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