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.
25 #include <nouveau_drm.h>
26 #include "util/format/u_format.h"
27 #include "util/format/u_format_s3tc.h"
28 #include "util/u_screen.h"
29 #include "pipe/p_screen.h"
30 #include "compiler/nir/nir.h"
32 #include "nv50/nv50_context.h"
33 #include "nv50/nv50_screen.h"
35 #include "nouveau_vp3_video.h"
37 #include "nv_object.xml.h"
39 /* affected by LOCAL_WARPS_LOG_ALLOC / LOCAL_WARPS_NO_CLAMP */
40 #define LOCAL_WARPS_ALLOC 32
41 /* affected by STACK_WARPS_LOG_ALLOC / STACK_WARPS_NO_CLAMP */
42 #define STACK_WARPS_ALLOC 32
44 #define THREADS_IN_WARP 32
47 nv50_screen_is_format_supported(struct pipe_screen
*pscreen
,
48 enum pipe_format format
,
49 enum pipe_texture_target target
,
50 unsigned sample_count
,
51 unsigned storage_sample_count
,
56 if (!(0x117 & (1 << sample_count
))) /* 0, 1, 2, 4 or 8 */
58 if (sample_count
== 8 && util_format_get_blocksizebits(format
) >= 128)
61 if (MAX2(1, sample_count
) != MAX2(1, storage_sample_count
))
65 case PIPE_FORMAT_Z16_UNORM
:
66 if (nv50_screen(pscreen
)->tesla
->oclass
< NVA0_3D_CLASS
)
73 if (bindings
& PIPE_BIND_LINEAR
)
74 if (util_format_is_depth_or_stencil(format
) ||
75 (target
!= PIPE_TEXTURE_1D
&&
76 target
!= PIPE_TEXTURE_2D
&&
77 target
!= PIPE_TEXTURE_RECT
) ||
81 /* shared is always supported */
82 bindings
&= ~(PIPE_BIND_LINEAR
|
85 return (( nv50_format_table
[format
].usage
|
86 nv50_vertex_format
[format
].usage
) & bindings
) == bindings
;
90 nv50_screen_get_param(struct pipe_screen
*pscreen
, enum pipe_cap param
)
92 const uint16_t class_3d
= nouveau_screen(pscreen
)->class_3d
;
93 struct nouveau_device
*dev
= nouveau_screen(pscreen
)->device
;
96 /* non-boolean caps */
97 case PIPE_CAP_MAX_TEXTURE_2D_SIZE
:
99 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS
:
101 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS
:
103 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS
:
105 case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET
:
106 case PIPE_CAP_MIN_TEXEL_OFFSET
:
108 case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET
:
109 case PIPE_CAP_MAX_TEXEL_OFFSET
:
111 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE
:
112 return 128 * 1024 * 1024;
113 case PIPE_CAP_GLSL_FEATURE_LEVEL
:
115 case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY
:
117 case PIPE_CAP_MAX_RENDER_TARGETS
:
119 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS
:
121 case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS
:
122 case PIPE_CAP_RASTERIZER_SUBPIXEL_BITS
:
124 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS
:
126 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS
:
127 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS
:
129 case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES
:
130 case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS
:
132 case PIPE_CAP_MAX_VERTEX_STREAMS
:
134 case PIPE_CAP_MAX_GS_INVOCATIONS
:
136 case PIPE_CAP_MAX_SHADER_BUFFER_SIZE
:
138 case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE
:
140 case PIPE_CAP_MAX_VERTEX_ELEMENT_SRC_OFFSET
:
142 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT
:
144 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT
:
145 return 16; /* 256 for binding as RT, but that's not possible in GL */
146 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT
:
147 return NOUVEAU_MIN_BUFFER_MAP_ALIGN
;
148 case PIPE_CAP_MAX_VIEWPORTS
:
149 return NV50_MAX_VIEWPORTS
;
150 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK
:
151 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50
;
152 case PIPE_CAP_ENDIANNESS
:
153 return PIPE_ENDIAN_LITTLE
;
154 case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS
:
155 return (class_3d
>= NVA3_3D_CLASS
) ? 4 : 0;
156 case PIPE_CAP_MAX_WINDOW_RECTANGLES
:
157 return NV50_MAX_WINDOW_RECTANGLES
;
158 case PIPE_CAP_MAX_TEXTURE_UPLOAD_MEMORY_BUDGET
:
159 return 16 * 1024 * 1024;
160 case PIPE_CAP_MAX_VARYINGS
:
162 case PIPE_CAP_MAX_VERTEX_BUFFERS
:
166 case PIPE_CAP_TEXTURE_MIRROR_CLAMP
:
167 case PIPE_CAP_TEXTURE_MIRROR_CLAMP_TO_EDGE
:
168 case PIPE_CAP_TEXTURE_SWIZZLE
:
169 case PIPE_CAP_TEXTURE_SHADOW_MAP
:
170 case PIPE_CAP_NPOT_TEXTURES
:
171 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES
:
172 case PIPE_CAP_MIXED_COLOR_DEPTH_BITS
:
173 case PIPE_CAP_ANISOTROPIC_FILTER
:
174 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS
:
175 case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT
:
176 case PIPE_CAP_DEPTH_CLIP_DISABLE
:
177 case PIPE_CAP_POINT_SPRITE
:
178 case PIPE_CAP_FRAGMENT_SHADER_TEXTURE_LOD
:
179 case PIPE_CAP_FRAGMENT_SHADER_DERIVATIVES
:
180 case PIPE_CAP_VERTEX_SHADER_SATURATE
:
181 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED
:
182 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED
:
183 case PIPE_CAP_VERTEX_COLOR_CLAMPED
:
184 case PIPE_CAP_QUERY_TIMESTAMP
:
185 case PIPE_CAP_QUERY_TIME_ELAPSED
:
186 case PIPE_CAP_OCCLUSION_QUERY
:
187 case PIPE_CAP_BLEND_EQUATION_SEPARATE
:
188 case PIPE_CAP_INDEP_BLEND_ENABLE
:
189 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT
:
190 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER
:
191 case PIPE_CAP_PRIMITIVE_RESTART
:
192 case PIPE_CAP_TGSI_INSTANCEID
:
193 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR
:
194 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS
:
195 case PIPE_CAP_CONDITIONAL_RENDER
:
196 case PIPE_CAP_TEXTURE_BARRIER
:
197 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION
:
198 case PIPE_CAP_START_INSTANCE
:
199 case PIPE_CAP_USER_VERTEX_BUFFERS
:
200 case PIPE_CAP_TEXTURE_MULTISAMPLE
:
201 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER
:
202 case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE
:
203 case PIPE_CAP_SAMPLER_VIEW_TARGET
:
204 case PIPE_CAP_CONDITIONAL_RENDER_INVERTED
:
205 case PIPE_CAP_CLIP_HALFZ
:
206 case PIPE_CAP_POLYGON_OFFSET_CLAMP
:
207 case PIPE_CAP_QUERY_PIPELINE_STATISTICS
:
208 case PIPE_CAP_TEXTURE_FLOAT_LINEAR
:
209 case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR
:
210 case PIPE_CAP_DEPTH_BOUNDS_TEST
:
211 case PIPE_CAP_TGSI_TXQS
:
212 case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS
:
213 case PIPE_CAP_SHAREABLE_SHADERS
:
214 case PIPE_CAP_CLEAR_TEXTURE
:
215 case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL
:
216 case PIPE_CAP_INVALIDATE_BUFFER
:
217 case PIPE_CAP_STRING_MARKER
:
218 case PIPE_CAP_CULL_DISTANCE
:
219 case PIPE_CAP_TGSI_ARRAY_COMPONENTS
:
220 case PIPE_CAP_TGSI_MUL_ZERO_WINS
:
221 case PIPE_CAP_TGSI_TEX_TXF_LZ
:
222 case PIPE_CAP_TGSI_CLOCK
:
223 case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX
:
224 case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION
:
225 case PIPE_CAP_DEST_SURFACE_SRGB_CONTROL
:
226 case PIPE_CAP_TGSI_DIV
:
227 case PIPE_CAP_PREFER_IMM_ARRAYS_AS_CONSTBUF
:
228 case PIPE_CAP_FLATSHADE
:
229 case PIPE_CAP_ALPHA_TEST
:
230 case PIPE_CAP_POINT_SIZE_FIXED
:
231 case PIPE_CAP_TWO_SIDED_COLOR
:
232 case PIPE_CAP_CLIP_PLANES
:
234 case PIPE_CAP_SEAMLESS_CUBE_MAP
:
235 return 1; /* class_3d >= NVA0_3D_CLASS; */
236 /* supported on nva0+ */
237 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME
:
238 return class_3d
>= NVA0_3D_CLASS
;
239 /* supported on nva3+ */
240 case PIPE_CAP_CUBE_MAP_ARRAY
:
241 case PIPE_CAP_INDEP_BLEND_FUNC
:
242 case PIPE_CAP_TEXTURE_QUERY_LOD
:
243 case PIPE_CAP_SAMPLE_SHADING
:
244 case PIPE_CAP_FORCE_PERSAMPLE_INTERP
:
245 return class_3d
>= NVA3_3D_CLASS
;
247 /* unsupported caps */
248 case PIPE_CAP_DEPTH_CLIP_DISABLE_SEPARATE
:
249 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE
:
250 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT
:
251 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER
:
252 case PIPE_CAP_SHADER_STENCIL_EXPORT
:
253 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS
:
254 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY
:
255 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY
:
256 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY
:
257 case PIPE_CAP_TGSI_TEXCOORD
:
258 case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT
:
259 case PIPE_CAP_TEXTURE_GATHER_SM5
:
260 case PIPE_CAP_FAKE_SW_MSAA
:
261 case PIPE_CAP_TEXTURE_GATHER_OFFSETS
:
262 case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION
:
263 case PIPE_CAP_DRAW_INDIRECT
:
264 case PIPE_CAP_MULTI_DRAW_INDIRECT
:
265 case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS
:
266 case PIPE_CAP_VERTEXID_NOBASE
:
267 case PIPE_CAP_MULTISAMPLE_Z_RESOLVE
: /* potentially supported on some hw */
268 case PIPE_CAP_RESOURCE_FROM_USER_MEMORY
:
269 case PIPE_CAP_DEVICE_RESET_STATUS_QUERY
:
270 case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS
:
271 case PIPE_CAP_DRAW_PARAMETERS
:
272 case PIPE_CAP_TGSI_PACK_HALF_FLOAT
:
273 case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL
:
274 case PIPE_CAP_TGSI_FS_POINT_IS_SYSVAL
:
275 case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT
:
276 case PIPE_CAP_GENERATE_MIPMAP
:
277 case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY
:
278 case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS
:
279 case PIPE_CAP_QUERY_BUFFER_OBJECT
:
280 case PIPE_CAP_QUERY_MEMORY_INFO
:
281 case PIPE_CAP_PCI_GROUP
:
282 case PIPE_CAP_PCI_BUS
:
283 case PIPE_CAP_PCI_DEVICE
:
284 case PIPE_CAP_PCI_FUNCTION
:
285 case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT
:
286 case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR
:
287 case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES
:
288 case PIPE_CAP_TGSI_VOTE
:
289 case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED
:
290 case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS
:
291 case PIPE_CAP_TGSI_CAN_READ_OUTPUTS
:
292 case PIPE_CAP_NATIVE_FENCE_FD
:
293 case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY
:
294 case PIPE_CAP_FBFETCH
:
295 case PIPE_CAP_DOUBLES
:
297 case PIPE_CAP_INT64_DIVMOD
:
298 case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE
:
299 case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE
:
300 case PIPE_CAP_TGSI_BALLOT
:
301 case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT
:
302 case PIPE_CAP_POST_DEPTH_COVERAGE
:
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_FRAMEBUFFER_MSAA_CONSTRAINTS
:
312 case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET
:
313 case PIPE_CAP_CONTEXT_PRIORITY_MASK
:
314 case PIPE_CAP_FENCE_SIGNAL
:
315 case PIPE_CAP_CONSTBUF0_FLAGS
:
316 case PIPE_CAP_PACKED_UNIFORMS
:
317 case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_TRIANGLES
:
318 case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_POINTS_LINES
:
319 case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES
:
320 case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_POINTS_LINES
:
321 case PIPE_CAP_CONSERVATIVE_RASTER_POST_DEPTH_COVERAGE
:
322 case PIPE_CAP_MAX_CONSERVATIVE_RASTER_SUBPIXEL_PRECISION_BIAS
:
323 case PIPE_CAP_PROGRAMMABLE_SAMPLE_LOCATIONS
:
324 case PIPE_CAP_MAX_COMBINED_SHADER_BUFFERS
:
325 case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTERS
:
326 case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTER_BUFFERS
:
327 case PIPE_CAP_SURFACE_SAMPLE_COUNT
:
328 case PIPE_CAP_TGSI_ATOMFADD
:
329 case PIPE_CAP_QUERY_PIPELINE_STATISTICS_SINGLE
:
330 case PIPE_CAP_RGB_OVERRIDE_DST_ALPHA_BLEND
:
331 case PIPE_CAP_GLSL_TESS_LEVELS_AS_INPUTS
:
332 case PIPE_CAP_NIR_COMPACT_ARRAYS
:
333 case PIPE_CAP_COMPUTE
:
334 case PIPE_CAP_IMAGE_LOAD_FORMATTED
:
335 case PIPE_CAP_COMPUTE_SHADER_DERIVATIVES
:
336 case PIPE_CAP_ATOMIC_FLOAT_MINMAX
:
337 case PIPE_CAP_CONSERVATIVE_RASTER_INNER_COVERAGE
:
338 case PIPE_CAP_FRAGMENT_SHADER_INTERLOCK
:
339 case PIPE_CAP_CS_DERIVED_SYSTEM_VALUES_SUPPORTED
:
340 case PIPE_CAP_FBFETCH_COHERENT
:
341 case PIPE_CAP_TGSI_SKIP_SHRINK_IO_ARRAYS
:
342 case PIPE_CAP_TGSI_ATOMINC_WRAP
:
343 case PIPE_CAP_DEMOTE_TO_HELPER_INVOCATION
:
344 case PIPE_CAP_TGSI_TG4_COMPONENT_IN_SWIZZLE
:
345 case PIPE_CAP_OPENCL_INTEGER_FUNCTIONS
:
346 case PIPE_CAP_INTEGER_MULTIPLY_32X16
: /* could be done */
347 case PIPE_CAP_FRONTEND_NOOP
:
348 case PIPE_CAP_GL_SPIRV
:
349 case PIPE_CAP_SHADER_SAMPLES_IDENTICAL
:
350 case PIPE_CAP_TEXTURE_SHADOW_LOD
: /* investigate */
353 case PIPE_CAP_VENDOR_ID
:
355 case PIPE_CAP_DEVICE_ID
: {
357 if (nouveau_getparam(dev
, NOUVEAU_GETPARAM_PCI_DEVICE
, &device_id
)) {
358 NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
363 case PIPE_CAP_ACCELERATED
:
365 case PIPE_CAP_VIDEO_MEMORY
:
366 return dev
->vram_size
>> 20;
371 debug_printf("%s: unhandled cap %d\n", __func__
, param
);
373 /* caps where we want the default value */
374 case PIPE_CAP_DMABUF
:
375 case PIPE_CAP_ESSL_FEATURE_LEVEL
:
376 case PIPE_CAP_THROTTLE
:
377 return u_pipe_screen_get_param_defaults(pscreen
, param
);
382 nv50_screen_get_shader_param(struct pipe_screen
*pscreen
,
383 enum pipe_shader_type shader
,
384 enum pipe_shader_cap param
)
386 const struct nouveau_screen
*screen
= nouveau_screen(pscreen
);
389 case PIPE_SHADER_VERTEX
:
390 case PIPE_SHADER_GEOMETRY
:
391 case PIPE_SHADER_FRAGMENT
:
393 case PIPE_SHADER_COMPUTE
:
399 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS
:
400 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS
:
401 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS
:
402 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS
:
404 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH
:
406 case PIPE_SHADER_CAP_MAX_INPUTS
:
407 if (shader
== PIPE_SHADER_VERTEX
)
410 case PIPE_SHADER_CAP_MAX_OUTPUTS
:
412 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE
:
414 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS
:
415 return NV50_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 nv50_screen(pscreen
)->max_tls_space
/ ONE_TEMP_SIZE
;
424 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED
:
426 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED
:
428 case PIPE_SHADER_CAP_INT64_ATOMICS
:
429 case PIPE_SHADER_CAP_FP16
:
430 case PIPE_SHADER_CAP_SUBROUTINES
:
431 return 0; /* please inline, or provide function declarations */
432 case PIPE_SHADER_CAP_INTEGERS
:
434 case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS
:
436 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS
:
437 /* The chip could handle more sampler views than samplers */
438 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS
:
439 return MIN2(16, PIPE_MAX_SAMPLERS
);
440 case PIPE_SHADER_CAP_PREFERRED_IR
:
441 return screen
->prefer_nir
? PIPE_SHADER_IR_NIR
: PIPE_SHADER_IR_TGSI
;
442 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT
:
444 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED
:
445 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED
:
446 case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED
:
447 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED
:
448 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE
:
449 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS
:
450 case PIPE_SHADER_CAP_SUPPORTED_IRS
:
451 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES
:
452 case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD
:
453 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS
:
454 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS
:
457 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param
);
463 nv50_screen_get_paramf(struct pipe_screen
*pscreen
, enum pipe_capf param
)
466 case PIPE_CAPF_MAX_LINE_WIDTH
:
467 case PIPE_CAPF_MAX_LINE_WIDTH_AA
:
469 case PIPE_CAPF_MAX_POINT_WIDTH
:
470 case PIPE_CAPF_MAX_POINT_WIDTH_AA
:
472 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY
:
474 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS
:
476 case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE
:
477 case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE
:
478 case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY
:
482 NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param
);
487 nv50_screen_get_compute_param(struct pipe_screen
*pscreen
,
488 enum pipe_shader_ir ir_type
,
489 enum pipe_compute_cap param
, void *data
)
491 struct nv50_screen
*screen
= nv50_screen(pscreen
);
493 #define RET(x) do { \
495 memcpy(data, x, sizeof(x)); \
500 case PIPE_COMPUTE_CAP_GRID_DIMENSION
:
501 RET((uint64_t []) { 2 });
502 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE
:
503 RET(((uint64_t []) { 65535, 65535 }));
504 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE
:
505 RET(((uint64_t []) { 512, 512, 64 }));
506 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK
:
507 RET((uint64_t []) { 512 });
508 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE
: /* g0-15[] */
509 RET((uint64_t []) { 1ULL << 32 });
510 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE
: /* s[] */
511 RET((uint64_t []) { 16 << 10 });
512 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE
: /* l[] */
513 RET((uint64_t []) { 16 << 10 });
514 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE
: /* c[], arbitrary limit */
515 RET((uint64_t []) { 4096 });
516 case PIPE_COMPUTE_CAP_SUBGROUP_SIZE
:
517 RET((uint32_t []) { 32 });
518 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE
:
519 RET((uint64_t []) { 1ULL << 40 });
520 case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED
:
521 RET((uint32_t []) { 0 });
522 case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS
:
523 RET((uint32_t []) { screen
->mp_count
});
524 case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY
:
525 RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
526 case PIPE_COMPUTE_CAP_ADDRESS_BITS
:
527 RET((uint32_t []) { 32 });
528 case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK
:
529 RET((uint64_t []) { 0 });
538 nv50_screen_destroy(struct pipe_screen
*pscreen
)
540 struct nv50_screen
*screen
= nv50_screen(pscreen
);
542 if (!nouveau_drm_screen_unref(&screen
->base
))
545 if (screen
->base
.fence
.current
) {
546 struct nouveau_fence
*current
= NULL
;
548 /* nouveau_fence_wait will create a new current fence, so wait on the
549 * _current_ one, and remove both.
551 nouveau_fence_ref(screen
->base
.fence
.current
, ¤t
);
552 nouveau_fence_wait(current
, NULL
);
553 nouveau_fence_ref(NULL
, ¤t
);
554 nouveau_fence_ref(NULL
, &screen
->base
.fence
.current
);
556 if (screen
->base
.pushbuf
)
557 screen
->base
.pushbuf
->user_priv
= NULL
;
560 nv50_blitter_destroy(screen
);
561 if (screen
->pm
.prog
) {
562 screen
->pm
.prog
->code
= NULL
; /* hardcoded, don't FREE */
563 nv50_program_destroy(NULL
, screen
->pm
.prog
);
564 FREE(screen
->pm
.prog
);
567 nouveau_bo_ref(NULL
, &screen
->code
);
568 nouveau_bo_ref(NULL
, &screen
->tls_bo
);
569 nouveau_bo_ref(NULL
, &screen
->stack_bo
);
570 nouveau_bo_ref(NULL
, &screen
->txc
);
571 nouveau_bo_ref(NULL
, &screen
->uniforms
);
572 nouveau_bo_ref(NULL
, &screen
->fence
.bo
);
574 nouveau_heap_destroy(&screen
->vp_code_heap
);
575 nouveau_heap_destroy(&screen
->gp_code_heap
);
576 nouveau_heap_destroy(&screen
->fp_code_heap
);
578 FREE(screen
->tic
.entries
);
580 nouveau_object_del(&screen
->tesla
);
581 nouveau_object_del(&screen
->eng2d
);
582 nouveau_object_del(&screen
->m2mf
);
583 nouveau_object_del(&screen
->compute
);
584 nouveau_object_del(&screen
->sync
);
586 nouveau_screen_fini(&screen
->base
);
592 nv50_screen_fence_emit(struct pipe_screen
*pscreen
, u32
*sequence
)
594 struct nv50_screen
*screen
= nv50_screen(pscreen
);
595 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
597 /* we need to do it after possible flush in MARK_RING */
598 *sequence
= ++screen
->base
.fence
.sequence
;
600 assert(PUSH_AVAIL(push
) + push
->rsvd_kick
>= 5);
601 PUSH_DATA (push
, NV50_FIFO_PKHDR(NV50_3D(QUERY_ADDRESS_HIGH
), 4));
602 PUSH_DATAh(push
, screen
->fence
.bo
->offset
);
603 PUSH_DATA (push
, screen
->fence
.bo
->offset
);
604 PUSH_DATA (push
, *sequence
);
605 PUSH_DATA (push
, NV50_3D_QUERY_GET_MODE_WRITE_UNK0
|
606 NV50_3D_QUERY_GET_UNK4
|
607 NV50_3D_QUERY_GET_UNIT_CROP
|
608 NV50_3D_QUERY_GET_TYPE_QUERY
|
609 NV50_3D_QUERY_GET_QUERY_SELECT_ZERO
|
610 NV50_3D_QUERY_GET_SHORT
);
614 nv50_screen_fence_update(struct pipe_screen
*pscreen
)
616 return nv50_screen(pscreen
)->fence
.map
[0];
620 nv50_screen_init_hwctx(struct nv50_screen
*screen
)
622 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
623 struct nv04_fifo
*fifo
;
626 fifo
= (struct nv04_fifo
*)screen
->base
.channel
->data
;
628 BEGIN_NV04(push
, SUBC_M2MF(NV01_SUBCHAN_OBJECT
), 1);
629 PUSH_DATA (push
, screen
->m2mf
->handle
);
630 BEGIN_NV04(push
, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY
), 3);
631 PUSH_DATA (push
, screen
->sync
->handle
);
632 PUSH_DATA (push
, fifo
->vram
);
633 PUSH_DATA (push
, fifo
->vram
);
635 BEGIN_NV04(push
, SUBC_2D(NV01_SUBCHAN_OBJECT
), 1);
636 PUSH_DATA (push
, screen
->eng2d
->handle
);
637 BEGIN_NV04(push
, NV50_2D(DMA_NOTIFY
), 4);
638 PUSH_DATA (push
, screen
->sync
->handle
);
639 PUSH_DATA (push
, fifo
->vram
);
640 PUSH_DATA (push
, fifo
->vram
);
641 PUSH_DATA (push
, fifo
->vram
);
642 BEGIN_NV04(push
, NV50_2D(OPERATION
), 1);
643 PUSH_DATA (push
, NV50_2D_OPERATION_SRCCOPY
);
644 BEGIN_NV04(push
, NV50_2D(CLIP_ENABLE
), 1);
646 BEGIN_NV04(push
, NV50_2D(COLOR_KEY_ENABLE
), 1);
648 BEGIN_NV04(push
, SUBC_2D(0x0888), 1);
650 BEGIN_NV04(push
, NV50_2D(COND_MODE
), 1);
651 PUSH_DATA (push
, NV50_2D_COND_MODE_ALWAYS
);
653 BEGIN_NV04(push
, SUBC_3D(NV01_SUBCHAN_OBJECT
), 1);
654 PUSH_DATA (push
, screen
->tesla
->handle
);
656 BEGIN_NV04(push
, NV50_3D(COND_MODE
), 1);
657 PUSH_DATA (push
, NV50_3D_COND_MODE_ALWAYS
);
659 BEGIN_NV04(push
, NV50_3D(DMA_NOTIFY
), 1);
660 PUSH_DATA (push
, screen
->sync
->handle
);
661 BEGIN_NV04(push
, NV50_3D(DMA_ZETA
), 11);
662 for (i
= 0; i
< 11; ++i
)
663 PUSH_DATA(push
, fifo
->vram
);
664 BEGIN_NV04(push
, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN
);
665 for (i
= 0; i
< NV50_3D_DMA_COLOR__LEN
; ++i
)
666 PUSH_DATA(push
, fifo
->vram
);
668 BEGIN_NV04(push
, NV50_3D(REG_MODE
), 1);
669 PUSH_DATA (push
, NV50_3D_REG_MODE_STRIPED
);
670 BEGIN_NV04(push
, NV50_3D(UNK1400_LANES
), 1);
671 PUSH_DATA (push
, 0xf);
673 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
674 BEGIN_NV04(push
, NV50_3D(WATCHDOG_TIMER
), 1);
675 PUSH_DATA (push
, 0x18);
678 BEGIN_NV04(push
, NV50_3D(ZETA_COMP_ENABLE
), 1);
679 PUSH_DATA(push
, screen
->base
.drm
->version
>= 0x01000101);
681 BEGIN_NV04(push
, NV50_3D(RT_COMP_ENABLE(0)), 8);
682 for (i
= 0; i
< 8; ++i
)
683 PUSH_DATA(push
, screen
->base
.drm
->version
>= 0x01000101);
685 BEGIN_NV04(push
, NV50_3D(RT_CONTROL
), 1);
688 BEGIN_NV04(push
, NV50_3D(CSAA_ENABLE
), 1);
690 BEGIN_NV04(push
, NV50_3D(MULTISAMPLE_ENABLE
), 1);
692 BEGIN_NV04(push
, NV50_3D(MULTISAMPLE_MODE
), 1);
693 PUSH_DATA (push
, NV50_3D_MULTISAMPLE_MODE_MS1
);
694 BEGIN_NV04(push
, NV50_3D(MULTISAMPLE_CTRL
), 1);
696 BEGIN_NV04(push
, NV50_3D(PRIM_RESTART_WITH_DRAW_ARRAYS
), 1);
698 BEGIN_NV04(push
, NV50_3D(BLEND_SEPARATE_ALPHA
), 1);
701 if (screen
->tesla
->oclass
>= NVA0_3D_CLASS
) {
702 BEGIN_NV04(push
, SUBC_3D(NVA0_3D_TEX_MISC
), 1);
706 BEGIN_NV04(push
, NV50_3D(SCREEN_Y_CONTROL
), 1);
708 BEGIN_NV04(push
, NV50_3D(WINDOW_OFFSET_X
), 2);
711 BEGIN_NV04(push
, NV50_3D(ZCULL_REGION
), 1);
712 PUSH_DATA (push
, 0x3f);
714 BEGIN_NV04(push
, NV50_3D(VP_ADDRESS_HIGH
), 2);
715 PUSH_DATAh(push
, screen
->code
->offset
+ (0 << NV50_CODE_BO_SIZE_LOG2
));
716 PUSH_DATA (push
, screen
->code
->offset
+ (0 << NV50_CODE_BO_SIZE_LOG2
));
718 BEGIN_NV04(push
, NV50_3D(FP_ADDRESS_HIGH
), 2);
719 PUSH_DATAh(push
, screen
->code
->offset
+ (1 << NV50_CODE_BO_SIZE_LOG2
));
720 PUSH_DATA (push
, screen
->code
->offset
+ (1 << NV50_CODE_BO_SIZE_LOG2
));
722 BEGIN_NV04(push
, NV50_3D(GP_ADDRESS_HIGH
), 2);
723 PUSH_DATAh(push
, screen
->code
->offset
+ (2 << NV50_CODE_BO_SIZE_LOG2
));
724 PUSH_DATA (push
, screen
->code
->offset
+ (2 << NV50_CODE_BO_SIZE_LOG2
));
726 BEGIN_NV04(push
, NV50_3D(LOCAL_ADDRESS_HIGH
), 3);
727 PUSH_DATAh(push
, screen
->tls_bo
->offset
);
728 PUSH_DATA (push
, screen
->tls_bo
->offset
);
729 PUSH_DATA (push
, util_logbase2(screen
->cur_tls_space
/ 8));
731 BEGIN_NV04(push
, NV50_3D(STACK_ADDRESS_HIGH
), 3);
732 PUSH_DATAh(push
, screen
->stack_bo
->offset
);
733 PUSH_DATA (push
, screen
->stack_bo
->offset
);
736 BEGIN_NV04(push
, NV50_3D(CB_DEF_ADDRESS_HIGH
), 3);
737 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (0 << 16));
738 PUSH_DATA (push
, screen
->uniforms
->offset
+ (0 << 16));
739 PUSH_DATA (push
, (NV50_CB_PVP
<< 16) | 0x0000);
741 BEGIN_NV04(push
, NV50_3D(CB_DEF_ADDRESS_HIGH
), 3);
742 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (1 << 16));
743 PUSH_DATA (push
, screen
->uniforms
->offset
+ (1 << 16));
744 PUSH_DATA (push
, (NV50_CB_PGP
<< 16) | 0x0000);
746 BEGIN_NV04(push
, NV50_3D(CB_DEF_ADDRESS_HIGH
), 3);
747 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (2 << 16));
748 PUSH_DATA (push
, screen
->uniforms
->offset
+ (2 << 16));
749 PUSH_DATA (push
, (NV50_CB_PFP
<< 16) | 0x0000);
751 BEGIN_NV04(push
, NV50_3D(CB_DEF_ADDRESS_HIGH
), 3);
752 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (3 << 16));
753 PUSH_DATA (push
, screen
->uniforms
->offset
+ (3 << 16));
754 PUSH_DATA (push
, (NV50_CB_AUX
<< 16) | (NV50_CB_AUX_SIZE
& 0xffff));
756 BEGIN_NI04(push
, NV50_3D(SET_PROGRAM_CB
), 3);
757 PUSH_DATA (push
, (NV50_CB_AUX
<< 12) | 0xf01);
758 PUSH_DATA (push
, (NV50_CB_AUX
<< 12) | 0xf21);
759 PUSH_DATA (push
, (NV50_CB_AUX
<< 12) | 0xf31);
761 /* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
762 BEGIN_NV04(push
, NV50_3D(CB_ADDR
), 1);
763 PUSH_DATA (push
, (NV50_CB_AUX_RUNOUT_OFFSET
<< (8 - 2)) | NV50_CB_AUX
);
764 BEGIN_NI04(push
, NV50_3D(CB_DATA(0)), 4);
765 PUSH_DATAf(push
, 0.0f
);
766 PUSH_DATAf(push
, 0.0f
);
767 PUSH_DATAf(push
, 0.0f
);
768 PUSH_DATAf(push
, 0.0f
);
769 BEGIN_NV04(push
, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH
), 2);
770 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET
);
771 PUSH_DATA (push
, screen
->uniforms
->offset
+ (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET
);
773 nv50_upload_ms_info(push
);
775 /* max TIC (bits 4:8) & TSC bindings, per program type */
776 for (i
= 0; i
< 3; ++i
) {
777 BEGIN_NV04(push
, NV50_3D(TEX_LIMITS(i
)), 1);
778 PUSH_DATA (push
, 0x54);
781 BEGIN_NV04(push
, NV50_3D(TIC_ADDRESS_HIGH
), 3);
782 PUSH_DATAh(push
, screen
->txc
->offset
);
783 PUSH_DATA (push
, screen
->txc
->offset
);
784 PUSH_DATA (push
, NV50_TIC_MAX_ENTRIES
- 1);
786 BEGIN_NV04(push
, NV50_3D(TSC_ADDRESS_HIGH
), 3);
787 PUSH_DATAh(push
, screen
->txc
->offset
+ 65536);
788 PUSH_DATA (push
, screen
->txc
->offset
+ 65536);
789 PUSH_DATA (push
, NV50_TSC_MAX_ENTRIES
- 1);
791 BEGIN_NV04(push
, NV50_3D(LINKED_TSC
), 1);
794 BEGIN_NV04(push
, NV50_3D(CLIP_RECTS_EN
), 1);
796 BEGIN_NV04(push
, NV50_3D(CLIP_RECTS_MODE
), 1);
797 PUSH_DATA (push
, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY
);
798 BEGIN_NV04(push
, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
799 for (i
= 0; i
< 8 * 2; ++i
)
801 BEGIN_NV04(push
, NV50_3D(CLIPID_ENABLE
), 1);
804 BEGIN_NV04(push
, NV50_3D(VIEWPORT_TRANSFORM_EN
), 1);
806 for (i
= 0; i
< NV50_MAX_VIEWPORTS
; i
++) {
807 BEGIN_NV04(push
, NV50_3D(DEPTH_RANGE_NEAR(i
)), 2);
808 PUSH_DATAf(push
, 0.0f
);
809 PUSH_DATAf(push
, 1.0f
);
810 BEGIN_NV04(push
, NV50_3D(VIEWPORT_HORIZ(i
)), 2);
811 PUSH_DATA (push
, 8192 << 16);
812 PUSH_DATA (push
, 8192 << 16);
815 BEGIN_NV04(push
, NV50_3D(VIEW_VOLUME_CLIP_CTRL
), 1);
816 #ifdef NV50_SCISSORS_CLIPPING
817 PUSH_DATA (push
, 0x0000);
819 PUSH_DATA (push
, 0x1080);
822 BEGIN_NV04(push
, NV50_3D(CLEAR_FLAGS
), 1);
823 PUSH_DATA (push
, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT
);
825 /* We use scissors instead of exact view volume clipping,
826 * so they're always enabled.
828 for (i
= 0; i
< NV50_MAX_VIEWPORTS
; i
++) {
829 BEGIN_NV04(push
, NV50_3D(SCISSOR_ENABLE(i
)), 3);
831 PUSH_DATA (push
, 8192 << 16);
832 PUSH_DATA (push
, 8192 << 16);
835 BEGIN_NV04(push
, NV50_3D(RASTERIZE_ENABLE
), 1);
837 BEGIN_NV04(push
, NV50_3D(POINT_RASTER_RULES
), 1);
838 PUSH_DATA (push
, NV50_3D_POINT_RASTER_RULES_OGL
);
839 BEGIN_NV04(push
, NV50_3D(FRAG_COLOR_CLAMP_EN
), 1);
840 PUSH_DATA (push
, 0x11111111);
841 BEGIN_NV04(push
, NV50_3D(EDGEFLAG
), 1);
844 BEGIN_NV04(push
, NV50_3D(VB_ELEMENT_BASE
), 1);
846 if (screen
->base
.class_3d
>= NV84_3D_CLASS
) {
847 BEGIN_NV04(push
, NV84_3D(VERTEX_ID_BASE
), 1);
851 BEGIN_NV04(push
, NV50_3D(UNK0FDC
), 1);
853 BEGIN_NV04(push
, NV50_3D(UNK19C0
), 1);
859 static int nv50_tls_alloc(struct nv50_screen
*screen
, unsigned tls_space
,
862 struct nouveau_device
*dev
= screen
->base
.device
;
865 screen
->cur_tls_space
= util_next_power_of_two(tls_space
/ ONE_TEMP_SIZE
) *
867 if (nouveau_mesa_debug
)
868 debug_printf("allocating space for %u temps\n",
869 util_next_power_of_two(tls_space
/ ONE_TEMP_SIZE
));
870 *tls_size
= screen
->cur_tls_space
* util_next_power_of_two(screen
->TPs
) *
871 screen
->MPsInTP
* LOCAL_WARPS_ALLOC
* THREADS_IN_WARP
;
873 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16,
874 *tls_size
, NULL
, &screen
->tls_bo
);
876 NOUVEAU_ERR("Failed to allocate local bo: %d\n", ret
);
883 int nv50_tls_realloc(struct nv50_screen
*screen
, unsigned tls_space
)
885 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
889 if (tls_space
< screen
->cur_tls_space
)
891 if (tls_space
> screen
->max_tls_space
) {
892 /* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
893 * LOCAL_WARPS_NO_CLAMP) */
894 NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
895 (unsigned)(tls_space
/ ONE_TEMP_SIZE
),
896 (unsigned)(screen
->max_tls_space
/ ONE_TEMP_SIZE
));
900 nouveau_bo_ref(NULL
, &screen
->tls_bo
);
901 ret
= nv50_tls_alloc(screen
, tls_space
, &tls_size
);
905 BEGIN_NV04(push
, NV50_3D(LOCAL_ADDRESS_HIGH
), 3);
906 PUSH_DATAh(push
, screen
->tls_bo
->offset
);
907 PUSH_DATA (push
, screen
->tls_bo
->offset
);
908 PUSH_DATA (push
, util_logbase2(screen
->cur_tls_space
/ 8));
913 static const nir_shader_compiler_options nir_options
= {
914 .fuse_ffma
= false, /* nir doesn't track mad vs fma */
915 .lower_flrp32
= true,
916 .lower_flrp64
= true,
918 .lower_uadd_carry
= true,
919 .lower_usub_borrow
= true,
921 .lower_ffract
= true,
922 .lower_pack_half_2x16
= true,
923 .lower_pack_unorm_2x16
= true,
924 .lower_pack_snorm_2x16
= true,
925 .lower_pack_unorm_4x8
= true,
926 .lower_pack_snorm_4x8
= true,
927 .lower_unpack_half_2x16
= true,
928 .lower_unpack_unorm_2x16
= true,
929 .lower_unpack_snorm_2x16
= true,
930 .lower_unpack_unorm_4x8
= true,
931 .lower_unpack_snorm_4x8
= true,
932 .lower_extract_byte
= true,
933 .lower_extract_word
= true,
934 .lower_all_io_to_temps
= false,
935 .lower_cs_local_index_from_id
= true,
936 .lower_rotate
= true,
937 .lower_to_scalar
= true,
938 .use_interpolated_input_intrinsics
= true,
939 .max_unroll_iterations
= 32,
943 nv50_screen_get_compiler_options(struct pipe_screen
*pscreen
,
944 enum pipe_shader_ir ir
,
945 enum pipe_shader_type shader
)
947 if (ir
== PIPE_SHADER_IR_NIR
)
952 struct nouveau_screen
*
953 nv50_screen_create(struct nouveau_device
*dev
)
955 struct nv50_screen
*screen
;
956 struct pipe_screen
*pscreen
;
957 struct nouveau_object
*chan
;
959 uint32_t tesla_class
;
963 screen
= CALLOC_STRUCT(nv50_screen
);
966 pscreen
= &screen
->base
.base
;
967 pscreen
->destroy
= nv50_screen_destroy
;
969 ret
= nouveau_screen_init(&screen
->base
, dev
);
971 NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret
);
975 /* TODO: Prevent FIFO prefetch before transfer of index buffers and
976 * admit them to VRAM.
978 screen
->base
.vidmem_bindings
|= PIPE_BIND_CONSTANT_BUFFER
|
979 PIPE_BIND_VERTEX_BUFFER
;
980 screen
->base
.sysmem_bindings
|=
981 PIPE_BIND_VERTEX_BUFFER
| PIPE_BIND_INDEX_BUFFER
;
983 screen
->base
.pushbuf
->user_priv
= screen
;
984 screen
->base
.pushbuf
->rsvd_kick
= 5;
986 chan
= screen
->base
.channel
;
988 pscreen
->context_create
= nv50_create
;
989 pscreen
->is_format_supported
= nv50_screen_is_format_supported
;
990 pscreen
->get_param
= nv50_screen_get_param
;
991 pscreen
->get_shader_param
= nv50_screen_get_shader_param
;
992 pscreen
->get_paramf
= nv50_screen_get_paramf
;
993 pscreen
->get_compute_param
= nv50_screen_get_compute_param
;
994 pscreen
->get_driver_query_info
= nv50_screen_get_driver_query_info
;
995 pscreen
->get_driver_query_group_info
= nv50_screen_get_driver_query_group_info
;
998 pscreen
->get_compiler_options
= nv50_screen_get_compiler_options
;
1000 nv50_screen_init_resource_functions(pscreen
);
1002 if (screen
->base
.device
->chipset
< 0x84 ||
1003 debug_get_bool_option("NOUVEAU_PMPEG", false)) {
1005 nouveau_screen_init_vdec(&screen
->base
);
1006 } else if (screen
->base
.device
->chipset
< 0x98 ||
1007 screen
->base
.device
->chipset
== 0xa0) {
1009 screen
->base
.base
.get_video_param
= nv84_screen_get_video_param
;
1010 screen
->base
.base
.is_video_format_supported
= nv84_screen_video_supported
;
1013 screen
->base
.base
.get_video_param
= nouveau_vp3_screen_get_video_param
;
1014 screen
->base
.base
.is_video_format_supported
= nouveau_vp3_screen_video_supported
;
1017 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_GART
| NOUVEAU_BO_MAP
, 0, 4096,
1018 NULL
, &screen
->fence
.bo
);
1020 NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret
);
1024 nouveau_bo_map(screen
->fence
.bo
, 0, NULL
);
1025 screen
->fence
.map
= screen
->fence
.bo
->map
;
1026 screen
->base
.fence
.emit
= nv50_screen_fence_emit
;
1027 screen
->base
.fence
.update
= nv50_screen_fence_update
;
1029 ret
= nouveau_object_new(chan
, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS
,
1030 &(struct nv04_notify
){ .length
= 32 },
1031 sizeof(struct nv04_notify
), &screen
->sync
);
1033 NOUVEAU_ERR("Failed to allocate notifier: %d\n", ret
);
1037 ret
= nouveau_object_new(chan
, 0xbeef5039, NV50_M2MF_CLASS
,
1038 NULL
, 0, &screen
->m2mf
);
1040 NOUVEAU_ERR("Failed to allocate PGRAPH context for M2MF: %d\n", ret
);
1044 ret
= nouveau_object_new(chan
, 0xbeef502d, NV50_2D_CLASS
,
1045 NULL
, 0, &screen
->eng2d
);
1047 NOUVEAU_ERR("Failed to allocate PGRAPH context for 2D: %d\n", ret
);
1051 switch (dev
->chipset
& 0xf0) {
1053 tesla_class
= NV50_3D_CLASS
;
1057 tesla_class
= NV84_3D_CLASS
;
1060 switch (dev
->chipset
) {
1064 tesla_class
= NVA0_3D_CLASS
;
1067 tesla_class
= NVAF_3D_CLASS
;
1070 tesla_class
= NVA3_3D_CLASS
;
1075 NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", dev
->chipset
);
1078 screen
->base
.class_3d
= tesla_class
;
1080 ret
= nouveau_object_new(chan
, 0xbeef5097, tesla_class
,
1081 NULL
, 0, &screen
->tesla
);
1083 NOUVEAU_ERR("Failed to allocate PGRAPH context for 3D: %d\n", ret
);
1087 /* This over-allocates by a page. The GP, which would execute at the end of
1088 * the last page, would trigger faults. The going theory is that it
1089 * prefetches up to a certain amount.
1091 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16,
1092 (3 << NV50_CODE_BO_SIZE_LOG2
) + 0x1000,
1093 NULL
, &screen
->code
);
1095 NOUVEAU_ERR("Failed to allocate code bo: %d\n", ret
);
1099 nouveau_heap_init(&screen
->vp_code_heap
, 0, 1 << NV50_CODE_BO_SIZE_LOG2
);
1100 nouveau_heap_init(&screen
->gp_code_heap
, 0, 1 << NV50_CODE_BO_SIZE_LOG2
);
1101 nouveau_heap_init(&screen
->fp_code_heap
, 0, 1 << NV50_CODE_BO_SIZE_LOG2
);
1103 nouveau_getparam(dev
, NOUVEAU_GETPARAM_GRAPH_UNITS
, &value
);
1105 screen
->TPs
= util_bitcount(value
& 0xffff);
1106 screen
->MPsInTP
= util_bitcount(value
& 0x0f000000);
1108 screen
->mp_count
= screen
->TPs
* screen
->MPsInTP
;
1110 stack_size
= util_next_power_of_two(screen
->TPs
) * screen
->MPsInTP
*
1111 STACK_WARPS_ALLOC
* 64 * 8;
1113 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16, stack_size
, NULL
,
1116 NOUVEAU_ERR("Failed to allocate stack bo: %d\n", ret
);
1120 uint64_t size_of_one_temp
= util_next_power_of_two(screen
->TPs
) *
1121 screen
->MPsInTP
* LOCAL_WARPS_ALLOC
* THREADS_IN_WARP
*
1123 screen
->max_tls_space
= dev
->vram_size
/ size_of_one_temp
* ONE_TEMP_SIZE
;
1124 screen
->max_tls_space
/= 2; /* half of vram */
1126 /* hw can address max 64 KiB */
1127 screen
->max_tls_space
= MIN2(screen
->max_tls_space
, 64 << 10);
1130 unsigned tls_space
= 4/*temps*/ * ONE_TEMP_SIZE
;
1131 ret
= nv50_tls_alloc(screen
, tls_space
, &tls_size
);
1135 if (nouveau_mesa_debug
)
1136 debug_printf("TPs = %u, MPsInTP = %u, VRAM = %"PRIu64
" MiB, tls_size = %"PRIu64
" KiB\n",
1137 screen
->TPs
, screen
->MPsInTP
, dev
->vram_size
>> 20, tls_size
>> 10);
1139 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16, 4 << 16, NULL
,
1142 NOUVEAU_ERR("Failed to allocate uniforms bo: %d\n", ret
);
1146 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16, 3 << 16, NULL
,
1149 NOUVEAU_ERR("Failed to allocate TIC/TSC bo: %d\n", ret
);
1153 screen
->tic
.entries
= CALLOC(4096, sizeof(void *));
1154 screen
->tsc
.entries
= screen
->tic
.entries
+ 2048;
1156 if (!nv50_blitter_create(screen
))
1159 nv50_screen_init_hwctx(screen
);
1161 ret
= nv50_screen_compute_setup(screen
, screen
->base
.pushbuf
);
1163 NOUVEAU_ERR("Failed to init compute context: %d\n", ret
);
1167 nouveau_fence_new(&screen
->base
, &screen
->base
.fence
.current
);
1169 return &screen
->base
;
1172 screen
->base
.base
.context_create
= NULL
;
1173 return &screen
->base
;
1177 nv50_screen_tic_alloc(struct nv50_screen
*screen
, void *entry
)
1179 int i
= screen
->tic
.next
;
1181 while (screen
->tic
.lock
[i
/ 32] & (1 << (i
% 32)))
1182 i
= (i
+ 1) & (NV50_TIC_MAX_ENTRIES
- 1);
1184 screen
->tic
.next
= (i
+ 1) & (NV50_TIC_MAX_ENTRIES
- 1);
1186 if (screen
->tic
.entries
[i
])
1187 nv50_tic_entry(screen
->tic
.entries
[i
])->id
= -1;
1189 screen
->tic
.entries
[i
] = entry
;
1194 nv50_screen_tsc_alloc(struct nv50_screen
*screen
, void *entry
)
1196 int i
= screen
->tsc
.next
;
1198 while (screen
->tsc
.lock
[i
/ 32] & (1 << (i
% 32)))
1199 i
= (i
+ 1) & (NV50_TSC_MAX_ENTRIES
- 1);
1201 screen
->tsc
.next
= (i
+ 1) & (NV50_TSC_MAX_ENTRIES
- 1);
1203 if (screen
->tsc
.entries
[i
])
1204 nv50_tsc_entry(screen
->tsc
.entries
[i
])->id
= -1;
1206 screen
->tsc
.entries
[i
] = entry
;