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
:
270 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER
:
271 return nouveau_screen(pscreen
)->vram_domain
& NOUVEAU_BO_VRAM
? 1 : 0;
272 case PIPE_CAP_TGSI_FS_FBFETCH
:
273 return class_3d
>= NVE4_3D_CLASS
; /* needs testing on fermi */
274 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE
:
275 case PIPE_CAP_TGSI_BALLOT
:
276 case PIPE_CAP_BINDLESS_TEXTURE
:
277 return class_3d
>= NVE4_3D_CLASS
;
278 case PIPE_CAP_TGSI_ATOMFADD
:
279 return class_3d
< GM107_3D_CLASS
; /* needs additional lowering */
280 case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE
:
281 case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT
:
282 case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT
:
283 case PIPE_CAP_POST_DEPTH_COVERAGE
:
284 case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_TRIANGLES
:
285 case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_POINTS_LINES
:
286 case PIPE_CAP_CONSERVATIVE_RASTER_POST_DEPTH_COVERAGE
:
287 case PIPE_CAP_PROGRAMMABLE_SAMPLE_LOCATIONS
:
288 return class_3d
>= GM200_3D_CLASS
;
289 case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES
:
290 return class_3d
>= GP100_3D_CLASS
;
292 /* unsupported caps */
293 case PIPE_CAP_DEPTH_CLIP_DISABLE_SEPARATE
:
294 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT
:
295 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER
:
296 case PIPE_CAP_SHADER_STENCIL_EXPORT
:
297 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS
:
298 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY
:
299 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY
:
300 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY
:
301 case PIPE_CAP_FAKE_SW_MSAA
:
302 case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION
:
303 case PIPE_CAP_VERTEXID_NOBASE
:
304 case PIPE_CAP_RESOURCE_FROM_USER_MEMORY
:
305 case PIPE_CAP_DEVICE_RESET_STATUS_QUERY
:
306 case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL
:
307 case PIPE_CAP_GENERATE_MIPMAP
:
308 case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY
:
309 case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS
:
310 case PIPE_CAP_QUERY_MEMORY_INFO
:
311 case PIPE_CAP_PCI_GROUP
:
312 case PIPE_CAP_PCI_BUS
:
313 case PIPE_CAP_PCI_DEVICE
:
314 case PIPE_CAP_PCI_FUNCTION
:
315 case PIPE_CAP_TGSI_CAN_READ_OUTPUTS
:
316 case PIPE_CAP_NATIVE_FENCE_FD
:
317 case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY
:
318 case PIPE_CAP_INT64_DIVMOD
:
319 case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE
:
320 case PIPE_CAP_NIR_SAMPLERS_AS_DEREF
:
321 case PIPE_CAP_MEMOBJ
:
322 case PIPE_CAP_LOAD_CONSTBUF
:
323 case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS
:
324 case PIPE_CAP_TILE_RASTER_ORDER
:
325 case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES
:
326 case PIPE_CAP_FRAMEBUFFER_MSAA_CONSTRAINTS
:
327 case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET
:
328 case PIPE_CAP_CONTEXT_PRIORITY_MASK
:
329 case PIPE_CAP_FENCE_SIGNAL
:
330 case PIPE_CAP_CONSTBUF0_FLAGS
:
331 case PIPE_CAP_PACKED_UNIFORMS
:
332 case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_POINTS_LINES
:
333 case PIPE_CAP_MAX_COMBINED_SHADER_BUFFERS
:
334 case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTERS
:
335 case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTER_BUFFERS
:
336 case PIPE_CAP_SURFACE_SAMPLE_COUNT
:
339 case PIPE_CAP_VENDOR_ID
:
341 case PIPE_CAP_DEVICE_ID
: {
343 if (nouveau_getparam(dev
, NOUVEAU_GETPARAM_PCI_DEVICE
, &device_id
)) {
344 NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
349 case PIPE_CAP_ACCELERATED
:
351 case PIPE_CAP_VIDEO_MEMORY
:
352 return dev
->vram_size
>> 20;
356 debug_printf("%s: unhandled cap %d\n", __func__
, param
);
357 return u_pipe_screen_get_param_defaults(pscreen
, param
);
362 nvc0_screen_get_shader_param(struct pipe_screen
*pscreen
,
363 enum pipe_shader_type shader
,
364 enum pipe_shader_cap param
)
366 const uint16_t class_3d
= nouveau_screen(pscreen
)->class_3d
;
369 case PIPE_SHADER_VERTEX
:
370 case PIPE_SHADER_GEOMETRY
:
371 case PIPE_SHADER_FRAGMENT
:
372 case PIPE_SHADER_COMPUTE
:
373 case PIPE_SHADER_TESS_CTRL
:
374 case PIPE_SHADER_TESS_EVAL
:
381 case PIPE_SHADER_CAP_PREFERRED_IR
:
382 return PIPE_SHADER_IR_TGSI
;
383 case PIPE_SHADER_CAP_SUPPORTED_IRS
:
384 return 1 << PIPE_SHADER_IR_TGSI
;
385 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS
:
386 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS
:
387 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS
:
388 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS
:
390 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH
:
392 case PIPE_SHADER_CAP_MAX_INPUTS
:
393 if (shader
== PIPE_SHADER_VERTEX
)
395 /* NOTE: These only count our slots for GENERIC varyings.
396 * The address space may be larger, but the actual hard limit seems to be
397 * less than what the address space layout permits, so don't add TEXCOORD,
400 if (shader
== PIPE_SHADER_FRAGMENT
)
402 /* Actually this counts CLIPVERTEX, which occupies the last generic slot,
403 * and excludes 0x60 per-patch inputs.
406 case PIPE_SHADER_CAP_MAX_OUTPUTS
:
408 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE
:
409 return NVC0_MAX_CONSTBUF_SIZE
;
410 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS
:
411 return NVC0_MAX_PIPE_CONSTBUFS
;
412 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR
:
413 return shader
!= PIPE_SHADER_FRAGMENT
;
414 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR
:
415 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR
:
416 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR
:
418 case PIPE_SHADER_CAP_MAX_TEMPS
:
419 return NVC0_CAP_MAX_PROGRAM_TEMPS
;
420 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED
:
422 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED
:
424 case PIPE_SHADER_CAP_SUBROUTINES
:
426 case PIPE_SHADER_CAP_INTEGERS
:
428 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED
:
430 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED
:
432 case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS
:
434 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED
:
435 case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED
:
436 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE
:
437 case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD
:
438 case PIPE_SHADER_CAP_INT64_ATOMICS
:
439 case PIPE_SHADER_CAP_FP16
:
440 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS
:
441 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS
:
443 case PIPE_SHADER_CAP_SCALAR_ISA
:
445 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS
:
446 return NVC0_MAX_BUFFERS
;
447 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS
:
448 return (class_3d
>= NVE4_3D_CLASS
) ? 32 : 16;
449 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS
:
450 return (class_3d
>= NVE4_3D_CLASS
) ? 32 : 16;
451 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT
:
453 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES
:
454 if (class_3d
>= NVE4_3D_CLASS
)
455 return NVC0_MAX_IMAGES
;
456 if (shader
== PIPE_SHADER_FRAGMENT
|| shader
== PIPE_SHADER_COMPUTE
)
457 return NVC0_MAX_IMAGES
;
460 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param
);
466 nvc0_screen_get_paramf(struct pipe_screen
*pscreen
, enum pipe_capf param
)
468 const uint16_t class_3d
= nouveau_screen(pscreen
)->class_3d
;
471 case PIPE_CAPF_MAX_LINE_WIDTH
:
472 case PIPE_CAPF_MAX_LINE_WIDTH_AA
:
474 case PIPE_CAPF_MAX_POINT_WIDTH
:
476 case PIPE_CAPF_MAX_POINT_WIDTH_AA
:
478 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY
:
480 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS
:
482 case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE
:
484 case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE
:
485 return class_3d
>= GM200_3D_CLASS
? 0.75f
: 0.0f
;
486 case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY
:
487 return class_3d
>= GM200_3D_CLASS
? 0.25f
: 0.0f
;
490 NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param
);
495 nvc0_screen_get_compute_param(struct pipe_screen
*pscreen
,
496 enum pipe_shader_ir ir_type
,
497 enum pipe_compute_cap param
, void *data
)
499 struct nvc0_screen
*screen
= nvc0_screen(pscreen
);
500 const uint16_t obj_class
= screen
->compute
->oclass
;
502 #define RET(x) do { \
504 memcpy(data, x, sizeof(x)); \
509 case PIPE_COMPUTE_CAP_GRID_DIMENSION
:
510 RET((uint64_t []) { 3 });
511 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE
:
512 if (obj_class
>= NVE4_COMPUTE_CLASS
) {
513 RET(((uint64_t []) { 0x7fffffff, 65535, 65535 }));
515 RET(((uint64_t []) { 65535, 65535, 65535 }));
517 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE
:
518 RET(((uint64_t []) { 1024, 1024, 64 }));
519 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK
:
520 RET((uint64_t []) { 1024 });
521 case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK
:
522 if (obj_class
>= NVE4_COMPUTE_CLASS
) {
523 RET((uint64_t []) { 1024 });
525 RET((uint64_t []) { 512 });
527 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE
: /* g[] */
528 RET((uint64_t []) { 1ULL << 40 });
529 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE
: /* s[] */
531 case GM200_COMPUTE_CLASS
:
532 RET((uint64_t []) { 96 << 10 });
534 case GM107_COMPUTE_CLASS
:
535 RET((uint64_t []) { 64 << 10 });
538 RET((uint64_t []) { 48 << 10 });
541 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE
: /* l[] */
542 RET((uint64_t []) { 512 << 10 });
543 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE
: /* c[], arbitrary limit */
544 RET((uint64_t []) { 4096 });
545 case PIPE_COMPUTE_CAP_SUBGROUP_SIZE
:
546 RET((uint32_t []) { 32 });
547 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE
:
548 RET((uint64_t []) { 1ULL << 40 });
549 case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED
:
550 RET((uint32_t []) { 0 });
551 case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS
:
552 RET((uint32_t []) { screen
->mp_count_compute
});
553 case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY
:
554 RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
555 case PIPE_COMPUTE_CAP_ADDRESS_BITS
:
556 RET((uint32_t []) { 64 });
565 nvc0_screen_get_sample_pixel_grid(struct pipe_screen
*pscreen
,
566 unsigned sample_count
,
567 unsigned *width
, unsigned *height
)
569 switch (sample_count
) {
572 /* this could be 4x4, but the GL state tracker makes it difficult to
573 * create a 1x MSAA texture and smaller grids save CB space */
595 nvc0_screen_destroy(struct pipe_screen
*pscreen
)
597 struct nvc0_screen
*screen
= nvc0_screen(pscreen
);
599 if (!nouveau_drm_screen_unref(&screen
->base
))
602 if (screen
->base
.fence
.current
) {
603 struct nouveau_fence
*current
= NULL
;
605 /* nouveau_fence_wait will create a new current fence, so wait on the
606 * _current_ one, and remove both.
608 nouveau_fence_ref(screen
->base
.fence
.current
, ¤t
);
609 nouveau_fence_wait(current
, NULL
);
610 nouveau_fence_ref(NULL
, ¤t
);
611 nouveau_fence_ref(NULL
, &screen
->base
.fence
.current
);
613 if (screen
->base
.pushbuf
)
614 screen
->base
.pushbuf
->user_priv
= NULL
;
617 nvc0_blitter_destroy(screen
);
618 if (screen
->pm
.prog
) {
619 screen
->pm
.prog
->code
= NULL
; /* hardcoded, don't FREE */
620 nvc0_program_destroy(NULL
, screen
->pm
.prog
);
621 FREE(screen
->pm
.prog
);
624 nouveau_bo_ref(NULL
, &screen
->text
);
625 nouveau_bo_ref(NULL
, &screen
->uniform_bo
);
626 nouveau_bo_ref(NULL
, &screen
->tls
);
627 nouveau_bo_ref(NULL
, &screen
->txc
);
628 nouveau_bo_ref(NULL
, &screen
->fence
.bo
);
629 nouveau_bo_ref(NULL
, &screen
->poly_cache
);
631 nouveau_heap_destroy(&screen
->lib_code
);
632 nouveau_heap_destroy(&screen
->text_heap
);
634 FREE(screen
->tic
.entries
);
636 nouveau_object_del(&screen
->eng3d
);
637 nouveau_object_del(&screen
->eng2d
);
638 nouveau_object_del(&screen
->m2mf
);
639 nouveau_object_del(&screen
->compute
);
640 nouveau_object_del(&screen
->nvsw
);
642 nouveau_screen_fini(&screen
->base
);
648 nvc0_graph_set_macro(struct nvc0_screen
*screen
, uint32_t m
, unsigned pos
,
649 unsigned size
, const uint32_t *data
)
651 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
655 assert((pos
+ size
) <= 0x800);
657 BEGIN_NVC0(push
, SUBC_3D(NVC0_GRAPH_MACRO_ID
), 2);
658 PUSH_DATA (push
, (m
- 0x3800) / 8);
659 PUSH_DATA (push
, pos
);
660 BEGIN_1IC0(push
, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS
), size
+ 1);
661 PUSH_DATA (push
, pos
);
662 PUSH_DATAp(push
, data
, size
);
668 nvc0_magic_3d_init(struct nouveau_pushbuf
*push
, uint16_t obj_class
)
670 BEGIN_NVC0(push
, SUBC_3D(0x10cc), 1);
671 PUSH_DATA (push
, 0xff);
672 BEGIN_NVC0(push
, SUBC_3D(0x10e0), 2);
673 PUSH_DATA (push
, 0xff);
674 PUSH_DATA (push
, 0xff);
675 BEGIN_NVC0(push
, SUBC_3D(0x10ec), 2);
676 PUSH_DATA (push
, 0xff);
677 PUSH_DATA (push
, 0xff);
678 BEGIN_NVC0(push
, SUBC_3D(0x074c), 1);
679 PUSH_DATA (push
, 0x3f);
681 BEGIN_NVC0(push
, SUBC_3D(0x16a8), 1);
682 PUSH_DATA (push
, (3 << 16) | 3);
683 BEGIN_NVC0(push
, SUBC_3D(0x1794), 1);
684 PUSH_DATA (push
, (2 << 16) | 2);
686 if (obj_class
< GM107_3D_CLASS
) {
687 BEGIN_NVC0(push
, SUBC_3D(0x12ac), 1);
690 BEGIN_NVC0(push
, SUBC_3D(0x0218), 1);
691 PUSH_DATA (push
, 0x10);
692 BEGIN_NVC0(push
, SUBC_3D(0x10fc), 1);
693 PUSH_DATA (push
, 0x10);
694 BEGIN_NVC0(push
, SUBC_3D(0x1290), 1);
695 PUSH_DATA (push
, 0x10);
696 BEGIN_NVC0(push
, SUBC_3D(0x12d8), 2);
697 PUSH_DATA (push
, 0x10);
698 PUSH_DATA (push
, 0x10);
699 BEGIN_NVC0(push
, SUBC_3D(0x1140), 1);
700 PUSH_DATA (push
, 0x10);
701 BEGIN_NVC0(push
, SUBC_3D(0x1610), 1);
702 PUSH_DATA (push
, 0xe);
704 BEGIN_NVC0(push
, NVC0_3D(VERTEX_ID_GEN_MODE
), 1);
705 PUSH_DATA (push
, NVC0_3D_VERTEX_ID_GEN_MODE_DRAW_ARRAYS_ADD_START
);
706 BEGIN_NVC0(push
, SUBC_3D(0x030c), 1);
708 BEGIN_NVC0(push
, SUBC_3D(0x0300), 1);
711 BEGIN_NVC0(push
, SUBC_3D(0x02d0), 1);
712 PUSH_DATA (push
, 0x3fffff);
713 BEGIN_NVC0(push
, SUBC_3D(0x0fdc), 1);
715 BEGIN_NVC0(push
, SUBC_3D(0x19c0), 1);
718 if (obj_class
< GM107_3D_CLASS
) {
719 BEGIN_NVC0(push
, SUBC_3D(0x075c), 1);
722 if (obj_class
>= NVE4_3D_CLASS
) {
723 BEGIN_NVC0(push
, SUBC_3D(0x07fc), 1);
728 /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
729 * are supposed to do */
733 nvc0_screen_fence_emit(struct pipe_screen
*pscreen
, u32
*sequence
)
735 struct nvc0_screen
*screen
= nvc0_screen(pscreen
);
736 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
738 /* we need to do it after possible flush in MARK_RING */
739 *sequence
= ++screen
->base
.fence
.sequence
;
741 assert(PUSH_AVAIL(push
) + push
->rsvd_kick
>= 5);
742 PUSH_DATA (push
, NVC0_FIFO_PKHDR_SQ(NVC0_3D(QUERY_ADDRESS_HIGH
), 4));
743 PUSH_DATAh(push
, screen
->fence
.bo
->offset
);
744 PUSH_DATA (push
, screen
->fence
.bo
->offset
);
745 PUSH_DATA (push
, *sequence
);
746 PUSH_DATA (push
, NVC0_3D_QUERY_GET_FENCE
| NVC0_3D_QUERY_GET_SHORT
|
747 (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT
));
751 nvc0_screen_fence_update(struct pipe_screen
*pscreen
)
753 struct nvc0_screen
*screen
= nvc0_screen(pscreen
);
754 return screen
->fence
.map
[0];
758 nvc0_screen_init_compute(struct nvc0_screen
*screen
)
760 screen
->base
.base
.get_compute_param
= nvc0_screen_get_compute_param
;
762 switch (screen
->base
.device
->chipset
& ~0xf) {
765 return nvc0_screen_compute_setup(screen
, screen
->base
.pushbuf
);
772 return nve4_screen_compute_setup(screen
, screen
->base
.pushbuf
);
779 nvc0_screen_resize_tls_area(struct nvc0_screen
*screen
,
780 uint32_t lpos
, uint32_t lneg
, uint32_t cstack
)
782 struct nouveau_bo
*bo
= NULL
;
784 uint64_t size
= (lpos
+ lneg
) * 32 + cstack
;
786 if (size
>= (1 << 20)) {
787 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64
"\n", size
);
791 size
*= (screen
->base
.device
->chipset
>= 0xe0) ? 64 : 48; /* max warps */
792 size
= align(size
, 0x8000);
793 size
*= screen
->mp_count
;
795 size
= align(size
, 1 << 17);
797 ret
= nouveau_bo_new(screen
->base
.device
, NV_VRAM_DOMAIN(&screen
->base
), 1 << 17, size
,
802 /* Make sure that the pushbuf has acquired a reference to the old tls
803 * segment, as it may have commands that will reference it.
806 PUSH_REFN(screen
->base
.pushbuf
, screen
->tls
,
807 NV_VRAM_DOMAIN(&screen
->base
) | NOUVEAU_BO_RDWR
);
808 nouveau_bo_ref(NULL
, &screen
->tls
);
814 nvc0_screen_resize_text_area(struct nvc0_screen
*screen
, uint64_t size
)
816 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
817 struct nouveau_bo
*bo
;
820 ret
= nouveau_bo_new(screen
->base
.device
, NV_VRAM_DOMAIN(&screen
->base
),
821 1 << 17, size
, NULL
, &bo
);
825 /* Make sure that the pushbuf has acquired a reference to the old text
826 * segment, as it may have commands that will reference it.
829 PUSH_REFN(push
, screen
->text
,
830 NV_VRAM_DOMAIN(&screen
->base
) | NOUVEAU_BO_RD
);
831 nouveau_bo_ref(NULL
, &screen
->text
);
834 nouveau_heap_destroy(&screen
->lib_code
);
835 nouveau_heap_destroy(&screen
->text_heap
);
837 /* XXX: getting a page fault at the end of the code buffer every few
838 * launches, don't use the last 256 bytes to work around them - prefetch ?
840 nouveau_heap_init(&screen
->text_heap
, 0, size
- 0x100);
842 /* update the code segment setup */
843 BEGIN_NVC0(push
, NVC0_3D(CODE_ADDRESS_HIGH
), 2);
844 PUSH_DATAh(push
, screen
->text
->offset
);
845 PUSH_DATA (push
, screen
->text
->offset
);
846 if (screen
->compute
) {
847 BEGIN_NVC0(push
, NVC0_CP(CODE_ADDRESS_HIGH
), 2);
848 PUSH_DATAh(push
, screen
->text
->offset
);
849 PUSH_DATA (push
, screen
->text
->offset
);
856 nvc0_screen_bind_cb_3d(struct nvc0_screen
*screen
, bool *can_serialize
,
857 int stage
, int index
, int size
, uint64_t addr
)
861 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
863 if (screen
->base
.class_3d
>= GM107_3D_CLASS
) {
864 struct nvc0_cb_binding
*binding
= &screen
->cb_bindings
[stage
][index
];
866 // TODO: Better figure out the conditions in which this is needed
867 bool serialize
= binding
->addr
== addr
&& binding
->size
!= size
;
869 serialize
= serialize
&& *can_serialize
;
871 IMMED_NVC0(push
, NVC0_3D(SERIALIZE
), 0);
873 *can_serialize
= false;
876 binding
->addr
= addr
;
877 binding
->size
= size
;
881 BEGIN_NVC0(push
, NVC0_3D(CB_SIZE
), 3);
882 PUSH_DATA (push
, size
);
883 PUSH_DATAh(push
, addr
);
884 PUSH_DATA (push
, addr
);
886 IMMED_NVC0(push
, NVC0_3D(CB_BIND(stage
)), (index
<< 4) | (size
>= 0));
889 #define FAIL_SCREEN_INIT(str, err) \
891 NOUVEAU_ERR(str, err); \
895 struct nouveau_screen
*
896 nvc0_screen_create(struct nouveau_device
*dev
)
898 struct nvc0_screen
*screen
;
899 struct pipe_screen
*pscreen
;
900 struct nouveau_object
*chan
;
901 struct nouveau_pushbuf
*push
;
908 switch (dev
->chipset
& ~0xf) {
922 screen
= CALLOC_STRUCT(nvc0_screen
);
925 pscreen
= &screen
->base
.base
;
926 pscreen
->destroy
= nvc0_screen_destroy
;
928 ret
= nouveau_screen_init(&screen
->base
, dev
);
930 FAIL_SCREEN_INIT("Base screen init failed: %d\n", ret
);
931 chan
= screen
->base
.channel
;
932 push
= screen
->base
.pushbuf
;
933 push
->user_priv
= screen
;
936 /* TODO: could this be higher on Kepler+? how does reclocking vs no
937 * reclocking affect performance?
938 * TODO: could this be higher on Fermi?
940 if (dev
->chipset
>= 0xe0)
941 screen
->base
.transfer_pushbuf_threshold
= 1024;
943 screen
->base
.vidmem_bindings
|= PIPE_BIND_CONSTANT_BUFFER
|
944 PIPE_BIND_SHADER_BUFFER
|
945 PIPE_BIND_VERTEX_BUFFER
| PIPE_BIND_INDEX_BUFFER
|
946 PIPE_BIND_COMMAND_ARGS_BUFFER
| PIPE_BIND_QUERY_BUFFER
;
947 screen
->base
.sysmem_bindings
|=
948 PIPE_BIND_VERTEX_BUFFER
| PIPE_BIND_INDEX_BUFFER
;
950 if (screen
->base
.vram_domain
& NOUVEAU_BO_GART
) {
951 screen
->base
.sysmem_bindings
|= screen
->base
.vidmem_bindings
;
952 screen
->base
.vidmem_bindings
= 0;
955 pscreen
->context_create
= nvc0_create
;
956 pscreen
->is_format_supported
= nvc0_screen_is_format_supported
;
957 pscreen
->get_param
= nvc0_screen_get_param
;
958 pscreen
->get_shader_param
= nvc0_screen_get_shader_param
;
959 pscreen
->get_paramf
= nvc0_screen_get_paramf
;
960 pscreen
->get_sample_pixel_grid
= nvc0_screen_get_sample_pixel_grid
;
961 pscreen
->get_driver_query_info
= nvc0_screen_get_driver_query_info
;
962 pscreen
->get_driver_query_group_info
= nvc0_screen_get_driver_query_group_info
;
964 nvc0_screen_init_resource_functions(pscreen
);
966 screen
->base
.base
.get_video_param
= nouveau_vp3_screen_get_video_param
;
967 screen
->base
.base
.is_video_format_supported
= nouveau_vp3_screen_video_supported
;
969 flags
= NOUVEAU_BO_GART
| NOUVEAU_BO_MAP
;
970 if (screen
->base
.drm
->version
>= 0x01000202)
971 flags
|= NOUVEAU_BO_COHERENT
;
973 ret
= nouveau_bo_new(dev
, flags
, 0, 4096, NULL
, &screen
->fence
.bo
);
975 FAIL_SCREEN_INIT("Error allocating fence BO: %d\n", ret
);
976 nouveau_bo_map(screen
->fence
.bo
, 0, NULL
);
977 screen
->fence
.map
= screen
->fence
.bo
->map
;
978 screen
->base
.fence
.emit
= nvc0_screen_fence_emit
;
979 screen
->base
.fence
.update
= nvc0_screen_fence_update
;
982 ret
= nouveau_object_new(chan
, (dev
->chipset
< 0xe0) ? 0x1f906e : 0x906e,
983 NVIF_CLASS_SW_GF100
, NULL
, 0, &screen
->nvsw
);
985 FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret
);
987 BEGIN_NVC0(push
, SUBC_SW(NV01_SUBCHAN_OBJECT
), 1);
988 PUSH_DATA (push
, screen
->nvsw
->handle
);
990 switch (dev
->chipset
& ~0xf) {
996 obj_class
= NVF0_P2MF_CLASS
;
999 obj_class
= NVE4_P2MF_CLASS
;
1002 obj_class
= NVC0_M2MF_CLASS
;
1005 ret
= nouveau_object_new(chan
, 0xbeef323f, obj_class
, NULL
, 0,
1008 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret
);
1010 BEGIN_NVC0(push
, SUBC_M2MF(NV01_SUBCHAN_OBJECT
), 1);
1011 PUSH_DATA (push
, screen
->m2mf
->oclass
);
1012 if (screen
->m2mf
->oclass
== NVE4_P2MF_CLASS
) {
1013 BEGIN_NVC0(push
, SUBC_COPY(NV01_SUBCHAN_OBJECT
), 1);
1014 PUSH_DATA (push
, 0xa0b5);
1017 ret
= nouveau_object_new(chan
, 0xbeef902d, NVC0_2D_CLASS
, NULL
, 0,
1020 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret
);
1022 BEGIN_NVC0(push
, SUBC_2D(NV01_SUBCHAN_OBJECT
), 1);
1023 PUSH_DATA (push
, screen
->eng2d
->oclass
);
1024 BEGIN_NVC0(push
, SUBC_2D(NVC0_2D_SINGLE_GPC
), 1);
1025 PUSH_DATA (push
, 0);
1026 BEGIN_NVC0(push
, NVC0_2D(OPERATION
), 1);
1027 PUSH_DATA (push
, NV50_2D_OPERATION_SRCCOPY
);
1028 BEGIN_NVC0(push
, NVC0_2D(CLIP_ENABLE
), 1);
1029 PUSH_DATA (push
, 0);
1030 BEGIN_NVC0(push
, NVC0_2D(COLOR_KEY_ENABLE
), 1);
1031 PUSH_DATA (push
, 0);
1032 BEGIN_NVC0(push
, SUBC_2D(0x0884), 1);
1033 PUSH_DATA (push
, 0x3f);
1034 BEGIN_NVC0(push
, SUBC_2D(0x0888), 1);
1035 PUSH_DATA (push
, 1);
1036 BEGIN_NVC0(push
, NVC0_2D(COND_MODE
), 1);
1037 PUSH_DATA (push
, NV50_2D_COND_MODE_ALWAYS
);
1039 BEGIN_NVC0(push
, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH
), 2);
1040 PUSH_DATAh(push
, screen
->fence
.bo
->offset
+ 16);
1041 PUSH_DATA (push
, screen
->fence
.bo
->offset
+ 16);
1043 switch (dev
->chipset
& ~0xf) {
1045 switch (dev
->chipset
) {
1048 obj_class
= GP100_3D_CLASS
;
1051 obj_class
= GP102_3D_CLASS
;
1056 obj_class
= GM200_3D_CLASS
;
1059 obj_class
= GM107_3D_CLASS
;
1063 obj_class
= NVF0_3D_CLASS
;
1066 switch (dev
->chipset
) {
1068 obj_class
= NVEA_3D_CLASS
;
1071 obj_class
= NVE4_3D_CLASS
;
1076 obj_class
= NVC8_3D_CLASS
;
1080 switch (dev
->chipset
) {
1082 obj_class
= NVC8_3D_CLASS
;
1085 obj_class
= NVC1_3D_CLASS
;
1088 obj_class
= NVC0_3D_CLASS
;
1093 ret
= nouveau_object_new(chan
, 0xbeef003d, obj_class
, NULL
, 0,
1096 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret
);
1097 screen
->base
.class_3d
= obj_class
;
1099 BEGIN_NVC0(push
, SUBC_3D(NV01_SUBCHAN_OBJECT
), 1);
1100 PUSH_DATA (push
, screen
->eng3d
->oclass
);
1102 BEGIN_NVC0(push
, NVC0_3D(COND_MODE
), 1);
1103 PUSH_DATA (push
, NVC0_3D_COND_MODE_ALWAYS
);
1105 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
1106 /* kill shaders after about 1 second (at 100 MHz) */
1107 BEGIN_NVC0(push
, NVC0_3D(WATCHDOG_TIMER
), 1);
1108 PUSH_DATA (push
, 0x17);
1111 IMMED_NVC0(push
, NVC0_3D(ZETA_COMP_ENABLE
),
1112 screen
->base
.drm
->version
>= 0x01000101);
1113 BEGIN_NVC0(push
, NVC0_3D(RT_COMP_ENABLE(0)), 8);
1114 for (i
= 0; i
< 8; ++i
)
1115 PUSH_DATA(push
, screen
->base
.drm
->version
>= 0x01000101);
1117 BEGIN_NVC0(push
, NVC0_3D(RT_CONTROL
), 1);
1118 PUSH_DATA (push
, 1);
1120 BEGIN_NVC0(push
, NVC0_3D(CSAA_ENABLE
), 1);
1121 PUSH_DATA (push
, 0);
1122 BEGIN_NVC0(push
, NVC0_3D(MULTISAMPLE_ENABLE
), 1);
1123 PUSH_DATA (push
, 0);
1124 BEGIN_NVC0(push
, NVC0_3D(MULTISAMPLE_MODE
), 1);
1125 PUSH_DATA (push
, NVC0_3D_MULTISAMPLE_MODE_MS1
);
1126 BEGIN_NVC0(push
, NVC0_3D(MULTISAMPLE_CTRL
), 1);
1127 PUSH_DATA (push
, 0);
1128 BEGIN_NVC0(push
, NVC0_3D(LINE_WIDTH_SEPARATE
), 1);
1129 PUSH_DATA (push
, 1);
1130 BEGIN_NVC0(push
, NVC0_3D(PRIM_RESTART_WITH_DRAW_ARRAYS
), 1);
1131 PUSH_DATA (push
, 1);
1132 BEGIN_NVC0(push
, NVC0_3D(BLEND_SEPARATE_ALPHA
), 1);
1133 PUSH_DATA (push
, 1);
1134 BEGIN_NVC0(push
, NVC0_3D(BLEND_ENABLE_COMMON
), 1);
1135 PUSH_DATA (push
, 0);
1136 BEGIN_NVC0(push
, NVC0_3D(SHADE_MODEL
), 1);
1137 PUSH_DATA (push
, NVC0_3D_SHADE_MODEL_SMOOTH
);
1138 if (screen
->eng3d
->oclass
< NVE4_3D_CLASS
) {
1139 IMMED_NVC0(push
, NVC0_3D(TEX_MISC
), 0);
1141 BEGIN_NVC0(push
, NVE4_3D(TEX_CB_INDEX
), 1);
1142 PUSH_DATA (push
, 15);
1144 BEGIN_NVC0(push
, NVC0_3D(CALL_LIMIT_LOG
), 1);
1145 PUSH_DATA (push
, 8); /* 128 */
1146 BEGIN_NVC0(push
, NVC0_3D(ZCULL_STATCTRS_ENABLE
), 1);
1147 PUSH_DATA (push
, 1);
1148 if (screen
->eng3d
->oclass
>= NVC1_3D_CLASS
) {
1149 BEGIN_NVC0(push
, NVC0_3D(CACHE_SPLIT
), 1);
1150 PUSH_DATA (push
, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1
);
1153 nvc0_magic_3d_init(push
, screen
->eng3d
->oclass
);
1155 ret
= nvc0_screen_resize_text_area(screen
, 1 << 19);
1157 FAIL_SCREEN_INIT("Error allocating TEXT area: %d\n", ret
);
1159 /* 6 user uniform areas, 6 driver areas, and 1 for the runout */
1160 ret
= nouveau_bo_new(dev
, NV_VRAM_DOMAIN(&screen
->base
), 1 << 12, 13 << 16, NULL
,
1161 &screen
->uniform_bo
);
1163 FAIL_SCREEN_INIT("Error allocating uniform BO: %d\n", ret
);
1165 PUSH_REFN (push
, screen
->uniform_bo
, NV_VRAM_DOMAIN(&screen
->base
) | NOUVEAU_BO_WR
);
1167 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
1168 BEGIN_NVC0(push
, NVC0_3D(CB_SIZE
), 3);
1169 PUSH_DATA (push
, 256);
1170 PUSH_DATAh(push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_RUNOUT_INFO
);
1171 PUSH_DATA (push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_RUNOUT_INFO
);
1172 BEGIN_1IC0(push
, NVC0_3D(CB_POS
), 5);
1173 PUSH_DATA (push
, 0);
1174 PUSH_DATAf(push
, 0.0f
);
1175 PUSH_DATAf(push
, 0.0f
);
1176 PUSH_DATAf(push
, 0.0f
);
1177 PUSH_DATAf(push
, 0.0f
);
1178 BEGIN_NVC0(push
, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH
), 2);
1179 PUSH_DATAh(push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_RUNOUT_INFO
);
1180 PUSH_DATA (push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_RUNOUT_INFO
);
1182 if (screen
->base
.drm
->version
>= 0x01000101) {
1183 ret
= nouveau_getparam(dev
, NOUVEAU_GETPARAM_GRAPH_UNITS
, &value
);
1185 FAIL_SCREEN_INIT("NOUVEAU_GETPARAM_GRAPH_UNITS failed: %d\n", ret
);
1187 if (dev
->chipset
>= 0xe0 && dev
->chipset
< 0xf0)
1188 value
= (8 << 8) | 4;
1190 value
= (16 << 8) | 4;
1192 screen
->gpc_count
= value
& 0x000000ff;
1193 screen
->mp_count
= value
>> 8;
1194 screen
->mp_count_compute
= screen
->mp_count
;
1196 ret
= nvc0_screen_resize_tls_area(screen
, 128 * 16, 0, 0x200);
1198 FAIL_SCREEN_INIT("Error allocating TLS area: %d\n", ret
);
1200 BEGIN_NVC0(push
, NVC0_3D(TEMP_ADDRESS_HIGH
), 4);
1201 PUSH_DATAh(push
, screen
->tls
->offset
);
1202 PUSH_DATA (push
, screen
->tls
->offset
);
1203 PUSH_DATA (push
, screen
->tls
->size
>> 32);
1204 PUSH_DATA (push
, screen
->tls
->size
);
1205 BEGIN_NVC0(push
, NVC0_3D(WARP_TEMP_ALLOC
), 1);
1206 PUSH_DATA (push
, 0);
1207 /* Reduce likelihood of collision with real buffers by placing the hole at
1208 * the top of the 4G area. This will have to be dealt with for real
1209 * eventually by blocking off that area from the VM.
1211 BEGIN_NVC0(push
, NVC0_3D(LOCAL_BASE
), 1);
1212 PUSH_DATA (push
, 0xff << 24);
1214 if (screen
->eng3d
->oclass
< GM107_3D_CLASS
) {
1215 ret
= nouveau_bo_new(dev
, NV_VRAM_DOMAIN(&screen
->base
), 1 << 17, 1 << 20, NULL
,
1216 &screen
->poly_cache
);
1218 FAIL_SCREEN_INIT("Error allocating poly cache BO: %d\n", ret
);
1220 BEGIN_NVC0(push
, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH
), 3);
1221 PUSH_DATAh(push
, screen
->poly_cache
->offset
);
1222 PUSH_DATA (push
, screen
->poly_cache
->offset
);
1223 PUSH_DATA (push
, 3);
1226 ret
= nouveau_bo_new(dev
, NV_VRAM_DOMAIN(&screen
->base
), 1 << 17, 1 << 17, NULL
,
1229 FAIL_SCREEN_INIT("Error allocating txc BO: %d\n", ret
);
1231 BEGIN_NVC0(push
, NVC0_3D(TIC_ADDRESS_HIGH
), 3);
1232 PUSH_DATAh(push
, screen
->txc
->offset
);
1233 PUSH_DATA (push
, screen
->txc
->offset
);
1234 PUSH_DATA (push
, NVC0_TIC_MAX_ENTRIES
- 1);
1235 if (screen
->eng3d
->oclass
>= GM107_3D_CLASS
) {
1236 screen
->tic
.maxwell
= true;
1237 if (screen
->eng3d
->oclass
== GM107_3D_CLASS
) {
1238 screen
->tic
.maxwell
=
1239 debug_get_bool_option("NOUVEAU_MAXWELL_TIC", true);
1240 IMMED_NVC0(push
, SUBC_3D(0x0f10), screen
->tic
.maxwell
);
1244 BEGIN_NVC0(push
, NVC0_3D(TSC_ADDRESS_HIGH
), 3);
1245 PUSH_DATAh(push
, screen
->txc
->offset
+ 65536);
1246 PUSH_DATA (push
, screen
->txc
->offset
+ 65536);
1247 PUSH_DATA (push
, NVC0_TSC_MAX_ENTRIES
- 1);
1249 BEGIN_NVC0(push
, NVC0_3D(SCREEN_Y_CONTROL
), 1);
1250 PUSH_DATA (push
, 0);
1251 BEGIN_NVC0(push
, NVC0_3D(WINDOW_OFFSET_X
), 2);
1252 PUSH_DATA (push
, 0);
1253 PUSH_DATA (push
, 0);
1254 BEGIN_NVC0(push
, NVC0_3D(ZCULL_REGION
), 1); /* deactivate ZCULL */
1255 PUSH_DATA (push
, 0x3f);
1257 BEGIN_NVC0(push
, NVC0_3D(CLIP_RECTS_MODE
), 1);
1258 PUSH_DATA (push
, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY
);
1259 BEGIN_NVC0(push
, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
1260 for (i
= 0; i
< 8 * 2; ++i
)
1262 BEGIN_NVC0(push
, NVC0_3D(CLIP_RECTS_EN
), 1);
1263 PUSH_DATA (push
, 0);
1264 BEGIN_NVC0(push
, NVC0_3D(CLIPID_ENABLE
), 1);
1265 PUSH_DATA (push
, 0);
1267 /* neither scissors, viewport nor stencil mask should affect clears */
1268 BEGIN_NVC0(push
, NVC0_3D(CLEAR_FLAGS
), 1);
1269 PUSH_DATA (push
, 0);
1271 BEGIN_NVC0(push
, NVC0_3D(VIEWPORT_TRANSFORM_EN
), 1);
1272 PUSH_DATA (push
, 1);
1273 for (i
= 0; i
< NVC0_MAX_VIEWPORTS
; i
++) {
1274 BEGIN_NVC0(push
, NVC0_3D(DEPTH_RANGE_NEAR(i
)), 2);
1275 PUSH_DATAf(push
, 0.0f
);
1276 PUSH_DATAf(push
, 1.0f
);
1278 BEGIN_NVC0(push
, NVC0_3D(VIEW_VOLUME_CLIP_CTRL
), 1);
1279 PUSH_DATA (push
, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1
);
1281 /* We use scissors instead of exact view volume clipping,
1282 * so they're always enabled.
1284 for (i
= 0; i
< NVC0_MAX_VIEWPORTS
; i
++) {
1285 BEGIN_NVC0(push
, NVC0_3D(SCISSOR_ENABLE(i
)), 3);
1286 PUSH_DATA (push
, 1);
1287 PUSH_DATA (push
, 8192 << 16);
1288 PUSH_DATA (push
, 8192 << 16);
1291 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
1294 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE
, mme9097_per_instance_bf
);
1295 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES
, mme9097_blend_enables
);
1296 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT
, mme9097_vertex_array_select
);
1297 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT
, mme9097_tep_select
);
1298 MK_MACRO(NVC0_3D_MACRO_GP_SELECT
, mme9097_gp_select
);
1299 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT
, mme9097_poly_mode_front
);
1300 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK
, mme9097_poly_mode_back
);
1301 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT
, mme9097_draw_arrays_indirect
);
1302 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT
, mme9097_draw_elts_indirect
);
1303 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT
, mme9097_draw_arrays_indirect_count
);
1304 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT
, mme9097_draw_elts_indirect_count
);
1305 MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE
, mme9097_query_buffer_write
);
1306 MK_MACRO(NVC0_3D_MACRO_CONSERVATIVE_RASTER_STATE
, mme9097_conservative_raster_state
);
1307 MK_MACRO(NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT
, mme90c0_launch_grid_indirect
);
1309 BEGIN_NVC0(push
, NVC0_3D(RASTERIZE_ENABLE
), 1);
1310 PUSH_DATA (push
, 1);
1311 BEGIN_NVC0(push
, NVC0_3D(RT_SEPARATE_FRAG_DATA
), 1);
1312 PUSH_DATA (push
, 1);
1313 BEGIN_NVC0(push
, NVC0_3D(MACRO_GP_SELECT
), 1);
1314 PUSH_DATA (push
, 0x40);
1315 BEGIN_NVC0(push
, NVC0_3D(LAYER
), 1);
1316 PUSH_DATA (push
, 0);
1317 BEGIN_NVC0(push
, NVC0_3D(MACRO_TEP_SELECT
), 1);
1318 PUSH_DATA (push
, 0x30);
1319 BEGIN_NVC0(push
, NVC0_3D(PATCH_VERTICES
), 1);
1320 PUSH_DATA (push
, 3);
1321 BEGIN_NVC0(push
, NVC0_3D(SP_SELECT(2)), 1);
1322 PUSH_DATA (push
, 0x20);
1323 BEGIN_NVC0(push
, NVC0_3D(SP_SELECT(0)), 1);
1324 PUSH_DATA (push
, 0x00);
1325 screen
->save_state
.patch_vertices
= 3;
1327 BEGIN_NVC0(push
, NVC0_3D(POINT_COORD_REPLACE
), 1);
1328 PUSH_DATA (push
, 0);
1329 BEGIN_NVC0(push
, NVC0_3D(POINT_RASTER_RULES
), 1);
1330 PUSH_DATA (push
, NVC0_3D_POINT_RASTER_RULES_OGL
);
1332 IMMED_NVC0(push
, NVC0_3D(EDGEFLAG
), 1);
1334 if (nvc0_screen_init_compute(screen
))
1337 /* XXX: Compute and 3D are somehow aliased on Fermi. */
1338 for (i
= 0; i
< 5; ++i
) {
1340 for (j
= 0; j
< 16; j
++)
1341 screen
->cb_bindings
[i
][j
].size
= -1;
1343 /* TIC and TSC entries for each unit (nve4+ only) */
1344 /* auxiliary constants (6 user clip planes, base instance id) */
1345 nvc0_screen_bind_cb_3d(screen
, NULL
, i
, 15, NVC0_CB_AUX_SIZE
,
1346 screen
->uniform_bo
->offset
+ NVC0_CB_AUX_INFO(i
));
1347 if (screen
->eng3d
->oclass
>= NVE4_3D_CLASS
) {
1349 BEGIN_1IC0(push
, NVC0_3D(CB_POS
), 9);
1350 PUSH_DATA (push
, NVC0_CB_AUX_UNK_INFO
);
1351 for (j
= 0; j
< 8; ++j
)
1354 BEGIN_NVC0(push
, NVC0_3D(TEX_LIMITS(i
)), 1);
1355 PUSH_DATA (push
, 0x54);
1358 /* MS sample coordinate offsets: these do not work with _ALT modes ! */
1359 BEGIN_1IC0(push
, NVC0_3D(CB_POS
), 1 + 2 * 8);
1360 PUSH_DATA (push
, NVC0_CB_AUX_MS_INFO
);
1361 PUSH_DATA (push
, 0); /* 0 */
1362 PUSH_DATA (push
, 0);
1363 PUSH_DATA (push
, 1); /* 1 */
1364 PUSH_DATA (push
, 0);
1365 PUSH_DATA (push
, 0); /* 2 */
1366 PUSH_DATA (push
, 1);
1367 PUSH_DATA (push
, 1); /* 3 */
1368 PUSH_DATA (push
, 1);
1369 PUSH_DATA (push
, 2); /* 4 */
1370 PUSH_DATA (push
, 0);
1371 PUSH_DATA (push
, 3); /* 5 */
1372 PUSH_DATA (push
, 0);
1373 PUSH_DATA (push
, 2); /* 6 */
1374 PUSH_DATA (push
, 1);
1375 PUSH_DATA (push
, 3); /* 7 */
1376 PUSH_DATA (push
, 1);
1378 BEGIN_NVC0(push
, NVC0_3D(LINKED_TSC
), 1);
1379 PUSH_DATA (push
, 0);
1383 screen
->tic
.entries
= CALLOC(
1384 NVC0_TIC_MAX_ENTRIES
+ NVC0_TSC_MAX_ENTRIES
+ NVE4_IMG_MAX_HANDLES
,
1386 screen
->tsc
.entries
= screen
->tic
.entries
+ NVC0_TIC_MAX_ENTRIES
;
1387 screen
->img
.entries
= (void *)(screen
->tsc
.entries
+ NVC0_TSC_MAX_ENTRIES
);
1389 if (!nvc0_blitter_create(screen
))
1392 nouveau_fence_new(&screen
->base
, &screen
->base
.fence
.current
);
1394 return &screen
->base
;
1397 screen
->base
.base
.context_create
= NULL
;
1398 return &screen
->base
;
1402 nvc0_screen_tic_alloc(struct nvc0_screen
*screen
, void *entry
)
1404 int i
= screen
->tic
.next
;
1406 while (screen
->tic
.lock
[i
/ 32] & (1 << (i
% 32)))
1407 i
= (i
+ 1) & (NVC0_TIC_MAX_ENTRIES
- 1);
1409 screen
->tic
.next
= (i
+ 1) & (NVC0_TIC_MAX_ENTRIES
- 1);
1411 if (screen
->tic
.entries
[i
])
1412 nv50_tic_entry(screen
->tic
.entries
[i
])->id
= -1;
1414 screen
->tic
.entries
[i
] = entry
;
1419 nvc0_screen_tsc_alloc(struct nvc0_screen
*screen
, void *entry
)
1421 int i
= screen
->tsc
.next
;
1423 while (screen
->tsc
.lock
[i
/ 32] & (1 << (i
% 32)))
1424 i
= (i
+ 1) & (NVC0_TSC_MAX_ENTRIES
- 1);
1426 screen
->tsc
.next
= (i
+ 1) & (NVC0_TSC_MAX_ENTRIES
- 1);
1428 if (screen
->tsc
.entries
[i
])
1429 nv50_tsc_entry(screen
->tsc
.entries
[i
])->id
= -1;
1431 screen
->tsc
.entries
[i
] = entry
;