2 * Copyright 2010 Christoph Bumiller
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
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"
31 #include "nouveau_vp3_video.h"
33 #include "nvc0/nvc0_context.h"
34 #include "nvc0/nvc0_screen.h"
36 #include "nvc0/mme/com9097.mme.h"
37 #include "nvc0/mme/com90c0.mme.h"
39 #include "nv50/g80_texture.xml.h"
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
,
49 const struct util_format_description
*desc
= util_format_description(format
);
53 if (!(0x117 & (1 << sample_count
))) /* 0, 1, 2, 4 or 8 */
56 if (MAX2(1, sample_count
) != MAX2(1, storage_sample_count
))
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.
62 if (format
== PIPE_FORMAT_NONE
&& bindings
& PIPE_BIND_RENDER_TARGET
)
65 if ((bindings
& PIPE_BIND_SAMPLER_VIEW
) && (target
!= PIPE_BUFFER
))
66 if (util_format_get_blocksizebits(format
) == 3 * 32)
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
) ||
77 /* Restrict ETC2 and ASTC formats here. These are only supported on GK20A.
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 &&
85 nouveau_screen(pscreen
)->class_3d
!= NVEA_3D_CLASS
)
88 /* shared is always supported */
89 bindings
&= ~(PIPE_BIND_LINEAR
|
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. */
101 return (( nvc0_format_table
[format
].usage
|
102 nvc0_vertex_format
[format
].usage
) & bindings
) == bindings
;
106 nvc0_screen_get_param(struct pipe_screen
*pscreen
, enum pipe_cap param
)
108 const uint16_t class_3d
= nouveau_screen(pscreen
)->class_3d
;
109 struct nouveau_device
*dev
= nouveau_screen(pscreen
)->device
;
112 /* non-boolean caps */
113 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS
:
114 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS
:
116 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS
:
118 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS
:
120 case PIPE_CAP_MIN_TEXEL_OFFSET
:
122 case PIPE_CAP_MAX_TEXEL_OFFSET
:
124 case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET
:
126 case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET
:
128 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE
:
129 return 128 * 1024 * 1024;
130 case PIPE_CAP_GLSL_FEATURE_LEVEL
:
132 case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY
:
134 case PIPE_CAP_MAX_RENDER_TARGETS
:
136 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS
:
138 case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS
:
139 case PIPE_CAP_RASTERIZER_SUBPIXEL_BITS
:
141 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS
:
143 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS
:
144 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS
:
146 case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES
:
147 case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS
:
149 case PIPE_CAP_MAX_VERTEX_STREAMS
:
151 case PIPE_CAP_MAX_GS_INVOCATIONS
:
153 case PIPE_CAP_MAX_SHADER_BUFFER_SIZE
:
155 case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE
:
157 case PIPE_CAP_MAX_VERTEX_ELEMENT_SRC_OFFSET
:
159 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT
:
161 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT
:
162 if (class_3d
< GM107_3D_CLASS
)
163 return 256; /* IMAGE bindings require alignment to 256 */
165 case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT
:
167 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT
:
168 return NOUVEAU_MIN_BUFFER_MAP_ALIGN
;
169 case PIPE_CAP_MAX_VIEWPORTS
:
170 return NVC0_MAX_VIEWPORTS
;
171 case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS
:
173 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK
:
174 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50
;
175 case PIPE_CAP_ENDIANNESS
:
176 return PIPE_ENDIAN_LITTLE
;
177 case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS
:
179 case PIPE_CAP_MAX_WINDOW_RECTANGLES
:
180 return NVC0_MAX_WINDOW_RECTANGLES
;
181 case PIPE_CAP_MAX_CONSERVATIVE_RASTER_SUBPIXEL_PRECISION_BIAS
:
182 return class_3d
>= GM200_3D_CLASS
? 8 : 0;
183 case PIPE_CAP_MAX_TEXTURE_UPLOAD_MEMORY_BUDGET
:
184 return 64 * 1024 * 1024;
187 case PIPE_CAP_TEXTURE_MIRROR_CLAMP
:
188 case PIPE_CAP_TEXTURE_MIRROR_CLAMP_TO_EDGE
:
189 case PIPE_CAP_TEXTURE_SWIZZLE
:
190 case PIPE_CAP_NPOT_TEXTURES
:
191 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES
:
192 case PIPE_CAP_MIXED_COLOR_DEPTH_BITS
:
193 case PIPE_CAP_ANISOTROPIC_FILTER
:
194 case PIPE_CAP_SEAMLESS_CUBE_MAP
:
195 case PIPE_CAP_CUBE_MAP_ARRAY
:
196 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS
:
197 case PIPE_CAP_TEXTURE_MULTISAMPLE
:
198 case PIPE_CAP_DEPTH_CLIP_DISABLE
:
199 case PIPE_CAP_POINT_SPRITE
:
200 case PIPE_CAP_TGSI_TEXCOORD
:
202 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED
:
203 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED
:
204 case PIPE_CAP_VERTEX_COLOR_CLAMPED
:
205 case PIPE_CAP_QUERY_TIMESTAMP
:
206 case PIPE_CAP_QUERY_TIME_ELAPSED
:
207 case PIPE_CAP_OCCLUSION_QUERY
:
208 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME
:
209 case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS
:
210 case PIPE_CAP_QUERY_PIPELINE_STATISTICS
:
211 case PIPE_CAP_BLEND_EQUATION_SEPARATE
:
212 case PIPE_CAP_INDEP_BLEND_ENABLE
:
213 case PIPE_CAP_INDEP_BLEND_FUNC
:
214 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT
:
215 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER
:
216 case PIPE_CAP_PRIMITIVE_RESTART
:
217 case PIPE_CAP_TGSI_INSTANCEID
:
218 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR
:
219 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS
:
220 case PIPE_CAP_CONDITIONAL_RENDER
:
221 case PIPE_CAP_TEXTURE_BARRIER
:
222 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION
:
223 case PIPE_CAP_START_INSTANCE
:
224 case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT
:
225 case PIPE_CAP_DRAW_INDIRECT
:
226 case PIPE_CAP_USER_VERTEX_BUFFERS
:
227 case PIPE_CAP_TEXTURE_QUERY_LOD
:
228 case PIPE_CAP_SAMPLE_SHADING
:
229 case PIPE_CAP_TEXTURE_GATHER_OFFSETS
:
230 case PIPE_CAP_TEXTURE_GATHER_SM5
:
231 case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE
:
232 case PIPE_CAP_CONDITIONAL_RENDER_INVERTED
:
233 case PIPE_CAP_SAMPLER_VIEW_TARGET
:
234 case PIPE_CAP_CLIP_HALFZ
:
235 case PIPE_CAP_POLYGON_OFFSET_CLAMP
:
236 case PIPE_CAP_MULTISAMPLE_Z_RESOLVE
:
237 case PIPE_CAP_TEXTURE_FLOAT_LINEAR
:
238 case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR
:
239 case PIPE_CAP_DEPTH_BOUNDS_TEST
:
240 case PIPE_CAP_TGSI_TXQS
:
241 case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS
:
242 case PIPE_CAP_FORCE_PERSAMPLE_INTERP
:
243 case PIPE_CAP_SHAREABLE_SHADERS
:
244 case PIPE_CAP_CLEAR_TEXTURE
:
245 case PIPE_CAP_DRAW_PARAMETERS
:
246 case PIPE_CAP_TGSI_PACK_HALF_FLOAT
:
247 case PIPE_CAP_MULTI_DRAW_INDIRECT
:
248 case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS
:
249 case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL
:
250 case PIPE_CAP_QUERY_BUFFER_OBJECT
:
251 case PIPE_CAP_INVALIDATE_BUFFER
:
252 case PIPE_CAP_STRING_MARKER
:
253 case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT
:
254 case PIPE_CAP_CULL_DISTANCE
:
255 case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES
:
256 case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR
:
257 case PIPE_CAP_TGSI_VOTE
:
258 case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED
:
259 case PIPE_CAP_TGSI_ARRAY_COMPONENTS
:
260 case PIPE_CAP_TGSI_MUL_ZERO_WINS
:
261 case PIPE_CAP_DOUBLES
:
263 case PIPE_CAP_TGSI_TEX_TXF_LZ
:
264 case PIPE_CAP_TGSI_CLOCK
:
265 case PIPE_CAP_COMPUTE
:
266 case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX
:
267 case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION
:
268 case PIPE_CAP_QUERY_SO_OVERFLOW
:
269 case PIPE_CAP_DEST_SURFACE_SRGB_CONTROL
:
271 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER
:
272 return nouveau_screen(pscreen
)->vram_domain
& NOUVEAU_BO_VRAM
? 1 : 0;
273 case PIPE_CAP_TGSI_FS_FBFETCH
:
274 return class_3d
>= NVE4_3D_CLASS
; /* needs testing on fermi */
275 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE
:
276 case PIPE_CAP_TGSI_BALLOT
:
277 case PIPE_CAP_BINDLESS_TEXTURE
:
278 return class_3d
>= NVE4_3D_CLASS
;
279 case PIPE_CAP_TGSI_ATOMFADD
:
280 return class_3d
< GM107_3D_CLASS
; /* needs additional lowering */
281 case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE
:
282 case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT
:
283 case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT
:
284 case PIPE_CAP_POST_DEPTH_COVERAGE
:
285 case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_TRIANGLES
:
286 case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_POINTS_LINES
:
287 case PIPE_CAP_CONSERVATIVE_RASTER_POST_DEPTH_COVERAGE
:
288 case PIPE_CAP_PROGRAMMABLE_SAMPLE_LOCATIONS
:
289 return class_3d
>= GM200_3D_CLASS
;
290 case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES
:
291 return class_3d
>= GP100_3D_CLASS
;
293 /* unsupported caps */
294 case PIPE_CAP_DEPTH_CLIP_DISABLE_SEPARATE
:
295 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT
:
296 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER
:
297 case PIPE_CAP_SHADER_STENCIL_EXPORT
:
298 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS
:
299 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY
:
300 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY
:
301 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY
:
302 case PIPE_CAP_FAKE_SW_MSAA
:
303 case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION
:
304 case PIPE_CAP_VERTEXID_NOBASE
:
305 case PIPE_CAP_RESOURCE_FROM_USER_MEMORY
:
306 case PIPE_CAP_DEVICE_RESET_STATUS_QUERY
:
307 case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL
:
308 case PIPE_CAP_GENERATE_MIPMAP
:
309 case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY
:
310 case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS
:
311 case PIPE_CAP_QUERY_MEMORY_INFO
:
312 case PIPE_CAP_PCI_GROUP
:
313 case PIPE_CAP_PCI_BUS
:
314 case PIPE_CAP_PCI_DEVICE
:
315 case PIPE_CAP_PCI_FUNCTION
:
316 case PIPE_CAP_TGSI_CAN_READ_OUTPUTS
:
317 case PIPE_CAP_NATIVE_FENCE_FD
:
318 case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY
:
319 case PIPE_CAP_INT64_DIVMOD
:
320 case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE
:
321 case PIPE_CAP_NIR_SAMPLERS_AS_DEREF
:
322 case PIPE_CAP_MEMOBJ
:
323 case PIPE_CAP_LOAD_CONSTBUF
:
324 case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS
:
325 case PIPE_CAP_TILE_RASTER_ORDER
:
326 case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES
:
327 case PIPE_CAP_FRAMEBUFFER_MSAA_CONSTRAINTS
:
328 case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET
:
329 case PIPE_CAP_CONTEXT_PRIORITY_MASK
:
330 case PIPE_CAP_FENCE_SIGNAL
:
331 case PIPE_CAP_CONSTBUF0_FLAGS
:
332 case PIPE_CAP_PACKED_UNIFORMS
:
333 case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_POINTS_LINES
:
334 case PIPE_CAP_MAX_COMBINED_SHADER_BUFFERS
:
335 case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTERS
:
336 case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTER_BUFFERS
:
337 case PIPE_CAP_SURFACE_SAMPLE_COUNT
:
338 case PIPE_CAP_QUERY_PIPELINE_STATISTICS_SINGLE
:
339 case PIPE_CAP_RGB_OVERRIDE_DST_ALPHA_BLEND
:
340 case PIPE_CAP_GLSL_TESS_LEVELS_AS_INPUTS
:
343 case PIPE_CAP_VENDOR_ID
:
345 case PIPE_CAP_DEVICE_ID
: {
347 if (nouveau_getparam(dev
, NOUVEAU_GETPARAM_PCI_DEVICE
, &device_id
)) {
348 NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
353 case PIPE_CAP_ACCELERATED
:
355 case PIPE_CAP_VIDEO_MEMORY
:
356 return dev
->vram_size
>> 20;
360 debug_printf("%s: unhandled cap %d\n", __func__
, param
);
361 return u_pipe_screen_get_param_defaults(pscreen
, param
);
366 nvc0_screen_get_shader_param(struct pipe_screen
*pscreen
,
367 enum pipe_shader_type shader
,
368 enum pipe_shader_cap param
)
370 const uint16_t class_3d
= nouveau_screen(pscreen
)->class_3d
;
373 case PIPE_SHADER_VERTEX
:
374 case PIPE_SHADER_GEOMETRY
:
375 case PIPE_SHADER_FRAGMENT
:
376 case PIPE_SHADER_COMPUTE
:
377 case PIPE_SHADER_TESS_CTRL
:
378 case PIPE_SHADER_TESS_EVAL
:
385 case PIPE_SHADER_CAP_PREFERRED_IR
:
386 return PIPE_SHADER_IR_TGSI
;
387 case PIPE_SHADER_CAP_SUPPORTED_IRS
:
388 return 1 << PIPE_SHADER_IR_TGSI
;
389 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS
:
390 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS
:
391 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS
:
392 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS
:
394 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH
:
396 case PIPE_SHADER_CAP_MAX_INPUTS
:
397 if (shader
== PIPE_SHADER_VERTEX
)
399 /* NOTE: These only count our slots for GENERIC varyings.
400 * The address space may be larger, but the actual hard limit seems to be
401 * less than what the address space layout permits, so don't add TEXCOORD,
404 if (shader
== PIPE_SHADER_FRAGMENT
)
406 /* Actually this counts CLIPVERTEX, which occupies the last generic slot,
407 * and excludes 0x60 per-patch inputs.
410 case PIPE_SHADER_CAP_MAX_OUTPUTS
:
412 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE
:
413 return NVC0_MAX_CONSTBUF_SIZE
;
414 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS
:
415 return NVC0_MAX_PIPE_CONSTBUFS
;
416 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR
:
417 return shader
!= PIPE_SHADER_FRAGMENT
;
418 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR
:
419 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR
:
420 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR
:
422 case PIPE_SHADER_CAP_MAX_TEMPS
:
423 return NVC0_CAP_MAX_PROGRAM_TEMPS
;
424 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED
:
426 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED
:
428 case PIPE_SHADER_CAP_SUBROUTINES
:
430 case PIPE_SHADER_CAP_INTEGERS
:
432 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED
:
434 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED
:
436 case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS
:
438 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED
:
439 case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED
:
440 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE
:
441 case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD
:
442 case PIPE_SHADER_CAP_INT64_ATOMICS
:
443 case PIPE_SHADER_CAP_FP16
:
444 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS
:
445 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS
:
447 case PIPE_SHADER_CAP_SCALAR_ISA
:
449 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS
:
450 return NVC0_MAX_BUFFERS
;
451 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS
:
452 return (class_3d
>= NVE4_3D_CLASS
) ? 32 : 16;
453 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS
:
454 return (class_3d
>= NVE4_3D_CLASS
) ? 32 : 16;
455 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT
:
457 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES
:
458 if (class_3d
>= NVE4_3D_CLASS
)
459 return NVC0_MAX_IMAGES
;
460 if (shader
== PIPE_SHADER_FRAGMENT
|| shader
== PIPE_SHADER_COMPUTE
)
461 return NVC0_MAX_IMAGES
;
464 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param
);
470 nvc0_screen_get_paramf(struct pipe_screen
*pscreen
, enum pipe_capf param
)
472 const uint16_t class_3d
= nouveau_screen(pscreen
)->class_3d
;
475 case PIPE_CAPF_MAX_LINE_WIDTH
:
476 case PIPE_CAPF_MAX_LINE_WIDTH_AA
:
478 case PIPE_CAPF_MAX_POINT_WIDTH
:
480 case PIPE_CAPF_MAX_POINT_WIDTH_AA
:
482 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY
:
484 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS
:
486 case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE
:
488 case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE
:
489 return class_3d
>= GM200_3D_CLASS
? 0.75f
: 0.0f
;
490 case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY
:
491 return class_3d
>= GM200_3D_CLASS
? 0.25f
: 0.0f
;
494 NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param
);
499 nvc0_screen_get_compute_param(struct pipe_screen
*pscreen
,
500 enum pipe_shader_ir ir_type
,
501 enum pipe_compute_cap param
, void *data
)
503 struct nvc0_screen
*screen
= nvc0_screen(pscreen
);
504 const uint16_t obj_class
= screen
->compute
->oclass
;
506 #define RET(x) do { \
508 memcpy(data, x, sizeof(x)); \
513 case PIPE_COMPUTE_CAP_GRID_DIMENSION
:
514 RET((uint64_t []) { 3 });
515 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE
:
516 if (obj_class
>= NVE4_COMPUTE_CLASS
) {
517 RET(((uint64_t []) { 0x7fffffff, 65535, 65535 }));
519 RET(((uint64_t []) { 65535, 65535, 65535 }));
521 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE
:
522 RET(((uint64_t []) { 1024, 1024, 64 }));
523 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK
:
524 RET((uint64_t []) { 1024 });
525 case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK
:
526 if (obj_class
>= NVE4_COMPUTE_CLASS
) {
527 RET((uint64_t []) { 1024 });
529 RET((uint64_t []) { 512 });
531 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE
: /* g[] */
532 RET((uint64_t []) { 1ULL << 40 });
533 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE
: /* s[] */
535 case GM200_COMPUTE_CLASS
:
536 RET((uint64_t []) { 96 << 10 });
538 case GM107_COMPUTE_CLASS
:
539 RET((uint64_t []) { 64 << 10 });
542 RET((uint64_t []) { 48 << 10 });
545 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE
: /* l[] */
546 RET((uint64_t []) { 512 << 10 });
547 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE
: /* c[], arbitrary limit */
548 RET((uint64_t []) { 4096 });
549 case PIPE_COMPUTE_CAP_SUBGROUP_SIZE
:
550 RET((uint32_t []) { 32 });
551 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE
:
552 RET((uint64_t []) { 1ULL << 40 });
553 case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED
:
554 RET((uint32_t []) { 0 });
555 case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS
:
556 RET((uint32_t []) { screen
->mp_count_compute
});
557 case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY
:
558 RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
559 case PIPE_COMPUTE_CAP_ADDRESS_BITS
:
560 RET((uint32_t []) { 64 });
569 nvc0_screen_get_sample_pixel_grid(struct pipe_screen
*pscreen
,
570 unsigned sample_count
,
571 unsigned *width
, unsigned *height
)
573 switch (sample_count
) {
576 /* this could be 4x4, but the GL state tracker makes it difficult to
577 * create a 1x MSAA texture and smaller grids save CB space */
599 nvc0_screen_destroy(struct pipe_screen
*pscreen
)
601 struct nvc0_screen
*screen
= nvc0_screen(pscreen
);
603 if (!nouveau_drm_screen_unref(&screen
->base
))
606 if (screen
->base
.fence
.current
) {
607 struct nouveau_fence
*current
= NULL
;
609 /* nouveau_fence_wait will create a new current fence, so wait on the
610 * _current_ one, and remove both.
612 nouveau_fence_ref(screen
->base
.fence
.current
, ¤t
);
613 nouveau_fence_wait(current
, NULL
);
614 nouveau_fence_ref(NULL
, ¤t
);
615 nouveau_fence_ref(NULL
, &screen
->base
.fence
.current
);
617 if (screen
->base
.pushbuf
)
618 screen
->base
.pushbuf
->user_priv
= NULL
;
621 nvc0_blitter_destroy(screen
);
622 if (screen
->pm
.prog
) {
623 screen
->pm
.prog
->code
= NULL
; /* hardcoded, don't FREE */
624 nvc0_program_destroy(NULL
, screen
->pm
.prog
);
625 FREE(screen
->pm
.prog
);
628 nouveau_bo_ref(NULL
, &screen
->text
);
629 nouveau_bo_ref(NULL
, &screen
->uniform_bo
);
630 nouveau_bo_ref(NULL
, &screen
->tls
);
631 nouveau_bo_ref(NULL
, &screen
->txc
);
632 nouveau_bo_ref(NULL
, &screen
->fence
.bo
);
633 nouveau_bo_ref(NULL
, &screen
->poly_cache
);
635 nouveau_heap_destroy(&screen
->lib_code
);
636 nouveau_heap_destroy(&screen
->text_heap
);
638 FREE(screen
->tic
.entries
);
640 nouveau_object_del(&screen
->eng3d
);
641 nouveau_object_del(&screen
->eng2d
);
642 nouveau_object_del(&screen
->m2mf
);
643 nouveau_object_del(&screen
->compute
);
644 nouveau_object_del(&screen
->nvsw
);
646 nouveau_screen_fini(&screen
->base
);
652 nvc0_graph_set_macro(struct nvc0_screen
*screen
, uint32_t m
, unsigned pos
,
653 unsigned size
, const uint32_t *data
)
655 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
659 assert((pos
+ size
) <= 0x800);
661 BEGIN_NVC0(push
, SUBC_3D(NVC0_GRAPH_MACRO_ID
), 2);
662 PUSH_DATA (push
, (m
- 0x3800) / 8);
663 PUSH_DATA (push
, pos
);
664 BEGIN_1IC0(push
, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS
), size
+ 1);
665 PUSH_DATA (push
, pos
);
666 PUSH_DATAp(push
, data
, size
);
672 nvc0_magic_3d_init(struct nouveau_pushbuf
*push
, uint16_t obj_class
)
674 BEGIN_NVC0(push
, SUBC_3D(0x10cc), 1);
675 PUSH_DATA (push
, 0xff);
676 BEGIN_NVC0(push
, SUBC_3D(0x10e0), 2);
677 PUSH_DATA (push
, 0xff);
678 PUSH_DATA (push
, 0xff);
679 BEGIN_NVC0(push
, SUBC_3D(0x10ec), 2);
680 PUSH_DATA (push
, 0xff);
681 PUSH_DATA (push
, 0xff);
682 BEGIN_NVC0(push
, SUBC_3D(0x074c), 1);
683 PUSH_DATA (push
, 0x3f);
685 BEGIN_NVC0(push
, SUBC_3D(0x16a8), 1);
686 PUSH_DATA (push
, (3 << 16) | 3);
687 BEGIN_NVC0(push
, SUBC_3D(0x1794), 1);
688 PUSH_DATA (push
, (2 << 16) | 2);
690 if (obj_class
< GM107_3D_CLASS
) {
691 BEGIN_NVC0(push
, SUBC_3D(0x12ac), 1);
694 BEGIN_NVC0(push
, SUBC_3D(0x0218), 1);
695 PUSH_DATA (push
, 0x10);
696 BEGIN_NVC0(push
, SUBC_3D(0x10fc), 1);
697 PUSH_DATA (push
, 0x10);
698 BEGIN_NVC0(push
, SUBC_3D(0x1290), 1);
699 PUSH_DATA (push
, 0x10);
700 BEGIN_NVC0(push
, SUBC_3D(0x12d8), 2);
701 PUSH_DATA (push
, 0x10);
702 PUSH_DATA (push
, 0x10);
703 BEGIN_NVC0(push
, SUBC_3D(0x1140), 1);
704 PUSH_DATA (push
, 0x10);
705 BEGIN_NVC0(push
, SUBC_3D(0x1610), 1);
706 PUSH_DATA (push
, 0xe);
708 BEGIN_NVC0(push
, NVC0_3D(VERTEX_ID_GEN_MODE
), 1);
709 PUSH_DATA (push
, NVC0_3D_VERTEX_ID_GEN_MODE_DRAW_ARRAYS_ADD_START
);
710 BEGIN_NVC0(push
, SUBC_3D(0x030c), 1);
712 BEGIN_NVC0(push
, SUBC_3D(0x0300), 1);
715 BEGIN_NVC0(push
, SUBC_3D(0x02d0), 1);
716 PUSH_DATA (push
, 0x3fffff);
717 BEGIN_NVC0(push
, SUBC_3D(0x0fdc), 1);
719 BEGIN_NVC0(push
, SUBC_3D(0x19c0), 1);
722 if (obj_class
< GM107_3D_CLASS
) {
723 BEGIN_NVC0(push
, SUBC_3D(0x075c), 1);
726 if (obj_class
>= NVE4_3D_CLASS
) {
727 BEGIN_NVC0(push
, SUBC_3D(0x07fc), 1);
732 /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
733 * are supposed to do */
737 nvc0_screen_fence_emit(struct pipe_screen
*pscreen
, u32
*sequence
)
739 struct nvc0_screen
*screen
= nvc0_screen(pscreen
);
740 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
742 /* we need to do it after possible flush in MARK_RING */
743 *sequence
= ++screen
->base
.fence
.sequence
;
745 assert(PUSH_AVAIL(push
) + push
->rsvd_kick
>= 5);
746 PUSH_DATA (push
, NVC0_FIFO_PKHDR_SQ(NVC0_3D(QUERY_ADDRESS_HIGH
), 4));
747 PUSH_DATAh(push
, screen
->fence
.bo
->offset
);
748 PUSH_DATA (push
, screen
->fence
.bo
->offset
);
749 PUSH_DATA (push
, *sequence
);
750 PUSH_DATA (push
, NVC0_3D_QUERY_GET_FENCE
| NVC0_3D_QUERY_GET_SHORT
|
751 (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT
));
755 nvc0_screen_fence_update(struct pipe_screen
*pscreen
)
757 struct nvc0_screen
*screen
= nvc0_screen(pscreen
);
758 return screen
->fence
.map
[0];
762 nvc0_screen_init_compute(struct nvc0_screen
*screen
)
764 screen
->base
.base
.get_compute_param
= nvc0_screen_get_compute_param
;
766 switch (screen
->base
.device
->chipset
& ~0xf) {
769 return nvc0_screen_compute_setup(screen
, screen
->base
.pushbuf
);
776 return nve4_screen_compute_setup(screen
, screen
->base
.pushbuf
);
783 nvc0_screen_resize_tls_area(struct nvc0_screen
*screen
,
784 uint32_t lpos
, uint32_t lneg
, uint32_t cstack
)
786 struct nouveau_bo
*bo
= NULL
;
788 uint64_t size
= (lpos
+ lneg
) * 32 + cstack
;
790 if (size
>= (1 << 20)) {
791 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64
"\n", size
);
795 size
*= (screen
->base
.device
->chipset
>= 0xe0) ? 64 : 48; /* max warps */
796 size
= align(size
, 0x8000);
797 size
*= screen
->mp_count
;
799 size
= align(size
, 1 << 17);
801 ret
= nouveau_bo_new(screen
->base
.device
, NV_VRAM_DOMAIN(&screen
->base
), 1 << 17, size
,
806 /* Make sure that the pushbuf has acquired a reference to the old tls
807 * segment, as it may have commands that will reference it.
810 PUSH_REFN(screen
->base
.pushbuf
, screen
->tls
,
811 NV_VRAM_DOMAIN(&screen
->base
) | NOUVEAU_BO_RDWR
);
812 nouveau_bo_ref(NULL
, &screen
->tls
);
818 nvc0_screen_resize_text_area(struct nvc0_screen
*screen
, uint64_t size
)
820 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
821 struct nouveau_bo
*bo
;
824 ret
= nouveau_bo_new(screen
->base
.device
, NV_VRAM_DOMAIN(&screen
->base
),
825 1 << 17, size
, NULL
, &bo
);
829 /* Make sure that the pushbuf has acquired a reference to the old text
830 * segment, as it may have commands that will reference it.
833 PUSH_REFN(push
, screen
->text
,
834 NV_VRAM_DOMAIN(&screen
->base
) | NOUVEAU_BO_RD
);
835 nouveau_bo_ref(NULL
, &screen
->text
);
838 nouveau_heap_destroy(&screen
->lib_code
);
839 nouveau_heap_destroy(&screen
->text_heap
);
841 /* XXX: getting a page fault at the end of the code buffer every few
842 * launches, don't use the last 256 bytes to work around them - prefetch ?
844 nouveau_heap_init(&screen
->text_heap
, 0, size
- 0x100);
846 /* update the code segment setup */
847 BEGIN_NVC0(push
, NVC0_3D(CODE_ADDRESS_HIGH
), 2);
848 PUSH_DATAh(push
, screen
->text
->offset
);
849 PUSH_DATA (push
, screen
->text
->offset
);
850 if (screen
->compute
) {
851 BEGIN_NVC0(push
, NVC0_CP(CODE_ADDRESS_HIGH
), 2);
852 PUSH_DATAh(push
, screen
->text
->offset
);
853 PUSH_DATA (push
, screen
->text
->offset
);
860 nvc0_screen_bind_cb_3d(struct nvc0_screen
*screen
, bool *can_serialize
,
861 int stage
, int index
, int size
, uint64_t addr
)
865 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
867 if (screen
->base
.class_3d
>= GM107_3D_CLASS
) {
868 struct nvc0_cb_binding
*binding
= &screen
->cb_bindings
[stage
][index
];
870 // TODO: Better figure out the conditions in which this is needed
871 bool serialize
= binding
->addr
== addr
&& binding
->size
!= size
;
873 serialize
= serialize
&& *can_serialize
;
875 IMMED_NVC0(push
, NVC0_3D(SERIALIZE
), 0);
877 *can_serialize
= false;
880 binding
->addr
= addr
;
881 binding
->size
= size
;
885 BEGIN_NVC0(push
, NVC0_3D(CB_SIZE
), 3);
886 PUSH_DATA (push
, size
);
887 PUSH_DATAh(push
, addr
);
888 PUSH_DATA (push
, addr
);
890 IMMED_NVC0(push
, NVC0_3D(CB_BIND(stage
)), (index
<< 4) | (size
>= 0));
893 #define FAIL_SCREEN_INIT(str, err) \
895 NOUVEAU_ERR(str, err); \
899 struct nouveau_screen
*
900 nvc0_screen_create(struct nouveau_device
*dev
)
902 struct nvc0_screen
*screen
;
903 struct pipe_screen
*pscreen
;
904 struct nouveau_object
*chan
;
905 struct nouveau_pushbuf
*push
;
912 switch (dev
->chipset
& ~0xf) {
926 screen
= CALLOC_STRUCT(nvc0_screen
);
929 pscreen
= &screen
->base
.base
;
930 pscreen
->destroy
= nvc0_screen_destroy
;
932 ret
= nouveau_screen_init(&screen
->base
, dev
);
934 FAIL_SCREEN_INIT("Base screen init failed: %d\n", ret
);
935 chan
= screen
->base
.channel
;
936 push
= screen
->base
.pushbuf
;
937 push
->user_priv
= screen
;
940 /* TODO: could this be higher on Kepler+? how does reclocking vs no
941 * reclocking affect performance?
942 * TODO: could this be higher on Fermi?
944 if (dev
->chipset
>= 0xe0)
945 screen
->base
.transfer_pushbuf_threshold
= 1024;
947 screen
->base
.vidmem_bindings
|= PIPE_BIND_CONSTANT_BUFFER
|
948 PIPE_BIND_SHADER_BUFFER
|
949 PIPE_BIND_VERTEX_BUFFER
| PIPE_BIND_INDEX_BUFFER
|
950 PIPE_BIND_COMMAND_ARGS_BUFFER
| PIPE_BIND_QUERY_BUFFER
;
951 screen
->base
.sysmem_bindings
|=
952 PIPE_BIND_VERTEX_BUFFER
| PIPE_BIND_INDEX_BUFFER
;
954 if (screen
->base
.vram_domain
& NOUVEAU_BO_GART
) {
955 screen
->base
.sysmem_bindings
|= screen
->base
.vidmem_bindings
;
956 screen
->base
.vidmem_bindings
= 0;
959 pscreen
->context_create
= nvc0_create
;
960 pscreen
->is_format_supported
= nvc0_screen_is_format_supported
;
961 pscreen
->get_param
= nvc0_screen_get_param
;
962 pscreen
->get_shader_param
= nvc0_screen_get_shader_param
;
963 pscreen
->get_paramf
= nvc0_screen_get_paramf
;
964 pscreen
->get_sample_pixel_grid
= nvc0_screen_get_sample_pixel_grid
;
965 pscreen
->get_driver_query_info
= nvc0_screen_get_driver_query_info
;
966 pscreen
->get_driver_query_group_info
= nvc0_screen_get_driver_query_group_info
;
968 nvc0_screen_init_resource_functions(pscreen
);
970 screen
->base
.base
.get_video_param
= nouveau_vp3_screen_get_video_param
;
971 screen
->base
.base
.is_video_format_supported
= nouveau_vp3_screen_video_supported
;
973 flags
= NOUVEAU_BO_GART
| NOUVEAU_BO_MAP
;
974 if (screen
->base
.drm
->version
>= 0x01000202)
975 flags
|= NOUVEAU_BO_COHERENT
;
977 ret
= nouveau_bo_new(dev
, flags
, 0, 4096, NULL
, &screen
->fence
.bo
);
979 FAIL_SCREEN_INIT("Error allocating fence BO: %d\n", ret
);
980 nouveau_bo_map(screen
->fence
.bo
, 0, NULL
);
981 screen
->fence
.map
= screen
->fence
.bo
->map
;
982 screen
->base
.fence
.emit
= nvc0_screen_fence_emit
;
983 screen
->base
.fence
.update
= nvc0_screen_fence_update
;
986 ret
= nouveau_object_new(chan
, (dev
->chipset
< 0xe0) ? 0x1f906e : 0x906e,
987 NVIF_CLASS_SW_GF100
, NULL
, 0, &screen
->nvsw
);
989 FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret
);
991 BEGIN_NVC0(push
, SUBC_SW(NV01_SUBCHAN_OBJECT
), 1);
992 PUSH_DATA (push
, screen
->nvsw
->handle
);
994 switch (dev
->chipset
& ~0xf) {
1000 obj_class
= NVF0_P2MF_CLASS
;
1003 obj_class
= NVE4_P2MF_CLASS
;
1006 obj_class
= NVC0_M2MF_CLASS
;
1009 ret
= nouveau_object_new(chan
, 0xbeef323f, obj_class
, NULL
, 0,
1012 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret
);
1014 BEGIN_NVC0(push
, SUBC_M2MF(NV01_SUBCHAN_OBJECT
), 1);
1015 PUSH_DATA (push
, screen
->m2mf
->oclass
);
1016 if (screen
->m2mf
->oclass
== NVE4_P2MF_CLASS
) {
1017 BEGIN_NVC0(push
, SUBC_COPY(NV01_SUBCHAN_OBJECT
), 1);
1018 PUSH_DATA (push
, 0xa0b5);
1021 ret
= nouveau_object_new(chan
, 0xbeef902d, NVC0_2D_CLASS
, NULL
, 0,
1024 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret
);
1026 BEGIN_NVC0(push
, SUBC_2D(NV01_SUBCHAN_OBJECT
), 1);
1027 PUSH_DATA (push
, screen
->eng2d
->oclass
);
1028 BEGIN_NVC0(push
, SUBC_2D(NVC0_2D_SINGLE_GPC
), 1);
1029 PUSH_DATA (push
, 0);
1030 BEGIN_NVC0(push
, NVC0_2D(OPERATION
), 1);
1031 PUSH_DATA (push
, NV50_2D_OPERATION_SRCCOPY
);
1032 BEGIN_NVC0(push
, NVC0_2D(CLIP_ENABLE
), 1);
1033 PUSH_DATA (push
, 0);
1034 BEGIN_NVC0(push
, NVC0_2D(COLOR_KEY_ENABLE
), 1);
1035 PUSH_DATA (push
, 0);
1036 BEGIN_NVC0(push
, SUBC_2D(0x0884), 1);
1037 PUSH_DATA (push
, 0x3f);
1038 BEGIN_NVC0(push
, SUBC_2D(0x0888), 1);
1039 PUSH_DATA (push
, 1);
1040 BEGIN_NVC0(push
, NVC0_2D(COND_MODE
), 1);
1041 PUSH_DATA (push
, NV50_2D_COND_MODE_ALWAYS
);
1043 BEGIN_NVC0(push
, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH
), 2);
1044 PUSH_DATAh(push
, screen
->fence
.bo
->offset
+ 16);
1045 PUSH_DATA (push
, screen
->fence
.bo
->offset
+ 16);
1047 switch (dev
->chipset
& ~0xf) {
1049 switch (dev
->chipset
) {
1052 obj_class
= GP100_3D_CLASS
;
1055 obj_class
= GP102_3D_CLASS
;
1060 obj_class
= GM200_3D_CLASS
;
1063 obj_class
= GM107_3D_CLASS
;
1067 obj_class
= NVF0_3D_CLASS
;
1070 switch (dev
->chipset
) {
1072 obj_class
= NVEA_3D_CLASS
;
1075 obj_class
= NVE4_3D_CLASS
;
1080 obj_class
= NVC8_3D_CLASS
;
1084 switch (dev
->chipset
) {
1086 obj_class
= NVC8_3D_CLASS
;
1089 obj_class
= NVC1_3D_CLASS
;
1092 obj_class
= NVC0_3D_CLASS
;
1097 ret
= nouveau_object_new(chan
, 0xbeef003d, obj_class
, NULL
, 0,
1100 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret
);
1101 screen
->base
.class_3d
= obj_class
;
1103 BEGIN_NVC0(push
, SUBC_3D(NV01_SUBCHAN_OBJECT
), 1);
1104 PUSH_DATA (push
, screen
->eng3d
->oclass
);
1106 BEGIN_NVC0(push
, NVC0_3D(COND_MODE
), 1);
1107 PUSH_DATA (push
, NVC0_3D_COND_MODE_ALWAYS
);
1109 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
1110 /* kill shaders after about 1 second (at 100 MHz) */
1111 BEGIN_NVC0(push
, NVC0_3D(WATCHDOG_TIMER
), 1);
1112 PUSH_DATA (push
, 0x17);
1115 IMMED_NVC0(push
, NVC0_3D(ZETA_COMP_ENABLE
),
1116 screen
->base
.drm
->version
>= 0x01000101);
1117 BEGIN_NVC0(push
, NVC0_3D(RT_COMP_ENABLE(0)), 8);
1118 for (i
= 0; i
< 8; ++i
)
1119 PUSH_DATA(push
, screen
->base
.drm
->version
>= 0x01000101);
1121 BEGIN_NVC0(push
, NVC0_3D(RT_CONTROL
), 1);
1122 PUSH_DATA (push
, 1);
1124 BEGIN_NVC0(push
, NVC0_3D(CSAA_ENABLE
), 1);
1125 PUSH_DATA (push
, 0);
1126 BEGIN_NVC0(push
, NVC0_3D(MULTISAMPLE_ENABLE
), 1);
1127 PUSH_DATA (push
, 0);
1128 BEGIN_NVC0(push
, NVC0_3D(MULTISAMPLE_MODE
), 1);
1129 PUSH_DATA (push
, NVC0_3D_MULTISAMPLE_MODE_MS1
);
1130 BEGIN_NVC0(push
, NVC0_3D(MULTISAMPLE_CTRL
), 1);
1131 PUSH_DATA (push
, 0);
1132 BEGIN_NVC0(push
, NVC0_3D(LINE_WIDTH_SEPARATE
), 1);
1133 PUSH_DATA (push
, 1);
1134 BEGIN_NVC0(push
, NVC0_3D(PRIM_RESTART_WITH_DRAW_ARRAYS
), 1);
1135 PUSH_DATA (push
, 1);
1136 BEGIN_NVC0(push
, NVC0_3D(BLEND_SEPARATE_ALPHA
), 1);
1137 PUSH_DATA (push
, 1);
1138 BEGIN_NVC0(push
, NVC0_3D(BLEND_ENABLE_COMMON
), 1);
1139 PUSH_DATA (push
, 0);
1140 BEGIN_NVC0(push
, NVC0_3D(SHADE_MODEL
), 1);
1141 PUSH_DATA (push
, NVC0_3D_SHADE_MODEL_SMOOTH
);
1142 if (screen
->eng3d
->oclass
< NVE4_3D_CLASS
) {
1143 IMMED_NVC0(push
, NVC0_3D(TEX_MISC
), 0);
1145 BEGIN_NVC0(push
, NVE4_3D(TEX_CB_INDEX
), 1);
1146 PUSH_DATA (push
, 15);
1148 BEGIN_NVC0(push
, NVC0_3D(CALL_LIMIT_LOG
), 1);
1149 PUSH_DATA (push
, 8); /* 128 */
1150 BEGIN_NVC0(push
, NVC0_3D(ZCULL_STATCTRS_ENABLE
), 1);
1151 PUSH_DATA (push
, 1);
1152 if (screen
->eng3d
->oclass
>= NVC1_3D_CLASS
) {
1153 BEGIN_NVC0(push
, NVC0_3D(CACHE_SPLIT
), 1);
1154 PUSH_DATA (push
, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1
);
1157 nvc0_magic_3d_init(push
, screen
->eng3d
->oclass
);
1159 ret
= nvc0_screen_resize_text_area(screen
, 1 << 19);
1161 FAIL_SCREEN_INIT("Error allocating TEXT area: %d\n", ret
);
1163 /* 6 user uniform areas, 6 driver areas, and 1 for the runout */
1164 ret
= nouveau_bo_new(dev
, NV_VRAM_DOMAIN(&screen
->base
), 1 << 12, 13 << 16, NULL
,
1165 &screen
->uniform_bo
);
1167 FAIL_SCREEN_INIT("Error allocating uniform BO: %d\n", ret
);
1169 PUSH_REFN (push
, screen
->uniform_bo
, NV_VRAM_DOMAIN(&screen
->base
) | NOUVEAU_BO_WR
);
1171 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
1172 BEGIN_NVC0(push
, NVC0_3D(CB_SIZE
), 3);
1173 PUSH_DATA (push
, 256);
1174 PUSH_DATAh(push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_RUNOUT_INFO
);
1175 PUSH_DATA (push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_RUNOUT_INFO
);
1176 BEGIN_1IC0(push
, NVC0_3D(CB_POS
), 5);
1177 PUSH_DATA (push
, 0);
1178 PUSH_DATAf(push
, 0.0f
);
1179 PUSH_DATAf(push
, 0.0f
);
1180 PUSH_DATAf(push
, 0.0f
);
1181 PUSH_DATAf(push
, 0.0f
);
1182 BEGIN_NVC0(push
, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH
), 2);
1183 PUSH_DATAh(push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_RUNOUT_INFO
);
1184 PUSH_DATA (push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_RUNOUT_INFO
);
1186 if (screen
->base
.drm
->version
>= 0x01000101) {
1187 ret
= nouveau_getparam(dev
, NOUVEAU_GETPARAM_GRAPH_UNITS
, &value
);
1189 FAIL_SCREEN_INIT("NOUVEAU_GETPARAM_GRAPH_UNITS failed: %d\n", ret
);
1191 if (dev
->chipset
>= 0xe0 && dev
->chipset
< 0xf0)
1192 value
= (8 << 8) | 4;
1194 value
= (16 << 8) | 4;
1196 screen
->gpc_count
= value
& 0x000000ff;
1197 screen
->mp_count
= value
>> 8;
1198 screen
->mp_count_compute
= screen
->mp_count
;
1200 ret
= nvc0_screen_resize_tls_area(screen
, 128 * 16, 0, 0x200);
1202 FAIL_SCREEN_INIT("Error allocating TLS area: %d\n", ret
);
1204 BEGIN_NVC0(push
, NVC0_3D(TEMP_ADDRESS_HIGH
), 4);
1205 PUSH_DATAh(push
, screen
->tls
->offset
);
1206 PUSH_DATA (push
, screen
->tls
->offset
);
1207 PUSH_DATA (push
, screen
->tls
->size
>> 32);
1208 PUSH_DATA (push
, screen
->tls
->size
);
1209 BEGIN_NVC0(push
, NVC0_3D(WARP_TEMP_ALLOC
), 1);
1210 PUSH_DATA (push
, 0);
1211 /* Reduce likelihood of collision with real buffers by placing the hole at
1212 * the top of the 4G area. This will have to be dealt with for real
1213 * eventually by blocking off that area from the VM.
1215 BEGIN_NVC0(push
, NVC0_3D(LOCAL_BASE
), 1);
1216 PUSH_DATA (push
, 0xff << 24);
1218 if (screen
->eng3d
->oclass
< GM107_3D_CLASS
) {
1219 ret
= nouveau_bo_new(dev
, NV_VRAM_DOMAIN(&screen
->base
), 1 << 17, 1 << 20, NULL
,
1220 &screen
->poly_cache
);
1222 FAIL_SCREEN_INIT("Error allocating poly cache BO: %d\n", ret
);
1224 BEGIN_NVC0(push
, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH
), 3);
1225 PUSH_DATAh(push
, screen
->poly_cache
->offset
);
1226 PUSH_DATA (push
, screen
->poly_cache
->offset
);
1227 PUSH_DATA (push
, 3);
1230 ret
= nouveau_bo_new(dev
, NV_VRAM_DOMAIN(&screen
->base
), 1 << 17, 1 << 17, NULL
,
1233 FAIL_SCREEN_INIT("Error allocating txc BO: %d\n", ret
);
1235 BEGIN_NVC0(push
, NVC0_3D(TIC_ADDRESS_HIGH
), 3);
1236 PUSH_DATAh(push
, screen
->txc
->offset
);
1237 PUSH_DATA (push
, screen
->txc
->offset
);
1238 PUSH_DATA (push
, NVC0_TIC_MAX_ENTRIES
- 1);
1239 if (screen
->eng3d
->oclass
>= GM107_3D_CLASS
) {
1240 screen
->tic
.maxwell
= true;
1241 if (screen
->eng3d
->oclass
== GM107_3D_CLASS
) {
1242 screen
->tic
.maxwell
=
1243 debug_get_bool_option("NOUVEAU_MAXWELL_TIC", true);
1244 IMMED_NVC0(push
, SUBC_3D(0x0f10), screen
->tic
.maxwell
);
1248 BEGIN_NVC0(push
, NVC0_3D(TSC_ADDRESS_HIGH
), 3);
1249 PUSH_DATAh(push
, screen
->txc
->offset
+ 65536);
1250 PUSH_DATA (push
, screen
->txc
->offset
+ 65536);
1251 PUSH_DATA (push
, NVC0_TSC_MAX_ENTRIES
- 1);
1253 BEGIN_NVC0(push
, NVC0_3D(SCREEN_Y_CONTROL
), 1);
1254 PUSH_DATA (push
, 0);
1255 BEGIN_NVC0(push
, NVC0_3D(WINDOW_OFFSET_X
), 2);
1256 PUSH_DATA (push
, 0);
1257 PUSH_DATA (push
, 0);
1258 BEGIN_NVC0(push
, NVC0_3D(ZCULL_REGION
), 1); /* deactivate ZCULL */
1259 PUSH_DATA (push
, 0x3f);
1261 BEGIN_NVC0(push
, NVC0_3D(CLIP_RECTS_MODE
), 1);
1262 PUSH_DATA (push
, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY
);
1263 BEGIN_NVC0(push
, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
1264 for (i
= 0; i
< 8 * 2; ++i
)
1266 BEGIN_NVC0(push
, NVC0_3D(CLIP_RECTS_EN
), 1);
1267 PUSH_DATA (push
, 0);
1268 BEGIN_NVC0(push
, NVC0_3D(CLIPID_ENABLE
), 1);
1269 PUSH_DATA (push
, 0);
1271 /* neither scissors, viewport nor stencil mask should affect clears */
1272 BEGIN_NVC0(push
, NVC0_3D(CLEAR_FLAGS
), 1);
1273 PUSH_DATA (push
, 0);
1275 BEGIN_NVC0(push
, NVC0_3D(VIEWPORT_TRANSFORM_EN
), 1);
1276 PUSH_DATA (push
, 1);
1277 for (i
= 0; i
< NVC0_MAX_VIEWPORTS
; i
++) {
1278 BEGIN_NVC0(push
, NVC0_3D(DEPTH_RANGE_NEAR(i
)), 2);
1279 PUSH_DATAf(push
, 0.0f
);
1280 PUSH_DATAf(push
, 1.0f
);
1282 BEGIN_NVC0(push
, NVC0_3D(VIEW_VOLUME_CLIP_CTRL
), 1);
1283 PUSH_DATA (push
, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1
);
1285 /* We use scissors instead of exact view volume clipping,
1286 * so they're always enabled.
1288 for (i
= 0; i
< NVC0_MAX_VIEWPORTS
; i
++) {
1289 BEGIN_NVC0(push
, NVC0_3D(SCISSOR_ENABLE(i
)), 3);
1290 PUSH_DATA (push
, 1);
1291 PUSH_DATA (push
, 8192 << 16);
1292 PUSH_DATA (push
, 8192 << 16);
1295 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
1298 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE
, mme9097_per_instance_bf
);
1299 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES
, mme9097_blend_enables
);
1300 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT
, mme9097_vertex_array_select
);
1301 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT
, mme9097_tep_select
);
1302 MK_MACRO(NVC0_3D_MACRO_GP_SELECT
, mme9097_gp_select
);
1303 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT
, mme9097_poly_mode_front
);
1304 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK
, mme9097_poly_mode_back
);
1305 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT
, mme9097_draw_arrays_indirect
);
1306 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT
, mme9097_draw_elts_indirect
);
1307 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT
, mme9097_draw_arrays_indirect_count
);
1308 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT
, mme9097_draw_elts_indirect_count
);
1309 MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE
, mme9097_query_buffer_write
);
1310 MK_MACRO(NVC0_3D_MACRO_CONSERVATIVE_RASTER_STATE
, mme9097_conservative_raster_state
);
1311 MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER
, mme9097_compute_counter
);
1312 MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER_TO_QUERY
, mme9097_compute_counter_to_query
);
1313 MK_MACRO(NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT
, mme90c0_launch_grid_indirect
);
1315 BEGIN_NVC0(push
, NVC0_3D(RASTERIZE_ENABLE
), 1);
1316 PUSH_DATA (push
, 1);
1317 BEGIN_NVC0(push
, NVC0_3D(RT_SEPARATE_FRAG_DATA
), 1);
1318 PUSH_DATA (push
, 1);
1319 BEGIN_NVC0(push
, NVC0_3D(MACRO_GP_SELECT
), 1);
1320 PUSH_DATA (push
, 0x40);
1321 BEGIN_NVC0(push
, NVC0_3D(LAYER
), 1);
1322 PUSH_DATA (push
, 0);
1323 BEGIN_NVC0(push
, NVC0_3D(MACRO_TEP_SELECT
), 1);
1324 PUSH_DATA (push
, 0x30);
1325 BEGIN_NVC0(push
, NVC0_3D(PATCH_VERTICES
), 1);
1326 PUSH_DATA (push
, 3);
1327 BEGIN_NVC0(push
, NVC0_3D(SP_SELECT(2)), 1);
1328 PUSH_DATA (push
, 0x20);
1329 BEGIN_NVC0(push
, NVC0_3D(SP_SELECT(0)), 1);
1330 PUSH_DATA (push
, 0x00);
1331 screen
->save_state
.patch_vertices
= 3;
1333 BEGIN_NVC0(push
, NVC0_3D(POINT_COORD_REPLACE
), 1);
1334 PUSH_DATA (push
, 0);
1335 BEGIN_NVC0(push
, NVC0_3D(POINT_RASTER_RULES
), 1);
1336 PUSH_DATA (push
, NVC0_3D_POINT_RASTER_RULES_OGL
);
1338 IMMED_NVC0(push
, NVC0_3D(EDGEFLAG
), 1);
1340 if (nvc0_screen_init_compute(screen
))
1343 /* XXX: Compute and 3D are somehow aliased on Fermi. */
1344 for (i
= 0; i
< 5; ++i
) {
1346 for (j
= 0; j
< 16; j
++)
1347 screen
->cb_bindings
[i
][j
].size
= -1;
1349 /* TIC and TSC entries for each unit (nve4+ only) */
1350 /* auxiliary constants (6 user clip planes, base instance id) */
1351 nvc0_screen_bind_cb_3d(screen
, NULL
, i
, 15, NVC0_CB_AUX_SIZE
,
1352 screen
->uniform_bo
->offset
+ NVC0_CB_AUX_INFO(i
));
1353 if (screen
->eng3d
->oclass
>= NVE4_3D_CLASS
) {
1355 BEGIN_1IC0(push
, NVC0_3D(CB_POS
), 9);
1356 PUSH_DATA (push
, NVC0_CB_AUX_UNK_INFO
);
1357 for (j
= 0; j
< 8; ++j
)
1360 BEGIN_NVC0(push
, NVC0_3D(TEX_LIMITS(i
)), 1);
1361 PUSH_DATA (push
, 0x54);
1364 /* MS sample coordinate offsets: these do not work with _ALT modes ! */
1365 BEGIN_1IC0(push
, NVC0_3D(CB_POS
), 1 + 2 * 8);
1366 PUSH_DATA (push
, NVC0_CB_AUX_MS_INFO
);
1367 PUSH_DATA (push
, 0); /* 0 */
1368 PUSH_DATA (push
, 0);
1369 PUSH_DATA (push
, 1); /* 1 */
1370 PUSH_DATA (push
, 0);
1371 PUSH_DATA (push
, 0); /* 2 */
1372 PUSH_DATA (push
, 1);
1373 PUSH_DATA (push
, 1); /* 3 */
1374 PUSH_DATA (push
, 1);
1375 PUSH_DATA (push
, 2); /* 4 */
1376 PUSH_DATA (push
, 0);
1377 PUSH_DATA (push
, 3); /* 5 */
1378 PUSH_DATA (push
, 0);
1379 PUSH_DATA (push
, 2); /* 6 */
1380 PUSH_DATA (push
, 1);
1381 PUSH_DATA (push
, 3); /* 7 */
1382 PUSH_DATA (push
, 1);
1384 BEGIN_NVC0(push
, NVC0_3D(LINKED_TSC
), 1);
1385 PUSH_DATA (push
, 0);
1389 screen
->tic
.entries
= CALLOC(
1390 NVC0_TIC_MAX_ENTRIES
+ NVC0_TSC_MAX_ENTRIES
+ NVE4_IMG_MAX_HANDLES
,
1392 screen
->tsc
.entries
= screen
->tic
.entries
+ NVC0_TIC_MAX_ENTRIES
;
1393 screen
->img
.entries
= (void *)(screen
->tsc
.entries
+ NVC0_TSC_MAX_ENTRIES
);
1395 if (!nvc0_blitter_create(screen
))
1398 nouveau_fence_new(&screen
->base
, &screen
->base
.fence
.current
);
1400 return &screen
->base
;
1403 screen
->base
.base
.context_create
= NULL
;
1404 return &screen
->base
;
1408 nvc0_screen_tic_alloc(struct nvc0_screen
*screen
, void *entry
)
1410 int i
= screen
->tic
.next
;
1412 while (screen
->tic
.lock
[i
/ 32] & (1 << (i
% 32)))
1413 i
= (i
+ 1) & (NVC0_TIC_MAX_ENTRIES
- 1);
1415 screen
->tic
.next
= (i
+ 1) & (NVC0_TIC_MAX_ENTRIES
- 1);
1417 if (screen
->tic
.entries
[i
])
1418 nv50_tic_entry(screen
->tic
.entries
[i
])->id
= -1;
1420 screen
->tic
.entries
[i
] = entry
;
1425 nvc0_screen_tsc_alloc(struct nvc0_screen
*screen
, void *entry
)
1427 int i
= screen
->tsc
.next
;
1429 while (screen
->tsc
.lock
[i
/ 32] & (1 << (i
% 32)))
1430 i
= (i
+ 1) & (NVC0_TSC_MAX_ENTRIES
- 1);
1432 screen
->tsc
.next
= (i
+ 1) & (NVC0_TSC_MAX_ENTRIES
- 1);
1434 if (screen
->tsc
.entries
[i
])
1435 nv50_tsc_entry(screen
->tsc
.entries
[i
])->id
= -1;
1437 screen
->tsc
.entries
[i
] = entry
;