gallium: add PIPE_CAP_MAX_VARYINGS
[mesa.git] / src / gallium / drivers / nouveau / nvc0 / nvc0_screen.c
1 /*
2 * Copyright 2010 Christoph Bumiller
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 */
22
23 #include <xf86drm.h>
24 #include <nouveau_drm.h>
25 #include <nvif/class.h>
26 #include "util/u_format.h"
27 #include "util/u_format_s3tc.h"
28 #include "util/u_screen.h"
29 #include "pipe/p_screen.h"
30
31 #include "nouveau_vp3_video.h"
32
33 #include "nvc0/nvc0_context.h"
34 #include "nvc0/nvc0_screen.h"
35
36 #include "nvc0/mme/com9097.mme.h"
37 #include "nvc0/mme/com90c0.mme.h"
38
39 #include "nv50/g80_texture.xml.h"
40
41 static boolean
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,
47 unsigned bindings)
48 {
49 const struct util_format_description *desc = util_format_description(format);
50
51 if (sample_count > 8)
52 return false;
53 if (!(0x117 & (1 << sample_count))) /* 0, 1, 2, 4 or 8 */
54 return false;
55
56 if (MAX2(1, sample_count) != MAX2(1, storage_sample_count))
57 return false;
58
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.
61 */
62 if (format == PIPE_FORMAT_NONE && bindings & PIPE_BIND_RENDER_TARGET)
63 return true;
64
65 if ((bindings & PIPE_BIND_SAMPLER_VIEW) && (target != PIPE_BUFFER))
66 if (util_format_get_blocksizebits(format) == 3 * 32)
67 return false;
68
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) ||
74 sample_count > 1)
75 return false;
76
77 /* Restrict ETC2 and ASTC formats here. These are only supported on GK20A.
78 */
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 &&
84 */
85 nouveau_screen(pscreen)->class_3d != NVEA_3D_CLASS)
86 return false;
87
88 /* shared is always supported */
89 bindings &= ~(PIPE_BIND_LINEAR |
90 PIPE_BIND_SHARED);
91
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. */
97 return false;
98 }
99 }
100
101 return (( nvc0_format_table[format].usage |
102 nvc0_vertex_format[format].usage) & bindings) == bindings;
103 }
104
105 static int
106 nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
107 {
108 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
109 struct nouveau_device *dev = nouveau_screen(pscreen)->device;
110
111 switch (param) {
112 /* non-boolean caps */
113 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
114 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
115 return 15;
116 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
117 return 12;
118 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
119 return 2048;
120 case PIPE_CAP_MIN_TEXEL_OFFSET:
121 return -8;
122 case PIPE_CAP_MAX_TEXEL_OFFSET:
123 return 7;
124 case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
125 return -32;
126 case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
127 return 31;
128 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
129 return 128 * 1024 * 1024;
130 case PIPE_CAP_GLSL_FEATURE_LEVEL:
131 return 430;
132 case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY:
133 return 430;
134 case PIPE_CAP_MAX_RENDER_TARGETS:
135 return 8;
136 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
137 return 1;
138 case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
139 case PIPE_CAP_RASTERIZER_SUBPIXEL_BITS:
140 return 8;
141 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
142 return 4;
143 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
144 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
145 return 128;
146 case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
147 case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
148 return 1024;
149 case PIPE_CAP_MAX_VERTEX_STREAMS:
150 return 4;
151 case PIPE_CAP_MAX_GS_INVOCATIONS:
152 return 32;
153 case PIPE_CAP_MAX_SHADER_BUFFER_SIZE:
154 return 1 << 27;
155 case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
156 return 2048;
157 case PIPE_CAP_MAX_VERTEX_ELEMENT_SRC_OFFSET:
158 return 2047;
159 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
160 return 256;
161 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
162 if (class_3d < GM107_3D_CLASS)
163 return 256; /* IMAGE bindings require alignment to 256 */
164 return 16;
165 case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
166 return 16;
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:
172 return 4;
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:
178 return 30;
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;
185 case PIPE_CAP_MAX_VARYINGS:
186 /* NOTE: These only count our slots for GENERIC varyings.
187 * The address space may be larger, but the actual hard limit seems to be
188 * less than what the address space layout permits, so don't add TEXCOORD,
189 * COLOR, etc. here.
190 */
191 return 0x1f0 / 16;
192
193 /* supported caps */
194 case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
195 case PIPE_CAP_TEXTURE_MIRROR_CLAMP_TO_EDGE:
196 case PIPE_CAP_TEXTURE_SWIZZLE:
197 case PIPE_CAP_NPOT_TEXTURES:
198 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
199 case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
200 case PIPE_CAP_ANISOTROPIC_FILTER:
201 case PIPE_CAP_SEAMLESS_CUBE_MAP:
202 case PIPE_CAP_CUBE_MAP_ARRAY:
203 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
204 case PIPE_CAP_TEXTURE_MULTISAMPLE:
205 case PIPE_CAP_DEPTH_CLIP_DISABLE:
206 case PIPE_CAP_POINT_SPRITE:
207 case PIPE_CAP_TGSI_TEXCOORD:
208 case PIPE_CAP_SM3:
209 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
210 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
211 case PIPE_CAP_VERTEX_COLOR_CLAMPED:
212 case PIPE_CAP_QUERY_TIMESTAMP:
213 case PIPE_CAP_QUERY_TIME_ELAPSED:
214 case PIPE_CAP_OCCLUSION_QUERY:
215 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
216 case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
217 case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
218 case PIPE_CAP_BLEND_EQUATION_SEPARATE:
219 case PIPE_CAP_INDEP_BLEND_ENABLE:
220 case PIPE_CAP_INDEP_BLEND_FUNC:
221 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
222 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
223 case PIPE_CAP_PRIMITIVE_RESTART:
224 case PIPE_CAP_TGSI_INSTANCEID:
225 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
226 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
227 case PIPE_CAP_CONDITIONAL_RENDER:
228 case PIPE_CAP_TEXTURE_BARRIER:
229 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
230 case PIPE_CAP_START_INSTANCE:
231 case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
232 case PIPE_CAP_DRAW_INDIRECT:
233 case PIPE_CAP_USER_VERTEX_BUFFERS:
234 case PIPE_CAP_TEXTURE_QUERY_LOD:
235 case PIPE_CAP_SAMPLE_SHADING:
236 case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
237 case PIPE_CAP_TEXTURE_GATHER_SM5:
238 case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
239 case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
240 case PIPE_CAP_SAMPLER_VIEW_TARGET:
241 case PIPE_CAP_CLIP_HALFZ:
242 case PIPE_CAP_POLYGON_OFFSET_CLAMP:
243 case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
244 case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
245 case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
246 case PIPE_CAP_DEPTH_BOUNDS_TEST:
247 case PIPE_CAP_TGSI_TXQS:
248 case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
249 case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
250 case PIPE_CAP_SHAREABLE_SHADERS:
251 case PIPE_CAP_CLEAR_TEXTURE:
252 case PIPE_CAP_DRAW_PARAMETERS:
253 case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
254 case PIPE_CAP_MULTI_DRAW_INDIRECT:
255 case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
256 case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
257 case PIPE_CAP_QUERY_BUFFER_OBJECT:
258 case PIPE_CAP_INVALIDATE_BUFFER:
259 case PIPE_CAP_STRING_MARKER:
260 case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
261 case PIPE_CAP_CULL_DISTANCE:
262 case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
263 case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
264 case PIPE_CAP_TGSI_VOTE:
265 case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
266 case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
267 case PIPE_CAP_TGSI_MUL_ZERO_WINS:
268 case PIPE_CAP_DOUBLES:
269 case PIPE_CAP_INT64:
270 case PIPE_CAP_TGSI_TEX_TXF_LZ:
271 case PIPE_CAP_TGSI_CLOCK:
272 case PIPE_CAP_COMPUTE:
273 case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
274 case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
275 case PIPE_CAP_QUERY_SO_OVERFLOW:
276 case PIPE_CAP_DEST_SURFACE_SRGB_CONTROL:
277 return 1;
278 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
279 return nouveau_screen(pscreen)->vram_domain & NOUVEAU_BO_VRAM ? 1 : 0;
280 case PIPE_CAP_TGSI_FS_FBFETCH:
281 return class_3d >= NVE4_3D_CLASS; /* needs testing on fermi */
282 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
283 case PIPE_CAP_TGSI_BALLOT:
284 case PIPE_CAP_BINDLESS_TEXTURE:
285 return class_3d >= NVE4_3D_CLASS;
286 case PIPE_CAP_TGSI_ATOMFADD:
287 return class_3d < GM107_3D_CLASS; /* needs additional lowering */
288 case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
289 case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
290 case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
291 case PIPE_CAP_POST_DEPTH_COVERAGE:
292 case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_TRIANGLES:
293 case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_POINTS_LINES:
294 case PIPE_CAP_CONSERVATIVE_RASTER_POST_DEPTH_COVERAGE:
295 case PIPE_CAP_PROGRAMMABLE_SAMPLE_LOCATIONS:
296 return class_3d >= GM200_3D_CLASS;
297 case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES:
298 return class_3d >= GP100_3D_CLASS;
299
300 /* unsupported caps */
301 case PIPE_CAP_DEPTH_CLIP_DISABLE_SEPARATE:
302 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
303 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
304 case PIPE_CAP_SHADER_STENCIL_EXPORT:
305 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
306 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
307 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
308 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
309 case PIPE_CAP_FAKE_SW_MSAA:
310 case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
311 case PIPE_CAP_VERTEXID_NOBASE:
312 case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
313 case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
314 case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
315 case PIPE_CAP_GENERATE_MIPMAP:
316 case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
317 case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
318 case PIPE_CAP_QUERY_MEMORY_INFO:
319 case PIPE_CAP_PCI_GROUP:
320 case PIPE_CAP_PCI_BUS:
321 case PIPE_CAP_PCI_DEVICE:
322 case PIPE_CAP_PCI_FUNCTION:
323 case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
324 case PIPE_CAP_NATIVE_FENCE_FD:
325 case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
326 case PIPE_CAP_INT64_DIVMOD:
327 case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
328 case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
329 case PIPE_CAP_MEMOBJ:
330 case PIPE_CAP_LOAD_CONSTBUF:
331 case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS:
332 case PIPE_CAP_TILE_RASTER_ORDER:
333 case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
334 case PIPE_CAP_FRAMEBUFFER_MSAA_CONSTRAINTS:
335 case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
336 case PIPE_CAP_CONTEXT_PRIORITY_MASK:
337 case PIPE_CAP_FENCE_SIGNAL:
338 case PIPE_CAP_CONSTBUF0_FLAGS:
339 case PIPE_CAP_PACKED_UNIFORMS:
340 case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_POINTS_LINES:
341 case PIPE_CAP_MAX_COMBINED_SHADER_BUFFERS:
342 case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTERS:
343 case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTER_BUFFERS:
344 case PIPE_CAP_SURFACE_SAMPLE_COUNT:
345 case PIPE_CAP_QUERY_PIPELINE_STATISTICS_SINGLE:
346 case PIPE_CAP_RGB_OVERRIDE_DST_ALPHA_BLEND:
347 case PIPE_CAP_GLSL_TESS_LEVELS_AS_INPUTS:
348 return 0;
349
350 case PIPE_CAP_VENDOR_ID:
351 return 0x10de;
352 case PIPE_CAP_DEVICE_ID: {
353 uint64_t device_id;
354 if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
355 NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
356 return -1;
357 }
358 return device_id;
359 }
360 case PIPE_CAP_ACCELERATED:
361 return 1;
362 case PIPE_CAP_VIDEO_MEMORY:
363 return dev->vram_size >> 20;
364 case PIPE_CAP_UMA:
365 return 0;
366 default:
367 debug_printf("%s: unhandled cap %d\n", __func__, param);
368 return u_pipe_screen_get_param_defaults(pscreen, param);
369 }
370 }
371
372 static int
373 nvc0_screen_get_shader_param(struct pipe_screen *pscreen,
374 enum pipe_shader_type shader,
375 enum pipe_shader_cap param)
376 {
377 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
378
379 switch (shader) {
380 case PIPE_SHADER_VERTEX:
381 case PIPE_SHADER_GEOMETRY:
382 case PIPE_SHADER_FRAGMENT:
383 case PIPE_SHADER_COMPUTE:
384 case PIPE_SHADER_TESS_CTRL:
385 case PIPE_SHADER_TESS_EVAL:
386 break;
387 default:
388 return 0;
389 }
390
391 switch (param) {
392 case PIPE_SHADER_CAP_PREFERRED_IR:
393 return PIPE_SHADER_IR_TGSI;
394 case PIPE_SHADER_CAP_SUPPORTED_IRS:
395 return 1 << PIPE_SHADER_IR_TGSI;
396 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
397 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
398 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
399 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
400 return 16384;
401 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
402 return 16;
403 case PIPE_SHADER_CAP_MAX_INPUTS:
404 return 0x200 / 16;
405 case PIPE_SHADER_CAP_MAX_OUTPUTS:
406 return 32;
407 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
408 return NVC0_MAX_CONSTBUF_SIZE;
409 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
410 return NVC0_MAX_PIPE_CONSTBUFS;
411 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
412 return shader != PIPE_SHADER_FRAGMENT;
413 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
414 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
415 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
416 return 1;
417 case PIPE_SHADER_CAP_MAX_TEMPS:
418 return NVC0_CAP_MAX_PROGRAM_TEMPS;
419 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
420 return 1;
421 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
422 return 1;
423 case PIPE_SHADER_CAP_SUBROUTINES:
424 return 1;
425 case PIPE_SHADER_CAP_INTEGERS:
426 return 1;
427 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
428 return 1;
429 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
430 return 1;
431 case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
432 return 1;
433 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
434 case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
435 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
436 case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
437 case PIPE_SHADER_CAP_INT64_ATOMICS:
438 case PIPE_SHADER_CAP_FP16:
439 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
440 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
441 return 0;
442 case PIPE_SHADER_CAP_SCALAR_ISA:
443 return 1;
444 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
445 return NVC0_MAX_BUFFERS;
446 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
447 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
448 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
449 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
450 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
451 return 32;
452 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
453 if (class_3d >= NVE4_3D_CLASS)
454 return NVC0_MAX_IMAGES;
455 if (shader == PIPE_SHADER_FRAGMENT || shader == PIPE_SHADER_COMPUTE)
456 return NVC0_MAX_IMAGES;
457 return 0;
458 default:
459 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
460 return 0;
461 }
462 }
463
464 static float
465 nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
466 {
467 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
468
469 switch (param) {
470 case PIPE_CAPF_MAX_LINE_WIDTH:
471 case PIPE_CAPF_MAX_LINE_WIDTH_AA:
472 return 10.0f;
473 case PIPE_CAPF_MAX_POINT_WIDTH:
474 return 63.0f;
475 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
476 return 63.375f;
477 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
478 return 16.0f;
479 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
480 return 15.0f;
481 case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE:
482 return 0.0f;
483 case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE:
484 return class_3d >= GM200_3D_CLASS ? 0.75f : 0.0f;
485 case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY:
486 return class_3d >= GM200_3D_CLASS ? 0.25f : 0.0f;
487 }
488
489 NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
490 return 0.0f;
491 }
492
493 static int
494 nvc0_screen_get_compute_param(struct pipe_screen *pscreen,
495 enum pipe_shader_ir ir_type,
496 enum pipe_compute_cap param, void *data)
497 {
498 struct nvc0_screen *screen = nvc0_screen(pscreen);
499 const uint16_t obj_class = screen->compute->oclass;
500
501 #define RET(x) do { \
502 if (data) \
503 memcpy(data, x, sizeof(x)); \
504 return sizeof(x); \
505 } while (0)
506
507 switch (param) {
508 case PIPE_COMPUTE_CAP_GRID_DIMENSION:
509 RET((uint64_t []) { 3 });
510 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
511 if (obj_class >= NVE4_COMPUTE_CLASS) {
512 RET(((uint64_t []) { 0x7fffffff, 65535, 65535 }));
513 } else {
514 RET(((uint64_t []) { 65535, 65535, 65535 }));
515 }
516 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
517 RET(((uint64_t []) { 1024, 1024, 64 }));
518 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
519 RET((uint64_t []) { 1024 });
520 case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
521 if (obj_class >= NVE4_COMPUTE_CLASS) {
522 RET((uint64_t []) { 1024 });
523 } else {
524 RET((uint64_t []) { 512 });
525 }
526 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */
527 RET((uint64_t []) { 1ULL << 40 });
528 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
529 switch (obj_class) {
530 case GM200_COMPUTE_CLASS:
531 RET((uint64_t []) { 96 << 10 });
532 break;
533 case GM107_COMPUTE_CLASS:
534 RET((uint64_t []) { 64 << 10 });
535 break;
536 default:
537 RET((uint64_t []) { 48 << 10 });
538 break;
539 }
540 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
541 RET((uint64_t []) { 512 << 10 });
542 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
543 RET((uint64_t []) { 4096 });
544 case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
545 RET((uint32_t []) { 32 });
546 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
547 RET((uint64_t []) { 1ULL << 40 });
548 case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
549 RET((uint32_t []) { 0 });
550 case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
551 RET((uint32_t []) { screen->mp_count_compute });
552 case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
553 RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
554 case PIPE_COMPUTE_CAP_ADDRESS_BITS:
555 RET((uint32_t []) { 64 });
556 default:
557 return 0;
558 }
559
560 #undef RET
561 }
562
563 static void
564 nvc0_screen_get_sample_pixel_grid(struct pipe_screen *pscreen,
565 unsigned sample_count,
566 unsigned *width, unsigned *height)
567 {
568 switch (sample_count) {
569 case 0:
570 case 1:
571 /* this could be 4x4, but the GL state tracker makes it difficult to
572 * create a 1x MSAA texture and smaller grids save CB space */
573 *width = 2;
574 *height = 4;
575 break;
576 case 2:
577 *width = 2;
578 *height = 4;
579 break;
580 case 4:
581 *width = 2;
582 *height = 2;
583 break;
584 case 8:
585 *width = 1;
586 *height = 2;
587 break;
588 default:
589 assert(0);
590 }
591 }
592
593 static void
594 nvc0_screen_destroy(struct pipe_screen *pscreen)
595 {
596 struct nvc0_screen *screen = nvc0_screen(pscreen);
597
598 if (!nouveau_drm_screen_unref(&screen->base))
599 return;
600
601 if (screen->base.fence.current) {
602 struct nouveau_fence *current = NULL;
603
604 /* nouveau_fence_wait will create a new current fence, so wait on the
605 * _current_ one, and remove both.
606 */
607 nouveau_fence_ref(screen->base.fence.current, &current);
608 nouveau_fence_wait(current, NULL);
609 nouveau_fence_ref(NULL, &current);
610 nouveau_fence_ref(NULL, &screen->base.fence.current);
611 }
612 if (screen->base.pushbuf)
613 screen->base.pushbuf->user_priv = NULL;
614
615 if (screen->blitter)
616 nvc0_blitter_destroy(screen);
617 if (screen->pm.prog) {
618 screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
619 nvc0_program_destroy(NULL, screen->pm.prog);
620 FREE(screen->pm.prog);
621 }
622
623 nouveau_bo_ref(NULL, &screen->text);
624 nouveau_bo_ref(NULL, &screen->uniform_bo);
625 nouveau_bo_ref(NULL, &screen->tls);
626 nouveau_bo_ref(NULL, &screen->txc);
627 nouveau_bo_ref(NULL, &screen->fence.bo);
628 nouveau_bo_ref(NULL, &screen->poly_cache);
629
630 nouveau_heap_destroy(&screen->lib_code);
631 nouveau_heap_destroy(&screen->text_heap);
632
633 FREE(screen->tic.entries);
634
635 nouveau_object_del(&screen->eng3d);
636 nouveau_object_del(&screen->eng2d);
637 nouveau_object_del(&screen->m2mf);
638 nouveau_object_del(&screen->compute);
639 nouveau_object_del(&screen->nvsw);
640
641 nouveau_screen_fini(&screen->base);
642
643 FREE(screen);
644 }
645
646 static int
647 nvc0_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos,
648 unsigned size, const uint32_t *data)
649 {
650 struct nouveau_pushbuf *push = screen->base.pushbuf;
651
652 size /= 4;
653
654 assert((pos + size) <= 0x800);
655
656 BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2);
657 PUSH_DATA (push, (m - 0x3800) / 8);
658 PUSH_DATA (push, pos);
659 BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1);
660 PUSH_DATA (push, pos);
661 PUSH_DATAp(push, data, size);
662
663 return pos + size;
664 }
665
666 static void
667 nvc0_magic_3d_init(struct nouveau_pushbuf *push, uint16_t obj_class)
668 {
669 BEGIN_NVC0(push, SUBC_3D(0x10cc), 1);
670 PUSH_DATA (push, 0xff);
671 BEGIN_NVC0(push, SUBC_3D(0x10e0), 2);
672 PUSH_DATA (push, 0xff);
673 PUSH_DATA (push, 0xff);
674 BEGIN_NVC0(push, SUBC_3D(0x10ec), 2);
675 PUSH_DATA (push, 0xff);
676 PUSH_DATA (push, 0xff);
677 BEGIN_NVC0(push, SUBC_3D(0x074c), 1);
678 PUSH_DATA (push, 0x3f);
679
680 BEGIN_NVC0(push, SUBC_3D(0x16a8), 1);
681 PUSH_DATA (push, (3 << 16) | 3);
682 BEGIN_NVC0(push, SUBC_3D(0x1794), 1);
683 PUSH_DATA (push, (2 << 16) | 2);
684
685 if (obj_class < GM107_3D_CLASS) {
686 BEGIN_NVC0(push, SUBC_3D(0x12ac), 1);
687 PUSH_DATA (push, 0);
688 }
689 BEGIN_NVC0(push, SUBC_3D(0x0218), 1);
690 PUSH_DATA (push, 0x10);
691 BEGIN_NVC0(push, SUBC_3D(0x10fc), 1);
692 PUSH_DATA (push, 0x10);
693 BEGIN_NVC0(push, SUBC_3D(0x1290), 1);
694 PUSH_DATA (push, 0x10);
695 BEGIN_NVC0(push, SUBC_3D(0x12d8), 2);
696 PUSH_DATA (push, 0x10);
697 PUSH_DATA (push, 0x10);
698 BEGIN_NVC0(push, SUBC_3D(0x1140), 1);
699 PUSH_DATA (push, 0x10);
700 BEGIN_NVC0(push, SUBC_3D(0x1610), 1);
701 PUSH_DATA (push, 0xe);
702
703 BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_GEN_MODE), 1);
704 PUSH_DATA (push, NVC0_3D_VERTEX_ID_GEN_MODE_DRAW_ARRAYS_ADD_START);
705 BEGIN_NVC0(push, SUBC_3D(0x030c), 1);
706 PUSH_DATA (push, 0);
707 BEGIN_NVC0(push, SUBC_3D(0x0300), 1);
708 PUSH_DATA (push, 3);
709
710 BEGIN_NVC0(push, SUBC_3D(0x02d0), 1);
711 PUSH_DATA (push, 0x3fffff);
712 BEGIN_NVC0(push, SUBC_3D(0x0fdc), 1);
713 PUSH_DATA (push, 1);
714 BEGIN_NVC0(push, SUBC_3D(0x19c0), 1);
715 PUSH_DATA (push, 1);
716
717 if (obj_class < GM107_3D_CLASS) {
718 BEGIN_NVC0(push, SUBC_3D(0x075c), 1);
719 PUSH_DATA (push, 3);
720
721 if (obj_class >= NVE4_3D_CLASS) {
722 BEGIN_NVC0(push, SUBC_3D(0x07fc), 1);
723 PUSH_DATA (push, 1);
724 }
725 }
726
727 /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
728 * are supposed to do */
729 }
730
731 static void
732 nvc0_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
733 {
734 struct nvc0_screen *screen = nvc0_screen(pscreen);
735 struct nouveau_pushbuf *push = screen->base.pushbuf;
736
737 /* we need to do it after possible flush in MARK_RING */
738 *sequence = ++screen->base.fence.sequence;
739
740 assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
741 PUSH_DATA (push, NVC0_FIFO_PKHDR_SQ(NVC0_3D(QUERY_ADDRESS_HIGH), 4));
742 PUSH_DATAh(push, screen->fence.bo->offset);
743 PUSH_DATA (push, screen->fence.bo->offset);
744 PUSH_DATA (push, *sequence);
745 PUSH_DATA (push, NVC0_3D_QUERY_GET_FENCE | NVC0_3D_QUERY_GET_SHORT |
746 (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT));
747 }
748
749 static u32
750 nvc0_screen_fence_update(struct pipe_screen *pscreen)
751 {
752 struct nvc0_screen *screen = nvc0_screen(pscreen);
753 return screen->fence.map[0];
754 }
755
756 static int
757 nvc0_screen_init_compute(struct nvc0_screen *screen)
758 {
759 screen->base.base.get_compute_param = nvc0_screen_get_compute_param;
760
761 switch (screen->base.device->chipset & ~0xf) {
762 case 0xc0:
763 case 0xd0:
764 return nvc0_screen_compute_setup(screen, screen->base.pushbuf);
765 case 0xe0:
766 case 0xf0:
767 case 0x100:
768 case 0x110:
769 case 0x120:
770 case 0x130:
771 return nve4_screen_compute_setup(screen, screen->base.pushbuf);
772 default:
773 return -1;
774 }
775 }
776
777 static int
778 nvc0_screen_resize_tls_area(struct nvc0_screen *screen,
779 uint32_t lpos, uint32_t lneg, uint32_t cstack)
780 {
781 struct nouveau_bo *bo = NULL;
782 int ret;
783 uint64_t size = (lpos + lneg) * 32 + cstack;
784
785 if (size >= (1 << 20)) {
786 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64"\n", size);
787 return -1;
788 }
789
790 size *= (screen->base.device->chipset >= 0xe0) ? 64 : 48; /* max warps */
791 size = align(size, 0x8000);
792 size *= screen->mp_count;
793
794 size = align(size, 1 << 17);
795
796 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base), 1 << 17, size,
797 NULL, &bo);
798 if (ret)
799 return ret;
800
801 /* Make sure that the pushbuf has acquired a reference to the old tls
802 * segment, as it may have commands that will reference it.
803 */
804 if (screen->tls)
805 PUSH_REFN(screen->base.pushbuf, screen->tls,
806 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RDWR);
807 nouveau_bo_ref(NULL, &screen->tls);
808 screen->tls = bo;
809 return 0;
810 }
811
812 int
813 nvc0_screen_resize_text_area(struct nvc0_screen *screen, uint64_t size)
814 {
815 struct nouveau_pushbuf *push = screen->base.pushbuf;
816 struct nouveau_bo *bo;
817 int ret;
818
819 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base),
820 1 << 17, size, NULL, &bo);
821 if (ret)
822 return ret;
823
824 /* Make sure that the pushbuf has acquired a reference to the old text
825 * segment, as it may have commands that will reference it.
826 */
827 if (screen->text)
828 PUSH_REFN(push, screen->text,
829 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RD);
830 nouveau_bo_ref(NULL, &screen->text);
831 screen->text = bo;
832
833 nouveau_heap_destroy(&screen->lib_code);
834 nouveau_heap_destroy(&screen->text_heap);
835
836 /* XXX: getting a page fault at the end of the code buffer every few
837 * launches, don't use the last 256 bytes to work around them - prefetch ?
838 */
839 nouveau_heap_init(&screen->text_heap, 0, size - 0x100);
840
841 /* update the code segment setup */
842 BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
843 PUSH_DATAh(push, screen->text->offset);
844 PUSH_DATA (push, screen->text->offset);
845 if (screen->compute) {
846 BEGIN_NVC0(push, NVC0_CP(CODE_ADDRESS_HIGH), 2);
847 PUSH_DATAh(push, screen->text->offset);
848 PUSH_DATA (push, screen->text->offset);
849 }
850
851 return 0;
852 }
853
854 void
855 nvc0_screen_bind_cb_3d(struct nvc0_screen *screen, bool *can_serialize,
856 int stage, int index, int size, uint64_t addr)
857 {
858 assert(stage != 5);
859
860 struct nouveau_pushbuf *push = screen->base.pushbuf;
861
862 if (screen->base.class_3d >= GM107_3D_CLASS) {
863 struct nvc0_cb_binding *binding = &screen->cb_bindings[stage][index];
864
865 // TODO: Better figure out the conditions in which this is needed
866 bool serialize = binding->addr == addr && binding->size != size;
867 if (can_serialize)
868 serialize = serialize && *can_serialize;
869 if (serialize) {
870 IMMED_NVC0(push, NVC0_3D(SERIALIZE), 0);
871 if (can_serialize)
872 *can_serialize = false;
873 }
874
875 binding->addr = addr;
876 binding->size = size;
877 }
878
879 if (size >= 0) {
880 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
881 PUSH_DATA (push, size);
882 PUSH_DATAh(push, addr);
883 PUSH_DATA (push, addr);
884 }
885 IMMED_NVC0(push, NVC0_3D(CB_BIND(stage)), (index << 4) | (size >= 0));
886 }
887
888 #define FAIL_SCREEN_INIT(str, err) \
889 do { \
890 NOUVEAU_ERR(str, err); \
891 goto fail; \
892 } while(0)
893
894 struct nouveau_screen *
895 nvc0_screen_create(struct nouveau_device *dev)
896 {
897 struct nvc0_screen *screen;
898 struct pipe_screen *pscreen;
899 struct nouveau_object *chan;
900 struct nouveau_pushbuf *push;
901 uint64_t value;
902 uint32_t obj_class;
903 uint32_t flags;
904 int ret;
905 unsigned i;
906
907 switch (dev->chipset & ~0xf) {
908 case 0xc0:
909 case 0xd0:
910 case 0xe0:
911 case 0xf0:
912 case 0x100:
913 case 0x110:
914 case 0x120:
915 case 0x130:
916 break;
917 default:
918 return NULL;
919 }
920
921 screen = CALLOC_STRUCT(nvc0_screen);
922 if (!screen)
923 return NULL;
924 pscreen = &screen->base.base;
925 pscreen->destroy = nvc0_screen_destroy;
926
927 ret = nouveau_screen_init(&screen->base, dev);
928 if (ret)
929 FAIL_SCREEN_INIT("Base screen init failed: %d\n", ret);
930 chan = screen->base.channel;
931 push = screen->base.pushbuf;
932 push->user_priv = screen;
933 push->rsvd_kick = 5;
934
935 /* TODO: could this be higher on Kepler+? how does reclocking vs no
936 * reclocking affect performance?
937 * TODO: could this be higher on Fermi?
938 */
939 if (dev->chipset >= 0xe0)
940 screen->base.transfer_pushbuf_threshold = 1024;
941
942 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
943 PIPE_BIND_SHADER_BUFFER |
944 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
945 PIPE_BIND_COMMAND_ARGS_BUFFER | PIPE_BIND_QUERY_BUFFER;
946 screen->base.sysmem_bindings |=
947 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
948
949 if (screen->base.vram_domain & NOUVEAU_BO_GART) {
950 screen->base.sysmem_bindings |= screen->base.vidmem_bindings;
951 screen->base.vidmem_bindings = 0;
952 }
953
954 pscreen->context_create = nvc0_create;
955 pscreen->is_format_supported = nvc0_screen_is_format_supported;
956 pscreen->get_param = nvc0_screen_get_param;
957 pscreen->get_shader_param = nvc0_screen_get_shader_param;
958 pscreen->get_paramf = nvc0_screen_get_paramf;
959 pscreen->get_sample_pixel_grid = nvc0_screen_get_sample_pixel_grid;
960 pscreen->get_driver_query_info = nvc0_screen_get_driver_query_info;
961 pscreen->get_driver_query_group_info = nvc0_screen_get_driver_query_group_info;
962
963 nvc0_screen_init_resource_functions(pscreen);
964
965 screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
966 screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
967
968 flags = NOUVEAU_BO_GART | NOUVEAU_BO_MAP;
969 if (screen->base.drm->version >= 0x01000202)
970 flags |= NOUVEAU_BO_COHERENT;
971
972 ret = nouveau_bo_new(dev, flags, 0, 4096, NULL, &screen->fence.bo);
973 if (ret)
974 FAIL_SCREEN_INIT("Error allocating fence BO: %d\n", ret);
975 nouveau_bo_map(screen->fence.bo, 0, NULL);
976 screen->fence.map = screen->fence.bo->map;
977 screen->base.fence.emit = nvc0_screen_fence_emit;
978 screen->base.fence.update = nvc0_screen_fence_update;
979
980
981 ret = nouveau_object_new(chan, (dev->chipset < 0xe0) ? 0x1f906e : 0x906e,
982 NVIF_CLASS_SW_GF100, NULL, 0, &screen->nvsw);
983 if (ret)
984 FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret);
985
986 BEGIN_NVC0(push, SUBC_SW(NV01_SUBCHAN_OBJECT), 1);
987 PUSH_DATA (push, screen->nvsw->handle);
988
989 switch (dev->chipset & ~0xf) {
990 case 0x130:
991 case 0x120:
992 case 0x110:
993 case 0x100:
994 case 0xf0:
995 obj_class = NVF0_P2MF_CLASS;
996 break;
997 case 0xe0:
998 obj_class = NVE4_P2MF_CLASS;
999 break;
1000 default:
1001 obj_class = NVC0_M2MF_CLASS;
1002 break;
1003 }
1004 ret = nouveau_object_new(chan, 0xbeef323f, obj_class, NULL, 0,
1005 &screen->m2mf);
1006 if (ret)
1007 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
1008
1009 BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
1010 PUSH_DATA (push, screen->m2mf->oclass);
1011 if (screen->m2mf->oclass == NVE4_P2MF_CLASS) {
1012 BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
1013 PUSH_DATA (push, 0xa0b5);
1014 }
1015
1016 ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
1017 &screen->eng2d);
1018 if (ret)
1019 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
1020
1021 BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
1022 PUSH_DATA (push, screen->eng2d->oclass);
1023 BEGIN_NVC0(push, SUBC_2D(NVC0_2D_SINGLE_GPC), 1);
1024 PUSH_DATA (push, 0);
1025 BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
1026 PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
1027 BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
1028 PUSH_DATA (push, 0);
1029 BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
1030 PUSH_DATA (push, 0);
1031 BEGIN_NVC0(push, SUBC_2D(0x0884), 1);
1032 PUSH_DATA (push, 0x3f);
1033 BEGIN_NVC0(push, SUBC_2D(0x0888), 1);
1034 PUSH_DATA (push, 1);
1035 BEGIN_NVC0(push, NVC0_2D(COND_MODE), 1);
1036 PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
1037
1038 BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
1039 PUSH_DATAh(push, screen->fence.bo->offset + 16);
1040 PUSH_DATA (push, screen->fence.bo->offset + 16);
1041
1042 switch (dev->chipset & ~0xf) {
1043 case 0x130:
1044 switch (dev->chipset) {
1045 case 0x130:
1046 case 0x13b:
1047 obj_class = GP100_3D_CLASS;
1048 break;
1049 default:
1050 obj_class = GP102_3D_CLASS;
1051 break;
1052 }
1053 break;
1054 case 0x120:
1055 obj_class = GM200_3D_CLASS;
1056 break;
1057 case 0x110:
1058 obj_class = GM107_3D_CLASS;
1059 break;
1060 case 0x100:
1061 case 0xf0:
1062 obj_class = NVF0_3D_CLASS;
1063 break;
1064 case 0xe0:
1065 switch (dev->chipset) {
1066 case 0xea:
1067 obj_class = NVEA_3D_CLASS;
1068 break;
1069 default:
1070 obj_class = NVE4_3D_CLASS;
1071 break;
1072 }
1073 break;
1074 case 0xd0:
1075 obj_class = NVC8_3D_CLASS;
1076 break;
1077 case 0xc0:
1078 default:
1079 switch (dev->chipset) {
1080 case 0xc8:
1081 obj_class = NVC8_3D_CLASS;
1082 break;
1083 case 0xc1:
1084 obj_class = NVC1_3D_CLASS;
1085 break;
1086 default:
1087 obj_class = NVC0_3D_CLASS;
1088 break;
1089 }
1090 break;
1091 }
1092 ret = nouveau_object_new(chan, 0xbeef003d, obj_class, NULL, 0,
1093 &screen->eng3d);
1094 if (ret)
1095 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
1096 screen->base.class_3d = obj_class;
1097
1098 BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
1099 PUSH_DATA (push, screen->eng3d->oclass);
1100
1101 BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
1102 PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
1103
1104 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
1105 /* kill shaders after about 1 second (at 100 MHz) */
1106 BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
1107 PUSH_DATA (push, 0x17);
1108 }
1109
1110 IMMED_NVC0(push, NVC0_3D(ZETA_COMP_ENABLE),
1111 screen->base.drm->version >= 0x01000101);
1112 BEGIN_NVC0(push, NVC0_3D(RT_COMP_ENABLE(0)), 8);
1113 for (i = 0; i < 8; ++i)
1114 PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
1115
1116 BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
1117 PUSH_DATA (push, 1);
1118
1119 BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
1120 PUSH_DATA (push, 0);
1121 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
1122 PUSH_DATA (push, 0);
1123 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
1124 PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
1125 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
1126 PUSH_DATA (push, 0);
1127 BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
1128 PUSH_DATA (push, 1);
1129 BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
1130 PUSH_DATA (push, 1);
1131 BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
1132 PUSH_DATA (push, 1);
1133 BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
1134 PUSH_DATA (push, 0);
1135 BEGIN_NVC0(push, NVC0_3D(SHADE_MODEL), 1);
1136 PUSH_DATA (push, NVC0_3D_SHADE_MODEL_SMOOTH);
1137 if (screen->eng3d->oclass < NVE4_3D_CLASS) {
1138 IMMED_NVC0(push, NVC0_3D(TEX_MISC), 0);
1139 } else {
1140 BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
1141 PUSH_DATA (push, 15);
1142 }
1143 BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
1144 PUSH_DATA (push, 8); /* 128 */
1145 BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
1146 PUSH_DATA (push, 1);
1147 if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
1148 BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
1149 PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
1150 }
1151
1152 nvc0_magic_3d_init(push, screen->eng3d->oclass);
1153
1154 ret = nvc0_screen_resize_text_area(screen, 1 << 19);
1155 if (ret)
1156 FAIL_SCREEN_INIT("Error allocating TEXT area: %d\n", ret);
1157
1158 /* 6 user uniform areas, 6 driver areas, and 1 for the runout */
1159 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 12, 13 << 16, NULL,
1160 &screen->uniform_bo);
1161 if (ret)
1162 FAIL_SCREEN_INIT("Error allocating uniform BO: %d\n", ret);
1163
1164 PUSH_REFN (push, screen->uniform_bo, NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_WR);
1165
1166 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
1167 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1168 PUSH_DATA (push, 256);
1169 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1170 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1171 BEGIN_1IC0(push, NVC0_3D(CB_POS), 5);
1172 PUSH_DATA (push, 0);
1173 PUSH_DATAf(push, 0.0f);
1174 PUSH_DATAf(push, 0.0f);
1175 PUSH_DATAf(push, 0.0f);
1176 PUSH_DATAf(push, 0.0f);
1177 BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
1178 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1179 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1180
1181 if (screen->base.drm->version >= 0x01000101) {
1182 ret = nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
1183 if (ret)
1184 FAIL_SCREEN_INIT("NOUVEAU_GETPARAM_GRAPH_UNITS failed: %d\n", ret);
1185 } else {
1186 if (dev->chipset >= 0xe0 && dev->chipset < 0xf0)
1187 value = (8 << 8) | 4;
1188 else
1189 value = (16 << 8) | 4;
1190 }
1191 screen->gpc_count = value & 0x000000ff;
1192 screen->mp_count = value >> 8;
1193 screen->mp_count_compute = screen->mp_count;
1194
1195 ret = nvc0_screen_resize_tls_area(screen, 128 * 16, 0, 0x200);
1196 if (ret)
1197 FAIL_SCREEN_INIT("Error allocating TLS area: %d\n", ret);
1198
1199 BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
1200 PUSH_DATAh(push, screen->tls->offset);
1201 PUSH_DATA (push, screen->tls->offset);
1202 PUSH_DATA (push, screen->tls->size >> 32);
1203 PUSH_DATA (push, screen->tls->size);
1204 BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
1205 PUSH_DATA (push, 0);
1206 /* Reduce likelihood of collision with real buffers by placing the hole at
1207 * the top of the 4G area. This will have to be dealt with for real
1208 * eventually by blocking off that area from the VM.
1209 */
1210 BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
1211 PUSH_DATA (push, 0xff << 24);
1212
1213 if (screen->eng3d->oclass < GM107_3D_CLASS) {
1214 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 20, NULL,
1215 &screen->poly_cache);
1216 if (ret)
1217 FAIL_SCREEN_INIT("Error allocating poly cache BO: %d\n", ret);
1218
1219 BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
1220 PUSH_DATAh(push, screen->poly_cache->offset);
1221 PUSH_DATA (push, screen->poly_cache->offset);
1222 PUSH_DATA (push, 3);
1223 }
1224
1225 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 17, NULL,
1226 &screen->txc);
1227 if (ret)
1228 FAIL_SCREEN_INIT("Error allocating txc BO: %d\n", ret);
1229
1230 BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
1231 PUSH_DATAh(push, screen->txc->offset);
1232 PUSH_DATA (push, screen->txc->offset);
1233 PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
1234 if (screen->eng3d->oclass >= GM107_3D_CLASS) {
1235 screen->tic.maxwell = true;
1236 if (screen->eng3d->oclass == GM107_3D_CLASS) {
1237 screen->tic.maxwell =
1238 debug_get_bool_option("NOUVEAU_MAXWELL_TIC", true);
1239 IMMED_NVC0(push, SUBC_3D(0x0f10), screen->tic.maxwell);
1240 }
1241 }
1242
1243 BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
1244 PUSH_DATAh(push, screen->txc->offset + 65536);
1245 PUSH_DATA (push, screen->txc->offset + 65536);
1246 PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
1247
1248 BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
1249 PUSH_DATA (push, 0);
1250 BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
1251 PUSH_DATA (push, 0);
1252 PUSH_DATA (push, 0);
1253 BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
1254 PUSH_DATA (push, 0x3f);
1255
1256 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
1257 PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
1258 BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
1259 for (i = 0; i < 8 * 2; ++i)
1260 PUSH_DATA(push, 0);
1261 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
1262 PUSH_DATA (push, 0);
1263 BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
1264 PUSH_DATA (push, 0);
1265
1266 /* neither scissors, viewport nor stencil mask should affect clears */
1267 BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
1268 PUSH_DATA (push, 0);
1269
1270 BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
1271 PUSH_DATA (push, 1);
1272 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1273 BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(i)), 2);
1274 PUSH_DATAf(push, 0.0f);
1275 PUSH_DATAf(push, 1.0f);
1276 }
1277 BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
1278 PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
1279
1280 /* We use scissors instead of exact view volume clipping,
1281 * so they're always enabled.
1282 */
1283 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1284 BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(i)), 3);
1285 PUSH_DATA (push, 1);
1286 PUSH_DATA (push, 8192 << 16);
1287 PUSH_DATA (push, 8192 << 16);
1288 }
1289
1290 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
1291
1292 i = 0;
1293 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, mme9097_per_instance_bf);
1294 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, mme9097_blend_enables);
1295 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, mme9097_vertex_array_select);
1296 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, mme9097_tep_select);
1297 MK_MACRO(NVC0_3D_MACRO_GP_SELECT, mme9097_gp_select);
1298 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, mme9097_poly_mode_front);
1299 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, mme9097_poly_mode_back);
1300 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT, mme9097_draw_arrays_indirect);
1301 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT, mme9097_draw_elts_indirect);
1302 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT, mme9097_draw_arrays_indirect_count);
1303 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT, mme9097_draw_elts_indirect_count);
1304 MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE, mme9097_query_buffer_write);
1305 MK_MACRO(NVC0_3D_MACRO_CONSERVATIVE_RASTER_STATE, mme9097_conservative_raster_state);
1306 MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER, mme9097_compute_counter);
1307 MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER_TO_QUERY, mme9097_compute_counter_to_query);
1308 MK_MACRO(NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT, mme90c0_launch_grid_indirect);
1309
1310 BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
1311 PUSH_DATA (push, 1);
1312 BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
1313 PUSH_DATA (push, 1);
1314 BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
1315 PUSH_DATA (push, 0x40);
1316 BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
1317 PUSH_DATA (push, 0);
1318 BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
1319 PUSH_DATA (push, 0x30);
1320 BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
1321 PUSH_DATA (push, 3);
1322 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
1323 PUSH_DATA (push, 0x20);
1324 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
1325 PUSH_DATA (push, 0x00);
1326 screen->save_state.patch_vertices = 3;
1327
1328 BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
1329 PUSH_DATA (push, 0);
1330 BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
1331 PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
1332
1333 IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
1334
1335 if (nvc0_screen_init_compute(screen))
1336 goto fail;
1337
1338 /* XXX: Compute and 3D are somehow aliased on Fermi. */
1339 for (i = 0; i < 5; ++i) {
1340 unsigned j = 0;
1341 for (j = 0; j < 16; j++)
1342 screen->cb_bindings[i][j].size = -1;
1343
1344 /* TIC and TSC entries for each unit (nve4+ only) */
1345 /* auxiliary constants (6 user clip planes, base instance id) */
1346 nvc0_screen_bind_cb_3d(screen, NULL, i, 15, NVC0_CB_AUX_SIZE,
1347 screen->uniform_bo->offset + NVC0_CB_AUX_INFO(i));
1348 if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
1349 unsigned j;
1350 BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
1351 PUSH_DATA (push, NVC0_CB_AUX_UNK_INFO);
1352 for (j = 0; j < 8; ++j)
1353 PUSH_DATA(push, j);
1354 } else {
1355 BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
1356 PUSH_DATA (push, 0x54);
1357 }
1358
1359 /* MS sample coordinate offsets: these do not work with _ALT modes ! */
1360 BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 2 * 8);
1361 PUSH_DATA (push, NVC0_CB_AUX_MS_INFO);
1362 PUSH_DATA (push, 0); /* 0 */
1363 PUSH_DATA (push, 0);
1364 PUSH_DATA (push, 1); /* 1 */
1365 PUSH_DATA (push, 0);
1366 PUSH_DATA (push, 0); /* 2 */
1367 PUSH_DATA (push, 1);
1368 PUSH_DATA (push, 1); /* 3 */
1369 PUSH_DATA (push, 1);
1370 PUSH_DATA (push, 2); /* 4 */
1371 PUSH_DATA (push, 0);
1372 PUSH_DATA (push, 3); /* 5 */
1373 PUSH_DATA (push, 0);
1374 PUSH_DATA (push, 2); /* 6 */
1375 PUSH_DATA (push, 1);
1376 PUSH_DATA (push, 3); /* 7 */
1377 PUSH_DATA (push, 1);
1378 }
1379 BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
1380 PUSH_DATA (push, 0);
1381
1382 PUSH_KICK (push);
1383
1384 screen->tic.entries = CALLOC(
1385 NVC0_TIC_MAX_ENTRIES + NVC0_TSC_MAX_ENTRIES + NVE4_IMG_MAX_HANDLES,
1386 sizeof(void *));
1387 screen->tsc.entries = screen->tic.entries + NVC0_TIC_MAX_ENTRIES;
1388 screen->img.entries = (void *)(screen->tsc.entries + NVC0_TSC_MAX_ENTRIES);
1389
1390 if (!nvc0_blitter_create(screen))
1391 goto fail;
1392
1393 nouveau_fence_new(&screen->base, &screen->base.fence.current);
1394
1395 return &screen->base;
1396
1397 fail:
1398 screen->base.base.context_create = NULL;
1399 return &screen->base;
1400 }
1401
1402 int
1403 nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
1404 {
1405 int i = screen->tic.next;
1406
1407 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1408 i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1409
1410 screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1411
1412 if (screen->tic.entries[i])
1413 nv50_tic_entry(screen->tic.entries[i])->id = -1;
1414
1415 screen->tic.entries[i] = entry;
1416 return i;
1417 }
1418
1419 int
1420 nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
1421 {
1422 int i = screen->tsc.next;
1423
1424 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1425 i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1426
1427 screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1428
1429 if (screen->tsc.entries[i])
1430 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1431
1432 screen->tsc.entries[i] = entry;
1433 return i;
1434 }