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 "pipe/p_screen.h"
30 #include "nouveau_vp3_video.h"
32 #include "nvc0/nvc0_context.h"
33 #include "nvc0/nvc0_screen.h"
35 #include "nvc0/mme/com9097.mme.h"
36 #include "nvc0/mme/com90c0.mme.h"
38 #include "nv50/g80_texture.xml.h"
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
,
47 const struct util_format_description
*desc
= util_format_description(format
);
51 if (!(0x117 & (1 << sample_count
))) /* 0, 1, 2, 4 or 8 */
54 /* Short-circuit the rest of the logic -- this is used by the state tracker
55 * to determine valid MS levels in a no-attachments scenario.
57 if (format
== PIPE_FORMAT_NONE
&& bindings
& PIPE_BIND_RENDER_TARGET
)
60 if (!util_format_is_supported(format
, bindings
))
63 if ((bindings
& PIPE_BIND_SAMPLER_VIEW
) && (target
!= PIPE_BUFFER
))
64 if (util_format_get_blocksizebits(format
) == 3 * 32)
67 if (bindings
& PIPE_BIND_LINEAR
)
68 if (util_format_is_depth_or_stencil(format
) ||
69 (target
!= PIPE_TEXTURE_1D
&&
70 target
!= PIPE_TEXTURE_2D
&&
71 target
!= PIPE_TEXTURE_RECT
) ||
75 /* Restrict ETC2 and ASTC formats here. These are only supported on GK20A.
77 if ((desc
->layout
== UTIL_FORMAT_LAYOUT_ETC
||
78 desc
->layout
== UTIL_FORMAT_LAYOUT_ASTC
) &&
79 /* The claim is that this should work on GM107 but it doesn't. Need to
80 * test further and figure out if it's a nouveau issue or a HW one.
81 nouveau_screen(pscreen)->class_3d < GM107_3D_CLASS &&
83 nouveau_screen(pscreen
)->class_3d
!= NVEA_3D_CLASS
)
86 /* shared is always supported */
87 bindings
&= ~(PIPE_BIND_LINEAR
|
90 if (bindings
& PIPE_BIND_SHADER_IMAGE
) {
91 if (sample_count
> 1 &&
92 nouveau_screen(pscreen
)->class_3d
>= GM107_3D_CLASS
) {
93 /* MS images are currently unsupported on Maxwell because they have to
94 * be handled explicitly. */
98 if (format
== PIPE_FORMAT_B8G8R8A8_UNORM
&&
99 nouveau_screen(pscreen
)->class_3d
< NVE4_3D_CLASS
) {
100 /* This should work on Fermi, but for currently unknown reasons it
101 * does not and results in breaking reads from pbos. */
106 return (( nvc0_format_table
[format
].usage
|
107 nvc0_vertex_format
[format
].usage
) & bindings
) == bindings
;
111 nvc0_screen_get_param(struct pipe_screen
*pscreen
, enum pipe_cap param
)
113 const uint16_t class_3d
= nouveau_screen(pscreen
)->class_3d
;
114 struct nouveau_device
*dev
= nouveau_screen(pscreen
)->device
;
117 /* non-boolean caps */
118 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS
:
119 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS
:
121 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS
:
123 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS
:
125 case PIPE_CAP_MIN_TEXEL_OFFSET
:
127 case PIPE_CAP_MAX_TEXEL_OFFSET
:
129 case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET
:
131 case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET
:
133 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE
:
134 return 128 * 1024 * 1024;
135 case PIPE_CAP_GLSL_FEATURE_LEVEL
:
137 case PIPE_CAP_MAX_RENDER_TARGETS
:
139 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS
:
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_VERTEX_ATTRIB_STRIDE
:
153 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT
:
155 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT
:
156 if (class_3d
< GM107_3D_CLASS
)
157 return 256; /* IMAGE bindings require alignment to 256 */
159 case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT
:
161 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT
:
162 return NOUVEAU_MIN_BUFFER_MAP_ALIGN
;
163 case PIPE_CAP_MAX_VIEWPORTS
:
164 return NVC0_MAX_VIEWPORTS
;
165 case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS
:
167 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK
:
168 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50
;
169 case PIPE_CAP_ENDIANNESS
:
170 return PIPE_ENDIAN_LITTLE
;
171 case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS
:
173 case PIPE_CAP_MAX_WINDOW_RECTANGLES
:
174 return NVC0_MAX_WINDOW_RECTANGLES
;
177 case PIPE_CAP_TEXTURE_MIRROR_CLAMP
:
178 case PIPE_CAP_TEXTURE_SWIZZLE
:
179 case PIPE_CAP_TEXTURE_SHADOW_MAP
:
180 case PIPE_CAP_NPOT_TEXTURES
:
181 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES
:
182 case PIPE_CAP_MIXED_COLOR_DEPTH_BITS
:
183 case PIPE_CAP_ANISOTROPIC_FILTER
:
184 case PIPE_CAP_SEAMLESS_CUBE_MAP
:
185 case PIPE_CAP_CUBE_MAP_ARRAY
:
186 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS
:
187 case PIPE_CAP_TEXTURE_MULTISAMPLE
:
188 case PIPE_CAP_TWO_SIDED_STENCIL
:
189 case PIPE_CAP_DEPTH_CLIP_DISABLE
:
190 case PIPE_CAP_POINT_SPRITE
:
191 case PIPE_CAP_TGSI_TEXCOORD
:
193 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED
:
194 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED
:
195 case PIPE_CAP_VERTEX_COLOR_CLAMPED
:
196 case PIPE_CAP_QUERY_TIMESTAMP
:
197 case PIPE_CAP_QUERY_TIME_ELAPSED
:
198 case PIPE_CAP_OCCLUSION_QUERY
:
199 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME
:
200 case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS
:
201 case PIPE_CAP_QUERY_PIPELINE_STATISTICS
:
202 case PIPE_CAP_BLEND_EQUATION_SEPARATE
:
203 case PIPE_CAP_INDEP_BLEND_ENABLE
:
204 case PIPE_CAP_INDEP_BLEND_FUNC
:
205 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT
:
206 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER
:
207 case PIPE_CAP_PRIMITIVE_RESTART
:
208 case PIPE_CAP_TGSI_INSTANCEID
:
209 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR
:
210 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS
:
211 case PIPE_CAP_CONDITIONAL_RENDER
:
212 case PIPE_CAP_TEXTURE_BARRIER
:
213 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION
:
214 case PIPE_CAP_START_INSTANCE
:
215 case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT
:
216 case PIPE_CAP_DRAW_INDIRECT
:
217 case PIPE_CAP_USER_CONSTANT_BUFFERS
:
218 case PIPE_CAP_USER_VERTEX_BUFFERS
:
219 case PIPE_CAP_TEXTURE_QUERY_LOD
:
220 case PIPE_CAP_SAMPLE_SHADING
:
221 case PIPE_CAP_TEXTURE_GATHER_OFFSETS
:
222 case PIPE_CAP_TEXTURE_GATHER_SM5
:
223 case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE
:
224 case PIPE_CAP_CONDITIONAL_RENDER_INVERTED
:
225 case PIPE_CAP_SAMPLER_VIEW_TARGET
:
226 case PIPE_CAP_CLIP_HALFZ
:
227 case PIPE_CAP_POLYGON_OFFSET_CLAMP
:
228 case PIPE_CAP_MULTISAMPLE_Z_RESOLVE
:
229 case PIPE_CAP_TEXTURE_FLOAT_LINEAR
:
230 case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR
:
231 case PIPE_CAP_DEPTH_BOUNDS_TEST
:
232 case PIPE_CAP_TGSI_TXQS
:
233 case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS
:
234 case PIPE_CAP_FORCE_PERSAMPLE_INTERP
:
235 case PIPE_CAP_SHAREABLE_SHADERS
:
236 case PIPE_CAP_CLEAR_TEXTURE
:
237 case PIPE_CAP_DRAW_PARAMETERS
:
238 case PIPE_CAP_TGSI_PACK_HALF_FLOAT
:
239 case PIPE_CAP_MULTI_DRAW_INDIRECT
:
240 case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS
:
241 case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL
:
242 case PIPE_CAP_QUERY_BUFFER_OBJECT
:
243 case PIPE_CAP_INVALIDATE_BUFFER
:
244 case PIPE_CAP_STRING_MARKER
:
245 case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT
:
246 case PIPE_CAP_CULL_DISTANCE
:
247 case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES
:
248 case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR
:
249 case PIPE_CAP_TGSI_VOTE
:
250 case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED
:
251 case PIPE_CAP_TGSI_ARRAY_COMPONENTS
:
252 case PIPE_CAP_TGSI_MUL_ZERO_WINS
:
253 case PIPE_CAP_DOUBLES
:
255 case PIPE_CAP_TGSI_TEX_TXF_LZ
:
256 case PIPE_CAP_TGSI_CLOCK
:
257 case PIPE_CAP_COMPUTE
:
258 case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX
:
259 case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION
:
261 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE
:
262 return (class_3d
>= NVE4_3D_CLASS
) ? 1 : 0;
263 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER
:
264 return nouveau_screen(pscreen
)->vram_domain
& NOUVEAU_BO_VRAM
? 1 : 0;
265 case PIPE_CAP_TGSI_FS_FBFETCH
:
266 return class_3d
>= NVE4_3D_CLASS
; /* needs testing on fermi */
267 case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE
:
268 case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT
:
269 case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT
:
270 case PIPE_CAP_POST_DEPTH_COVERAGE
:
271 return class_3d
>= GM200_3D_CLASS
;
272 case PIPE_CAP_TGSI_BALLOT
:
273 return class_3d
>= NVE4_3D_CLASS
;
275 /* unsupported caps */
276 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT
:
277 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER
:
278 case PIPE_CAP_SHADER_STENCIL_EXPORT
:
279 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS
:
280 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY
:
281 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY
:
282 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY
:
283 case PIPE_CAP_FAKE_SW_MSAA
:
284 case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION
:
285 case PIPE_CAP_VERTEXID_NOBASE
:
286 case PIPE_CAP_RESOURCE_FROM_USER_MEMORY
:
287 case PIPE_CAP_DEVICE_RESET_STATUS_QUERY
:
288 case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL
:
289 case PIPE_CAP_GENERATE_MIPMAP
:
290 case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY
:
291 case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS
:
292 case PIPE_CAP_QUERY_MEMORY_INFO
:
293 case PIPE_CAP_PCI_GROUP
:
294 case PIPE_CAP_PCI_BUS
:
295 case PIPE_CAP_PCI_DEVICE
:
296 case PIPE_CAP_PCI_FUNCTION
:
297 case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS
:
298 case PIPE_CAP_TGSI_CAN_READ_OUTPUTS
:
299 case PIPE_CAP_NATIVE_FENCE_FD
:
300 case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY
:
301 case PIPE_CAP_INT64_DIVMOD
:
302 case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE
:
303 case PIPE_CAP_BINDLESS_TEXTURE
:
304 case PIPE_CAP_NIR_SAMPLERS_AS_DEREF
:
305 case PIPE_CAP_QUERY_SO_OVERFLOW
:
306 case PIPE_CAP_MEMOBJ
:
307 case PIPE_CAP_LOAD_CONSTBUF
:
308 case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS
:
309 case PIPE_CAP_TILE_RASTER_ORDER
:
310 case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES
:
311 case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET
:
314 case PIPE_CAP_VENDOR_ID
:
316 case PIPE_CAP_DEVICE_ID
: {
318 if (nouveau_getparam(dev
, NOUVEAU_GETPARAM_PCI_DEVICE
, &device_id
)) {
319 NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
324 case PIPE_CAP_ACCELERATED
:
326 case PIPE_CAP_VIDEO_MEMORY
:
327 return dev
->vram_size
>> 20;
332 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param
);
337 nvc0_screen_get_shader_param(struct pipe_screen
*pscreen
,
338 enum pipe_shader_type shader
,
339 enum pipe_shader_cap param
)
341 const uint16_t class_3d
= nouveau_screen(pscreen
)->class_3d
;
344 case PIPE_SHADER_VERTEX
:
345 case PIPE_SHADER_GEOMETRY
:
346 case PIPE_SHADER_FRAGMENT
:
347 case PIPE_SHADER_COMPUTE
:
348 case PIPE_SHADER_TESS_CTRL
:
349 case PIPE_SHADER_TESS_EVAL
:
356 case PIPE_SHADER_CAP_PREFERRED_IR
:
357 return PIPE_SHADER_IR_TGSI
;
358 case PIPE_SHADER_CAP_SUPPORTED_IRS
:
359 return 1 << PIPE_SHADER_IR_TGSI
;
360 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS
:
361 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS
:
362 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS
:
363 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS
:
365 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH
:
367 case PIPE_SHADER_CAP_MAX_INPUTS
:
368 if (shader
== PIPE_SHADER_VERTEX
)
370 /* NOTE: These only count our slots for GENERIC varyings.
371 * The address space may be larger, but the actual hard limit seems to be
372 * less than what the address space layout permits, so don't add TEXCOORD,
375 if (shader
== PIPE_SHADER_FRAGMENT
)
377 /* Actually this counts CLIPVERTEX, which occupies the last generic slot,
378 * and excludes 0x60 per-patch inputs.
381 case PIPE_SHADER_CAP_MAX_OUTPUTS
:
383 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE
:
385 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS
:
386 return NVC0_MAX_PIPE_CONSTBUFS
;
387 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR
:
388 return shader
!= PIPE_SHADER_FRAGMENT
;
389 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR
:
390 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR
:
391 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR
:
393 case PIPE_SHADER_CAP_MAX_TEMPS
:
394 return NVC0_CAP_MAX_PROGRAM_TEMPS
;
395 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED
:
397 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED
:
399 case PIPE_SHADER_CAP_SUBROUTINES
:
401 case PIPE_SHADER_CAP_INTEGERS
:
403 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED
:
405 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED
:
407 case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS
:
409 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED
:
410 case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED
:
411 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE
:
412 case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD
:
413 case PIPE_SHADER_CAP_INT64_ATOMICS
:
414 case PIPE_SHADER_CAP_FP16
:
416 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS
:
417 return NVC0_MAX_BUFFERS
;
418 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS
:
419 return (class_3d
>= NVE4_3D_CLASS
) ? 32 : 16;
420 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS
:
421 return (class_3d
>= NVE4_3D_CLASS
) ? 32 : 16;
422 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT
:
424 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES
:
425 if (class_3d
>= NVE4_3D_CLASS
)
426 return NVC0_MAX_IMAGES
;
427 if (shader
== PIPE_SHADER_FRAGMENT
|| shader
== PIPE_SHADER_COMPUTE
)
428 return NVC0_MAX_IMAGES
;
431 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param
);
437 nvc0_screen_get_paramf(struct pipe_screen
*pscreen
, enum pipe_capf param
)
440 case PIPE_CAPF_MAX_LINE_WIDTH
:
441 case PIPE_CAPF_MAX_LINE_WIDTH_AA
:
443 case PIPE_CAPF_MAX_POINT_WIDTH
:
445 case PIPE_CAPF_MAX_POINT_WIDTH_AA
:
447 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY
:
449 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS
:
451 case PIPE_CAPF_GUARD_BAND_LEFT
:
452 case PIPE_CAPF_GUARD_BAND_TOP
:
454 case PIPE_CAPF_GUARD_BAND_RIGHT
:
455 case PIPE_CAPF_GUARD_BAND_BOTTOM
:
456 return 0.0f
; /* that or infinity */
459 NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param
);
464 nvc0_screen_get_compute_param(struct pipe_screen
*pscreen
,
465 enum pipe_shader_ir ir_type
,
466 enum pipe_compute_cap param
, void *data
)
468 struct nvc0_screen
*screen
= nvc0_screen(pscreen
);
469 const uint16_t obj_class
= screen
->compute
->oclass
;
471 #define RET(x) do { \
473 memcpy(data, x, sizeof(x)); \
478 case PIPE_COMPUTE_CAP_GRID_DIMENSION
:
479 RET((uint64_t []) { 3 });
480 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE
:
481 if (obj_class
>= NVE4_COMPUTE_CLASS
) {
482 RET(((uint64_t []) { 0x7fffffff, 65535, 65535 }));
484 RET(((uint64_t []) { 65535, 65535, 65535 }));
486 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE
:
487 RET(((uint64_t []) { 1024, 1024, 64 }));
488 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK
:
489 RET((uint64_t []) { 1024 });
490 case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK
:
491 if (obj_class
>= NVE4_COMPUTE_CLASS
) {
492 RET((uint64_t []) { 1024 });
494 RET((uint64_t []) { 512 });
496 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE
: /* g[] */
497 RET((uint64_t []) { 1ULL << 40 });
498 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE
: /* s[] */
500 case GM200_COMPUTE_CLASS
:
501 RET((uint64_t []) { 96 << 10 });
503 case GM107_COMPUTE_CLASS
:
504 RET((uint64_t []) { 64 << 10 });
507 RET((uint64_t []) { 48 << 10 });
510 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE
: /* l[] */
511 RET((uint64_t []) { 512 << 10 });
512 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE
: /* c[], arbitrary limit */
513 RET((uint64_t []) { 4096 });
514 case PIPE_COMPUTE_CAP_SUBGROUP_SIZE
:
515 RET((uint32_t []) { 32 });
516 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE
:
517 RET((uint64_t []) { 1ULL << 40 });
518 case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED
:
519 RET((uint32_t []) { 0 });
520 case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS
:
521 RET((uint32_t []) { screen
->mp_count_compute
});
522 case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY
:
523 RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
524 case PIPE_COMPUTE_CAP_ADDRESS_BITS
:
525 RET((uint32_t []) { 64 });
534 nvc0_screen_destroy(struct pipe_screen
*pscreen
)
536 struct nvc0_screen
*screen
= nvc0_screen(pscreen
);
538 if (!nouveau_drm_screen_unref(&screen
->base
))
541 if (screen
->base
.fence
.current
) {
542 struct nouveau_fence
*current
= NULL
;
544 /* nouveau_fence_wait will create a new current fence, so wait on the
545 * _current_ one, and remove both.
547 nouveau_fence_ref(screen
->base
.fence
.current
, ¤t
);
548 nouveau_fence_wait(current
, NULL
);
549 nouveau_fence_ref(NULL
, ¤t
);
550 nouveau_fence_ref(NULL
, &screen
->base
.fence
.current
);
552 if (screen
->base
.pushbuf
)
553 screen
->base
.pushbuf
->user_priv
= NULL
;
556 nvc0_blitter_destroy(screen
);
557 if (screen
->pm
.prog
) {
558 screen
->pm
.prog
->code
= NULL
; /* hardcoded, don't FREE */
559 nvc0_program_destroy(NULL
, screen
->pm
.prog
);
560 FREE(screen
->pm
.prog
);
563 nouveau_bo_ref(NULL
, &screen
->text
);
564 nouveau_bo_ref(NULL
, &screen
->uniform_bo
);
565 nouveau_bo_ref(NULL
, &screen
->tls
);
566 nouveau_bo_ref(NULL
, &screen
->txc
);
567 nouveau_bo_ref(NULL
, &screen
->fence
.bo
);
568 nouveau_bo_ref(NULL
, &screen
->poly_cache
);
570 nouveau_heap_destroy(&screen
->lib_code
);
571 nouveau_heap_destroy(&screen
->text_heap
);
573 FREE(screen
->default_tsc
);
574 FREE(screen
->tic
.entries
);
576 nouveau_object_del(&screen
->eng3d
);
577 nouveau_object_del(&screen
->eng2d
);
578 nouveau_object_del(&screen
->m2mf
);
579 nouveau_object_del(&screen
->compute
);
580 nouveau_object_del(&screen
->nvsw
);
582 nouveau_screen_fini(&screen
->base
);
588 nvc0_graph_set_macro(struct nvc0_screen
*screen
, uint32_t m
, unsigned pos
,
589 unsigned size
, const uint32_t *data
)
591 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
595 assert((pos
+ size
) <= 0x800);
597 BEGIN_NVC0(push
, SUBC_3D(NVC0_GRAPH_MACRO_ID
), 2);
598 PUSH_DATA (push
, (m
- 0x3800) / 8);
599 PUSH_DATA (push
, pos
);
600 BEGIN_1IC0(push
, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS
), size
+ 1);
601 PUSH_DATA (push
, pos
);
602 PUSH_DATAp(push
, data
, size
);
608 nvc0_magic_3d_init(struct nouveau_pushbuf
*push
, uint16_t obj_class
)
610 BEGIN_NVC0(push
, SUBC_3D(0x10cc), 1);
611 PUSH_DATA (push
, 0xff);
612 BEGIN_NVC0(push
, SUBC_3D(0x10e0), 2);
613 PUSH_DATA (push
, 0xff);
614 PUSH_DATA (push
, 0xff);
615 BEGIN_NVC0(push
, SUBC_3D(0x10ec), 2);
616 PUSH_DATA (push
, 0xff);
617 PUSH_DATA (push
, 0xff);
618 BEGIN_NVC0(push
, SUBC_3D(0x074c), 1);
619 PUSH_DATA (push
, 0x3f);
621 BEGIN_NVC0(push
, SUBC_3D(0x16a8), 1);
622 PUSH_DATA (push
, (3 << 16) | 3);
623 BEGIN_NVC0(push
, SUBC_3D(0x1794), 1);
624 PUSH_DATA (push
, (2 << 16) | 2);
626 if (obj_class
< GM107_3D_CLASS
) {
627 BEGIN_NVC0(push
, SUBC_3D(0x12ac), 1);
630 BEGIN_NVC0(push
, SUBC_3D(0x0218), 1);
631 PUSH_DATA (push
, 0x10);
632 BEGIN_NVC0(push
, SUBC_3D(0x10fc), 1);
633 PUSH_DATA (push
, 0x10);
634 BEGIN_NVC0(push
, SUBC_3D(0x1290), 1);
635 PUSH_DATA (push
, 0x10);
636 BEGIN_NVC0(push
, SUBC_3D(0x12d8), 2);
637 PUSH_DATA (push
, 0x10);
638 PUSH_DATA (push
, 0x10);
639 BEGIN_NVC0(push
, SUBC_3D(0x1140), 1);
640 PUSH_DATA (push
, 0x10);
641 BEGIN_NVC0(push
, SUBC_3D(0x1610), 1);
642 PUSH_DATA (push
, 0xe);
644 BEGIN_NVC0(push
, NVC0_3D(VERTEX_ID_GEN_MODE
), 1);
645 PUSH_DATA (push
, NVC0_3D_VERTEX_ID_GEN_MODE_DRAW_ARRAYS_ADD_START
);
646 BEGIN_NVC0(push
, SUBC_3D(0x030c), 1);
648 BEGIN_NVC0(push
, SUBC_3D(0x0300), 1);
651 BEGIN_NVC0(push
, SUBC_3D(0x02d0), 1);
652 PUSH_DATA (push
, 0x3fffff);
653 BEGIN_NVC0(push
, SUBC_3D(0x0fdc), 1);
655 BEGIN_NVC0(push
, SUBC_3D(0x19c0), 1);
658 if (obj_class
< GM107_3D_CLASS
) {
659 BEGIN_NVC0(push
, SUBC_3D(0x075c), 1);
662 if (obj_class
>= NVE4_3D_CLASS
) {
663 BEGIN_NVC0(push
, SUBC_3D(0x07fc), 1);
668 /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
669 * are supposed to do */
673 nvc0_screen_fence_emit(struct pipe_screen
*pscreen
, u32
*sequence
)
675 struct nvc0_screen
*screen
= nvc0_screen(pscreen
);
676 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
678 /* we need to do it after possible flush in MARK_RING */
679 *sequence
= ++screen
->base
.fence
.sequence
;
681 assert(PUSH_AVAIL(push
) + push
->rsvd_kick
>= 5);
682 PUSH_DATA (push
, NVC0_FIFO_PKHDR_SQ(NVC0_3D(QUERY_ADDRESS_HIGH
), 4));
683 PUSH_DATAh(push
, screen
->fence
.bo
->offset
);
684 PUSH_DATA (push
, screen
->fence
.bo
->offset
);
685 PUSH_DATA (push
, *sequence
);
686 PUSH_DATA (push
, NVC0_3D_QUERY_GET_FENCE
| NVC0_3D_QUERY_GET_SHORT
|
687 (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT
));
691 nvc0_screen_fence_update(struct pipe_screen
*pscreen
)
693 struct nvc0_screen
*screen
= nvc0_screen(pscreen
);
694 return screen
->fence
.map
[0];
698 nvc0_screen_init_compute(struct nvc0_screen
*screen
)
700 screen
->base
.base
.get_compute_param
= nvc0_screen_get_compute_param
;
702 switch (screen
->base
.device
->chipset
& ~0xf) {
705 return nvc0_screen_compute_setup(screen
, screen
->base
.pushbuf
);
712 return nve4_screen_compute_setup(screen
, screen
->base
.pushbuf
);
719 nvc0_screen_resize_tls_area(struct nvc0_screen
*screen
,
720 uint32_t lpos
, uint32_t lneg
, uint32_t cstack
)
722 struct nouveau_bo
*bo
= NULL
;
724 uint64_t size
= (lpos
+ lneg
) * 32 + cstack
;
726 if (size
>= (1 << 20)) {
727 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64
"\n", size
);
731 size
*= (screen
->base
.device
->chipset
>= 0xe0) ? 64 : 48; /* max warps */
732 size
= align(size
, 0x8000);
733 size
*= screen
->mp_count
;
735 size
= align(size
, 1 << 17);
737 ret
= nouveau_bo_new(screen
->base
.device
, NV_VRAM_DOMAIN(&screen
->base
), 1 << 17, size
,
741 nouveau_bo_ref(NULL
, &screen
->tls
);
747 nvc0_screen_resize_text_area(struct nvc0_screen
*screen
, uint64_t size
)
749 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
750 struct nouveau_bo
*bo
;
753 ret
= nouveau_bo_new(screen
->base
.device
, NV_VRAM_DOMAIN(&screen
->base
),
754 1 << 17, size
, NULL
, &bo
);
758 nouveau_bo_ref(NULL
, &screen
->text
);
761 nouveau_heap_destroy(&screen
->lib_code
);
762 nouveau_heap_destroy(&screen
->text_heap
);
764 /* XXX: getting a page fault at the end of the code buffer every few
765 * launches, don't use the last 256 bytes to work around them - prefetch ?
767 nouveau_heap_init(&screen
->text_heap
, 0, size
- 0x100);
769 /* update the code segment setup */
770 BEGIN_NVC0(push
, NVC0_3D(CODE_ADDRESS_HIGH
), 2);
771 PUSH_DATAh(push
, screen
->text
->offset
);
772 PUSH_DATA (push
, screen
->text
->offset
);
773 if (screen
->compute
) {
774 BEGIN_NVC0(push
, NVC0_CP(CODE_ADDRESS_HIGH
), 2);
775 PUSH_DATAh(push
, screen
->text
->offset
);
776 PUSH_DATA (push
, screen
->text
->offset
);
782 #define FAIL_SCREEN_INIT(str, err) \
784 NOUVEAU_ERR(str, err); \
788 struct nouveau_screen
*
789 nvc0_screen_create(struct nouveau_device
*dev
)
791 struct nvc0_screen
*screen
;
792 struct pipe_screen
*pscreen
;
793 struct nouveau_object
*chan
;
794 struct nouveau_pushbuf
*push
;
801 switch (dev
->chipset
& ~0xf) {
815 screen
= CALLOC_STRUCT(nvc0_screen
);
818 pscreen
= &screen
->base
.base
;
819 pscreen
->destroy
= nvc0_screen_destroy
;
821 ret
= nouveau_screen_init(&screen
->base
, dev
);
823 FAIL_SCREEN_INIT("Base screen init failed: %d\n", ret
);
824 chan
= screen
->base
.channel
;
825 push
= screen
->base
.pushbuf
;
826 push
->user_priv
= screen
;
829 screen
->base
.vidmem_bindings
|= PIPE_BIND_CONSTANT_BUFFER
|
830 PIPE_BIND_SHADER_BUFFER
|
831 PIPE_BIND_VERTEX_BUFFER
| PIPE_BIND_INDEX_BUFFER
|
832 PIPE_BIND_COMMAND_ARGS_BUFFER
| PIPE_BIND_QUERY_BUFFER
;
833 screen
->base
.sysmem_bindings
|=
834 PIPE_BIND_VERTEX_BUFFER
| PIPE_BIND_INDEX_BUFFER
;
836 if (screen
->base
.vram_domain
& NOUVEAU_BO_GART
) {
837 screen
->base
.sysmem_bindings
|= screen
->base
.vidmem_bindings
;
838 screen
->base
.vidmem_bindings
= 0;
841 pscreen
->context_create
= nvc0_create
;
842 pscreen
->is_format_supported
= nvc0_screen_is_format_supported
;
843 pscreen
->get_param
= nvc0_screen_get_param
;
844 pscreen
->get_shader_param
= nvc0_screen_get_shader_param
;
845 pscreen
->get_paramf
= nvc0_screen_get_paramf
;
846 pscreen
->get_driver_query_info
= nvc0_screen_get_driver_query_info
;
847 pscreen
->get_driver_query_group_info
= nvc0_screen_get_driver_query_group_info
;
849 nvc0_screen_init_resource_functions(pscreen
);
851 screen
->base
.base
.get_video_param
= nouveau_vp3_screen_get_video_param
;
852 screen
->base
.base
.is_video_format_supported
= nouveau_vp3_screen_video_supported
;
854 flags
= NOUVEAU_BO_GART
| NOUVEAU_BO_MAP
;
855 if (screen
->base
.drm
->version
>= 0x01000202)
856 flags
|= NOUVEAU_BO_COHERENT
;
858 ret
= nouveau_bo_new(dev
, flags
, 0, 4096, NULL
, &screen
->fence
.bo
);
860 FAIL_SCREEN_INIT("Error allocating fence BO: %d\n", ret
);
861 nouveau_bo_map(screen
->fence
.bo
, 0, NULL
);
862 screen
->fence
.map
= screen
->fence
.bo
->map
;
863 screen
->base
.fence
.emit
= nvc0_screen_fence_emit
;
864 screen
->base
.fence
.update
= nvc0_screen_fence_update
;
867 ret
= nouveau_object_new(chan
, (dev
->chipset
< 0xe0) ? 0x1f906e : 0x906e,
868 NVIF_CLASS_SW_GF100
, NULL
, 0, &screen
->nvsw
);
870 FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret
);
872 BEGIN_NVC0(push
, SUBC_SW(NV01_SUBCHAN_OBJECT
), 1);
873 PUSH_DATA (push
, screen
->nvsw
->handle
);
875 switch (dev
->chipset
& ~0xf) {
881 obj_class
= NVF0_P2MF_CLASS
;
884 obj_class
= NVE4_P2MF_CLASS
;
887 obj_class
= NVC0_M2MF_CLASS
;
890 ret
= nouveau_object_new(chan
, 0xbeef323f, obj_class
, NULL
, 0,
893 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret
);
895 BEGIN_NVC0(push
, SUBC_M2MF(NV01_SUBCHAN_OBJECT
), 1);
896 PUSH_DATA (push
, screen
->m2mf
->oclass
);
897 if (screen
->m2mf
->oclass
== NVE4_P2MF_CLASS
) {
898 BEGIN_NVC0(push
, SUBC_COPY(NV01_SUBCHAN_OBJECT
), 1);
899 PUSH_DATA (push
, 0xa0b5);
902 ret
= nouveau_object_new(chan
, 0xbeef902d, NVC0_2D_CLASS
, NULL
, 0,
905 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret
);
907 BEGIN_NVC0(push
, SUBC_2D(NV01_SUBCHAN_OBJECT
), 1);
908 PUSH_DATA (push
, screen
->eng2d
->oclass
);
909 BEGIN_NVC0(push
, SUBC_2D(NVC0_2D_SINGLE_GPC
), 1);
911 BEGIN_NVC0(push
, NVC0_2D(OPERATION
), 1);
912 PUSH_DATA (push
, NV50_2D_OPERATION_SRCCOPY
);
913 BEGIN_NVC0(push
, NVC0_2D(CLIP_ENABLE
), 1);
915 BEGIN_NVC0(push
, NVC0_2D(COLOR_KEY_ENABLE
), 1);
917 BEGIN_NVC0(push
, SUBC_2D(0x0884), 1);
918 PUSH_DATA (push
, 0x3f);
919 BEGIN_NVC0(push
, SUBC_2D(0x0888), 1);
921 BEGIN_NVC0(push
, NVC0_2D(COND_MODE
), 1);
922 PUSH_DATA (push
, NV50_2D_COND_MODE_ALWAYS
);
924 BEGIN_NVC0(push
, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH
), 2);
925 PUSH_DATAh(push
, screen
->fence
.bo
->offset
+ 16);
926 PUSH_DATA (push
, screen
->fence
.bo
->offset
+ 16);
928 switch (dev
->chipset
& ~0xf) {
930 switch (dev
->chipset
) {
933 obj_class
= GP100_3D_CLASS
;
936 obj_class
= GP102_3D_CLASS
;
941 obj_class
= GM200_3D_CLASS
;
944 obj_class
= GM107_3D_CLASS
;
948 obj_class
= NVF0_3D_CLASS
;
951 switch (dev
->chipset
) {
953 obj_class
= NVEA_3D_CLASS
;
956 obj_class
= NVE4_3D_CLASS
;
961 obj_class
= NVC8_3D_CLASS
;
965 switch (dev
->chipset
) {
967 obj_class
= NVC8_3D_CLASS
;
970 obj_class
= NVC1_3D_CLASS
;
973 obj_class
= NVC0_3D_CLASS
;
978 ret
= nouveau_object_new(chan
, 0xbeef003d, obj_class
, NULL
, 0,
981 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret
);
982 screen
->base
.class_3d
= obj_class
;
984 BEGIN_NVC0(push
, SUBC_3D(NV01_SUBCHAN_OBJECT
), 1);
985 PUSH_DATA (push
, screen
->eng3d
->oclass
);
987 BEGIN_NVC0(push
, NVC0_3D(COND_MODE
), 1);
988 PUSH_DATA (push
, NVC0_3D_COND_MODE_ALWAYS
);
990 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
991 /* kill shaders after about 1 second (at 100 MHz) */
992 BEGIN_NVC0(push
, NVC0_3D(WATCHDOG_TIMER
), 1);
993 PUSH_DATA (push
, 0x17);
996 IMMED_NVC0(push
, NVC0_3D(ZETA_COMP_ENABLE
),
997 screen
->base
.drm
->version
>= 0x01000101);
998 BEGIN_NVC0(push
, NVC0_3D(RT_COMP_ENABLE(0)), 8);
999 for (i
= 0; i
< 8; ++i
)
1000 PUSH_DATA(push
, screen
->base
.drm
->version
>= 0x01000101);
1002 BEGIN_NVC0(push
, NVC0_3D(RT_CONTROL
), 1);
1003 PUSH_DATA (push
, 1);
1005 BEGIN_NVC0(push
, NVC0_3D(CSAA_ENABLE
), 1);
1006 PUSH_DATA (push
, 0);
1007 BEGIN_NVC0(push
, NVC0_3D(MULTISAMPLE_ENABLE
), 1);
1008 PUSH_DATA (push
, 0);
1009 BEGIN_NVC0(push
, NVC0_3D(MULTISAMPLE_MODE
), 1);
1010 PUSH_DATA (push
, NVC0_3D_MULTISAMPLE_MODE_MS1
);
1011 BEGIN_NVC0(push
, NVC0_3D(MULTISAMPLE_CTRL
), 1);
1012 PUSH_DATA (push
, 0);
1013 BEGIN_NVC0(push
, NVC0_3D(LINE_WIDTH_SEPARATE
), 1);
1014 PUSH_DATA (push
, 1);
1015 BEGIN_NVC0(push
, NVC0_3D(PRIM_RESTART_WITH_DRAW_ARRAYS
), 1);
1016 PUSH_DATA (push
, 1);
1017 BEGIN_NVC0(push
, NVC0_3D(BLEND_SEPARATE_ALPHA
), 1);
1018 PUSH_DATA (push
, 1);
1019 BEGIN_NVC0(push
, NVC0_3D(BLEND_ENABLE_COMMON
), 1);
1020 PUSH_DATA (push
, 0);
1021 BEGIN_NVC0(push
, NVC0_3D(SHADE_MODEL
), 1);
1022 PUSH_DATA (push
, NVC0_3D_SHADE_MODEL_SMOOTH
);
1023 if (screen
->eng3d
->oclass
< NVE4_3D_CLASS
) {
1024 IMMED_NVC0(push
, NVC0_3D(TEX_MISC
), 0);
1026 BEGIN_NVC0(push
, NVE4_3D(TEX_CB_INDEX
), 1);
1027 PUSH_DATA (push
, 15);
1029 BEGIN_NVC0(push
, NVC0_3D(CALL_LIMIT_LOG
), 1);
1030 PUSH_DATA (push
, 8); /* 128 */
1031 BEGIN_NVC0(push
, NVC0_3D(ZCULL_STATCTRS_ENABLE
), 1);
1032 PUSH_DATA (push
, 1);
1033 if (screen
->eng3d
->oclass
>= NVC1_3D_CLASS
) {
1034 BEGIN_NVC0(push
, NVC0_3D(CACHE_SPLIT
), 1);
1035 PUSH_DATA (push
, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1
);
1038 nvc0_magic_3d_init(push
, screen
->eng3d
->oclass
);
1040 ret
= nvc0_screen_resize_text_area(screen
, 1 << 19);
1042 FAIL_SCREEN_INIT("Error allocating TEXT area: %d\n", ret
);
1044 ret
= nouveau_bo_new(dev
, NV_VRAM_DOMAIN(&screen
->base
), 1 << 12, 7 << 16, NULL
,
1045 &screen
->uniform_bo
);
1047 FAIL_SCREEN_INIT("Error allocating uniform BO: %d\n", ret
);
1049 PUSH_REFN (push
, screen
->uniform_bo
, NV_VRAM_DOMAIN(&screen
->base
) | NOUVEAU_BO_WR
);
1051 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
1052 BEGIN_NVC0(push
, NVC0_3D(CB_SIZE
), 3);
1053 PUSH_DATA (push
, 256);
1054 PUSH_DATAh(push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_RUNOUT_INFO
);
1055 PUSH_DATA (push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_RUNOUT_INFO
);
1056 BEGIN_1IC0(push
, NVC0_3D(CB_POS
), 5);
1057 PUSH_DATA (push
, 0);
1058 PUSH_DATAf(push
, 0.0f
);
1059 PUSH_DATAf(push
, 0.0f
);
1060 PUSH_DATAf(push
, 0.0f
);
1061 PUSH_DATAf(push
, 0.0f
);
1062 BEGIN_NVC0(push
, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH
), 2);
1063 PUSH_DATAh(push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_RUNOUT_INFO
);
1064 PUSH_DATA (push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_RUNOUT_INFO
);
1066 if (screen
->base
.drm
->version
>= 0x01000101) {
1067 ret
= nouveau_getparam(dev
, NOUVEAU_GETPARAM_GRAPH_UNITS
, &value
);
1069 FAIL_SCREEN_INIT("NOUVEAU_GETPARAM_GRAPH_UNITS failed: %d\n", ret
);
1071 if (dev
->chipset
>= 0xe0 && dev
->chipset
< 0xf0)
1072 value
= (8 << 8) | 4;
1074 value
= (16 << 8) | 4;
1076 screen
->gpc_count
= value
& 0x000000ff;
1077 screen
->mp_count
= value
>> 8;
1078 screen
->mp_count_compute
= screen
->mp_count
;
1080 ret
= nvc0_screen_resize_tls_area(screen
, 128 * 16, 0, 0x200);
1082 FAIL_SCREEN_INIT("Error allocating TLS area: %d\n", ret
);
1084 BEGIN_NVC0(push
, NVC0_3D(TEMP_ADDRESS_HIGH
), 4);
1085 PUSH_DATAh(push
, screen
->tls
->offset
);
1086 PUSH_DATA (push
, screen
->tls
->offset
);
1087 PUSH_DATA (push
, screen
->tls
->size
>> 32);
1088 PUSH_DATA (push
, screen
->tls
->size
);
1089 BEGIN_NVC0(push
, NVC0_3D(WARP_TEMP_ALLOC
), 1);
1090 PUSH_DATA (push
, 0);
1091 /* Reduce likelihood of collision with real buffers by placing the hole at
1092 * the top of the 4G area. This will have to be dealt with for real
1093 * eventually by blocking off that area from the VM.
1095 BEGIN_NVC0(push
, NVC0_3D(LOCAL_BASE
), 1);
1096 PUSH_DATA (push
, 0xff << 24);
1098 if (screen
->eng3d
->oclass
< GM107_3D_CLASS
) {
1099 ret
= nouveau_bo_new(dev
, NV_VRAM_DOMAIN(&screen
->base
), 1 << 17, 1 << 20, NULL
,
1100 &screen
->poly_cache
);
1102 FAIL_SCREEN_INIT("Error allocating poly cache BO: %d\n", ret
);
1104 BEGIN_NVC0(push
, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH
), 3);
1105 PUSH_DATAh(push
, screen
->poly_cache
->offset
);
1106 PUSH_DATA (push
, screen
->poly_cache
->offset
);
1107 PUSH_DATA (push
, 3);
1110 ret
= nouveau_bo_new(dev
, NV_VRAM_DOMAIN(&screen
->base
), 1 << 17, 1 << 17, NULL
,
1113 FAIL_SCREEN_INIT("Error allocating txc BO: %d\n", ret
);
1115 BEGIN_NVC0(push
, NVC0_3D(TIC_ADDRESS_HIGH
), 3);
1116 PUSH_DATAh(push
, screen
->txc
->offset
);
1117 PUSH_DATA (push
, screen
->txc
->offset
);
1118 PUSH_DATA (push
, NVC0_TIC_MAX_ENTRIES
- 1);
1119 if (screen
->eng3d
->oclass
>= GM107_3D_CLASS
) {
1120 screen
->tic
.maxwell
= true;
1121 if (screen
->eng3d
->oclass
== GM107_3D_CLASS
) {
1122 screen
->tic
.maxwell
=
1123 debug_get_bool_option("NOUVEAU_MAXWELL_TIC", true);
1124 IMMED_NVC0(push
, SUBC_3D(0x0f10), screen
->tic
.maxwell
);
1128 BEGIN_NVC0(push
, NVC0_3D(TSC_ADDRESS_HIGH
), 3);
1129 PUSH_DATAh(push
, screen
->txc
->offset
+ 65536);
1130 PUSH_DATA (push
, screen
->txc
->offset
+ 65536);
1131 PUSH_DATA (push
, NVC0_TSC_MAX_ENTRIES
- 1);
1133 BEGIN_NVC0(push
, NVC0_3D(SCREEN_Y_CONTROL
), 1);
1134 PUSH_DATA (push
, 0);
1135 BEGIN_NVC0(push
, NVC0_3D(WINDOW_OFFSET_X
), 2);
1136 PUSH_DATA (push
, 0);
1137 PUSH_DATA (push
, 0);
1138 BEGIN_NVC0(push
, NVC0_3D(ZCULL_REGION
), 1); /* deactivate ZCULL */
1139 PUSH_DATA (push
, 0x3f);
1141 BEGIN_NVC0(push
, NVC0_3D(CLIP_RECTS_MODE
), 1);
1142 PUSH_DATA (push
, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY
);
1143 BEGIN_NVC0(push
, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
1144 for (i
= 0; i
< 8 * 2; ++i
)
1146 BEGIN_NVC0(push
, NVC0_3D(CLIP_RECTS_EN
), 1);
1147 PUSH_DATA (push
, 0);
1148 BEGIN_NVC0(push
, NVC0_3D(CLIPID_ENABLE
), 1);
1149 PUSH_DATA (push
, 0);
1151 /* neither scissors, viewport nor stencil mask should affect clears */
1152 BEGIN_NVC0(push
, NVC0_3D(CLEAR_FLAGS
), 1);
1153 PUSH_DATA (push
, 0);
1155 BEGIN_NVC0(push
, NVC0_3D(VIEWPORT_TRANSFORM_EN
), 1);
1156 PUSH_DATA (push
, 1);
1157 for (i
= 0; i
< NVC0_MAX_VIEWPORTS
; i
++) {
1158 BEGIN_NVC0(push
, NVC0_3D(DEPTH_RANGE_NEAR(i
)), 2);
1159 PUSH_DATAf(push
, 0.0f
);
1160 PUSH_DATAf(push
, 1.0f
);
1162 BEGIN_NVC0(push
, NVC0_3D(VIEW_VOLUME_CLIP_CTRL
), 1);
1163 PUSH_DATA (push
, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1
);
1165 /* We use scissors instead of exact view volume clipping,
1166 * so they're always enabled.
1168 for (i
= 0; i
< NVC0_MAX_VIEWPORTS
; i
++) {
1169 BEGIN_NVC0(push
, NVC0_3D(SCISSOR_ENABLE(i
)), 3);
1170 PUSH_DATA (push
, 1);
1171 PUSH_DATA (push
, 8192 << 16);
1172 PUSH_DATA (push
, 8192 << 16);
1175 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
1178 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE
, mme9097_per_instance_bf
);
1179 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES
, mme9097_blend_enables
);
1180 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT
, mme9097_vertex_array_select
);
1181 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT
, mme9097_tep_select
);
1182 MK_MACRO(NVC0_3D_MACRO_GP_SELECT
, mme9097_gp_select
);
1183 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT
, mme9097_poly_mode_front
);
1184 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK
, mme9097_poly_mode_back
);
1185 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT
, mme9097_draw_arrays_indirect
);
1186 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT
, mme9097_draw_elts_indirect
);
1187 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT
, mme9097_draw_arrays_indirect_count
);
1188 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT
, mme9097_draw_elts_indirect_count
);
1189 MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE
, mme9097_query_buffer_write
);
1190 MK_MACRO(NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT
, mme90c0_launch_grid_indirect
);
1192 BEGIN_NVC0(push
, NVC0_3D(RASTERIZE_ENABLE
), 1);
1193 PUSH_DATA (push
, 1);
1194 BEGIN_NVC0(push
, NVC0_3D(RT_SEPARATE_FRAG_DATA
), 1);
1195 PUSH_DATA (push
, 1);
1196 BEGIN_NVC0(push
, NVC0_3D(MACRO_GP_SELECT
), 1);
1197 PUSH_DATA (push
, 0x40);
1198 BEGIN_NVC0(push
, NVC0_3D(LAYER
), 1);
1199 PUSH_DATA (push
, 0);
1200 BEGIN_NVC0(push
, NVC0_3D(MACRO_TEP_SELECT
), 1);
1201 PUSH_DATA (push
, 0x30);
1202 BEGIN_NVC0(push
, NVC0_3D(PATCH_VERTICES
), 1);
1203 PUSH_DATA (push
, 3);
1204 BEGIN_NVC0(push
, NVC0_3D(SP_SELECT(2)), 1);
1205 PUSH_DATA (push
, 0x20);
1206 BEGIN_NVC0(push
, NVC0_3D(SP_SELECT(0)), 1);
1207 PUSH_DATA (push
, 0x00);
1208 screen
->save_state
.patch_vertices
= 3;
1210 BEGIN_NVC0(push
, NVC0_3D(POINT_COORD_REPLACE
), 1);
1211 PUSH_DATA (push
, 0);
1212 BEGIN_NVC0(push
, NVC0_3D(POINT_RASTER_RULES
), 1);
1213 PUSH_DATA (push
, NVC0_3D_POINT_RASTER_RULES_OGL
);
1215 IMMED_NVC0(push
, NVC0_3D(EDGEFLAG
), 1);
1217 if (nvc0_screen_init_compute(screen
))
1220 /* XXX: Compute and 3D are somehow aliased on Fermi. */
1221 for (i
= 0; i
< 5; ++i
) {
1222 /* TIC and TSC entries for each unit (nve4+ only) */
1223 /* auxiliary constants (6 user clip planes, base instance id) */
1224 BEGIN_NVC0(push
, NVC0_3D(CB_SIZE
), 3);
1225 PUSH_DATA (push
, NVC0_CB_AUX_SIZE
);
1226 PUSH_DATAh(push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_INFO(i
));
1227 PUSH_DATA (push
, screen
->uniform_bo
->offset
+ NVC0_CB_AUX_INFO(i
));
1228 BEGIN_NVC0(push
, NVC0_3D(CB_BIND(i
)), 1);
1229 PUSH_DATA (push
, (15 << 4) | 1);
1230 if (screen
->eng3d
->oclass
>= NVE4_3D_CLASS
) {
1232 BEGIN_1IC0(push
, NVC0_3D(CB_POS
), 9);
1233 PUSH_DATA (push
, NVC0_CB_AUX_UNK_INFO
);
1234 for (j
= 0; j
< 8; ++j
)
1237 BEGIN_NVC0(push
, NVC0_3D(TEX_LIMITS(i
)), 1);
1238 PUSH_DATA (push
, 0x54);
1241 /* MS sample coordinate offsets: these do not work with _ALT modes ! */
1242 BEGIN_1IC0(push
, NVC0_3D(CB_POS
), 1 + 2 * 8);
1243 PUSH_DATA (push
, NVC0_CB_AUX_MS_INFO
);
1244 PUSH_DATA (push
, 0); /* 0 */
1245 PUSH_DATA (push
, 0);
1246 PUSH_DATA (push
, 1); /* 1 */
1247 PUSH_DATA (push
, 0);
1248 PUSH_DATA (push
, 0); /* 2 */
1249 PUSH_DATA (push
, 1);
1250 PUSH_DATA (push
, 1); /* 3 */
1251 PUSH_DATA (push
, 1);
1252 PUSH_DATA (push
, 2); /* 4 */
1253 PUSH_DATA (push
, 0);
1254 PUSH_DATA (push
, 3); /* 5 */
1255 PUSH_DATA (push
, 0);
1256 PUSH_DATA (push
, 2); /* 6 */
1257 PUSH_DATA (push
, 1);
1258 PUSH_DATA (push
, 3); /* 7 */
1259 PUSH_DATA (push
, 1);
1261 BEGIN_NVC0(push
, NVC0_3D(LINKED_TSC
), 1);
1262 PUSH_DATA (push
, 0);
1266 screen
->tic
.entries
= CALLOC(4096, sizeof(void *));
1267 screen
->tsc
.entries
= screen
->tic
.entries
+ 2048;
1269 if (!nvc0_blitter_create(screen
))
1272 screen
->default_tsc
= CALLOC_STRUCT(nv50_tsc_entry
);
1273 screen
->default_tsc
->tsc
[0] = G80_TSC_0_SRGB_CONVERSION
;
1275 nouveau_fence_new(&screen
->base
, &screen
->base
.fence
.current
);
1277 return &screen
->base
;
1280 screen
->base
.base
.context_create
= NULL
;
1281 return &screen
->base
;
1285 nvc0_screen_tic_alloc(struct nvc0_screen
*screen
, void *entry
)
1287 int i
= screen
->tic
.next
;
1289 while (screen
->tic
.lock
[i
/ 32] & (1 << (i
% 32)))
1290 i
= (i
+ 1) & (NVC0_TIC_MAX_ENTRIES
- 1);
1292 screen
->tic
.next
= (i
+ 1) & (NVC0_TIC_MAX_ENTRIES
- 1);
1294 if (screen
->tic
.entries
[i
])
1295 nv50_tic_entry(screen
->tic
.entries
[i
])->id
= -1;
1297 screen
->tic
.entries
[i
] = entry
;
1302 nvc0_screen_tsc_alloc(struct nvc0_screen
*screen
, void *entry
)
1304 int i
= screen
->tsc
.next
;
1306 while (screen
->tsc
.lock
[i
/ 32] & (1 << (i
% 32)))
1307 i
= (i
+ 1) & (NVC0_TSC_MAX_ENTRIES
- 1);
1309 screen
->tsc
.next
= (i
+ 1) & (NVC0_TSC_MAX_ENTRIES
- 1);
1311 if (screen
->tsc
.entries
[i
])
1312 nv50_tsc_entry(screen
->tsc
.entries
[i
])->id
= -1;
1314 screen
->tsc
.entries
[i
] = entry
;