nvc0: we have 16k-sized framebuffers, fix default scissors
[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 case PIPE_CAP_NIR_COMPACT_ARRAYS:
349 return 0;
350
351 case PIPE_CAP_VENDOR_ID:
352 return 0x10de;
353 case PIPE_CAP_DEVICE_ID: {
354 uint64_t device_id;
355 if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
356 NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
357 return -1;
358 }
359 return device_id;
360 }
361 case PIPE_CAP_ACCELERATED:
362 return 1;
363 case PIPE_CAP_VIDEO_MEMORY:
364 return dev->vram_size >> 20;
365 case PIPE_CAP_UMA:
366 return 0;
367 default:
368 debug_printf("%s: unhandled cap %d\n", __func__, param);
369 return u_pipe_screen_get_param_defaults(pscreen, param);
370 }
371 }
372
373 static int
374 nvc0_screen_get_shader_param(struct pipe_screen *pscreen,
375 enum pipe_shader_type shader,
376 enum pipe_shader_cap param)
377 {
378 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
379
380 switch (shader) {
381 case PIPE_SHADER_VERTEX:
382 case PIPE_SHADER_GEOMETRY:
383 case PIPE_SHADER_FRAGMENT:
384 case PIPE_SHADER_COMPUTE:
385 case PIPE_SHADER_TESS_CTRL:
386 case PIPE_SHADER_TESS_EVAL:
387 break;
388 default:
389 return 0;
390 }
391
392 switch (param) {
393 case PIPE_SHADER_CAP_PREFERRED_IR:
394 return PIPE_SHADER_IR_TGSI;
395 case PIPE_SHADER_CAP_SUPPORTED_IRS:
396 return 1 << PIPE_SHADER_IR_TGSI;
397 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
398 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
399 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
400 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
401 return 16384;
402 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
403 return 16;
404 case PIPE_SHADER_CAP_MAX_INPUTS:
405 return 0x200 / 16;
406 case PIPE_SHADER_CAP_MAX_OUTPUTS:
407 return 32;
408 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
409 return NVC0_MAX_CONSTBUF_SIZE;
410 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
411 return NVC0_MAX_PIPE_CONSTBUFS;
412 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
413 return shader != PIPE_SHADER_FRAGMENT;
414 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
415 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
416 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
417 return 1;
418 case PIPE_SHADER_CAP_MAX_TEMPS:
419 return NVC0_CAP_MAX_PROGRAM_TEMPS;
420 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
421 return 1;
422 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
423 return 1;
424 case PIPE_SHADER_CAP_SUBROUTINES:
425 return 1;
426 case PIPE_SHADER_CAP_INTEGERS:
427 return 1;
428 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
429 return 1;
430 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
431 return 1;
432 case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
433 return 1;
434 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
435 case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
436 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
437 case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
438 case PIPE_SHADER_CAP_INT64_ATOMICS:
439 case PIPE_SHADER_CAP_FP16:
440 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
441 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
442 return 0;
443 case PIPE_SHADER_CAP_SCALAR_ISA:
444 return 1;
445 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
446 return NVC0_MAX_BUFFERS;
447 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
448 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
449 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
450 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
451 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
452 return 32;
453 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
454 if (class_3d >= NVE4_3D_CLASS)
455 return NVC0_MAX_IMAGES;
456 if (shader == PIPE_SHADER_FRAGMENT || shader == PIPE_SHADER_COMPUTE)
457 return NVC0_MAX_IMAGES;
458 return 0;
459 default:
460 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
461 return 0;
462 }
463 }
464
465 static float
466 nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
467 {
468 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
469
470 switch (param) {
471 case PIPE_CAPF_MAX_LINE_WIDTH:
472 case PIPE_CAPF_MAX_LINE_WIDTH_AA:
473 return 10.0f;
474 case PIPE_CAPF_MAX_POINT_WIDTH:
475 return 63.0f;
476 case PIPE_CAPF_MAX_POINT_WIDTH_AA:
477 return 63.375f;
478 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
479 return 16.0f;
480 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
481 return 15.0f;
482 case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE:
483 return 0.0f;
484 case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE:
485 return class_3d >= GM200_3D_CLASS ? 0.75f : 0.0f;
486 case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY:
487 return class_3d >= GM200_3D_CLASS ? 0.25f : 0.0f;
488 }
489
490 NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
491 return 0.0f;
492 }
493
494 static int
495 nvc0_screen_get_compute_param(struct pipe_screen *pscreen,
496 enum pipe_shader_ir ir_type,
497 enum pipe_compute_cap param, void *data)
498 {
499 struct nvc0_screen *screen = nvc0_screen(pscreen);
500 const uint16_t obj_class = screen->compute->oclass;
501
502 #define RET(x) do { \
503 if (data) \
504 memcpy(data, x, sizeof(x)); \
505 return sizeof(x); \
506 } while (0)
507
508 switch (param) {
509 case PIPE_COMPUTE_CAP_GRID_DIMENSION:
510 RET((uint64_t []) { 3 });
511 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
512 if (obj_class >= NVE4_COMPUTE_CLASS) {
513 RET(((uint64_t []) { 0x7fffffff, 65535, 65535 }));
514 } else {
515 RET(((uint64_t []) { 65535, 65535, 65535 }));
516 }
517 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
518 RET(((uint64_t []) { 1024, 1024, 64 }));
519 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
520 RET((uint64_t []) { 1024 });
521 case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
522 if (obj_class >= NVE4_COMPUTE_CLASS) {
523 RET((uint64_t []) { 1024 });
524 } else {
525 RET((uint64_t []) { 512 });
526 }
527 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */
528 RET((uint64_t []) { 1ULL << 40 });
529 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
530 switch (obj_class) {
531 case GM200_COMPUTE_CLASS:
532 RET((uint64_t []) { 96 << 10 });
533 break;
534 case GM107_COMPUTE_CLASS:
535 RET((uint64_t []) { 64 << 10 });
536 break;
537 default:
538 RET((uint64_t []) { 48 << 10 });
539 break;
540 }
541 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
542 RET((uint64_t []) { 512 << 10 });
543 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
544 RET((uint64_t []) { 4096 });
545 case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
546 RET((uint32_t []) { 32 });
547 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
548 RET((uint64_t []) { 1ULL << 40 });
549 case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
550 RET((uint32_t []) { 0 });
551 case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
552 RET((uint32_t []) { screen->mp_count_compute });
553 case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
554 RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
555 case PIPE_COMPUTE_CAP_ADDRESS_BITS:
556 RET((uint32_t []) { 64 });
557 default:
558 return 0;
559 }
560
561 #undef RET
562 }
563
564 static void
565 nvc0_screen_get_sample_pixel_grid(struct pipe_screen *pscreen,
566 unsigned sample_count,
567 unsigned *width, unsigned *height)
568 {
569 switch (sample_count) {
570 case 0:
571 case 1:
572 /* this could be 4x4, but the GL state tracker makes it difficult to
573 * create a 1x MSAA texture and smaller grids save CB space */
574 *width = 2;
575 *height = 4;
576 break;
577 case 2:
578 *width = 2;
579 *height = 4;
580 break;
581 case 4:
582 *width = 2;
583 *height = 2;
584 break;
585 case 8:
586 *width = 1;
587 *height = 2;
588 break;
589 default:
590 assert(0);
591 }
592 }
593
594 static void
595 nvc0_screen_destroy(struct pipe_screen *pscreen)
596 {
597 struct nvc0_screen *screen = nvc0_screen(pscreen);
598
599 if (!nouveau_drm_screen_unref(&screen->base))
600 return;
601
602 if (screen->base.fence.current) {
603 struct nouveau_fence *current = NULL;
604
605 /* nouveau_fence_wait will create a new current fence, so wait on the
606 * _current_ one, and remove both.
607 */
608 nouveau_fence_ref(screen->base.fence.current, &current);
609 nouveau_fence_wait(current, NULL);
610 nouveau_fence_ref(NULL, &current);
611 nouveau_fence_ref(NULL, &screen->base.fence.current);
612 }
613 if (screen->base.pushbuf)
614 screen->base.pushbuf->user_priv = NULL;
615
616 if (screen->blitter)
617 nvc0_blitter_destroy(screen);
618 if (screen->pm.prog) {
619 screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
620 nvc0_program_destroy(NULL, screen->pm.prog);
621 FREE(screen->pm.prog);
622 }
623
624 nouveau_bo_ref(NULL, &screen->text);
625 nouveau_bo_ref(NULL, &screen->uniform_bo);
626 nouveau_bo_ref(NULL, &screen->tls);
627 nouveau_bo_ref(NULL, &screen->txc);
628 nouveau_bo_ref(NULL, &screen->fence.bo);
629 nouveau_bo_ref(NULL, &screen->poly_cache);
630
631 nouveau_heap_destroy(&screen->lib_code);
632 nouveau_heap_destroy(&screen->text_heap);
633
634 FREE(screen->tic.entries);
635
636 nouveau_object_del(&screen->eng3d);
637 nouveau_object_del(&screen->eng2d);
638 nouveau_object_del(&screen->m2mf);
639 nouveau_object_del(&screen->compute);
640 nouveau_object_del(&screen->nvsw);
641
642 nouveau_screen_fini(&screen->base);
643
644 FREE(screen);
645 }
646
647 static int
648 nvc0_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos,
649 unsigned size, const uint32_t *data)
650 {
651 struct nouveau_pushbuf *push = screen->base.pushbuf;
652
653 size /= 4;
654
655 assert((pos + size) <= 0x800);
656
657 BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2);
658 PUSH_DATA (push, (m - 0x3800) / 8);
659 PUSH_DATA (push, pos);
660 BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1);
661 PUSH_DATA (push, pos);
662 PUSH_DATAp(push, data, size);
663
664 return pos + size;
665 }
666
667 static void
668 nvc0_magic_3d_init(struct nouveau_pushbuf *push, uint16_t obj_class)
669 {
670 BEGIN_NVC0(push, SUBC_3D(0x10cc), 1);
671 PUSH_DATA (push, 0xff);
672 BEGIN_NVC0(push, SUBC_3D(0x10e0), 2);
673 PUSH_DATA (push, 0xff);
674 PUSH_DATA (push, 0xff);
675 BEGIN_NVC0(push, SUBC_3D(0x10ec), 2);
676 PUSH_DATA (push, 0xff);
677 PUSH_DATA (push, 0xff);
678 BEGIN_NVC0(push, SUBC_3D(0x074c), 1);
679 PUSH_DATA (push, 0x3f);
680
681 BEGIN_NVC0(push, SUBC_3D(0x16a8), 1);
682 PUSH_DATA (push, (3 << 16) | 3);
683 BEGIN_NVC0(push, SUBC_3D(0x1794), 1);
684 PUSH_DATA (push, (2 << 16) | 2);
685
686 if (obj_class < GM107_3D_CLASS) {
687 BEGIN_NVC0(push, SUBC_3D(0x12ac), 1);
688 PUSH_DATA (push, 0);
689 }
690 BEGIN_NVC0(push, SUBC_3D(0x0218), 1);
691 PUSH_DATA (push, 0x10);
692 BEGIN_NVC0(push, SUBC_3D(0x10fc), 1);
693 PUSH_DATA (push, 0x10);
694 BEGIN_NVC0(push, SUBC_3D(0x1290), 1);
695 PUSH_DATA (push, 0x10);
696 BEGIN_NVC0(push, SUBC_3D(0x12d8), 2);
697 PUSH_DATA (push, 0x10);
698 PUSH_DATA (push, 0x10);
699 BEGIN_NVC0(push, SUBC_3D(0x1140), 1);
700 PUSH_DATA (push, 0x10);
701 BEGIN_NVC0(push, SUBC_3D(0x1610), 1);
702 PUSH_DATA (push, 0xe);
703
704 BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_GEN_MODE), 1);
705 PUSH_DATA (push, NVC0_3D_VERTEX_ID_GEN_MODE_DRAW_ARRAYS_ADD_START);
706 BEGIN_NVC0(push, SUBC_3D(0x030c), 1);
707 PUSH_DATA (push, 0);
708 BEGIN_NVC0(push, SUBC_3D(0x0300), 1);
709 PUSH_DATA (push, 3);
710
711 BEGIN_NVC0(push, SUBC_3D(0x02d0), 1);
712 PUSH_DATA (push, 0x3fffff);
713 BEGIN_NVC0(push, SUBC_3D(0x0fdc), 1);
714 PUSH_DATA (push, 1);
715 BEGIN_NVC0(push, SUBC_3D(0x19c0), 1);
716 PUSH_DATA (push, 1);
717
718 if (obj_class < GM107_3D_CLASS) {
719 BEGIN_NVC0(push, SUBC_3D(0x075c), 1);
720 PUSH_DATA (push, 3);
721
722 if (obj_class >= NVE4_3D_CLASS) {
723 BEGIN_NVC0(push, SUBC_3D(0x07fc), 1);
724 PUSH_DATA (push, 1);
725 }
726 }
727
728 /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
729 * are supposed to do */
730 }
731
732 static void
733 nvc0_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
734 {
735 struct nvc0_screen *screen = nvc0_screen(pscreen);
736 struct nouveau_pushbuf *push = screen->base.pushbuf;
737
738 /* we need to do it after possible flush in MARK_RING */
739 *sequence = ++screen->base.fence.sequence;
740
741 assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
742 PUSH_DATA (push, NVC0_FIFO_PKHDR_SQ(NVC0_3D(QUERY_ADDRESS_HIGH), 4));
743 PUSH_DATAh(push, screen->fence.bo->offset);
744 PUSH_DATA (push, screen->fence.bo->offset);
745 PUSH_DATA (push, *sequence);
746 PUSH_DATA (push, NVC0_3D_QUERY_GET_FENCE | NVC0_3D_QUERY_GET_SHORT |
747 (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT));
748 }
749
750 static u32
751 nvc0_screen_fence_update(struct pipe_screen *pscreen)
752 {
753 struct nvc0_screen *screen = nvc0_screen(pscreen);
754 return screen->fence.map[0];
755 }
756
757 static int
758 nvc0_screen_init_compute(struct nvc0_screen *screen)
759 {
760 screen->base.base.get_compute_param = nvc0_screen_get_compute_param;
761
762 switch (screen->base.device->chipset & ~0xf) {
763 case 0xc0:
764 case 0xd0:
765 return nvc0_screen_compute_setup(screen, screen->base.pushbuf);
766 case 0xe0:
767 case 0xf0:
768 case 0x100:
769 case 0x110:
770 case 0x120:
771 case 0x130:
772 return nve4_screen_compute_setup(screen, screen->base.pushbuf);
773 default:
774 return -1;
775 }
776 }
777
778 static int
779 nvc0_screen_resize_tls_area(struct nvc0_screen *screen,
780 uint32_t lpos, uint32_t lneg, uint32_t cstack)
781 {
782 struct nouveau_bo *bo = NULL;
783 int ret;
784 uint64_t size = (lpos + lneg) * 32 + cstack;
785
786 if (size >= (1 << 20)) {
787 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64"\n", size);
788 return -1;
789 }
790
791 size *= (screen->base.device->chipset >= 0xe0) ? 64 : 48; /* max warps */
792 size = align(size, 0x8000);
793 size *= screen->mp_count;
794
795 size = align(size, 1 << 17);
796
797 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base), 1 << 17, size,
798 NULL, &bo);
799 if (ret)
800 return ret;
801
802 /* Make sure that the pushbuf has acquired a reference to the old tls
803 * segment, as it may have commands that will reference it.
804 */
805 if (screen->tls)
806 PUSH_REFN(screen->base.pushbuf, screen->tls,
807 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RDWR);
808 nouveau_bo_ref(NULL, &screen->tls);
809 screen->tls = bo;
810 return 0;
811 }
812
813 int
814 nvc0_screen_resize_text_area(struct nvc0_screen *screen, uint64_t size)
815 {
816 struct nouveau_pushbuf *push = screen->base.pushbuf;
817 struct nouveau_bo *bo;
818 int ret;
819
820 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base),
821 1 << 17, size, NULL, &bo);
822 if (ret)
823 return ret;
824
825 /* Make sure that the pushbuf has acquired a reference to the old text
826 * segment, as it may have commands that will reference it.
827 */
828 if (screen->text)
829 PUSH_REFN(push, screen->text,
830 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RD);
831 nouveau_bo_ref(NULL, &screen->text);
832 screen->text = bo;
833
834 nouveau_heap_destroy(&screen->lib_code);
835 nouveau_heap_destroy(&screen->text_heap);
836
837 /* XXX: getting a page fault at the end of the code buffer every few
838 * launches, don't use the last 256 bytes to work around them - prefetch ?
839 */
840 nouveau_heap_init(&screen->text_heap, 0, size - 0x100);
841
842 /* update the code segment setup */
843 BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
844 PUSH_DATAh(push, screen->text->offset);
845 PUSH_DATA (push, screen->text->offset);
846 if (screen->compute) {
847 BEGIN_NVC0(push, NVC0_CP(CODE_ADDRESS_HIGH), 2);
848 PUSH_DATAh(push, screen->text->offset);
849 PUSH_DATA (push, screen->text->offset);
850 }
851
852 return 0;
853 }
854
855 void
856 nvc0_screen_bind_cb_3d(struct nvc0_screen *screen, bool *can_serialize,
857 int stage, int index, int size, uint64_t addr)
858 {
859 assert(stage != 5);
860
861 struct nouveau_pushbuf *push = screen->base.pushbuf;
862
863 if (screen->base.class_3d >= GM107_3D_CLASS) {
864 struct nvc0_cb_binding *binding = &screen->cb_bindings[stage][index];
865
866 // TODO: Better figure out the conditions in which this is needed
867 bool serialize = binding->addr == addr && binding->size != size;
868 if (can_serialize)
869 serialize = serialize && *can_serialize;
870 if (serialize) {
871 IMMED_NVC0(push, NVC0_3D(SERIALIZE), 0);
872 if (can_serialize)
873 *can_serialize = false;
874 }
875
876 binding->addr = addr;
877 binding->size = size;
878 }
879
880 if (size >= 0) {
881 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
882 PUSH_DATA (push, size);
883 PUSH_DATAh(push, addr);
884 PUSH_DATA (push, addr);
885 }
886 IMMED_NVC0(push, NVC0_3D(CB_BIND(stage)), (index << 4) | (size >= 0));
887 }
888
889 #define FAIL_SCREEN_INIT(str, err) \
890 do { \
891 NOUVEAU_ERR(str, err); \
892 goto fail; \
893 } while(0)
894
895 struct nouveau_screen *
896 nvc0_screen_create(struct nouveau_device *dev)
897 {
898 struct nvc0_screen *screen;
899 struct pipe_screen *pscreen;
900 struct nouveau_object *chan;
901 struct nouveau_pushbuf *push;
902 uint64_t value;
903 uint32_t obj_class;
904 uint32_t flags;
905 int ret;
906 unsigned i;
907
908 switch (dev->chipset & ~0xf) {
909 case 0xc0:
910 case 0xd0:
911 case 0xe0:
912 case 0xf0:
913 case 0x100:
914 case 0x110:
915 case 0x120:
916 case 0x130:
917 break;
918 default:
919 return NULL;
920 }
921
922 screen = CALLOC_STRUCT(nvc0_screen);
923 if (!screen)
924 return NULL;
925 pscreen = &screen->base.base;
926 pscreen->destroy = nvc0_screen_destroy;
927
928 ret = nouveau_screen_init(&screen->base, dev);
929 if (ret)
930 FAIL_SCREEN_INIT("Base screen init failed: %d\n", ret);
931 chan = screen->base.channel;
932 push = screen->base.pushbuf;
933 push->user_priv = screen;
934 push->rsvd_kick = 5;
935
936 /* TODO: could this be higher on Kepler+? how does reclocking vs no
937 * reclocking affect performance?
938 * TODO: could this be higher on Fermi?
939 */
940 if (dev->chipset >= 0xe0)
941 screen->base.transfer_pushbuf_threshold = 1024;
942
943 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
944 PIPE_BIND_SHADER_BUFFER |
945 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
946 PIPE_BIND_COMMAND_ARGS_BUFFER | PIPE_BIND_QUERY_BUFFER;
947 screen->base.sysmem_bindings |=
948 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
949
950 if (screen->base.vram_domain & NOUVEAU_BO_GART) {
951 screen->base.sysmem_bindings |= screen->base.vidmem_bindings;
952 screen->base.vidmem_bindings = 0;
953 }
954
955 pscreen->context_create = nvc0_create;
956 pscreen->is_format_supported = nvc0_screen_is_format_supported;
957 pscreen->get_param = nvc0_screen_get_param;
958 pscreen->get_shader_param = nvc0_screen_get_shader_param;
959 pscreen->get_paramf = nvc0_screen_get_paramf;
960 pscreen->get_sample_pixel_grid = nvc0_screen_get_sample_pixel_grid;
961 pscreen->get_driver_query_info = nvc0_screen_get_driver_query_info;
962 pscreen->get_driver_query_group_info = nvc0_screen_get_driver_query_group_info;
963
964 nvc0_screen_init_resource_functions(pscreen);
965
966 screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
967 screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
968
969 flags = NOUVEAU_BO_GART | NOUVEAU_BO_MAP;
970 if (screen->base.drm->version >= 0x01000202)
971 flags |= NOUVEAU_BO_COHERENT;
972
973 ret = nouveau_bo_new(dev, flags, 0, 4096, NULL, &screen->fence.bo);
974 if (ret)
975 FAIL_SCREEN_INIT("Error allocating fence BO: %d\n", ret);
976 nouveau_bo_map(screen->fence.bo, 0, NULL);
977 screen->fence.map = screen->fence.bo->map;
978 screen->base.fence.emit = nvc0_screen_fence_emit;
979 screen->base.fence.update = nvc0_screen_fence_update;
980
981
982 ret = nouveau_object_new(chan, (dev->chipset < 0xe0) ? 0x1f906e : 0x906e,
983 NVIF_CLASS_SW_GF100, NULL, 0, &screen->nvsw);
984 if (ret)
985 FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret);
986
987 BEGIN_NVC0(push, SUBC_SW(NV01_SUBCHAN_OBJECT), 1);
988 PUSH_DATA (push, screen->nvsw->handle);
989
990 switch (dev->chipset & ~0xf) {
991 case 0x130:
992 case 0x120:
993 case 0x110:
994 case 0x100:
995 case 0xf0:
996 obj_class = NVF0_P2MF_CLASS;
997 break;
998 case 0xe0:
999 obj_class = NVE4_P2MF_CLASS;
1000 break;
1001 default:
1002 obj_class = NVC0_M2MF_CLASS;
1003 break;
1004 }
1005 ret = nouveau_object_new(chan, 0xbeef323f, obj_class, NULL, 0,
1006 &screen->m2mf);
1007 if (ret)
1008 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
1009
1010 BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
1011 PUSH_DATA (push, screen->m2mf->oclass);
1012 if (screen->m2mf->oclass == NVE4_P2MF_CLASS) {
1013 BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
1014 PUSH_DATA (push, 0xa0b5);
1015 }
1016
1017 ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
1018 &screen->eng2d);
1019 if (ret)
1020 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
1021
1022 BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
1023 PUSH_DATA (push, screen->eng2d->oclass);
1024 BEGIN_NVC0(push, SUBC_2D(NVC0_2D_SINGLE_GPC), 1);
1025 PUSH_DATA (push, 0);
1026 BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
1027 PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
1028 BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
1029 PUSH_DATA (push, 0);
1030 BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
1031 PUSH_DATA (push, 0);
1032 BEGIN_NVC0(push, SUBC_2D(0x0884), 1);
1033 PUSH_DATA (push, 0x3f);
1034 BEGIN_NVC0(push, SUBC_2D(0x0888), 1);
1035 PUSH_DATA (push, 1);
1036 BEGIN_NVC0(push, NVC0_2D(COND_MODE), 1);
1037 PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
1038
1039 BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
1040 PUSH_DATAh(push, screen->fence.bo->offset + 16);
1041 PUSH_DATA (push, screen->fence.bo->offset + 16);
1042
1043 switch (dev->chipset & ~0xf) {
1044 case 0x130:
1045 switch (dev->chipset) {
1046 case 0x130:
1047 case 0x13b:
1048 obj_class = GP100_3D_CLASS;
1049 break;
1050 default:
1051 obj_class = GP102_3D_CLASS;
1052 break;
1053 }
1054 break;
1055 case 0x120:
1056 obj_class = GM200_3D_CLASS;
1057 break;
1058 case 0x110:
1059 obj_class = GM107_3D_CLASS;
1060 break;
1061 case 0x100:
1062 case 0xf0:
1063 obj_class = NVF0_3D_CLASS;
1064 break;
1065 case 0xe0:
1066 switch (dev->chipset) {
1067 case 0xea:
1068 obj_class = NVEA_3D_CLASS;
1069 break;
1070 default:
1071 obj_class = NVE4_3D_CLASS;
1072 break;
1073 }
1074 break;
1075 case 0xd0:
1076 obj_class = NVC8_3D_CLASS;
1077 break;
1078 case 0xc0:
1079 default:
1080 switch (dev->chipset) {
1081 case 0xc8:
1082 obj_class = NVC8_3D_CLASS;
1083 break;
1084 case 0xc1:
1085 obj_class = NVC1_3D_CLASS;
1086 break;
1087 default:
1088 obj_class = NVC0_3D_CLASS;
1089 break;
1090 }
1091 break;
1092 }
1093 ret = nouveau_object_new(chan, 0xbeef003d, obj_class, NULL, 0,
1094 &screen->eng3d);
1095 if (ret)
1096 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
1097 screen->base.class_3d = obj_class;
1098
1099 BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
1100 PUSH_DATA (push, screen->eng3d->oclass);
1101
1102 BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
1103 PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
1104
1105 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
1106 /* kill shaders after about 1 second (at 100 MHz) */
1107 BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
1108 PUSH_DATA (push, 0x17);
1109 }
1110
1111 IMMED_NVC0(push, NVC0_3D(ZETA_COMP_ENABLE),
1112 screen->base.drm->version >= 0x01000101);
1113 BEGIN_NVC0(push, NVC0_3D(RT_COMP_ENABLE(0)), 8);
1114 for (i = 0; i < 8; ++i)
1115 PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
1116
1117 BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
1118 PUSH_DATA (push, 1);
1119
1120 BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
1121 PUSH_DATA (push, 0);
1122 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
1123 PUSH_DATA (push, 0);
1124 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
1125 PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
1126 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
1127 PUSH_DATA (push, 0);
1128 BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
1129 PUSH_DATA (push, 1);
1130 BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
1131 PUSH_DATA (push, 1);
1132 BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
1133 PUSH_DATA (push, 1);
1134 BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
1135 PUSH_DATA (push, 0);
1136 BEGIN_NVC0(push, NVC0_3D(SHADE_MODEL), 1);
1137 PUSH_DATA (push, NVC0_3D_SHADE_MODEL_SMOOTH);
1138 if (screen->eng3d->oclass < NVE4_3D_CLASS) {
1139 IMMED_NVC0(push, NVC0_3D(TEX_MISC), 0);
1140 } else {
1141 BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
1142 PUSH_DATA (push, 15);
1143 }
1144 BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
1145 PUSH_DATA (push, 8); /* 128 */
1146 BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
1147 PUSH_DATA (push, 1);
1148 if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
1149 BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
1150 PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
1151 }
1152
1153 nvc0_magic_3d_init(push, screen->eng3d->oclass);
1154
1155 ret = nvc0_screen_resize_text_area(screen, 1 << 19);
1156 if (ret)
1157 FAIL_SCREEN_INIT("Error allocating TEXT area: %d\n", ret);
1158
1159 /* 6 user uniform areas, 6 driver areas, and 1 for the runout */
1160 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 12, 13 << 16, NULL,
1161 &screen->uniform_bo);
1162 if (ret)
1163 FAIL_SCREEN_INIT("Error allocating uniform BO: %d\n", ret);
1164
1165 PUSH_REFN (push, screen->uniform_bo, NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_WR);
1166
1167 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
1168 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1169 PUSH_DATA (push, 256);
1170 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1171 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1172 BEGIN_1IC0(push, NVC0_3D(CB_POS), 5);
1173 PUSH_DATA (push, 0);
1174 PUSH_DATAf(push, 0.0f);
1175 PUSH_DATAf(push, 0.0f);
1176 PUSH_DATAf(push, 0.0f);
1177 PUSH_DATAf(push, 0.0f);
1178 BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
1179 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1180 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1181
1182 if (screen->base.drm->version >= 0x01000101) {
1183 ret = nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
1184 if (ret)
1185 FAIL_SCREEN_INIT("NOUVEAU_GETPARAM_GRAPH_UNITS failed: %d\n", ret);
1186 } else {
1187 if (dev->chipset >= 0xe0 && dev->chipset < 0xf0)
1188 value = (8 << 8) | 4;
1189 else
1190 value = (16 << 8) | 4;
1191 }
1192 screen->gpc_count = value & 0x000000ff;
1193 screen->mp_count = value >> 8;
1194 screen->mp_count_compute = screen->mp_count;
1195
1196 ret = nvc0_screen_resize_tls_area(screen, 128 * 16, 0, 0x200);
1197 if (ret)
1198 FAIL_SCREEN_INIT("Error allocating TLS area: %d\n", ret);
1199
1200 BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
1201 PUSH_DATAh(push, screen->tls->offset);
1202 PUSH_DATA (push, screen->tls->offset);
1203 PUSH_DATA (push, screen->tls->size >> 32);
1204 PUSH_DATA (push, screen->tls->size);
1205 BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
1206 PUSH_DATA (push, 0);
1207 /* Reduce likelihood of collision with real buffers by placing the hole at
1208 * the top of the 4G area. This will have to be dealt with for real
1209 * eventually by blocking off that area from the VM.
1210 */
1211 BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
1212 PUSH_DATA (push, 0xff << 24);
1213
1214 if (screen->eng3d->oclass < GM107_3D_CLASS) {
1215 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 20, NULL,
1216 &screen->poly_cache);
1217 if (ret)
1218 FAIL_SCREEN_INIT("Error allocating poly cache BO: %d\n", ret);
1219
1220 BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
1221 PUSH_DATAh(push, screen->poly_cache->offset);
1222 PUSH_DATA (push, screen->poly_cache->offset);
1223 PUSH_DATA (push, 3);
1224 }
1225
1226 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 17, NULL,
1227 &screen->txc);
1228 if (ret)
1229 FAIL_SCREEN_INIT("Error allocating txc BO: %d\n", ret);
1230
1231 BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
1232 PUSH_DATAh(push, screen->txc->offset);
1233 PUSH_DATA (push, screen->txc->offset);
1234 PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
1235 if (screen->eng3d->oclass >= GM107_3D_CLASS) {
1236 screen->tic.maxwell = true;
1237 if (screen->eng3d->oclass == GM107_3D_CLASS) {
1238 screen->tic.maxwell =
1239 debug_get_bool_option("NOUVEAU_MAXWELL_TIC", true);
1240 IMMED_NVC0(push, SUBC_3D(0x0f10), screen->tic.maxwell);
1241 }
1242 }
1243
1244 BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
1245 PUSH_DATAh(push, screen->txc->offset + 65536);
1246 PUSH_DATA (push, screen->txc->offset + 65536);
1247 PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
1248
1249 BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
1250 PUSH_DATA (push, 0);
1251 BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
1252 PUSH_DATA (push, 0);
1253 PUSH_DATA (push, 0);
1254 BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
1255 PUSH_DATA (push, 0x3f);
1256
1257 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
1258 PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
1259 BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
1260 for (i = 0; i < 8 * 2; ++i)
1261 PUSH_DATA(push, 0);
1262 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
1263 PUSH_DATA (push, 0);
1264 BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
1265 PUSH_DATA (push, 0);
1266
1267 /* neither scissors, viewport nor stencil mask should affect clears */
1268 BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
1269 PUSH_DATA (push, 0);
1270
1271 BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
1272 PUSH_DATA (push, 1);
1273 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1274 BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(i)), 2);
1275 PUSH_DATAf(push, 0.0f);
1276 PUSH_DATAf(push, 1.0f);
1277 }
1278 BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
1279 PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
1280
1281 /* We use scissors instead of exact view volume clipping,
1282 * so they're always enabled.
1283 */
1284 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1285 BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(i)), 3);
1286 PUSH_DATA (push, 1);
1287 PUSH_DATA (push, 16384 << 16);
1288 PUSH_DATA (push, 16384 << 16);
1289 }
1290
1291 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
1292
1293 i = 0;
1294 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, mme9097_per_instance_bf);
1295 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, mme9097_blend_enables);
1296 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, mme9097_vertex_array_select);
1297 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, mme9097_tep_select);
1298 MK_MACRO(NVC0_3D_MACRO_GP_SELECT, mme9097_gp_select);
1299 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, mme9097_poly_mode_front);
1300 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, mme9097_poly_mode_back);
1301 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT, mme9097_draw_arrays_indirect);
1302 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT, mme9097_draw_elts_indirect);
1303 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT, mme9097_draw_arrays_indirect_count);
1304 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT, mme9097_draw_elts_indirect_count);
1305 MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE, mme9097_query_buffer_write);
1306 MK_MACRO(NVC0_3D_MACRO_CONSERVATIVE_RASTER_STATE, mme9097_conservative_raster_state);
1307 MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER, mme9097_compute_counter);
1308 MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER_TO_QUERY, mme9097_compute_counter_to_query);
1309 MK_MACRO(NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT, mme90c0_launch_grid_indirect);
1310
1311 BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
1312 PUSH_DATA (push, 1);
1313 BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
1314 PUSH_DATA (push, 1);
1315 BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
1316 PUSH_DATA (push, 0x40);
1317 BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
1318 PUSH_DATA (push, 0);
1319 BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
1320 PUSH_DATA (push, 0x30);
1321 BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
1322 PUSH_DATA (push, 3);
1323 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
1324 PUSH_DATA (push, 0x20);
1325 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
1326 PUSH_DATA (push, 0x00);
1327 screen->save_state.patch_vertices = 3;
1328
1329 BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
1330 PUSH_DATA (push, 0);
1331 BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
1332 PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
1333
1334 IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
1335
1336 if (nvc0_screen_init_compute(screen))
1337 goto fail;
1338
1339 /* XXX: Compute and 3D are somehow aliased on Fermi. */
1340 for (i = 0; i < 5; ++i) {
1341 unsigned j = 0;
1342 for (j = 0; j < 16; j++)
1343 screen->cb_bindings[i][j].size = -1;
1344
1345 /* TIC and TSC entries for each unit (nve4+ only) */
1346 /* auxiliary constants (6 user clip planes, base instance id) */
1347 nvc0_screen_bind_cb_3d(screen, NULL, i, 15, NVC0_CB_AUX_SIZE,
1348 screen->uniform_bo->offset + NVC0_CB_AUX_INFO(i));
1349 if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
1350 unsigned j;
1351 BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
1352 PUSH_DATA (push, NVC0_CB_AUX_UNK_INFO);
1353 for (j = 0; j < 8; ++j)
1354 PUSH_DATA(push, j);
1355 } else {
1356 BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
1357 PUSH_DATA (push, 0x54);
1358 }
1359
1360 /* MS sample coordinate offsets: these do not work with _ALT modes ! */
1361 BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 2 * 8);
1362 PUSH_DATA (push, NVC0_CB_AUX_MS_INFO);
1363 PUSH_DATA (push, 0); /* 0 */
1364 PUSH_DATA (push, 0);
1365 PUSH_DATA (push, 1); /* 1 */
1366 PUSH_DATA (push, 0);
1367 PUSH_DATA (push, 0); /* 2 */
1368 PUSH_DATA (push, 1);
1369 PUSH_DATA (push, 1); /* 3 */
1370 PUSH_DATA (push, 1);
1371 PUSH_DATA (push, 2); /* 4 */
1372 PUSH_DATA (push, 0);
1373 PUSH_DATA (push, 3); /* 5 */
1374 PUSH_DATA (push, 0);
1375 PUSH_DATA (push, 2); /* 6 */
1376 PUSH_DATA (push, 1);
1377 PUSH_DATA (push, 3); /* 7 */
1378 PUSH_DATA (push, 1);
1379 }
1380 BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
1381 PUSH_DATA (push, 0);
1382
1383 PUSH_KICK (push);
1384
1385 screen->tic.entries = CALLOC(
1386 NVC0_TIC_MAX_ENTRIES + NVC0_TSC_MAX_ENTRIES + NVE4_IMG_MAX_HANDLES,
1387 sizeof(void *));
1388 screen->tsc.entries = screen->tic.entries + NVC0_TIC_MAX_ENTRIES;
1389 screen->img.entries = (void *)(screen->tsc.entries + NVC0_TSC_MAX_ENTRIES);
1390
1391 if (!nvc0_blitter_create(screen))
1392 goto fail;
1393
1394 nouveau_fence_new(&screen->base, &screen->base.fence.current);
1395
1396 return &screen->base;
1397
1398 fail:
1399 screen->base.base.context_create = NULL;
1400 return &screen->base;
1401 }
1402
1403 int
1404 nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
1405 {
1406 int i = screen->tic.next;
1407
1408 while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1409 i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1410
1411 screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1412
1413 if (screen->tic.entries[i])
1414 nv50_tic_entry(screen->tic.entries[i])->id = -1;
1415
1416 screen->tic.entries[i] = entry;
1417 return i;
1418 }
1419
1420 int
1421 nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
1422 {
1423 int i = screen->tsc.next;
1424
1425 while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1426 i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1427
1428 screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1429
1430 if (screen->tsc.entries[i])
1431 nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1432
1433 screen->tsc.entries[i] = entry;
1434 return i;
1435 }