• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/format/u_format.h"
27 #include "util/format/u_format_s3tc.h"
28 #include "util/u_screen.h"
29 #include "pipe/p_screen.h"
30 
31 #include "nouveau_vp3_video.h"
32 
33 #include "nv50_ir_driver.h"
34 
35 #include "nvc0/nvc0_context.h"
36 #include "nvc0/nvc0_screen.h"
37 
38 #include "nvc0/mme/com9097.mme.h"
39 #include "nvc0/mme/com90c0.mme.h"
40 #include "nvc0/mme/comc597.mme.h"
41 
42 #include "nv50/g80_texture.xml.h"
43 
44 static bool
nvc0_screen_is_format_supported(struct pipe_screen * pscreen,enum pipe_format format,enum pipe_texture_target target,unsigned sample_count,unsigned storage_sample_count,unsigned bindings)45 nvc0_screen_is_format_supported(struct pipe_screen *pscreen,
46                                 enum pipe_format format,
47                                 enum pipe_texture_target target,
48                                 unsigned sample_count,
49                                 unsigned storage_sample_count,
50                                 unsigned bindings)
51 {
52    const struct util_format_description *desc = util_format_description(format);
53 
54    if (sample_count > 8)
55       return false;
56    if (!(0x117 & (1 << sample_count))) /* 0, 1, 2, 4 or 8 */
57       return false;
58 
59    if (MAX2(1, sample_count) != MAX2(1, storage_sample_count))
60       return false;
61 
62    /* Short-circuit the rest of the logic -- this is used by the gallium frontend
63     * to determine valid MS levels in a no-attachments scenario.
64     */
65    if (format == PIPE_FORMAT_NONE && bindings & PIPE_BIND_RENDER_TARGET)
66       return true;
67 
68    if ((bindings & PIPE_BIND_SAMPLER_VIEW) && (target != PIPE_BUFFER))
69       if (util_format_get_blocksizebits(format) == 3 * 32)
70          return false;
71 
72    if (bindings & PIPE_BIND_LINEAR)
73       if (util_format_is_depth_or_stencil(format) ||
74           (target != PIPE_TEXTURE_1D &&
75            target != PIPE_TEXTURE_2D &&
76            target != PIPE_TEXTURE_RECT) ||
77           sample_count > 1)
78          return false;
79 
80    /* Restrict ETC2 and ASTC formats here. These are only supported on GK20A
81     * and GM20B.
82     */
83    if ((desc->layout == UTIL_FORMAT_LAYOUT_ETC ||
84         desc->layout == UTIL_FORMAT_LAYOUT_ASTC) &&
85        nouveau_screen(pscreen)->device->chipset != 0x12b &&
86        nouveau_screen(pscreen)->class_3d != NVEA_3D_CLASS)
87       return false;
88 
89    /* shared is always supported */
90    bindings &= ~(PIPE_BIND_LINEAR |
91                  PIPE_BIND_SHARED);
92 
93    if (bindings & PIPE_BIND_SHADER_IMAGE) {
94       if (format == PIPE_FORMAT_B8G8R8A8_UNORM &&
95           nouveau_screen(pscreen)->class_3d < NVE4_3D_CLASS) {
96          /* This should work on Fermi, but for currently unknown reasons it
97           * does not and results in breaking reads from pbos. */
98          return false;
99       }
100    }
101 
102    if (bindings & PIPE_BIND_INDEX_BUFFER) {
103       if (format != PIPE_FORMAT_R8_UINT &&
104           format != PIPE_FORMAT_R16_UINT &&
105           format != PIPE_FORMAT_R32_UINT)
106          return false;
107       bindings &= ~PIPE_BIND_INDEX_BUFFER;
108    }
109 
110    return (( nvc0_format_table[format].usage |
111             nvc0_vertex_format[format].usage) & bindings) == bindings;
112 }
113 
114 static int
nvc0_screen_get_param(struct pipe_screen * pscreen,enum pipe_cap param)115 nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
116 {
117    const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
118    const struct nouveau_screen *screen = nouveau_screen(pscreen);
119    struct nouveau_device *dev = screen->device;
120 
121    switch (param) {
122    /* non-boolean caps */
123    case PIPE_CAP_MAX_TEXTURE_2D_SIZE:
124       return 16384;
125    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
126       return 15;
127    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
128       return 12;
129    case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
130       return 2048;
131    case PIPE_CAP_MIN_TEXEL_OFFSET:
132       return -8;
133    case PIPE_CAP_MAX_TEXEL_OFFSET:
134       return 7;
135    case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
136       return -32;
137    case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
138       return 31;
139    case PIPE_CAP_MAX_TEXEL_BUFFER_ELEMENTS_UINT:
140       return 128 * 1024 * 1024;
141    case PIPE_CAP_GLSL_FEATURE_LEVEL:
142       return 430;
143    case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY:
144       return 430;
145    case PIPE_CAP_MAX_RENDER_TARGETS:
146       return 8;
147    case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
148       return 1;
149    case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
150    case PIPE_CAP_RASTERIZER_SUBPIXEL_BITS:
151       return 8;
152    case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
153       return 4;
154    case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
155    case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
156       return 128;
157    case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
158    case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
159       return 1024;
160    case PIPE_CAP_MAX_VERTEX_STREAMS:
161       return 4;
162    case PIPE_CAP_MAX_GS_INVOCATIONS:
163       return 32;
164    case PIPE_CAP_MAX_SHADER_BUFFER_SIZE_UINT:
165       return 1 << 27;
166    case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
167       return 2048;
168    case PIPE_CAP_MAX_VERTEX_ELEMENT_SRC_OFFSET:
169       return 2047;
170    case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
171       return 256;
172    case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
173       if (class_3d < GM107_3D_CLASS)
174          return 256; /* IMAGE bindings require alignment to 256 */
175       return 16;
176    case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
177       return 16;
178    case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
179       return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
180    case PIPE_CAP_MAX_VIEWPORTS:
181       return NVC0_MAX_VIEWPORTS;
182    case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
183       return 4;
184    case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
185       return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
186    case PIPE_CAP_ENDIANNESS:
187       return PIPE_ENDIAN_LITTLE;
188    case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
189       return 30;
190    case PIPE_CAP_MAX_WINDOW_RECTANGLES:
191       return NVC0_MAX_WINDOW_RECTANGLES;
192    case PIPE_CAP_MAX_CONSERVATIVE_RASTER_SUBPIXEL_PRECISION_BIAS:
193       return class_3d >= GM200_3D_CLASS ? 8 : 0;
194    case PIPE_CAP_MAX_TEXTURE_UPLOAD_MEMORY_BUDGET:
195       return 64 * 1024 * 1024;
196    case PIPE_CAP_MAX_VARYINGS:
197       /* NOTE: These only count our slots for GENERIC varyings.
198        * The address space may be larger, but the actual hard limit seems to be
199        * less than what the address space layout permits, so don't add TEXCOORD,
200        * COLOR, etc. here.
201        */
202       return 0x1f0 / 16;
203    case PIPE_CAP_MAX_VERTEX_BUFFERS:
204       return 16;
205    case PIPE_CAP_GL_BEGIN_END_BUFFER_SIZE:
206       return 512 * 1024; /* TODO: Investigate tuning this */
207    case PIPE_CAP_MAX_TEXTURE_MB:
208       return 0; /* TODO: use 1/2 of VRAM for this? */
209 
210    case PIPE_CAP_TIMER_RESOLUTION:
211       return 1000;
212 
213    case PIPE_CAP_SUPPORTED_PRIM_MODES_WITH_RESTART:
214    case PIPE_CAP_SUPPORTED_PRIM_MODES:
215       return BITFIELD_MASK(MESA_PRIM_COUNT);
216 
217    /* supported caps */
218    case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
219    case PIPE_CAP_TEXTURE_MIRROR_CLAMP_TO_EDGE:
220    case PIPE_CAP_TEXTURE_SWIZZLE:
221    case PIPE_CAP_NPOT_TEXTURES:
222    case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
223    case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
224    case PIPE_CAP_ANISOTROPIC_FILTER:
225    case PIPE_CAP_SEAMLESS_CUBE_MAP:
226    case PIPE_CAP_CUBE_MAP_ARRAY:
227    case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
228    case PIPE_CAP_TEXTURE_MULTISAMPLE:
229    case PIPE_CAP_DEPTH_CLIP_DISABLE:
230    case PIPE_CAP_TGSI_TEXCOORD:
231    case PIPE_CAP_FRAGMENT_SHADER_TEXTURE_LOD:
232    case PIPE_CAP_FRAGMENT_SHADER_DERIVATIVES:
233    case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
234    case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
235    case PIPE_CAP_VERTEX_COLOR_CLAMPED:
236    case PIPE_CAP_QUERY_TIMESTAMP:
237    case PIPE_CAP_QUERY_TIME_ELAPSED:
238    case PIPE_CAP_OCCLUSION_QUERY:
239    case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
240    case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
241    case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
242    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
243    case PIPE_CAP_INDEP_BLEND_ENABLE:
244    case PIPE_CAP_INDEP_BLEND_FUNC:
245    case PIPE_CAP_FS_COORD_ORIGIN_UPPER_LEFT:
246    case PIPE_CAP_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
247    case PIPE_CAP_PRIMITIVE_RESTART:
248    case PIPE_CAP_PRIMITIVE_RESTART_FIXED_INDEX:
249    case PIPE_CAP_VS_INSTANCEID:
250    case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
251    case PIPE_CAP_CONDITIONAL_RENDER:
252    case PIPE_CAP_TEXTURE_BARRIER:
253    case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
254    case PIPE_CAP_START_INSTANCE:
255    case PIPE_CAP_DRAW_INDIRECT:
256    case PIPE_CAP_USER_VERTEX_BUFFERS:
257    case PIPE_CAP_TEXTURE_QUERY_LOD:
258    case PIPE_CAP_SAMPLE_SHADING:
259    case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
260    case PIPE_CAP_TEXTURE_GATHER_SM5:
261    case PIPE_CAP_FS_FINE_DERIVATIVE:
262    case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
263    case PIPE_CAP_SAMPLER_VIEW_TARGET:
264    case PIPE_CAP_CLIP_HALFZ:
265    case PIPE_CAP_POLYGON_OFFSET_CLAMP:
266    case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
267    case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
268    case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
269    case PIPE_CAP_DEPTH_BOUNDS_TEST:
270    case PIPE_CAP_TEXTURE_QUERY_SAMPLES:
271    case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
272    case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
273    case PIPE_CAP_DRAW_PARAMETERS:
274    case PIPE_CAP_SHADER_PACK_HALF_FLOAT:
275    case PIPE_CAP_MULTI_DRAW_INDIRECT:
276    case PIPE_CAP_MEMOBJ:
277    case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
278    case PIPE_CAP_FS_FACE_IS_INTEGER_SYSVAL:
279    case PIPE_CAP_QUERY_BUFFER_OBJECT:
280    case PIPE_CAP_INVALIDATE_BUFFER:
281    case PIPE_CAP_STRING_MARKER:
282    case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
283    case PIPE_CAP_CULL_DISTANCE:
284    case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
285    case PIPE_CAP_SHADER_GROUP_VOTE:
286    case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
287    case PIPE_CAP_SHADER_ARRAY_COMPONENTS:
288    case PIPE_CAP_LEGACY_MATH_RULES:
289    case PIPE_CAP_DOUBLES:
290    case PIPE_CAP_INT64:
291    case PIPE_CAP_TGSI_TEX_TXF_LZ:
292    case PIPE_CAP_SHADER_CLOCK:
293    case PIPE_CAP_COMPUTE:
294    case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
295    case PIPE_CAP_QUERY_SO_OVERFLOW:
296    case PIPE_CAP_TGSI_DIV:
297    case PIPE_CAP_IMAGE_ATOMIC_INC_WRAP:
298    case PIPE_CAP_DEMOTE_TO_HELPER_INVOCATION:
299    case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
300    case PIPE_CAP_TEXTURE_SHADOW_LOD:
301    case PIPE_CAP_CLEAR_SCISSORED:
302    case PIPE_CAP_IMAGE_STORE_FORMATTED:
303    case PIPE_CAP_QUERY_MEMORY_INFO:
304       return 1;
305    case PIPE_CAP_TEXTURE_TRANSFER_MODES:
306       return nouveau_screen(pscreen)->vram_domain & NOUVEAU_BO_VRAM ? PIPE_TEXTURE_TRANSFER_BLIT : 0;
307    case PIPE_CAP_FBFETCH:
308       return class_3d >= NVE4_3D_CLASS ? 1 : 0; /* needs testing on fermi */
309    case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
310    case PIPE_CAP_SHADER_BALLOT:
311       return class_3d >= NVE4_3D_CLASS;
312    case PIPE_CAP_BINDLESS_TEXTURE:
313       return class_3d >= NVE4_3D_CLASS;
314    case PIPE_CAP_IMAGE_ATOMIC_FLOAT_ADD:
315       return class_3d < GM107_3D_CLASS; /* needs additional lowering */
316    case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
317    case PIPE_CAP_VS_LAYER_VIEWPORT:
318    case PIPE_CAP_TES_LAYER_VIEWPORT:
319    case PIPE_CAP_POST_DEPTH_COVERAGE:
320    case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_TRIANGLES:
321    case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_POINTS_LINES:
322    case PIPE_CAP_CONSERVATIVE_RASTER_POST_DEPTH_COVERAGE:
323    case PIPE_CAP_PROGRAMMABLE_SAMPLE_LOCATIONS:
324    case PIPE_CAP_VIEWPORT_SWIZZLE:
325    case PIPE_CAP_VIEWPORT_MASK:
326    case PIPE_CAP_SAMPLER_REDUCTION_MINMAX:
327       return class_3d >= GM200_3D_CLASS;
328    case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES:
329       return class_3d >= GP100_3D_CLASS;
330    case PIPE_CAP_RESOURCE_FROM_USER_MEMORY_COMPUTE_ONLY:
331    case PIPE_CAP_SYSTEM_SVM:
332       return screen->has_svm ? 1 : 0;
333 
334    case PIPE_CAP_GL_SPIRV:
335    case PIPE_CAP_GL_SPIRV_VARIABLE_POINTERS:
336       return 1;
337 
338    /* nir related caps */
339    case PIPE_CAP_NIR_IMAGES_AS_DEREF:
340       return 0;
341 
342    case PIPE_CAP_PCI_GROUP:
343    case PIPE_CAP_PCI_BUS:
344    case PIPE_CAP_PCI_DEVICE:
345    case PIPE_CAP_PCI_FUNCTION:
346       return 0;
347 
348    case PIPE_CAP_OPENCL_INTEGER_FUNCTIONS: /* could be done */
349    case PIPE_CAP_INTEGER_MULTIPLY_32X16: /* could be done */
350    case PIPE_CAP_MAP_UNSYNCHRONIZED_THREAD_SAFE: /* when we fix MT stuff */
351    case PIPE_CAP_ALPHA_TO_COVERAGE_DITHER_CONTROL: /* TODO */
352    case PIPE_CAP_SHADER_ATOMIC_INT64: /* TODO */
353    case PIPE_CAP_HARDWARE_GL_SELECT:
354       return 0;
355 
356    case PIPE_CAP_VENDOR_ID:
357       return 0x10de;
358    case PIPE_CAP_DEVICE_ID: {
359       uint64_t device_id;
360       if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
361          NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
362          return -1;
363       }
364       return device_id;
365    }
366    case PIPE_CAP_ACCELERATED:
367       return 1;
368    case PIPE_CAP_VIDEO_MEMORY:
369       return dev->vram_size >> 20;
370    case PIPE_CAP_UMA:
371       return nouveau_screen(pscreen)->is_uma;
372 
373    default:
374       return u_pipe_screen_get_param_defaults(pscreen, param);
375    }
376 }
377 
378 static int
nvc0_screen_get_shader_param(struct pipe_screen * pscreen,enum pipe_shader_type shader,enum pipe_shader_cap param)379 nvc0_screen_get_shader_param(struct pipe_screen *pscreen,
380                              enum pipe_shader_type shader,
381                              enum pipe_shader_cap param)
382 {
383    const struct nouveau_screen *screen = nouveau_screen(pscreen);
384    const uint16_t class_3d = screen->class_3d;
385 
386    switch (shader) {
387    case PIPE_SHADER_VERTEX:
388    case PIPE_SHADER_GEOMETRY:
389    case PIPE_SHADER_FRAGMENT:
390    case PIPE_SHADER_COMPUTE:
391    case PIPE_SHADER_TESS_CTRL:
392    case PIPE_SHADER_TESS_EVAL:
393       break;
394    default:
395       return 0;
396    }
397 
398    switch (param) {
399    case PIPE_SHADER_CAP_SUPPORTED_IRS: {
400       uint32_t irs = 1 << PIPE_SHADER_IR_NIR;
401       if (screen->force_enable_cl)
402          irs |= 1 << PIPE_SHADER_IR_NIR_SERIALIZED;
403       return irs;
404    }
405    case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
406    case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
407    case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
408    case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
409       return 16384;
410    case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
411       return 16;
412    case PIPE_SHADER_CAP_MAX_INPUTS:
413       return 0x200 / 16;
414    case PIPE_SHADER_CAP_MAX_OUTPUTS:
415       return 32;
416    case PIPE_SHADER_CAP_MAX_CONST_BUFFER0_SIZE:
417       return NVC0_MAX_CONSTBUF_SIZE;
418    case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
419       return NVC0_MAX_PIPE_CONSTBUFS;
420    case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
421       return shader != PIPE_SHADER_FRAGMENT;
422    case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
423       /* HW doesn't support indirect addressing of fragment program inputs
424        * on Volta.  The binary driver generates a function to handle every
425        * possible indirection, and indirectly calls the function to handle
426        * this instead.
427        */
428       if (class_3d >= GV100_3D_CLASS)
429          return shader != PIPE_SHADER_FRAGMENT;
430       return 1;
431    case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
432    case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
433       return 1;
434    case PIPE_SHADER_CAP_MAX_TEMPS:
435       return NVC0_CAP_MAX_PROGRAM_TEMPS;
436    case PIPE_SHADER_CAP_CONT_SUPPORTED:
437       return 1;
438    case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
439       return 1;
440    case PIPE_SHADER_CAP_SUBROUTINES:
441       return 1;
442    case PIPE_SHADER_CAP_INTEGERS:
443       return 1;
444    case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
445    case PIPE_SHADER_CAP_INT64_ATOMICS:
446    case PIPE_SHADER_CAP_FP16:
447    case PIPE_SHADER_CAP_FP16_DERIVATIVES:
448    case PIPE_SHADER_CAP_FP16_CONST_BUFFERS:
449    case PIPE_SHADER_CAP_INT16:
450    case PIPE_SHADER_CAP_GLSL_16BIT_CONSTS:
451    case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
452    case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
453       return 0;
454    case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
455       return NVC0_MAX_BUFFERS;
456    case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
457       return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
458    case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
459       return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
460    case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
461       if (class_3d >= NVE4_3D_CLASS)
462          return NVC0_MAX_IMAGES;
463       if (shader == PIPE_SHADER_FRAGMENT || shader == PIPE_SHADER_COMPUTE)
464          return NVC0_MAX_IMAGES;
465       return 0;
466    default:
467       NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
468       return 0;
469    }
470 }
471 
472 static float
nvc0_screen_get_paramf(struct pipe_screen * pscreen,enum pipe_capf param)473 nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
474 {
475    const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
476 
477    switch (param) {
478    case PIPE_CAPF_MIN_LINE_WIDTH:
479    case PIPE_CAPF_MIN_LINE_WIDTH_AA:
480    case PIPE_CAPF_MIN_POINT_SIZE:
481    case PIPE_CAPF_MIN_POINT_SIZE_AA:
482       return 1;
483    case PIPE_CAPF_POINT_SIZE_GRANULARITY:
484    case PIPE_CAPF_LINE_WIDTH_GRANULARITY:
485       return 0.1;
486    case PIPE_CAPF_MAX_LINE_WIDTH:
487    case PIPE_CAPF_MAX_LINE_WIDTH_AA:
488       return 10.0f;
489    case PIPE_CAPF_MAX_POINT_SIZE:
490       return 63.0f;
491    case PIPE_CAPF_MAX_POINT_SIZE_AA:
492       return 63.375f;
493    case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
494       return 16.0f;
495    case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
496       return 15.0f;
497    case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE:
498       return 0.0f;
499    case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE:
500       return class_3d >= GM200_3D_CLASS ? 0.75f : 0.0f;
501    case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY:
502       return class_3d >= GM200_3D_CLASS ? 0.25f : 0.0f;
503    }
504 
505    NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
506    return 0.0f;
507 }
508 
509 static int
nvc0_screen_get_compute_param(struct pipe_screen * pscreen,enum pipe_shader_ir ir_type,enum pipe_compute_cap param,void * data)510 nvc0_screen_get_compute_param(struct pipe_screen *pscreen,
511                               enum pipe_shader_ir ir_type,
512                               enum pipe_compute_cap param, void *data)
513 {
514    struct nvc0_screen *screen = nvc0_screen(pscreen);
515    struct nouveau_device *dev = screen->base.device;
516    const uint16_t obj_class = screen->compute->oclass;
517 
518 #define RET(x) do {                  \
519    if (data)                         \
520       memcpy(data, x, sizeof(x));    \
521    return sizeof(x);                 \
522 } while (0)
523 
524    switch (param) {
525    case PIPE_COMPUTE_CAP_GRID_DIMENSION:
526       RET((uint64_t []) { 3 });
527    case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
528       if (obj_class >= NVE4_COMPUTE_CLASS) {
529          RET(((uint64_t []) { 0x7fffffff, 65535, 65535 }));
530       } else {
531          RET(((uint64_t []) { 65535, 65535, 65535 }));
532       }
533    case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
534       RET(((uint64_t []) { 1024, 1024, 64 }));
535    case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
536       RET((uint64_t []) { 1024 });
537    case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
538       if (obj_class >= NVE4_COMPUTE_CLASS) {
539          RET((uint64_t []) { 1024 });
540       } else {
541          RET((uint64_t []) { 512 });
542       }
543    case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */
544       RET((uint64_t []) { nouveau_device_get_global_mem_size(dev) });
545    case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
546       switch (obj_class) {
547       case GM200_COMPUTE_CLASS:
548          RET((uint64_t []) { 96 << 10 });
549       case GM107_COMPUTE_CLASS:
550          RET((uint64_t []) { 64 << 10 });
551       default:
552          RET((uint64_t []) { 48 << 10 });
553       }
554    case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
555       RET((uint64_t []) { 512 << 10 });
556    case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
557       RET((uint64_t []) { 4096 });
558    case PIPE_COMPUTE_CAP_SUBGROUP_SIZES:
559       RET((uint32_t []) { 32 });
560    case PIPE_COMPUTE_CAP_MAX_SUBGROUPS:
561       RET((uint32_t []) { 0 });
562    case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
563       RET((uint64_t []) { nouveau_device_get_global_mem_size(dev) });
564    case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
565       RET((uint32_t []) { NVC0_MAX_IMAGES });
566    case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
567       RET((uint32_t []) { screen->mp_count_compute });
568    case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
569       RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
570    case PIPE_COMPUTE_CAP_ADDRESS_BITS:
571       RET((uint32_t []) { 64 });
572    default:
573       return 0;
574    }
575 
576 #undef RET
577 }
578 
579 static void
nvc0_screen_get_sample_pixel_grid(struct pipe_screen * pscreen,unsigned sample_count,unsigned * width,unsigned * height)580 nvc0_screen_get_sample_pixel_grid(struct pipe_screen *pscreen,
581                                   unsigned sample_count,
582                                   unsigned *width, unsigned *height)
583 {
584    switch (sample_count) {
585    case 0:
586    case 1:
587       /* this could be 4x4, but the GL state tracker makes it difficult to
588        * create a 1x MSAA texture and smaller grids save CB space */
589       *width = 2;
590       *height = 4;
591       break;
592    case 2:
593       *width = 2;
594       *height = 4;
595       break;
596    case 4:
597       *width = 2;
598       *height = 2;
599       break;
600    case 8:
601       *width = 1;
602       *height = 2;
603       break;
604    default:
605       assert(0);
606    }
607 }
608 
609 static void
nvc0_screen_destroy(struct pipe_screen * pscreen)610 nvc0_screen_destroy(struct pipe_screen *pscreen)
611 {
612    struct nvc0_screen *screen = nvc0_screen(pscreen);
613 
614    if (!nouveau_drm_screen_unref(&screen->base))
615       return;
616 
617    if (screen->blitter)
618       nvc0_blitter_destroy(screen);
619    if (screen->pm.prog) {
620       screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
621       nvc0_program_destroy(NULL, screen->pm.prog);
622       FREE(screen->pm.prog);
623    }
624 
625    nouveau_bo_ref(NULL, &screen->text);
626    nouveau_bo_ref(NULL, &screen->uniform_bo);
627    nouveau_bo_ref(NULL, &screen->tls);
628    nouveau_bo_ref(NULL, &screen->txc);
629    nouveau_bo_ref(NULL, &screen->fence.bo);
630    nouveau_bo_ref(NULL, &screen->poly_cache);
631 
632    nouveau_heap_free(&screen->lib_code);
633    nouveau_heap_destroy(&screen->text_heap);
634 
635    FREE(screen->tic.entries);
636 
637    nouveau_object_del(&screen->eng3d);
638    nouveau_object_del(&screen->eng2d);
639    nouveau_object_del(&screen->m2mf);
640    nouveau_object_del(&screen->copy);
641    nouveau_object_del(&screen->compute);
642    nouveau_object_del(&screen->nvsw);
643 
644    nouveau_screen_fini(&screen->base);
645    simple_mtx_destroy(&screen->state_lock);
646 
647    FREE(screen);
648 }
649 
650 static int
nvc0_graph_set_macro(struct nvc0_screen * screen,uint32_t m,unsigned pos,unsigned size,const uint32_t * data)651 nvc0_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos,
652                      unsigned size, const uint32_t *data)
653 {
654    struct nouveau_pushbuf *push = screen->base.pushbuf;
655 
656    size /= 4;
657 
658    assert((pos + size) <= 0x800);
659 
660    BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2);
661    PUSH_DATA (push, (m - 0x3800) / 8);
662    PUSH_DATA (push, pos);
663    BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1);
664    PUSH_DATA (push, pos);
665    PUSH_DATAp(push, data, size);
666 
667    return pos + size;
668 }
669 
670 static int
tu102_graph_set_macro(struct nvc0_screen * screen,uint32_t m,unsigned pos,unsigned size,const uint32_t * data)671 tu102_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos,
672                      unsigned size, const uint32_t *data)
673 {
674    struct nouveau_pushbuf *push = screen->base.pushbuf;
675 
676    size /= 4;
677 
678    assert((pos + size) <= 0x800);
679 
680    BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2);
681    PUSH_DATA (push, (m - 0x3800) / 8);
682    PUSH_DATA (push, pos);
683    BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1);
684    PUSH_DATA (push, pos);
685    PUSH_DATAp(push, data, size);
686 
687    return pos + (size / 3);
688 }
689 
690 static void
nvc0_magic_3d_init(struct nouveau_pushbuf * push,uint16_t obj_class)691 nvc0_magic_3d_init(struct nouveau_pushbuf *push, uint16_t obj_class)
692 {
693    BEGIN_NVC0(push, SUBC_3D(0x10cc), 1);
694    PUSH_DATA (push, 0xff);
695    BEGIN_NVC0(push, SUBC_3D(0x10e0), 2);
696    PUSH_DATA (push, 0xff);
697    PUSH_DATA (push, 0xff);
698    BEGIN_NVC0(push, SUBC_3D(0x10ec), 2);
699    PUSH_DATA (push, 0xff);
700    PUSH_DATA (push, 0xff);
701    if (obj_class < GV100_3D_CLASS) {
702       BEGIN_NVC0(push, SUBC_3D(0x074c), 1);
703       PUSH_DATA (push, 0x3f);
704    }
705 
706    BEGIN_NVC0(push, SUBC_3D(0x16a8), 1);
707    PUSH_DATA (push, (3 << 16) | 3);
708    BEGIN_NVC0(push, SUBC_3D(0x1794), 1);
709    PUSH_DATA (push, (2 << 16) | 2);
710 
711    if (obj_class < GM107_3D_CLASS) {
712       BEGIN_NVC0(push, SUBC_3D(0x12ac), 1);
713       PUSH_DATA (push, 0);
714    }
715    BEGIN_NVC0(push, SUBC_3D(0x0218), 1);
716    PUSH_DATA (push, 0x10);
717    BEGIN_NVC0(push, SUBC_3D(0x10fc), 1);
718    PUSH_DATA (push, 0x10);
719    BEGIN_NVC0(push, SUBC_3D(0x1290), 1);
720    PUSH_DATA (push, 0x10);
721    BEGIN_NVC0(push, SUBC_3D(0x12d8), 2);
722    PUSH_DATA (push, 0x10);
723    PUSH_DATA (push, 0x10);
724    BEGIN_NVC0(push, SUBC_3D(0x1140), 1);
725    PUSH_DATA (push, 0x10);
726    BEGIN_NVC0(push, SUBC_3D(0x1610), 1);
727    PUSH_DATA (push, 0xe);
728 
729    BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_GEN_MODE), 1);
730    PUSH_DATA (push, NVC0_3D_VERTEX_ID_GEN_MODE_DRAW_ARRAYS_ADD_START);
731    BEGIN_NVC0(push, SUBC_3D(0x030c), 1);
732    PUSH_DATA (push, 0);
733    BEGIN_NVC0(push, SUBC_3D(0x0300), 1);
734    PUSH_DATA (push, 3);
735 
736    if (obj_class < GV100_3D_CLASS) {
737       BEGIN_NVC0(push, SUBC_3D(0x02d0), 1);
738       PUSH_DATA (push, 0x3fffff);
739    }
740    BEGIN_NVC0(push, SUBC_3D(0x0fdc), 1);
741    PUSH_DATA (push, 1);
742    BEGIN_NVC0(push, SUBC_3D(0x19c0), 1);
743    PUSH_DATA (push, 1);
744 
745    if (obj_class < GM107_3D_CLASS) {
746       BEGIN_NVC0(push, SUBC_3D(0x075c), 1);
747       PUSH_DATA (push, 3);
748 
749       if (obj_class >= NVE4_3D_CLASS) {
750          BEGIN_NVC0(push, SUBC_3D(0x07fc), 1);
751          PUSH_DATA (push, 1);
752       }
753    }
754 
755    /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
756     * are supposed to do */
757 }
758 
759 static void
nvc0_screen_fence_emit(struct pipe_context * pcontext,u32 * sequence,struct nouveau_bo * wait)760 nvc0_screen_fence_emit(struct pipe_context *pcontext, u32 *sequence,
761                        struct nouveau_bo *wait)
762 {
763    struct nvc0_context *nvc0 = nvc0_context(pcontext);
764    struct nvc0_screen *screen = nvc0->screen;
765    struct nouveau_pushbuf *push = nvc0->base.pushbuf;
766    struct nouveau_pushbuf_refn ref = { wait, NOUVEAU_BO_GART | NOUVEAU_BO_RDWR };
767 
768    /* we need to do it after possible flush in MARK_RING */
769    *sequence = ++screen->base.fence.sequence;
770 
771    assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
772    PUSH_DATA (push, NVC0_FIFO_PKHDR_SQ(NVC0_3D(QUERY_ADDRESS_HIGH), 4));
773    PUSH_DATAh(push, screen->fence.bo->offset);
774    PUSH_DATA (push, screen->fence.bo->offset);
775    PUSH_DATA (push, *sequence);
776    PUSH_DATA (push, NVC0_3D_QUERY_GET_FENCE | NVC0_3D_QUERY_GET_SHORT |
777               (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT));
778 
779    nouveau_pushbuf_refn(push, &ref, 1);
780 }
781 
782 static u32
nvc0_screen_fence_update(struct pipe_screen * pscreen)783 nvc0_screen_fence_update(struct pipe_screen *pscreen)
784 {
785    struct nvc0_screen *screen = nvc0_screen(pscreen);
786    return screen->fence.map[0];
787 }
788 
789 static int
nvc0_screen_init_compute(struct nvc0_screen * screen)790 nvc0_screen_init_compute(struct nvc0_screen *screen)
791 {
792    const struct nouveau_mclass computes[] = {
793       { AD102_COMPUTE_CLASS, -1 },
794       { GA102_COMPUTE_CLASS, -1 },
795       { TU102_COMPUTE_CLASS, -1 },
796       { GV100_COMPUTE_CLASS, -1 },
797       { GP104_COMPUTE_CLASS, -1 },
798       { GP100_COMPUTE_CLASS, -1 },
799       { GM200_COMPUTE_CLASS, -1 },
800       { GM107_COMPUTE_CLASS, -1 },
801       {  NVF0_COMPUTE_CLASS, -1 },
802       {  NVE4_COMPUTE_CLASS, -1 },
803       /* In theory, GF110+ should also support NVC8_COMPUTE_CLASS but,
804        * in practice, a ILLEGAL_CLASS dmesg fail appears when using it. */
805 //      {  NVC8_COMPUTE_CLASS, -1 },
806       {  NVC0_COMPUTE_CLASS, -1 },
807       {}
808    };
809    struct nouveau_object *chan = screen->base.channel;
810    int ret;
811 
812    screen->base.base.get_compute_param = nvc0_screen_get_compute_param;
813 
814    ret = nouveau_object_mclass(chan, computes);
815    if (ret < 0) {
816       NOUVEAU_ERR("No supported compute class: %d\n", ret);
817       return ret;
818    }
819 
820    ret = nouveau_object_new(chan, 0xbeef00c0, computes[ret].oclass, NULL, 0, &screen->compute);
821    if (ret) {
822       NOUVEAU_ERR("Failed to allocate compute class: %d\n", ret);
823       return ret;
824    }
825 
826    if (screen->compute->oclass < NVE4_COMPUTE_CLASS)
827       return nvc0_screen_compute_setup(screen, screen->base.pushbuf);
828 
829    return nve4_screen_compute_setup(screen, screen->base.pushbuf);
830 }
831 
832 static int
nvc0_screen_resize_tls_area(struct nvc0_screen * screen,uint32_t lpos,uint32_t lneg,uint32_t cstack)833 nvc0_screen_resize_tls_area(struct nvc0_screen *screen,
834                             uint32_t lpos, uint32_t lneg, uint32_t cstack)
835 {
836    struct nouveau_bo *bo = NULL;
837    int ret;
838    uint64_t size = (lpos + lneg) * 32 + cstack;
839 
840    if (size >= (1 << 20)) {
841       NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64"\n", size);
842       return -1;
843    }
844 
845    size *= (screen->base.device->chipset >= 0xe0) ? 64 : 48; /* max warps */
846    size  = align(size, 0x8000);
847    size *= screen->mp_count;
848 
849    size = align(size, 1 << 17);
850 
851    ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base), 1 << 17, size,
852                         NULL, &bo);
853    if (ret)
854       return ret;
855 
856    /* Make sure that the pushbuf has acquired a reference to the old tls
857     * segment, as it may have commands that will reference it.
858     */
859    if (screen->tls)
860       PUSH_REF1(screen->base.pushbuf, screen->tls,
861                 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RDWR);
862    nouveau_bo_ref(NULL, &screen->tls);
863    screen->tls = bo;
864    return 0;
865 }
866 
867 int
nvc0_screen_resize_text_area(struct nvc0_screen * screen,struct nouveau_pushbuf * push,uint64_t size)868 nvc0_screen_resize_text_area(struct nvc0_screen *screen, struct nouveau_pushbuf *push,
869                              uint64_t size)
870 {
871    struct nouveau_bo *bo;
872    int ret;
873 
874    ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base),
875                         1 << 17, size, NULL, &bo);
876    if (ret)
877       return ret;
878 
879    /* Make sure that the pushbuf has acquired a reference to the old text
880     * segment, as it may have commands that will reference it.
881     */
882    if (screen->text)
883       PUSH_REF1(screen->base.pushbuf, screen->text,
884                 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RD);
885    nouveau_bo_ref(NULL, &screen->text);
886    screen->text = bo;
887 
888    nouveau_heap_free(&screen->lib_code);
889    nouveau_heap_destroy(&screen->text_heap);
890 
891    /* XXX: getting a page fault at the end of the code buffer every few
892     *  launches, don't use the last 256 bytes to work around them - prefetch ?
893     */
894    nouveau_heap_init(&screen->text_heap, 0, size - 0x100);
895 
896    /* update the code segment setup */
897    if (screen->eng3d->oclass < GV100_3D_CLASS) {
898       BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
899       PUSH_DATAh(push, screen->text->offset);
900       PUSH_DATA (push, screen->text->offset);
901       if (screen->compute) {
902          BEGIN_NVC0(push, NVC0_CP(CODE_ADDRESS_HIGH), 2);
903          PUSH_DATAh(push, screen->text->offset);
904          PUSH_DATA (push, screen->text->offset);
905       }
906    }
907 
908    return 0;
909 }
910 
911 void
nvc0_screen_bind_cb_3d(struct nvc0_screen * screen,struct nouveau_pushbuf * push,bool * can_serialize,int stage,int index,int size,uint64_t addr)912 nvc0_screen_bind_cb_3d(struct nvc0_screen *screen, struct nouveau_pushbuf *push,
913                        bool *can_serialize, int stage, int index, int size, uint64_t addr)
914 {
915    assert(stage != 5);
916 
917    if (screen->base.class_3d >= GM107_3D_CLASS) {
918       struct nvc0_cb_binding *binding = &screen->cb_bindings[stage][index];
919 
920       // TODO: Better figure out the conditions in which this is needed
921       bool serialize = binding->addr == addr && binding->size != size;
922       if (can_serialize)
923          serialize = serialize && *can_serialize;
924       if (serialize) {
925          IMMED_NVC0(push, NVC0_3D(SERIALIZE), 0);
926          if (can_serialize)
927             *can_serialize = false;
928       }
929 
930       binding->addr = addr;
931       binding->size = size;
932    }
933 
934    if (size >= 0) {
935       BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
936       PUSH_DATA (push, size);
937       PUSH_DATAh(push, addr);
938       PUSH_DATA (push, addr);
939    }
940    IMMED_NVC0(push, NVC0_3D(CB_BIND(stage)), (index << 4) | (size >= 0));
941 }
942 
943 static const void *
nvc0_screen_get_compiler_options(struct pipe_screen * pscreen,enum pipe_shader_ir ir,enum pipe_shader_type shader)944 nvc0_screen_get_compiler_options(struct pipe_screen *pscreen,
945                                  enum pipe_shader_ir ir,
946                                  enum pipe_shader_type shader)
947 {
948    struct nvc0_screen *screen = nvc0_screen(pscreen);
949    if (ir == PIPE_SHADER_IR_NIR)
950       return nv50_ir_nir_shader_compiler_options(screen->base.device->chipset, shader);
951    return NULL;
952 }
953 
954 #define FAIL_SCREEN_INIT(str, err)                    \
955    do {                                               \
956       NOUVEAU_ERR(str, err);                          \
957       goto fail;                                      \
958    } while(0)
959 
960 struct nouveau_screen *
nvc0_screen_create(struct nouveau_device * dev)961 nvc0_screen_create(struct nouveau_device *dev)
962 {
963    struct nvc0_screen *screen;
964    struct pipe_screen *pscreen;
965    struct nouveau_object *chan;
966 
967    struct nouveau_pushbuf *push;
968    uint64_t value;
969    uint32_t flags;
970    int ret;
971    unsigned i;
972 
973    switch (dev->chipset & ~0xf) {
974    case 0xc0:
975    case 0xd0:
976    case 0xe0:
977    case 0xf0:
978    case 0x100:
979    case 0x110:
980    case 0x120:
981    case 0x130:
982    case 0x140:
983    case 0x160:
984    case 0x170:
985    case 0x190:
986       break;
987    default:
988       return NULL;
989    }
990 
991    screen = CALLOC_STRUCT(nvc0_screen);
992    if (!screen)
993       return NULL;
994    pscreen = &screen->base.base;
995    pscreen->destroy = nvc0_screen_destroy;
996 
997    simple_mtx_init(&screen->state_lock, mtx_plain);
998 
999    ret = nouveau_screen_init(&screen->base, dev);
1000    if (ret)
1001       FAIL_SCREEN_INIT("Base screen init failed: %d\n", ret);
1002    chan = screen->base.channel;
1003    push = screen->base.pushbuf;
1004    push->rsvd_kick = 5;
1005 
1006    /* TODO: could this be higher on Kepler+? how does reclocking vs no
1007     * reclocking affect performance?
1008     * TODO: could this be higher on Fermi?
1009     */
1010    if (dev->chipset >= 0xe0)
1011       screen->base.transfer_pushbuf_threshold = 1024;
1012 
1013    screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
1014       PIPE_BIND_SHADER_BUFFER |
1015       PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
1016       PIPE_BIND_COMMAND_ARGS_BUFFER | PIPE_BIND_QUERY_BUFFER;
1017    screen->base.sysmem_bindings |=
1018       PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
1019 
1020    if (screen->base.vram_domain & NOUVEAU_BO_GART) {
1021       screen->base.sysmem_bindings |= screen->base.vidmem_bindings;
1022       screen->base.vidmem_bindings = 0;
1023    }
1024 
1025    pscreen->context_create = nvc0_create;
1026    pscreen->is_format_supported = nvc0_screen_is_format_supported;
1027    pscreen->get_param = nvc0_screen_get_param;
1028    pscreen->get_shader_param = nvc0_screen_get_shader_param;
1029    pscreen->get_paramf = nvc0_screen_get_paramf;
1030    pscreen->get_sample_pixel_grid = nvc0_screen_get_sample_pixel_grid;
1031    pscreen->get_driver_query_info = nvc0_screen_get_driver_query_info;
1032    pscreen->get_driver_query_group_info = nvc0_screen_get_driver_query_group_info;
1033    /* nir stuff */
1034    pscreen->get_compiler_options = nvc0_screen_get_compiler_options;
1035 
1036    nvc0_screen_init_resource_functions(pscreen);
1037 
1038    screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
1039    screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
1040 
1041    flags = NOUVEAU_BO_GART | NOUVEAU_BO_MAP;
1042    if (screen->base.drm->version >= 0x01000202)
1043       flags |= NOUVEAU_BO_COHERENT;
1044 
1045    ret = nouveau_bo_new(dev, flags, 0, 4096, NULL, &screen->fence.bo);
1046    if (ret)
1047       FAIL_SCREEN_INIT("Error allocating fence BO: %d\n", ret);
1048    BO_MAP(&screen->base, screen->fence.bo, 0, NULL);
1049    screen->fence.map = screen->fence.bo->map;
1050    screen->base.fence.emit = nvc0_screen_fence_emit;
1051    screen->base.fence.update = nvc0_screen_fence_update;
1052 
1053    if (dev->chipset < 0x140) {
1054       ret = nouveau_object_new(chan, (dev->chipset < 0xe0) ? 0x1f906e : 0x906e,
1055                                NVIF_CLASS_SW_GF100, NULL, 0, &screen->nvsw);
1056       if (ret)
1057          FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret);
1058 
1059       BEGIN_NVC0(push, SUBC_SW(NV01_SUBCHAN_OBJECT), 1);
1060       PUSH_DATA (push, screen->nvsw->handle);
1061    }
1062 
1063    const struct nouveau_mclass m2mfs[] = {
1064       { NVF0_P2MF_CLASS, -1 },
1065       { NVE4_P2MF_CLASS, -1 },
1066       { NVC0_M2MF_CLASS, -1 },
1067       {}
1068    };
1069 
1070    ret = nouveau_object_mclass(chan, m2mfs);
1071    if (ret < 0)
1072       FAIL_SCREEN_INIT("No supported m2mf class: %d\n", ret);
1073 
1074    ret = nouveau_object_new(chan, 0xbeef323f, m2mfs[ret].oclass, NULL, 0,
1075                             &screen->m2mf);
1076    if (ret)
1077       FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
1078 
1079    BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
1080    PUSH_DATA (push, screen->m2mf->oclass);
1081 
1082    if (screen->m2mf->oclass >= NVE4_P2MF_CLASS) {
1083       const struct nouveau_mclass copys[] = {
1084          {  AMPERE_DMA_COPY_B, -1 },
1085          {  AMPERE_DMA_COPY_A, -1 },
1086          {  TURING_DMA_COPY_A, -1 },
1087          {   VOLTA_DMA_COPY_A, -1 },
1088          {  PASCAL_DMA_COPY_B, -1 },
1089          {  PASCAL_DMA_COPY_A, -1 },
1090          { MAXWELL_DMA_COPY_A, -1 },
1091          {  KEPLER_DMA_COPY_A, -1 },
1092          {}
1093       };
1094 
1095       ret = nouveau_object_mclass(chan, copys);
1096       if (ret < 0)
1097          FAIL_SCREEN_INIT("No supported copy engine class: %d\n", ret);
1098 
1099       ret = nouveau_object_new(chan, 0, copys[ret].oclass, NULL, 0, &screen->copy);
1100       if (ret)
1101          FAIL_SCREEN_INIT("Error allocating copy engine class: %d\n", ret);
1102 
1103       BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
1104       PUSH_DATA (push, screen->copy->oclass);
1105    }
1106 
1107    ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
1108                             &screen->eng2d);
1109    if (ret)
1110       FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
1111 
1112    BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
1113    PUSH_DATA (push, screen->eng2d->oclass);
1114    BEGIN_NVC0(push, SUBC_2D(NVC0_2D_SINGLE_GPC), 1);
1115    PUSH_DATA (push, 0);
1116    BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
1117    PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
1118    BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
1119    PUSH_DATA (push, 0);
1120    BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
1121    PUSH_DATA (push, 0);
1122    BEGIN_NVC0(push, NVC0_2D(SET_PIXELS_FROM_MEMORY_CORRAL_SIZE), 1);
1123    PUSH_DATA (push, 0x3f);
1124    BEGIN_NVC0(push, NVC0_2D(SET_PIXELS_FROM_MEMORY_SAFE_OVERLAP), 1);
1125    PUSH_DATA (push, 1);
1126    BEGIN_NVC0(push, NVC0_2D(COND_MODE), 1);
1127    PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
1128 
1129    BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
1130    PUSH_DATAh(push, screen->fence.bo->offset + 16);
1131    PUSH_DATA (push, screen->fence.bo->offset + 16);
1132 
1133    const struct nouveau_mclass threeds[] = {
1134       { AD102_3D_CLASS, -1 },
1135       { GA102_3D_CLASS, -1 },
1136       { TU102_3D_CLASS, -1 },
1137       { GV100_3D_CLASS, -1 },
1138       { GP102_3D_CLASS, -1 },
1139       { GP100_3D_CLASS, -1 },
1140       { GM200_3D_CLASS, -1 },
1141       { GM107_3D_CLASS, -1 },
1142       {  NVF0_3D_CLASS, -1 },
1143       {  NVEA_3D_CLASS, -1 },
1144       {  NVE4_3D_CLASS, -1 },
1145       {  NVC8_3D_CLASS, -1 },
1146       {  NVC1_3D_CLASS, -1 },
1147       {  NVC0_3D_CLASS, -1 },
1148       {}
1149    };
1150 
1151    ret = nouveau_object_mclass(chan, threeds);
1152    if (ret < 0)
1153       FAIL_SCREEN_INIT("No supported 3d class: %d\n", ret);
1154 
1155    ret = nouveau_object_new(chan, 0xbeef003d, threeds[ret].oclass, NULL, 0,
1156                             &screen->eng3d);
1157    if (ret)
1158       FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
1159    screen->base.class_3d = screen->eng3d->oclass;
1160 
1161    BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
1162    PUSH_DATA (push, screen->eng3d->oclass);
1163 
1164    BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
1165    PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
1166 
1167    if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
1168       /* kill shaders after about 1 second (at 100 MHz) */
1169       BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
1170       PUSH_DATA (push, 0x17);
1171    }
1172 
1173    IMMED_NVC0(push, NVC0_3D(ZETA_COMP_ENABLE),
1174                     screen->base.drm->version >= 0x01000101);
1175    BEGIN_NVC0(push, NVC0_3D(RT_COMP_ENABLE(0)), 8);
1176    for (i = 0; i < 8; ++i)
1177       PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
1178 
1179    BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
1180    PUSH_DATA (push, 1);
1181 
1182    BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
1183    PUSH_DATA (push, 0);
1184    BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
1185    PUSH_DATA (push, 0);
1186    BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
1187    PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
1188    BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
1189    PUSH_DATA (push, 0);
1190    BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
1191    PUSH_DATA (push, 1);
1192    BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
1193    PUSH_DATA (push, 1);
1194    BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
1195    PUSH_DATA (push, 1);
1196    BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
1197    PUSH_DATA (push, 0);
1198    BEGIN_NVC0(push, NVC0_3D(SHADE_MODEL), 1);
1199    PUSH_DATA (push, NVC0_3D_SHADE_MODEL_SMOOTH);
1200    if (screen->eng3d->oclass < NVE4_3D_CLASS) {
1201       IMMED_NVC0(push, NVC0_3D(TEX_MISC), 0);
1202    } else if (screen->eng3d->oclass < GA102_3D_CLASS) {
1203       BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
1204       PUSH_DATA (push, 15);
1205    }
1206    BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
1207    PUSH_DATA (push, 8); /* 128 */
1208    BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
1209    PUSH_DATA (push, 1);
1210    if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
1211       BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
1212       PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
1213    }
1214 
1215    nvc0_magic_3d_init(push, screen->eng3d->oclass);
1216 
1217    ret = nvc0_screen_resize_text_area(screen, push, 1 << 19);
1218    if (ret)
1219       FAIL_SCREEN_INIT("Error allocating TEXT area: %d\n", ret);
1220 
1221    /* 6 user uniform areas, 6 driver areas, and 1 for the runout */
1222    ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 12, 13 << 16, NULL,
1223                         &screen->uniform_bo);
1224    if (ret)
1225       FAIL_SCREEN_INIT("Error allocating uniform BO: %d\n", ret);
1226 
1227    PUSH_REF1 (push, screen->uniform_bo, NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_WR);
1228 
1229    /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
1230    BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1231    PUSH_DATA (push, 256);
1232    PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1233    PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1234    BEGIN_1IC0(push, NVC0_3D(CB_POS), 5);
1235    PUSH_DATA (push, 0);
1236    PUSH_DATAf(push, 0.0f);
1237    PUSH_DATAf(push, 0.0f);
1238    PUSH_DATAf(push, 0.0f);
1239    PUSH_DATAf(push, 0.0f);
1240    BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
1241    PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1242    PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1243 
1244    if (screen->base.drm->version >= 0x01000101) {
1245       ret = nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
1246       if (ret)
1247          FAIL_SCREEN_INIT("NOUVEAU_GETPARAM_GRAPH_UNITS failed: %d\n", ret);
1248    } else {
1249       if (dev->chipset >= 0xe0 && dev->chipset < 0xf0)
1250          value = (8 << 8) | 4;
1251       else
1252          value = (16 << 8) | 4;
1253    }
1254    screen->gpc_count = value & 0x000000ff;
1255    screen->mp_count = value >> 8;
1256    screen->mp_count_compute = screen->mp_count;
1257 
1258    ret = nvc0_screen_resize_tls_area(screen, 128 * 16, 0, 0x200);
1259    if (ret)
1260       FAIL_SCREEN_INIT("Error allocating TLS area: %d\n", ret);
1261 
1262    BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
1263    PUSH_DATAh(push, screen->tls->offset);
1264    PUSH_DATA (push, screen->tls->offset);
1265    PUSH_DATA (push, screen->tls->size >> 32);
1266    PUSH_DATA (push, screen->tls->size);
1267    BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
1268    PUSH_DATA (push, 0);
1269    /* Reduce likelihood of collision with real buffers by placing the hole at
1270     * the top of the 4G area. This will have to be dealt with for real
1271     * eventually by blocking off that area from the VM.
1272     */
1273    BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
1274    PUSH_DATA (push, 0xff << 24);
1275 
1276    if (screen->eng3d->oclass < GM107_3D_CLASS) {
1277       ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 20, NULL,
1278                            &screen->poly_cache);
1279       if (ret)
1280          FAIL_SCREEN_INIT("Error allocating poly cache BO: %d\n", ret);
1281 
1282       BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
1283       PUSH_DATAh(push, screen->poly_cache->offset);
1284       PUSH_DATA (push, screen->poly_cache->offset);
1285       PUSH_DATA (push, 3);
1286    }
1287 
1288    ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 17, NULL,
1289                         &screen->txc);
1290    if (ret)
1291       FAIL_SCREEN_INIT("Error allocating txc BO: %d\n", ret);
1292 
1293    BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
1294    PUSH_DATAh(push, screen->txc->offset);
1295    PUSH_DATA (push, screen->txc->offset);
1296    PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
1297    if (screen->eng3d->oclass >= GM107_3D_CLASS) {
1298       screen->tic.maxwell = true;
1299       if (screen->eng3d->oclass == GM107_3D_CLASS) {
1300          screen->tic.maxwell =
1301             debug_get_bool_option("NOUVEAU_MAXWELL_TIC", true);
1302          IMMED_NVC0(push, SUBC_3D(0x0f10), screen->tic.maxwell);
1303       }
1304    }
1305 
1306    BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
1307    PUSH_DATAh(push, screen->txc->offset + 65536);
1308    PUSH_DATA (push, screen->txc->offset + 65536);
1309    PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
1310 
1311    BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
1312    PUSH_DATA (push, 0);
1313    BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
1314    PUSH_DATA (push, 0);
1315    PUSH_DATA (push, 0);
1316    BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
1317    PUSH_DATA (push, 0x3f);
1318 
1319    BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
1320    PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
1321    BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
1322    for (i = 0; i < 8 * 2; ++i)
1323       PUSH_DATA(push, 0);
1324    BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
1325    PUSH_DATA (push, 0);
1326    BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
1327    PUSH_DATA (push, 0);
1328 
1329    /* neither scissors, viewport nor stencil mask should affect clears */
1330    BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
1331    PUSH_DATA (push, 0);
1332 
1333    BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
1334    PUSH_DATA (push, 1);
1335    for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1336       BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(i)), 2);
1337       PUSH_DATAf(push, 0.0f);
1338       PUSH_DATAf(push, 1.0f);
1339    }
1340    BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
1341    PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
1342 
1343    /* We use scissors instead of exact view volume clipping,
1344     * so they're always enabled.
1345     */
1346    for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1347       BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(i)), 3);
1348       PUSH_DATA (push, 1);
1349       PUSH_DATA (push, 16384 << 16);
1350       PUSH_DATA (push, 16384 << 16);
1351    }
1352 
1353    if (screen->eng3d->oclass < TU102_3D_CLASS) {
1354 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
1355 
1356       i = 0;
1357       MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, mme9097_per_instance_bf);
1358       MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, mme9097_blend_enables);
1359       MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, mme9097_vertex_array_select);
1360       MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, mme9097_tep_select);
1361       MK_MACRO(NVC0_3D_MACRO_GP_SELECT, mme9097_gp_select);
1362       MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, mme9097_poly_mode_front);
1363       MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, mme9097_poly_mode_back);
1364       MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT, mme9097_draw_arrays_indirect);
1365       MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT, mme9097_draw_elts_indirect);
1366       MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT, mme9097_draw_arrays_indirect_count);
1367       MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT, mme9097_draw_elts_indirect_count);
1368       MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE, mme9097_query_buffer_write);
1369       MK_MACRO(NVC0_3D_MACRO_CONSERVATIVE_RASTER_STATE, mme9097_conservative_raster_state);
1370       MK_MACRO(NVC0_3D_MACRO_SET_PRIV_REG, mme9097_set_priv_reg);
1371       MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER, mme9097_compute_counter);
1372       MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER_TO_QUERY, mme9097_compute_counter_to_query);
1373       MK_MACRO(NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT, mme90c0_launch_grid_indirect);
1374    } else {
1375 #undef MK_MACRO
1376 #define MK_MACRO(m, n) i = tu102_graph_set_macro(screen, m, i, sizeof(n), n);
1377 
1378       i = 0;
1379       MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, mmec597_per_instance_bf);
1380       MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, mmec597_blend_enables);
1381       MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, mmec597_vertex_array_select);
1382       MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, mmec597_tep_select);
1383       MK_MACRO(NVC0_3D_MACRO_GP_SELECT, mmec597_gp_select);
1384       MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, mmec597_poly_mode_front);
1385       MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, mmec597_poly_mode_back);
1386       MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT, mmec597_draw_arrays_indirect);
1387       MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT, mmec597_draw_elts_indirect);
1388       MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT, mmec597_draw_arrays_indirect_count);
1389       MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT, mmec597_draw_elts_indirect_count);
1390       MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE, mmec597_query_buffer_write);
1391       MK_MACRO(NVC0_3D_MACRO_CONSERVATIVE_RASTER_STATE, mmec597_conservative_raster_state);
1392       MK_MACRO(NVC0_3D_MACRO_SET_PRIV_REG, mmec597_set_priv_reg);
1393       MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER, mmec597_compute_counter);
1394       MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER_TO_QUERY, mmec597_compute_counter_to_query);
1395    }
1396 
1397    BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
1398    PUSH_DATA (push, 1);
1399    BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
1400    PUSH_DATA (push, 1);
1401    BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
1402    PUSH_DATA (push, 0x40);
1403    BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
1404    PUSH_DATA (push, 0);
1405    BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
1406    PUSH_DATA (push, 0x30);
1407    BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
1408    PUSH_DATA (push, 3);
1409    BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
1410    PUSH_DATA (push, 0x20);
1411    BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
1412    PUSH_DATA (push, 0x00);
1413    screen->save_state.patch_vertices = 3;
1414 
1415    BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
1416    PUSH_DATA (push, 0);
1417    BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
1418    PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
1419 
1420    IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
1421 
1422    if (nvc0_screen_init_compute(screen))
1423       goto fail;
1424 
1425    /* XXX: Compute and 3D are somehow aliased on Fermi. */
1426    for (i = 0; i < 5; ++i) {
1427       unsigned j = 0;
1428       for (j = 0; j < 16; j++)
1429          screen->cb_bindings[i][j].size = -1;
1430 
1431       /* TIC and TSC entries for each unit (nve4+ only) */
1432       /* auxiliary constants (6 user clip planes, base instance id) */
1433       nvc0_screen_bind_cb_3d(screen, push, NULL, i, 15, NVC0_CB_AUX_SIZE,
1434                              screen->uniform_bo->offset + NVC0_CB_AUX_INFO(i));
1435       if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
1436          unsigned j;
1437          BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
1438          PUSH_DATA (push, NVC0_CB_AUX_UNK_INFO);
1439          for (j = 0; j < 8; ++j)
1440             PUSH_DATA(push, j);
1441       } else {
1442          BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
1443          PUSH_DATA (push, 0x54);
1444       }
1445 
1446       /* MS sample coordinate offsets: these do not work with _ALT modes ! */
1447       BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 2 * 8);
1448       PUSH_DATA (push, NVC0_CB_AUX_MS_INFO);
1449       PUSH_DATA (push, 0); /* 0 */
1450       PUSH_DATA (push, 0);
1451       PUSH_DATA (push, 1); /* 1 */
1452       PUSH_DATA (push, 0);
1453       PUSH_DATA (push, 0); /* 2 */
1454       PUSH_DATA (push, 1);
1455       PUSH_DATA (push, 1); /* 3 */
1456       PUSH_DATA (push, 1);
1457       PUSH_DATA (push, 2); /* 4 */
1458       PUSH_DATA (push, 0);
1459       PUSH_DATA (push, 3); /* 5 */
1460       PUSH_DATA (push, 0);
1461       PUSH_DATA (push, 2); /* 6 */
1462       PUSH_DATA (push, 1);
1463       PUSH_DATA (push, 3); /* 7 */
1464       PUSH_DATA (push, 1);
1465    }
1466    BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
1467    PUSH_DATA (push, 0);
1468 
1469    /* requires Nvidia provided firmware */
1470    if (screen->eng3d->oclass >= GM200_3D_CLASS) {
1471       unsigned reg = screen->eng3d->oclass >= GV100_3D_CLASS ? 0x419ba4 : 0x419f78;
1472       BEGIN_1IC0(push, NVC0_3D(MACRO_SET_PRIV_REG), 3);
1473       PUSH_DATA (push, reg);
1474       PUSH_DATA (push, 0x00000000);
1475       PUSH_DATA (push, 0x00000008);
1476    }
1477 
1478    PUSH_KICK (push);
1479 
1480    screen->tic.entries = CALLOC(
1481          NVC0_TIC_MAX_ENTRIES + NVC0_TSC_MAX_ENTRIES + NVE4_IMG_MAX_HANDLES,
1482          sizeof(void *));
1483    screen->tsc.entries = screen->tic.entries + NVC0_TIC_MAX_ENTRIES;
1484    screen->img.entries = (void *)(screen->tsc.entries + NVC0_TSC_MAX_ENTRIES);
1485 
1486    if (!nvc0_blitter_create(screen))
1487       goto fail;
1488 
1489    return &screen->base;
1490 
1491 fail:
1492    screen->base.base.context_create = NULL;
1493    return &screen->base;
1494 }
1495 
1496 int
nvc0_screen_tic_alloc(struct nvc0_screen * screen,void * entry)1497 nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
1498 {
1499    int i = screen->tic.next;
1500 
1501    while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1502       i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1503 
1504    screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1505 
1506    if (screen->tic.entries[i])
1507       nv50_tic_entry(screen->tic.entries[i])->id = -1;
1508 
1509    screen->tic.entries[i] = entry;
1510    return i;
1511 }
1512 
1513 int
nvc0_screen_tsc_alloc(struct nvc0_screen * screen,void * entry)1514 nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
1515 {
1516    int i = screen->tsc.next;
1517 
1518    while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1519       i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1520 
1521    screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1522 
1523    if (screen->tsc.entries[i])
1524       nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1525 
1526    screen->tsc.entries[i] = entry;
1527    return i;
1528 }
1529