width = 1;
581 *height = 2;
582 break;
583 default:
584 assert(0);
585 }
586 }
587
588 static void
589 nvc0_screen_destroy(struct pipe_screen *pscreen)
590 {
591 struct nvc0_screen *screen = nvc0_screen(pscreen);
592
593 if (!nouveau_drm_screen_unref(&screen->base))
594 return;
595
596 if (screen->base.fence.current) {
597 struct nouveau_fence *current = NULL;
598
599 /* nouveau_fence_wait will create a new current fence, so wait on the
600 * _current_ one, and remove both.
601 */
602 nouveau_fence_ref(screen->base.fence.current, &current);
603 nouveau_fence_wait(current, NULL);
604 nouveau_fence_ref(NULL, &current);
605 nouveau_fence_ref(NULL, &screen->base.fence.current);
606 }
607 if (screen->base.pushbuf)
608 screen->base.pushbuf->user_priv = NULL;
609
610 if (screen->blitter)
611 nvc0_blitter_destroy(screen);
612 if (screen->pm.prog) {
613 screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
614 nvc0_program_destroy(NULL, screen->pm.prog);
615 FREE(screen->pm.prog);
616 }
617
618 nouveau_bo_ref(NULL, &screen->text);
619 nouveau_bo_ref(NULL, &screen->uniform_bo);
620 nouveau_bo_ref(NULL, &screen->tls);
621 nouveau_bo_ref(NULL, &screen->txc);
622 nouveau_bo_ref(NULL, &screen->fence.bo);
623 nouveau_bo_ref(NULL, &screen->poly_cache);
624
625 nouveau_heap_destroy(&screen->lib_code);
626 nouveau_heap_destroy(&screen->text_heap);
627
628 FREE(screen->default_tsc);
629 FREE(screen->tic.entries);
630
631 nouveau_object_del(&screen->eng3d);
632 nouveau_object_del(&screen->eng2d);
633 nouveau_object_del(&screen->m2mf);
634 nouveau_object_del(&screen->compute);
635 nouveau_object_del(&screen->nvsw);
636
637 nouveau_screen_fini(&screen->base);
638
639 FREE(screen);
640 }
641
642 static int
643 nvc0_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos,
644 unsigned size, const uint32_t *data)
645 {
646 struct nouveau_pushbuf *push = screen->base.pushbuf;
647
648 size /= 4;
649
650 assert((pos + size) <= 0x800);
651
652 BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2);
653 PUSH_DATA (push, (m - 0x3800) / 8);
654 PUSH_DATA (push, pos);
655 BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1);
656 PUSH_DATA (push, pos);
657 PUSH_DATAp(push, data, size);
658
659 return pos + size;
660 }
661
662 static void
663 nvc0_magic_3d_init(struct nouveau_pushbuf *push, uint16_t obj_class)
664 {
665 BEGIN_NVC0(push, SUBC_3D(0x10cc), 1);
666 PUSH_DATA (push, 0xff);
667 BEGIN_NVC0(push, SUBC_3D(0x10e0), 2);
668 PUSH_DATA (push, 0xff);
669 PUSH_DATA (push, 0xff);
670 BEGIN_NVC0(push, SUBC_3D(0x10ec), 2);
671 PUSH_DATA (push, 0xff);
672 PUSH_DATA (push, 0xff);
673 BEGIN_NVC0(push, SUBC_3D(0x074c), 1);
674 PUSH_DATA (push, 0x3f);
675
676 BEGIN_NVC0(push, SUBC_3D(0x16a8), 1);
677 PUSH_DATA (push, (3 << 16) | 3);
678 BEGIN_NVC0(push, SUBC_3D(0x1794), 1);
679 PUSH_DATA (push, (2 << 16) | 2);
680
681 if (obj_class < GM107_3D_CLASS) {
682 BEGIN_NVC0(push, SUBC_3D(0x12ac), 1);
683 PUSH_DATA (push, 0);
684 }
685 BEGIN_NVC0(push, SUBC_3D(0x0218), 1);
686 PUSH_DATA (push, 0x10);
687 BEGIN_NVC0(push, SUBC_3D(0x10fc), 1);
688 PUSH_DATA (push, 0x10);
689 BEGIN_NVC0(push, SUBC_3D(0x1290), 1);
690 PUSH_DATA (push, 0x10);
691 BEGIN_NVC0(push, SUBC_3D(0x12d8), 2);
692 PUSH_DATA (push, 0x10);
693 PUSH_DATA (push, 0x10);
694 BEGIN_NVC0(push, SUBC_3D(0x1140), 1);
695 PUSH_DATA (push, 0x10);
696 BEGIN_NVC0(push, SUBC_3D(0x1610), 1);
697 PUSH_DATA (push, 0xe);
698
699 BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_GEN_MODE), 1);
700 PUSH_DATA (push, NVC0_3D_VERTEX_ID_GEN_MODE_DRAW_ARRAYS_ADD_START);
701 BEGIN_NVC0(push, SUBC_3D(0x030c), 1);
702 PUSH_DATA (push, 0);
703 BEGIN_NVC0(push, SUBC_3D(0x0300), 1);
704 PUSH_DATA (push, 3);
705
706 BEGIN_NVC0(push, SUBC_3D(0x02d0), 1);
707 PUSH_DATA (push, 0x3fffff);
708 BEGIN_NVC0(push, SUBC_3D(0x0fdc), 1);
709 PUSH_DATA (push, 1);
710 BEGIN_NVC0(push, SUBC_3D(0x19c0), 1);
711 PUSH_DATA (push, 1);
712
713 if (obj_class < GM107_3D_CLASS) {
714 BEGIN_NVC0(push, SUBC_3D(0x075c), 1);
715 PUSH_DATA (push, 3);
716
717 if (obj_class >= NVE4_3D_CLASS) {
718 BEGIN_NVC0(push, SUBC_3D(0x07fc), 1);
719 PUSH_DATA (push, 1);
720 }
721 }
722
723 /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
724 * are supposed to do */
725 }
726
727 static void
728 nvc0_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
729 {
730 struct nvc0_screen *screen = nvc0_screen(pscreen);
731 struct nouveau_pushbuf *push = screen->base.pushbuf;
732
733 /* we need to do it after possible flush in MARK_RING */
734 *sequence = ++screen->base.fence.sequence;
735
736 assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
737 PUSH_DATA (push, NVC0_FIFO_PKHDR_SQ(NVC0_3D(QUERY_ADDRESS_HIGH), 4));
738 PUSH_DATAh(push, screen->fence.bo->offset);
739 PUSH_DATA (push, screen->fence.bo->offset);
740 PUSH_DATA (push, *sequence);
741 PUSH_DATA (push, NVC0_3D_QUERY_GET_FENCE | NVC0_3D_QUERY_GET_SHORT |
742 (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT));
743 }
744
745 static u32
746 nvc0_screen_fence_update(struct pipe_screen *pscreen)
747 {
748 struct nvc0_screen *screen = nvc0_screen(pscreen);
749 return screen->fence.map[0];
750 }
751
752 static int
753 nvc0_screen_init_compute(struct nvc0_screen *screen)
754 {
755 screen->base.base.get_compute_param = nvc0_screen_get_compute_param;
756
757 switch (screen->base.device->chipset & ~0xf) {
758 case 0xc0:
759 case 0xd0:
760 return nvc0_screen_compute_setup(screen, screen->base.pushbuf);
761 case 0xe0:
762 case 0xf0:
763 case 0x100:
764 case 0x110:
765 case 0x120:
766 case 0x130:
767 return nve4_screen_compute_setup(screen, screen->base.pushbuf);
768 default:
769 return -1;
770 }
771 }
772
773 static int
774 nvc0_screen_resize_tls_area(struct nvc0_screen *screen,
775 uint32_t lpos, uint32_t lneg, uint32_t cstack)
776 {
777 struct nouveau_bo *bo = NULL;
778 int ret;
779 uint64_t size = (lpos + lneg) * 32 + cstack;
780
781 if (size >= (1 << 20)) {
782 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64"\n", size);
783 return -1;
784 }
785
786 size *= (screen->base.device->chipset >= 0xe0) ? 64 : 48; /* max warps */
787 size = align(size, 0x8000);
788 size *= screen->mp_count;
789
790 size = align(size, 1 << 17);
791
792 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base), 1 << 17, size,
793 NULL, &bo);
794 if (ret)
795 return ret;
796
797 /* Make sure that the pushbuf has acquired a reference to the old tls
798 * segment, as it may have commands that will reference it.
799 */
800 if (screen->tls)
801 PUSH_REFN(screen->base.pushbuf, screen->tls,
802 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RDWR);
803 nouveau_bo_ref(NULL, &screen->tls);
804 screen->tls = bo;
805 return 0;
806 }
807
808 int
809 nvc0_screen_resize_text_area(struct nvc0_screen *screen, uint64_t size)
810 {
811 struct nouveau_pushbuf *push = screen->base.pushbuf;
812 struct nouveau_bo *bo;
813 int ret;
814
815 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base),
816 1 << 17, size, NULL, &bo);
817 if (ret)
818 return ret;
819
820 /* Make sure that the pushbuf has acquired a reference to the old text
821 * segment, as it may have commands that will reference it.
822 */
823 if (screen->text)
824 PUSH_REFN(push, screen->text,
825 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RD);
826 nouveau_bo_ref(NULL, &screen->text);
827 screen->text = bo;
828
829 nouveau_heap_destroy(&screen->lib_code);
830 nouveau_heap_destroy(&screen->text_heap);
831
832 /* XXX: getting a page fault at the end of the code buffer every few
833 * launches, don't use the last 256 bytes to work around them - prefetch ?
834 */
835 nouveau_heap_init(&screen->text_heap, 0, size - 0x100);
836
837 /* update the code segment setup */
838 BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
839 PUSH_DATAh(push, screen->text->offset);
840 PUSH_DATA (push, screen->text->offset);
841 if (screen->compute) {
842 BEGIN_NVC0(push, NVC0_CP(CODE_ADDRESS_HIGH), 2);
843 PUSH_DATAh(push, screen->text->offset);
844 PUSH_DATA (push, screen->text->offset);
845 }
846
847 return 0;
848 }
849
850 void
851 nvc0_screen_bind_cb_3d(struct nvc0_screen *screen, bool *can_serialize,
852 int stage, int index, int size, uint64_t addr)
853 {
854 assert(stage != 5);
855
856 struct nouveau_pushbuf *push = screen->base.pushbuf;
857
858 if (screen->base.class_3d >= GM107_3D_CLASS) {
859 struct nvc0_cb_binding *binding = &screen->cb_bindings[stage][index];
860
861 // TODO: Better figure out the conditions in which this is needed
862 bool serialize = binding->addr == addr && binding->size != size;
863 if (can_serialize)
864 serialize = serialize && *can_serialize;
865 if (serialize) {
866 IMMED_NVC0(push, NVC0_3D(SERIALIZE), 0);
867 if (can_serialize)
868 *can_serialize = false;
869 }
870
871 binding->addr = addr;
872 binding->size = size;
873 }
874
875 if (size >= 0) {
876 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
877 PUSH_DATA (push, size);
878 PUSH_DATAh(push, addr);
879 PUSH_DATA (push, addr);
880 }
881 IMMED_NVC0(push, NVC0_3D(CB_BIND(stage)), (index << 4) | (size >= 0));
882 }
883
884 #define FAIL_SCREEN_INIT(str, err) \
885 do { \
886 NOUVEAU_ERR(str, err); \
887 goto fail; \
888 } while(0)
889
890 struct nouveau_screen *
891 nvc0_screen_create(struct nouveau_device *dev)
892 {
893 struct nvc0_screen *screen;
894 struct pipe_screen *pscreen;
895 struct nouveau_object *chan;
896 struct nouveau_pushbuf *push;
897 uint64_t value;
898 uint32_t obj_class;
899 uint32_t flags;
900 int ret;
901 unsigned i;
902
903 switch (dev->chipset & ~0xf) {
904 case 0xc0:
905 case 0xd0:
906 case 0xe0:
907 case 0xf0:
908 case 0x100:
909 case 0x110:
910 case 0x120:
911 case 0x130:
912 break;
913 default:
914 return NULL;
915 }
916
917 screen = CALLOC_STRUCT(nvc0_screen);
918 if (!screen)
919 return NULL;
920 pscreen = &screen->base.base;
921 pscreen->destroy = nvc0_screen_destroy;
922
923 ret = nouveau_screen_init(&screen->base, dev);
924 if (ret)
925 FAIL_SCREEN_INIT("Base screen init failed: %d\n", ret);
926 chan = screen->base.channel;
927 push = screen->base.pushbuf;
928 push->user_priv = screen;
929 push->rsvd_kick = 5;
930
931 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
932 PIPE_BIND_SHADER_BUFFER |
933 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
934 PIPE_BIND_COMMAND_ARGS_BUFFER | PIPE_BIND_QUERY_BUFFER;
935 screen->base.sysmem_bindings |=
936 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
937
938 if (screen->base.vram_domain & NOUVEAU_BO_GART) {
939 screen->base.sysmem_bindings |= screen->base.vidmem_bindings;
940 screen->base.vidmem_bindings = 0;
941 }
942
943 pscreen->context_create = nvc0_create;
944 pscreen->is_format_supported = nvc0_screen_is_format_supported;
945 pscreen->get_param = nvc0_screen_get_param;
946 pscreen->get_shader_param = nvc0_screen_get_shader_param;
947 pscreen->get_paramf = nvc0_screen_get_paramf;
948 pscreen->get_sample_pixel_grid = nvc0_screen_get_sample_pixel_grid;
949 pscreen->get_driver_query_info = nvc0_screen_get_driver_query_info;
950 pscreen->get_driver_query_group_info = nvc0_screen_get_driver_query_group_info;
951
952 nvc0_screen_init_resource_functions(pscreen);
953
954 screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
955 screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
956
957 flags = NOUVEAU_BO_GART | NOUVEAU_BO_MAP;
958 if (screen->base.drm->version >= 0x01000202)
959 flags |= NOUVEAU_BO_COHERENT;
960
961 ret = nouveau_bo_new(dev, flags, 0, 4096, NULL, &screen->fence.bo);
962 if (ret)
963 FAIL_SCREEN_INIT("Error allocating fence BO: %d\n", ret);
964 nouveau_bo_map(screen->fence.bo, 0, NULL);
965 screen->fence.map = screen->fence.bo->map;
966 screen->base.fence.emit = nvc0_screen_fence_emit;
967 screen->base.fence.update = nvc0_screen_fence_update;
968
969
970 ret = nouveau_object_new(chan, (dev->chipset < 0xe0) ? 0x1f906e : 0x906e,
971 NVIF_CLASS_SW_GF100, NULL, 0, &screen->nvsw);
972 if (ret)
973 FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret);
974
975 BEGIN_NVC0(push, SUBC_SW(NV01_SUBCHAN_OBJECT), 1);
976 PUSH_DATA (push, screen->nvsw->handle);
977
978 switch (dev->chipset & ~0xf) {
979 case 0x130:
980 case 0x120:
981 case 0x110:
982 case 0x100:
983 case 0xf0:
984 obj_class = NVF0_P2MF_CLASS;
985 break;
986 case 0xe0:
987 obj_class = NVE4_P2MF_CLASS;
988 break;
989 default:
990 obj_class = NVC0_M2MF_CLASS;
991 break;
992 }
993 ret = nouveau_object_new(chan, 0xbeef323f, obj_class, NULL, 0,
994 &screen->m2mf);
995 if (ret)
996 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
997
998 BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
999 PUSH_DATA (push, screen->m2mf->oclass);
1000 if (screen->m2mf->oclass == NVE4_P2MF_CLASS) {
1001 BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
1002 PUSH_DATA (push, 0xa0b5);
1003 }
1004
1005 ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
1006 &screen->eng2d);
1007 if (ret)
1008 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
1009
1010 BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
1011 PUSH_DATA (push, screen->eng2d->oclass);
1012 BEGIN_NVC0(push, SUBC_2D(NVC0_2D_SINGLE_GPC), 1);
1013 PUSH_DATA (push, 0);
1014 BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
1015 PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
1016 BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
1017 PUSH_DATA (push, 0);
1018 BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
1019 PUSH_DATA (push, 0);
1020 BEGIN_NVC0(push, SUBC_2D(0x0884), 1);
1021 PUSH_DATA (push, 0x3f);
1022 BEGIN_NVC0(push, SUBC_2D(0x0888), 1);
1023 PUSH_DATA (push, 1);
1024 BEGIN_NVC0(push, NVC0_2D(COND_MODE), 1);
1025 PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
1026
1027 BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
1028 PUSH_DATAh(push, screen->fence.bo->offset + 16);
1029 PUSH_DATA (push, screen->fence.bo->offset + 16);
1030
1031 switch (dev->chipset & ~0xf) {
1032 case 0x130:
1033 switch (dev->chipset) {
1034 case 0x130:
1035 case 0x13b:
1036 obj_class = GP100_3D_CLASS;
1037 break;
1038 default:
1039 obj_class = GP102_3D_CLASS;
1040 break;
1041 }
1042 break;
1043 case 0x120:
1044 obj_class = GM200_3D_CLASS;
1045 break;
1046 case 0x110:
1047 obj_class = GM107_3D_CLASS;
1048 break;
1049 case 0x100:
1050 case 0xf0:
1051 obj_class = NVF0_3D_CLASS;
1052 break;
1053 case 0xe0:
1054 switch (dev->chipset) {
1055 case 0xea:
1056 obj_class = NVEA_3D_CLASS;
1057 break;
1058 default:
1059 obj_class = NVE4_3D_CLASS;
1060 break;
1061 }
1062 break;
1063 case 0xd0:
1064 obj_class = NVC8_3D_CLASS;
1065 break;
1066 case 0xc0:
1067 default:
1068 switch (dev->chipset) {
1069 case 0xc8:
1070 obj_class = NVC8_3D_CLASS;
1071 break;
1072 case 0xc1:
1073 obj_class = NVC1_3D_CLASS;
1074 break;
1075 default:
1076 obj_class = NVC0_3D_CLASS;
1077 break;
1078 }
1079 break;
1080 }
1081 ret = nouveau_object_new(chan, 0xbeef003d, obj_class, NULL, 0,
1082 &screen->eng3d);
1083 if (ret)
1084 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
1085 screen->base.class_3d = obj_class;
1086
1087 BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
1088 PUSH_DATA (push, screen->eng3d->oclass);
1089
1090 BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
1091 PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
1092
1093 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
1094 /* kill shaders after about 1 second (at 100 MHz) */
1095 BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
1096 PUSH_DATA (push, 0x17);
1097 }
1098
1099 IMMED_NVC0(push, NVC0_3D(ZETA_COMP_ENABLE),
1100 screen->base.drm->version >= 0x01000101);
1101 BEGIN_NVC0(push, NVC0_3D(RT_COMP_ENABLE(0)), 8);
1102 for (i = 0; i < 8; ++i)
1103 PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
1104
1105 BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
1106 PUSH_DATA (push, 1);
1107
1108 BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
1109 PUSH_DATA (push, 0);
1110 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
1111 PUSH_DATA (push, 0);
1112 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
1113 PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
1114 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
1115 PUSH_DATA (push, 0);
1116 BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
1117 PUSH_DATA (push, 1);
1118 BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
1119 PUSH_DATA (push, 1);
1120 BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
1121 PUSH_DATA (push, 1);
1122 BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
1123 PUSH_DATA (push, 0);
1124 BEGIN_NVC0(push, NVC0_3D(SHADE_MODEL), 1);
1125 PUSH_DATA (push, NVC0_3D_SHADE_MODEL_SMOOTH);
1126 if (screen->eng3d->oclass < NVE4_3D_CLASS) {
1127 IMMED_NVC0(push, NVC0_3D(TEX_MISC), 0);
1128 } else {
1129 BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
1130 PUSH_DATA (push, 15);
1131 }
1132 BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
1133 PUSH_DATA (push, 8); /* 128 */
1134 BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
1135 PUSH_DATA (push, 1);
1136 if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
1137 BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
1138 PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
1139 }
1140
1141 nvc0_magic_3d_init(push, screen->eng3d->oclass);
1142
1143 ret = nvc0_screen_resize_text_area(screen, 1 << 19);
1144 if (ret)
1145 FAIL_SCREEN_INIT("Error allocating TEXT area: %d\n", ret);
1146
1147 /* 6 user uniform areas, 6 driver areas, and 1 for the runout */
1148 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 12, 13 << 16, NULL,
1149 &screen->uniform_bo);
1150 if (ret)
1151 FAIL_SCREEN_INIT("Error allocating uniform BO: %d\n", ret);
1152
1153 PUSH_REFN (push, screen->uniform_bo, NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_WR);
1154
1155 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
1156 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1157 PUSH_DATA (push, 256);
1158 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1159 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1160 BEGIN_1IC0(push, NVC0_3D(CB_POS), 5);
1161 PUSH_DATA (push, 0);
1162 PUSH_DATAf(push, 0.0f);
1163 PUSH_DATAf(push, 0.0f);
1164 PUSH_DATAf(push, 0.0f);
1165 PUSH_DATAf(push, 0.0f);
1166 BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
1167 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1168 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1169
1170 if (screen->base.drm->version >= 0x01000101) {
1171 ret = nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
1172 if (ret)
1173 FAIL_SCREEN_INIT("NOUVEAU_GETPARAM_GRAPH_UNITS failed: %d\n", ret);
1174 } else {
1175 if (dev->chipset >= 0xe0 && dev->chipset < 0xf0)
1176 value = (8 << 8) | 4;
1177 else
1178 value = (16 << 8) | 4;
1179 }
1180 screen->gpc_count = value & 0x000000ff;
1181 screen->mp_count = value >> 8;
1182 screen->mp_count_compute = screen->mp_count;
1183
1184 ret = nvc0_screen_resize_tls_area(screen, 128 * 16, 0, 0x200);
1185 if (ret)
1186 FAIL_SCREEN_INIT("Error allocating TLS area: %d\n", ret);
1187
1188 BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
1189 PUSH_DATAh(push, screen->tls->offset);
1190 PUSH_DATA (push, screen->tls->offset);
1191 PUSH_DATA (push, screen->tls->size >> 32);
1192 PUSH_DATA (push, screen->tls->size);
1193 BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
1194 PUSH_DATA (push, 0);
1195 /* Reduce likelihood of collision with real buffers by placing the hole at
1196 * the top of the 4G area. This will have to be dealt with for real
1197 * eventually by blocking off that area from the VM.
1198 */
1199 BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
1200 PUSH_DATA (push, 0xff << 24);
1201
1202 if (screen->eng3d->oclass < GM107_3D_CLASS) {
1203 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 20, NULL,
1204 &screen->poly_cache);
1205 if (ret)
1206 FAIL_SCREEN_INIT("Error allocating poly cache BO: %d\n", ret);
1207
1208 BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
1209 PUSH_DATAh(push, screen->poly_cache->offset);
1210 PUSH_DATA (push, screen->poly_cache->offset);
1211 PUSH_DATA (push, 3);
1212 }
1213
1214 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 17, NULL,
1215 &screen->txc);
1216 if (ret)
1217 FAIL_SCREEN_INIT("Error allocating txc BO: %d\n", ret);
1218
1219 BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
1220 PUSH_DATAh(push, screen->txc->offset);
1221 PUSH_DATA (push, screen->txc->offset);
1222 PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
1223 if (screen->eng3d->oclass >= GM107_3D_CLASS) {
1224 screen->tic.maxwell = true;
1225 if (screen->eng3d->oclass == GM107_3D_CLASS) {
1226 screen->tic.maxwell =
1227 debug_get_bool_option("NOUVEAU_MAXWELL_TIC", true);
1228 IMMED_NVC0(push, SUBC_3D(0x0f10), screen->tic.maxwell);
1229 }
1230 }
1231
1232 BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
1233 PUSH_DATAh(push, screen->txc->offset + 65536);
1234 PUSH_DATA (push, screen->txc->offset + 65536);
1235 PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
1236
1237 BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
1238 PUSH_DATA (push, 0);
1239 BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
1240 PUSH_DATA (push, 0);
1241 PUSH_DATA (push, 0);
1242 BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
1243 PUSH_DATA (push, 0x3f);
1244
1245 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
1246 PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
1247 BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
1248 for (i = 0; i < 8 * 2; ++i)
1249 PUSH_DATA(push, 0);
1250 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
1251 PUSH_DATA (push, 0);
1252 BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
1253 PUSH_DATA (push, 0);
1254
1255 /* neither scissors, viewport nor stencil mask should affect clears */
1256 BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
1257 PUSH_DATA (push, 0);
1258
1259 BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
1260 PUSH_DATA (push, 1);
1261 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1262 BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(i)), 2);
1263 PUSH_DATAf(push, 0.0f);
1264 PUSH_DATAf(push, 1.0f);
1265 }
1266 BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
1267 PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
1268
1269 /* We use scissors instead of exact view volume clipping,
1270 * so they're always enabled.
1271 */
1272 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1273 BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(i)), 3);
1274 PUSH_DATA (push, 1);
1275 PUSH_DATA (push, 8192 << 16);
1276 PUSH_DATA (push, 8192 << 16);
1277 }
1278
1279 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
1280
1281 i = 0;
1282 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, mme9097_per_instance_bf);
1283 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, mme9097_blend_enables);
1284 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, mme9097_vertex_array_select);
1285 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, mme9097_tep_select);
1286 MK_MACRO(NVC0_3D_MACRO_GP_SELECT, mme9097_gp_select);
1287 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, mme9097_poly_mode_front);
1288 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, mme9097_poly_mode_back);
1289 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT, mme9097_draw_arrays_indirect);
1290 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT, mme9097_draw_elts_indirect);
1291 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT, mme9097_draw_arrays_indirect_count);
1292 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT, mme9097_draw_elts_indirect_count);
1293 MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE, mme9097_query_buffer_write);
1294 MK_MACRO(NVC0_3D_MACRO_CONSERVATIVE_RASTER_STATE, mme9097_conservative_raster_state);
1295 MK_MACRO(NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT, mme90c0_launch_grid_indirect);
1296
1297 BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
1298 PUSH_DATA (push, 1);
1299 BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
1300 PUSH_DATA (push, 1);
1301 BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
1302 PUSH_DATA (push, 0x40);
1303 BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
1304 PUSH_DATA (push, 0);
1305 BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
1306 PUSH_DATA (push, 0x30);
1307 BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
1308 PUSH_DATA (push, 3);
1309 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
1310 PUSH_DATA (push, 0x20);
1311 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
1312 PUSH_DATA (push, 0x00);
1313 screen->save_state.patch_vertices = 3;
1314
1315 BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
1316 PUSH_DATA (push, 0);
1317 BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
1318 PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
1319
1320 IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
1321
1322 if (nvc0_screen_init_compute(screen))
1323 goto fail;
1324
1325 /* XXX: Compute and 3D are somehow aliased on Fermi. */
1326 for (i = 0; i < 5; ++i) {
1327 unsigned j = 0;
1328 for (j = 0; j < 16; j++)
1329 screen->cb_bindings[i][j].size = -1;
1330
1331 /* TIC and TSC entries for each unit (nve4+ only) */
1332 /* auxiliary constants (6 user clip planes, base instance id) */
1333 nvc0_screen_bind_cb_3d(screen, NULL, i, 15, NVC0_CB_AUX_SIZE,
1334 screen->uniform_bo->offset + NVC0_CB_AUX_INFO(i));
1335 if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
1336 unsigned j;
1337 BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
1338 PUSH_DATA (push, NVC0_CB_AUX_UNK_INFO);
1339 for (j = 0; j < 8; ++j)
1340 PUSH_DATA(push, j);
1341 } else {
1342 BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
1343 PUSH_DATA (push, 0x54);
1344 }
1345
1346 /* MS sample coordinate offsets: these do not work with _ALT modes ! */
1347 BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 2 * 8);
1348 PUSH_DATA (push, NVC0_CB_AUX_MS_INFO);
1349 PUSH_DATA (push, 0); /* 0 */
1350 PUSH_DATA (push, 0);
1351 PUSH_DATA (push, 1); /* 1 */
1352 PUSH_DATA (push, 0);
1353 PUSH_DATA (push, 0); /* 2 */
1354 PUSH_DATA (push, 1);
1355 PUSH_DATA (push, 1); /* 3 */
1356 PUSH_DATA (push, 1);
1357 PUSH_DATA (push, 2); /* 4 */
1358 PUSH_DATA (push, 0);
1359 PUSH_DATA (push, 3); /* 5 */
1360 PUSH_DATA (push, 0);
1361 PUSH_DATA (push, 2); /* 6 */
1362 PUSH_DATA (push, 1);
1363 PUSH_DATA (push, 3); /* 7 */
1364 PUSH_DATA (push, 1);
1365 }
1366 BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
1367 PUSH_DATA (push, 0);
1368
1369 PUSH_KICK (push);
1370
1371 screen->tic.entries = CALLOC(
1372 NVC0_TIC_MAX_ENTRIES + NVC0_TSC_MAX_ENTRIES + NVE4_IMG_MAX_HANDLES,
1373 sizeof(void *));
1374 screen->tsc.entries = screen->tic.entries + NVC0_TIC_MAX_ENTRIES;
1375 screen->img.entries = (void *)(screen->tsc.entries + NVC0_TSC_MAX_ENTRIES);
1376
1377 if (!nvc0_blitter_create(screen))
1378 goto fail;
1379
1380 screen->default_tsc = CALLOC_STRUCT(nv50_tsc_entry);
1381 screen->default_tsc->tsc[0] = G80_TSC_0_SRGB_CONVERSION;
1382
1383 nouveau_fence_new(&screen->base, &screen->base.fence.current);
1384
1385 return &screen->base;
1386
1387 fail:
1388 screen->base.base.context_create = NULL;
1389 return &screen->base;
1390 }
1391
1392 int
1393 nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
1394 {
1395 int i = screen->tic.next;
1396
1397 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1398 i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1399
1400 screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1401
1402 if (screen->tic.entries[i])
1403 nv50_tic_entry(screen->tic.entries[i])->id = -1;
1404
1405 screen->tic.entries[i] = entry;
1406 return i;
1407 }
1408
1409 int
1410 nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
1411 {
1412 int i = screen->tsc.next;
1413
1414 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1415 i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1416
1417 screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1418
1419 if (screen->tsc.entries[i])
1420 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1421
1422 screen->tsc.entries[i] = entry;
1423 return i;
1424 }