xref: /aosp_15_r20/external/mesa3d/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c (revision 6104692788411f58d303aa86923a9ff6ecaded22)
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 "drm-uapi/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       return dev->info.pci.domain;
344    case PIPE_CAP_PCI_BUS:
345       return dev->info.pci.bus;
346    case PIPE_CAP_PCI_DEVICE:
347       return dev->info.pci.dev;
348    case PIPE_CAP_PCI_FUNCTION:
349       return dev->info.pci.func;
350 
351    case PIPE_CAP_OPENCL_INTEGER_FUNCTIONS: /* could be done */
352    case PIPE_CAP_INTEGER_MULTIPLY_32X16: /* could be done */
353    case PIPE_CAP_MAP_UNSYNCHRONIZED_THREAD_SAFE: /* when we fix MT stuff */
354    case PIPE_CAP_ALPHA_TO_COVERAGE_DITHER_CONTROL: /* TODO */
355    case PIPE_CAP_SHADER_ATOMIC_INT64: /* TODO */
356    case PIPE_CAP_HARDWARE_GL_SELECT:
357       return 0;
358 
359    case PIPE_CAP_VENDOR_ID:
360       return 0x10de;
361    case PIPE_CAP_DEVICE_ID:
362       return dev->info.device_id;
363    case PIPE_CAP_ACCELERATED:
364       return 1;
365    case PIPE_CAP_VIDEO_MEMORY:
366       return dev->vram_size >> 20;
367    case PIPE_CAP_UMA:
368       return nouveau_screen(pscreen)->is_uma;
369 
370    default:
371       return u_pipe_screen_get_param_defaults(pscreen, param);
372    }
373 }
374 
375 static int
nvc0_screen_get_shader_param(struct pipe_screen * pscreen,enum pipe_shader_type shader,enum pipe_shader_cap param)376 nvc0_screen_get_shader_param(struct pipe_screen *pscreen,
377                              enum pipe_shader_type shader,
378                              enum pipe_shader_cap param)
379 {
380    const struct nouveau_screen *screen = nouveau_screen(pscreen);
381    const uint16_t class_3d = screen->class_3d;
382 
383    switch (shader) {
384    case PIPE_SHADER_VERTEX:
385    case PIPE_SHADER_GEOMETRY:
386    case PIPE_SHADER_FRAGMENT:
387    case PIPE_SHADER_COMPUTE:
388    case PIPE_SHADER_TESS_CTRL:
389    case PIPE_SHADER_TESS_EVAL:
390       break;
391    default:
392       return 0;
393    }
394 
395    switch (param) {
396    case PIPE_SHADER_CAP_SUPPORTED_IRS: {
397       uint32_t irs = 1 << PIPE_SHADER_IR_NIR;
398       if (screen->force_enable_cl)
399          irs |= 1 << PIPE_SHADER_IR_NIR_SERIALIZED;
400       return irs;
401    }
402    case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
403    case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
404    case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
405    case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
406       return 16384;
407    case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
408       return 16;
409    case PIPE_SHADER_CAP_MAX_INPUTS:
410       return 0x200 / 16;
411    case PIPE_SHADER_CAP_MAX_OUTPUTS:
412       return 32;
413    case PIPE_SHADER_CAP_MAX_CONST_BUFFER0_SIZE:
414       return NVC0_MAX_CONSTBUF_SIZE;
415    case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
416       return NVC0_MAX_PIPE_CONSTBUFS;
417    case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
418       return shader != PIPE_SHADER_FRAGMENT;
419    case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
420       /* HW doesn't support indirect addressing of fragment program inputs
421        * on Volta.  The binary driver generates a function to handle every
422        * possible indirection, and indirectly calls the function to handle
423        * this instead.
424        */
425       if (class_3d >= GV100_3D_CLASS)
426          return shader != PIPE_SHADER_FRAGMENT;
427       return 1;
428    case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
429    case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
430       return 1;
431    case PIPE_SHADER_CAP_MAX_TEMPS:
432       return NVC0_CAP_MAX_PROGRAM_TEMPS;
433    case PIPE_SHADER_CAP_CONT_SUPPORTED:
434       return 1;
435    case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
436       return 1;
437    case PIPE_SHADER_CAP_SUBROUTINES:
438       return 1;
439    case PIPE_SHADER_CAP_INTEGERS:
440       return 1;
441    case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
442    case PIPE_SHADER_CAP_INT64_ATOMICS:
443    case PIPE_SHADER_CAP_FP16:
444    case PIPE_SHADER_CAP_FP16_DERIVATIVES:
445    case PIPE_SHADER_CAP_FP16_CONST_BUFFERS:
446    case PIPE_SHADER_CAP_INT16:
447    case PIPE_SHADER_CAP_GLSL_16BIT_CONSTS:
448    case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
449    case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
450       return 0;
451    case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
452       return NVC0_MAX_BUFFERS;
453    case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
454       return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
455    case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
456       return (class_3d >= NVE4_3D_CLASS) ? 32 : 16;
457    case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
458       if (class_3d >= NVE4_3D_CLASS)
459          return NVC0_MAX_IMAGES;
460       if (shader == PIPE_SHADER_FRAGMENT || shader == PIPE_SHADER_COMPUTE)
461          return NVC0_MAX_IMAGES;
462       return 0;
463    default:
464       NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
465       return 0;
466    }
467 }
468 
469 static float
nvc0_screen_get_paramf(struct pipe_screen * pscreen,enum pipe_capf param)470 nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
471 {
472    const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
473 
474    switch (param) {
475    case PIPE_CAPF_MIN_LINE_WIDTH:
476    case PIPE_CAPF_MIN_LINE_WIDTH_AA:
477    case PIPE_CAPF_MIN_POINT_SIZE:
478    case PIPE_CAPF_MIN_POINT_SIZE_AA:
479       return 1;
480    case PIPE_CAPF_POINT_SIZE_GRANULARITY:
481    case PIPE_CAPF_LINE_WIDTH_GRANULARITY:
482       return 0.1;
483    case PIPE_CAPF_MAX_LINE_WIDTH:
484    case PIPE_CAPF_MAX_LINE_WIDTH_AA:
485       return 10.0f;
486    case PIPE_CAPF_MAX_POINT_SIZE:
487       return 63.0f;
488    case PIPE_CAPF_MAX_POINT_SIZE_AA:
489       return 63.375f;
490    case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
491       return 16.0f;
492    case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
493       return 15.0f;
494    case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE:
495       return 0.0f;
496    case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE:
497       return class_3d >= GM200_3D_CLASS ? 0.75f : 0.0f;
498    case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY:
499       return class_3d >= GM200_3D_CLASS ? 0.25f : 0.0f;
500    }
501 
502    NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
503    return 0.0f;
504 }
505 
506 static int
nvc0_screen_get_compute_param(struct pipe_screen * pscreen,enum pipe_shader_ir ir_type,enum pipe_compute_cap param,void * data)507 nvc0_screen_get_compute_param(struct pipe_screen *pscreen,
508                               enum pipe_shader_ir ir_type,
509                               enum pipe_compute_cap param, void *data)
510 {
511    struct nvc0_screen *screen = nvc0_screen(pscreen);
512    struct nouveau_device *dev = screen->base.device;
513    const uint16_t obj_class = screen->compute->oclass;
514 
515 #define RET(x) do {                  \
516    if (data)                         \
517       memcpy(data, x, sizeof(x));    \
518    return sizeof(x);                 \
519 } while (0)
520 
521    switch (param) {
522    case PIPE_COMPUTE_CAP_GRID_DIMENSION:
523       RET((uint64_t []) { 3 });
524    case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
525       if (obj_class >= NVE4_COMPUTE_CLASS) {
526          RET(((uint64_t []) { 0x7fffffff, 65535, 65535 }));
527       } else {
528          RET(((uint64_t []) { 65535, 65535, 65535 }));
529       }
530    case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
531       RET(((uint64_t []) { 1024, 1024, 64 }));
532    case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
533       RET((uint64_t []) { 1024 });
534    case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
535       if (obj_class >= NVE4_COMPUTE_CLASS) {
536          RET((uint64_t []) { 1024 });
537       } else {
538          RET((uint64_t []) { 512 });
539       }
540    case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */
541       RET((uint64_t []) { nouveau_device_get_global_mem_size(dev) });
542    case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
543       switch (obj_class) {
544       case GM200_COMPUTE_CLASS:
545          RET((uint64_t []) { 96 << 10 });
546       case GM107_COMPUTE_CLASS:
547          RET((uint64_t []) { 64 << 10 });
548       default:
549          RET((uint64_t []) { 48 << 10 });
550       }
551    case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
552       RET((uint64_t []) { 512 << 10 });
553    case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
554       RET((uint64_t []) { 4096 });
555    case PIPE_COMPUTE_CAP_SUBGROUP_SIZES:
556       RET((uint32_t []) { 32 });
557    case PIPE_COMPUTE_CAP_MAX_SUBGROUPS:
558       RET((uint32_t []) { 0 });
559    case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
560       RET((uint64_t []) { nouveau_device_get_global_mem_size(dev) });
561    case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
562       RET((uint32_t []) { NVC0_MAX_IMAGES });
563    case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
564       RET((uint32_t []) { screen->mp_count_compute });
565    case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
566       RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
567    case PIPE_COMPUTE_CAP_ADDRESS_BITS:
568       RET((uint32_t []) { 64 });
569    default:
570       return 0;
571    }
572 
573 #undef RET
574 }
575 
576 static void
nvc0_screen_get_sample_pixel_grid(struct pipe_screen * pscreen,unsigned sample_count,unsigned * width,unsigned * height)577 nvc0_screen_get_sample_pixel_grid(struct pipe_screen *pscreen,
578                                   unsigned sample_count,
579                                   unsigned *width, unsigned *height)
580 {
581    switch (sample_count) {
582    case 0:
583    case 1:
584       /* this could be 4x4, but the GL state tracker makes it difficult to
585        * create a 1x MSAA texture and smaller grids save CB space */
586       *width = 2;
587       *height = 4;
588       break;
589    case 2:
590       *width = 2;
591       *height = 4;
592       break;
593    case 4:
594       *width = 2;
595       *height = 2;
596       break;
597    case 8:
598       *width = 1;
599       *height = 2;
600       break;
601    default:
602       assert(0);
603    }
604 }
605 
606 static void
nvc0_screen_destroy(struct pipe_screen * pscreen)607 nvc0_screen_destroy(struct pipe_screen *pscreen)
608 {
609    struct nvc0_screen *screen = nvc0_screen(pscreen);
610 
611    if (!nouveau_drm_screen_unref(&screen->base))
612       return;
613 
614    if (screen->blitter)
615       nvc0_blitter_destroy(screen);
616    if (screen->pm.prog) {
617       screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
618       nvc0_program_destroy(NULL, screen->pm.prog);
619       FREE(screen->pm.prog);
620    }
621 
622    nouveau_bo_ref(NULL, &screen->text);
623    nouveau_bo_ref(NULL, &screen->uniform_bo);
624    nouveau_bo_ref(NULL, &screen->tls);
625    nouveau_bo_ref(NULL, &screen->txc);
626    nouveau_bo_ref(NULL, &screen->fence.bo);
627    nouveau_bo_ref(NULL, &screen->poly_cache);
628 
629    nouveau_heap_free(&screen->lib_code);
630    nouveau_heap_destroy(&screen->text_heap);
631 
632    FREE(screen->tic.entries);
633 
634    nouveau_object_del(&screen->eng3d);
635    nouveau_object_del(&screen->eng2d);
636    nouveau_object_del(&screen->m2mf);
637    nouveau_object_del(&screen->copy);
638    nouveau_object_del(&screen->compute);
639    nouveau_object_del(&screen->nvsw);
640 
641    nouveau_screen_fini(&screen->base);
642    simple_mtx_destroy(&screen->state_lock);
643 
644    FREE(screen);
645 }
646 
647 static int
nvc0_graph_set_macro(struct nvc0_screen * screen,uint32_t m,unsigned pos,unsigned size,const uint32_t * data)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 int
tu102_graph_set_macro(struct nvc0_screen * screen,uint32_t m,unsigned pos,unsigned size,const uint32_t * data)668 tu102_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos,
669                      unsigned size, const uint32_t *data)
670 {
671    struct nouveau_pushbuf *push = screen->base.pushbuf;
672 
673    size /= 4;
674 
675    assert((pos + size) <= 0x800);
676 
677    BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2);
678    PUSH_DATA (push, (m - 0x3800) / 8);
679    PUSH_DATA (push, pos);
680    BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1);
681    PUSH_DATA (push, pos);
682    PUSH_DATAp(push, data, size);
683 
684    return pos + (size / 3);
685 }
686 
687 static void
nvc0_magic_3d_init(struct nouveau_pushbuf * push,uint16_t obj_class)688 nvc0_magic_3d_init(struct nouveau_pushbuf *push, uint16_t obj_class)
689 {
690    BEGIN_NVC0(push, SUBC_3D(0x10cc), 1);
691    PUSH_DATA (push, 0xff);
692    BEGIN_NVC0(push, SUBC_3D(0x10e0), 2);
693    PUSH_DATA (push, 0xff);
694    PUSH_DATA (push, 0xff);
695    BEGIN_NVC0(push, SUBC_3D(0x10ec), 2);
696    PUSH_DATA (push, 0xff);
697    PUSH_DATA (push, 0xff);
698    if (obj_class < GV100_3D_CLASS) {
699       BEGIN_NVC0(push, SUBC_3D(0x074c), 1);
700       PUSH_DATA (push, 0x3f);
701    }
702 
703    BEGIN_NVC0(push, SUBC_3D(0x16a8), 1);
704    PUSH_DATA (push, (3 << 16) | 3);
705    BEGIN_NVC0(push, SUBC_3D(0x1794), 1);
706    PUSH_DATA (push, (2 << 16) | 2);
707 
708    if (obj_class < GM107_3D_CLASS) {
709       BEGIN_NVC0(push, SUBC_3D(0x12ac), 1);
710       PUSH_DATA (push, 0);
711    }
712    BEGIN_NVC0(push, SUBC_3D(0x0218), 1);
713    PUSH_DATA (push, 0x10);
714    BEGIN_NVC0(push, SUBC_3D(0x10fc), 1);
715    PUSH_DATA (push, 0x10);
716    BEGIN_NVC0(push, SUBC_3D(0x1290), 1);
717    PUSH_DATA (push, 0x10);
718    BEGIN_NVC0(push, SUBC_3D(0x12d8), 2);
719    PUSH_DATA (push, 0x10);
720    PUSH_DATA (push, 0x10);
721    BEGIN_NVC0(push, SUBC_3D(0x1140), 1);
722    PUSH_DATA (push, 0x10);
723    BEGIN_NVC0(push, SUBC_3D(0x1610), 1);
724    PUSH_DATA (push, 0xe);
725 
726    BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_GEN_MODE), 1);
727    PUSH_DATA (push, NVC0_3D_VERTEX_ID_GEN_MODE_DRAW_ARRAYS_ADD_START);
728    BEGIN_NVC0(push, SUBC_3D(0x030c), 1);
729    PUSH_DATA (push, 0);
730    BEGIN_NVC0(push, SUBC_3D(0x0300), 1);
731    PUSH_DATA (push, 3);
732 
733    if (obj_class < GV100_3D_CLASS) {
734       BEGIN_NVC0(push, SUBC_3D(0x02d0), 1);
735       PUSH_DATA (push, 0x3fffff);
736    }
737    BEGIN_NVC0(push, SUBC_3D(0x0fdc), 1);
738    PUSH_DATA (push, 1);
739    BEGIN_NVC0(push, SUBC_3D(0x19c0), 1);
740    PUSH_DATA (push, 1);
741 
742    if (obj_class < GM107_3D_CLASS) {
743       BEGIN_NVC0(push, SUBC_3D(0x075c), 1);
744       PUSH_DATA (push, 3);
745 
746       if (obj_class >= NVE4_3D_CLASS) {
747          BEGIN_NVC0(push, SUBC_3D(0x07fc), 1);
748          PUSH_DATA (push, 1);
749       }
750    }
751 
752    /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc
753     * are supposed to do */
754 }
755 
756 static void
nvc0_screen_fence_emit(struct pipe_context * pcontext,u32 * sequence,struct nouveau_bo * wait)757 nvc0_screen_fence_emit(struct pipe_context *pcontext, u32 *sequence,
758                        struct nouveau_bo *wait)
759 {
760    struct nvc0_context *nvc0 = nvc0_context(pcontext);
761    struct nvc0_screen *screen = nvc0->screen;
762    struct nouveau_pushbuf *push = nvc0->base.pushbuf;
763    struct nouveau_pushbuf_refn ref = { wait, NOUVEAU_BO_GART | NOUVEAU_BO_RDWR };
764 
765    /* we need to do it after possible flush in MARK_RING */
766    *sequence = ++screen->base.fence.sequence;
767 
768    assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
769    PUSH_DATA (push, NVC0_FIFO_PKHDR_SQ(NVC0_3D(QUERY_ADDRESS_HIGH), 4));
770    PUSH_DATAh(push, screen->fence.bo->offset);
771    PUSH_DATA (push, screen->fence.bo->offset);
772    PUSH_DATA (push, *sequence);
773    PUSH_DATA (push, NVC0_3D_QUERY_GET_FENCE | NVC0_3D_QUERY_GET_SHORT |
774               (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT));
775 
776    nouveau_pushbuf_refn(push, &ref, 1);
777 }
778 
779 static u32
nvc0_screen_fence_update(struct pipe_screen * pscreen)780 nvc0_screen_fence_update(struct pipe_screen *pscreen)
781 {
782    struct nvc0_screen *screen = nvc0_screen(pscreen);
783    return screen->fence.map[0];
784 }
785 
786 static int
nvc0_screen_init_compute(struct nvc0_screen * screen)787 nvc0_screen_init_compute(struct nvc0_screen *screen)
788 {
789    const struct nouveau_mclass computes[] = {
790       { AD102_COMPUTE_CLASS, -1 },
791       { GA102_COMPUTE_CLASS, -1 },
792       { TU102_COMPUTE_CLASS, -1 },
793       { GV100_COMPUTE_CLASS, -1 },
794       { GP104_COMPUTE_CLASS, -1 },
795       { GP100_COMPUTE_CLASS, -1 },
796       { GM200_COMPUTE_CLASS, -1 },
797       { GM107_COMPUTE_CLASS, -1 },
798       {  NVF0_COMPUTE_CLASS, -1 },
799       {  NVE4_COMPUTE_CLASS, -1 },
800       /* In theory, GF110+ should also support NVC8_COMPUTE_CLASS but,
801        * in practice, a ILLEGAL_CLASS dmesg fail appears when using it. */
802 //      {  NVC8_COMPUTE_CLASS, -1 },
803       {  NVC0_COMPUTE_CLASS, -1 },
804       {}
805    };
806    struct nouveau_object *chan = screen->base.channel;
807    int ret;
808 
809    screen->base.base.get_compute_param = nvc0_screen_get_compute_param;
810 
811    ret = nouveau_object_mclass(chan, computes);
812    if (ret < 0) {
813       NOUVEAU_ERR("No supported compute class: %d\n", ret);
814       return ret;
815    }
816 
817    ret = nouveau_object_new(chan, 0xbeef00c0, computes[ret].oclass, NULL, 0, &screen->compute);
818    if (ret) {
819       NOUVEAU_ERR("Failed to allocate compute class: %d\n", ret);
820       return ret;
821    }
822 
823    if (screen->compute->oclass < NVE4_COMPUTE_CLASS)
824       return nvc0_screen_compute_setup(screen, screen->base.pushbuf);
825 
826    return nve4_screen_compute_setup(screen, screen->base.pushbuf);
827 }
828 
829 static int
nvc0_screen_resize_tls_area(struct nvc0_screen * screen,uint32_t lpos,uint32_t lneg,uint32_t cstack)830 nvc0_screen_resize_tls_area(struct nvc0_screen *screen,
831                             uint32_t lpos, uint32_t lneg, uint32_t cstack)
832 {
833    struct nouveau_bo *bo = NULL;
834    int ret;
835    uint64_t size = (lpos + lneg) * 32 + cstack;
836 
837    if (size >= (1 << 20)) {
838       NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64"\n", size);
839       return -1;
840    }
841 
842    size *= (screen->base.device->chipset >= 0xe0) ? 64 : 48; /* max warps */
843    size  = align(size, 0x8000);
844    size *= screen->mp_count;
845 
846    size = align(size, 1 << 17);
847 
848    ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base), 1 << 17, size,
849                         NULL, &bo);
850    if (ret)
851       return ret;
852 
853    /* Make sure that the pushbuf has acquired a reference to the old tls
854     * segment, as it may have commands that will reference it.
855     */
856    if (screen->tls)
857       PUSH_REF1(screen->base.pushbuf, screen->tls,
858                 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RDWR);
859    nouveau_bo_ref(NULL, &screen->tls);
860    screen->tls = bo;
861    return 0;
862 }
863 
864 int
nvc0_screen_resize_text_area(struct nvc0_screen * screen,struct nouveau_pushbuf * push,uint64_t size)865 nvc0_screen_resize_text_area(struct nvc0_screen *screen, struct nouveau_pushbuf *push,
866                              uint64_t size)
867 {
868    struct nouveau_bo *bo;
869    int ret;
870 
871    ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base),
872                         1 << 17, size, NULL, &bo);
873    if (ret)
874       return ret;
875 
876    /* Make sure that the pushbuf has acquired a reference to the old text
877     * segment, as it may have commands that will reference it.
878     */
879    if (screen->text)
880       PUSH_REF1(screen->base.pushbuf, screen->text,
881                 NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RD);
882    nouveau_bo_ref(NULL, &screen->text);
883    screen->text = bo;
884 
885    nouveau_heap_free(&screen->lib_code);
886    nouveau_heap_destroy(&screen->text_heap);
887 
888    /*
889     * Shader storage needs a 2K (from NVIDIA) overallocations at the end
890     * to avoid prefetch bugs.
891     */
892    nouveau_heap_init(&screen->text_heap, 0, size - 0x800);
893 
894    /* update the code segment setup */
895    if (screen->eng3d->oclass < GV100_3D_CLASS) {
896       BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
897       PUSH_DATAh(push, screen->text->offset);
898       PUSH_DATA (push, screen->text->offset);
899       if (screen->compute) {
900          BEGIN_NVC0(push, NVC0_CP(CODE_ADDRESS_HIGH), 2);
901          PUSH_DATAh(push, screen->text->offset);
902          PUSH_DATA (push, screen->text->offset);
903       }
904    }
905 
906    return 0;
907 }
908 
909 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)910 nvc0_screen_bind_cb_3d(struct nvc0_screen *screen, struct nouveau_pushbuf *push,
911                        bool *can_serialize, int stage, int index, int size, uint64_t addr)
912 {
913    assert(stage != 5);
914 
915    if (screen->base.class_3d >= GM107_3D_CLASS) {
916       struct nvc0_cb_binding *binding = &screen->cb_bindings[stage][index];
917 
918       // TODO: Better figure out the conditions in which this is needed
919       bool serialize = binding->addr == addr && binding->size != size;
920       if (can_serialize)
921          serialize = serialize && *can_serialize;
922       if (serialize) {
923          IMMED_NVC0(push, NVC0_3D(SERIALIZE), 0);
924          if (can_serialize)
925             *can_serialize = false;
926       }
927 
928       binding->addr = addr;
929       binding->size = size;
930    }
931 
932    if (size >= 0) {
933       BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
934       PUSH_DATA (push, size);
935       PUSH_DATAh(push, addr);
936       PUSH_DATA (push, addr);
937    }
938    IMMED_NVC0(push, NVC0_3D(CB_BIND(stage)), (index << 4) | (size >= 0));
939 }
940 
941 static const void *
nvc0_screen_get_compiler_options(struct pipe_screen * pscreen,enum pipe_shader_ir ir,enum pipe_shader_type shader)942 nvc0_screen_get_compiler_options(struct pipe_screen *pscreen,
943                                  enum pipe_shader_ir ir,
944                                  enum pipe_shader_type shader)
945 {
946    struct nvc0_screen *screen = nvc0_screen(pscreen);
947    if (ir == PIPE_SHADER_IR_NIR)
948       return nv50_ir_nir_shader_compiler_options(screen->base.device->chipset, shader);
949    return NULL;
950 }
951 
952 #define FAIL_SCREEN_INIT(str, err)                    \
953    do {                                               \
954       NOUVEAU_ERR(str, err);                          \
955       goto fail;                                      \
956    } while(0)
957 
958 struct nouveau_screen *
nvc0_screen_create(struct nouveau_device * dev)959 nvc0_screen_create(struct nouveau_device *dev)
960 {
961    struct nvc0_screen *screen;
962    struct pipe_screen *pscreen;
963    struct nouveau_object *chan;
964 
965    struct nouveau_pushbuf *push;
966    uint64_t value;
967    uint32_t flags;
968    int ret;
969    unsigned i;
970 
971    switch (dev->chipset & ~0xf) {
972    case 0xc0:
973    case 0xd0:
974    case 0xe0:
975    case 0xf0:
976    case 0x100:
977    case 0x110:
978    case 0x120:
979    case 0x130:
980    case 0x140:
981    case 0x160:
982    case 0x170:
983    case 0x190:
984       break;
985    default:
986       return NULL;
987    }
988 
989    screen = CALLOC_STRUCT(nvc0_screen);
990    if (!screen)
991       return NULL;
992    pscreen = &screen->base.base;
993    pscreen->destroy = nvc0_screen_destroy;
994 
995    simple_mtx_init(&screen->state_lock, mtx_plain);
996 
997    ret = nouveau_screen_init(&screen->base, dev);
998    if (ret)
999       FAIL_SCREEN_INIT("Base screen init failed: %d\n", ret);
1000    chan = screen->base.channel;
1001    push = screen->base.pushbuf;
1002    push->rsvd_kick = 5;
1003 
1004    /* TODO: could this be higher on Kepler+? how does reclocking vs no
1005     * reclocking affect performance?
1006     * TODO: could this be higher on Fermi?
1007     */
1008    if (dev->chipset >= 0xe0)
1009       screen->base.transfer_pushbuf_threshold = 1024;
1010 
1011    screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
1012       PIPE_BIND_SHADER_BUFFER |
1013       PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
1014       PIPE_BIND_COMMAND_ARGS_BUFFER | PIPE_BIND_QUERY_BUFFER;
1015    screen->base.sysmem_bindings |=
1016       PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
1017 
1018    if (screen->base.vram_domain & NOUVEAU_BO_GART) {
1019       screen->base.sysmem_bindings |= screen->base.vidmem_bindings;
1020       screen->base.vidmem_bindings = 0;
1021    }
1022 
1023    pscreen->context_create = nvc0_create;
1024    pscreen->is_format_supported = nvc0_screen_is_format_supported;
1025    pscreen->get_param = nvc0_screen_get_param;
1026    pscreen->get_shader_param = nvc0_screen_get_shader_param;
1027    pscreen->get_paramf = nvc0_screen_get_paramf;
1028    pscreen->get_sample_pixel_grid = nvc0_screen_get_sample_pixel_grid;
1029    pscreen->get_driver_query_info = nvc0_screen_get_driver_query_info;
1030    pscreen->get_driver_query_group_info = nvc0_screen_get_driver_query_group_info;
1031    /* nir stuff */
1032    pscreen->get_compiler_options = nvc0_screen_get_compiler_options;
1033 
1034    nvc0_screen_init_resource_functions(pscreen);
1035 
1036    screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
1037    screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
1038 
1039    flags = NOUVEAU_BO_GART | NOUVEAU_BO_MAP;
1040    if (screen->base.drm->version >= 0x01000202)
1041       flags |= NOUVEAU_BO_COHERENT;
1042 
1043    ret = nouveau_bo_new(dev, flags, 0, 4096, NULL, &screen->fence.bo);
1044    if (ret)
1045       FAIL_SCREEN_INIT("Error allocating fence BO: %d\n", ret);
1046    BO_MAP(&screen->base, screen->fence.bo, 0, NULL);
1047    screen->fence.map = screen->fence.bo->map;
1048    screen->base.fence.emit = nvc0_screen_fence_emit;
1049    screen->base.fence.update = nvc0_screen_fence_update;
1050 
1051    if (dev->chipset < 0x140) {
1052       ret = nouveau_object_new(chan, (dev->chipset < 0xe0) ? 0x1f906e : 0x906e,
1053                                NVIF_CLASS_SW_GF100, NULL, 0, &screen->nvsw);
1054       if (ret)
1055          FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret);
1056 
1057       BEGIN_NVC0(push, SUBC_SW(NV01_SUBCHAN_OBJECT), 1);
1058       PUSH_DATA (push, screen->nvsw->handle);
1059    }
1060 
1061    const struct nouveau_mclass m2mfs[] = {
1062       { NVF0_P2MF_CLASS, -1 },
1063       { NVE4_P2MF_CLASS, -1 },
1064       { NVC0_M2MF_CLASS, -1 },
1065       {}
1066    };
1067 
1068    ret = nouveau_object_mclass(chan, m2mfs);
1069    if (ret < 0)
1070       FAIL_SCREEN_INIT("No supported m2mf class: %d\n", ret);
1071 
1072    ret = nouveau_object_new(chan, 0xbeef323f, m2mfs[ret].oclass, NULL, 0,
1073                             &screen->m2mf);
1074    if (ret)
1075       FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret);
1076 
1077    BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
1078    PUSH_DATA (push, screen->m2mf->oclass);
1079 
1080    if (screen->m2mf->oclass >= NVE4_P2MF_CLASS) {
1081       const struct nouveau_mclass copys[] = {
1082          {  AMPERE_DMA_COPY_B, -1 },
1083          {  AMPERE_DMA_COPY_A, -1 },
1084          {  TURING_DMA_COPY_A, -1 },
1085          {   VOLTA_DMA_COPY_A, -1 },
1086          {  PASCAL_DMA_COPY_B, -1 },
1087          {  PASCAL_DMA_COPY_A, -1 },
1088          { MAXWELL_DMA_COPY_A, -1 },
1089          {  KEPLER_DMA_COPY_A, -1 },
1090          {}
1091       };
1092 
1093       ret = nouveau_object_mclass(chan, copys);
1094       if (ret < 0)
1095          FAIL_SCREEN_INIT("No supported copy engine class: %d\n", ret);
1096 
1097       ret = nouveau_object_new(chan, 0, copys[ret].oclass, NULL, 0, &screen->copy);
1098       if (ret)
1099          FAIL_SCREEN_INIT("Error allocating copy engine class: %d\n", ret);
1100 
1101       BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
1102       PUSH_DATA (push, screen->copy->oclass);
1103    }
1104 
1105    ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0,
1106                             &screen->eng2d);
1107    if (ret)
1108       FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret);
1109 
1110    BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
1111    PUSH_DATA (push, screen->eng2d->oclass);
1112    BEGIN_NVC0(push, SUBC_2D(NVC0_2D_SINGLE_GPC), 1);
1113    PUSH_DATA (push, 0);
1114    BEGIN_NVC0(push, NVC0_2D(OPERATION), 1);
1115    PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
1116    BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1);
1117    PUSH_DATA (push, 0);
1118    BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1);
1119    PUSH_DATA (push, 0);
1120    BEGIN_NVC0(push, NVC0_2D(SET_PIXELS_FROM_MEMORY_CORRAL_SIZE), 1);
1121    PUSH_DATA (push, 0x3f);
1122    BEGIN_NVC0(push, NVC0_2D(SET_PIXELS_FROM_MEMORY_SAFE_OVERLAP), 1);
1123    PUSH_DATA (push, 1);
1124    BEGIN_NVC0(push, NVC0_2D(COND_MODE), 1);
1125    PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
1126 
1127    BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2);
1128    PUSH_DATAh(push, screen->fence.bo->offset + 16);
1129    PUSH_DATA (push, screen->fence.bo->offset + 16);
1130 
1131    const struct nouveau_mclass threeds[] = {
1132       { AD102_3D_CLASS, -1 },
1133       { GA102_3D_CLASS, -1 },
1134       { TU102_3D_CLASS, -1 },
1135       { GV100_3D_CLASS, -1 },
1136       { GP102_3D_CLASS, -1 },
1137       { GP100_3D_CLASS, -1 },
1138       { GM200_3D_CLASS, -1 },
1139       { GM107_3D_CLASS, -1 },
1140       {  NVF0_3D_CLASS, -1 },
1141       {  NVEA_3D_CLASS, -1 },
1142       {  NVE4_3D_CLASS, -1 },
1143       {  NVC8_3D_CLASS, -1 },
1144       {  NVC1_3D_CLASS, -1 },
1145       {  NVC0_3D_CLASS, -1 },
1146       {}
1147    };
1148 
1149    ret = nouveau_object_mclass(chan, threeds);
1150    if (ret < 0)
1151       FAIL_SCREEN_INIT("No supported 3d class: %d\n", ret);
1152 
1153    ret = nouveau_object_new(chan, 0xbeef003d, threeds[ret].oclass, NULL, 0,
1154                             &screen->eng3d);
1155    if (ret)
1156       FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret);
1157    screen->base.class_3d = screen->eng3d->oclass;
1158 
1159    BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
1160    PUSH_DATA (push, screen->eng3d->oclass);
1161 
1162    BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1);
1163    PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS);
1164 
1165    if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
1166       /* kill shaders after about 1 second (at 100 MHz) */
1167       BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1);
1168       PUSH_DATA (push, 0x17);
1169    }
1170 
1171    IMMED_NVC0(push, NVC0_3D(ZETA_COMP_ENABLE),
1172                     screen->base.drm->version >= 0x01000101);
1173    BEGIN_NVC0(push, NVC0_3D(RT_COMP_ENABLE(0)), 8);
1174    for (i = 0; i < 8; ++i)
1175       PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
1176 
1177    BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1);
1178    PUSH_DATA (push, 1);
1179 
1180    BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1);
1181    PUSH_DATA (push, 0);
1182    BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1);
1183    PUSH_DATA (push, 0);
1184    BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1);
1185    PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1);
1186    BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1);
1187    PUSH_DATA (push, 0);
1188    BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1);
1189    PUSH_DATA (push, 1);
1190    BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
1191    PUSH_DATA (push, 1);
1192    BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1);
1193    PUSH_DATA (push, 1);
1194    BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1);
1195    PUSH_DATA (push, 0);
1196    BEGIN_NVC0(push, NVC0_3D(SHADE_MODEL), 1);
1197    PUSH_DATA (push, NVC0_3D_SHADE_MODEL_SMOOTH);
1198    if (screen->eng3d->oclass < NVE4_3D_CLASS) {
1199       IMMED_NVC0(push, NVC0_3D(TEX_MISC), 0);
1200    } else if (screen->eng3d->oclass < GA102_3D_CLASS) {
1201       BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1);
1202       PUSH_DATA (push, 15);
1203    }
1204    BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1);
1205    PUSH_DATA (push, 8); /* 128 */
1206    BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1);
1207    PUSH_DATA (push, 1);
1208    if (screen->eng3d->oclass >= NVC1_3D_CLASS) {
1209       BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1);
1210       PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1);
1211    }
1212 
1213    nvc0_magic_3d_init(push, screen->eng3d->oclass);
1214 
1215    ret = nvc0_screen_resize_text_area(screen, push, 1 << 19);
1216    if (ret)
1217       FAIL_SCREEN_INIT("Error allocating TEXT area: %d\n", ret);
1218 
1219    /* 6 user uniform areas, 6 driver areas, and 1 for the runout */
1220    ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 12, 13 << 16, NULL,
1221                         &screen->uniform_bo);
1222    if (ret)
1223       FAIL_SCREEN_INIT("Error allocating uniform BO: %d\n", ret);
1224 
1225    PUSH_REF1 (push, screen->uniform_bo, NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_WR);
1226 
1227    /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */
1228    BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1229    PUSH_DATA (push, 256);
1230    PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1231    PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1232    BEGIN_1IC0(push, NVC0_3D(CB_POS), 5);
1233    PUSH_DATA (push, 0);
1234    PUSH_DATAf(push, 0.0f);
1235    PUSH_DATAf(push, 0.0f);
1236    PUSH_DATAf(push, 0.0f);
1237    PUSH_DATAf(push, 0.0f);
1238    BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
1239    PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1240    PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO);
1241 
1242    if (screen->base.drm->version >= 0x01000101) {
1243       ret = nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
1244       if (ret)
1245          FAIL_SCREEN_INIT("NOUVEAU_GETPARAM_GRAPH_UNITS failed: %d\n", ret);
1246    } else {
1247       if (dev->chipset >= 0xe0 && dev->chipset < 0xf0)
1248          value = (8 << 8) | 4;
1249       else
1250          value = (16 << 8) | 4;
1251    }
1252    screen->gpc_count = value & 0x000000ff;
1253    screen->mp_count = value >> 8;
1254    screen->mp_count_compute = screen->mp_count;
1255 
1256    ret = nvc0_screen_resize_tls_area(screen, 128 * 16, 0, 0x200);
1257    if (ret)
1258       FAIL_SCREEN_INIT("Error allocating TLS area: %d\n", ret);
1259 
1260    BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4);
1261    PUSH_DATAh(push, screen->tls->offset);
1262    PUSH_DATA (push, screen->tls->offset);
1263    PUSH_DATA (push, screen->tls->size >> 32);
1264    PUSH_DATA (push, screen->tls->size);
1265    BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1);
1266    PUSH_DATA (push, 0);
1267    /* Reduce likelihood of collision with real buffers by placing the hole at
1268     * the top of the 4G area. This will have to be dealt with for real
1269     * eventually by blocking off that area from the VM.
1270     */
1271    BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1);
1272    PUSH_DATA (push, 0xff << 24);
1273 
1274    if (screen->eng3d->oclass < GM107_3D_CLASS) {
1275       ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 20, NULL,
1276                            &screen->poly_cache);
1277       if (ret)
1278          FAIL_SCREEN_INIT("Error allocating poly cache BO: %d\n", ret);
1279 
1280       BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3);
1281       PUSH_DATAh(push, screen->poly_cache->offset);
1282       PUSH_DATA (push, screen->poly_cache->offset);
1283       PUSH_DATA (push, 3);
1284    }
1285 
1286    ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 17, NULL,
1287                         &screen->txc);
1288    if (ret)
1289       FAIL_SCREEN_INIT("Error allocating txc BO: %d\n", ret);
1290 
1291    BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3);
1292    PUSH_DATAh(push, screen->txc->offset);
1293    PUSH_DATA (push, screen->txc->offset);
1294    PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
1295    if (screen->eng3d->oclass >= GM107_3D_CLASS) {
1296       screen->tic.maxwell = true;
1297       if (screen->eng3d->oclass == GM107_3D_CLASS) {
1298          screen->tic.maxwell =
1299             debug_get_bool_option("NOUVEAU_MAXWELL_TIC", true);
1300          IMMED_NVC0(push, SUBC_3D(0x0f10), screen->tic.maxwell);
1301       }
1302    }
1303 
1304    BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3);
1305    PUSH_DATAh(push, screen->txc->offset + 65536);
1306    PUSH_DATA (push, screen->txc->offset + 65536);
1307    PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
1308 
1309    BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1);
1310    PUSH_DATA (push, 0);
1311    BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2);
1312    PUSH_DATA (push, 0);
1313    PUSH_DATA (push, 0);
1314    BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
1315    PUSH_DATA (push, 0x3f);
1316 
1317    BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1);
1318    PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY);
1319    BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
1320    for (i = 0; i < 8 * 2; ++i)
1321       PUSH_DATA(push, 0);
1322    BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1);
1323    PUSH_DATA (push, 0);
1324    BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1);
1325    PUSH_DATA (push, 0);
1326 
1327    /* neither scissors, viewport nor stencil mask should affect clears */
1328    BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1);
1329    PUSH_DATA (push, 0);
1330 
1331    BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1);
1332    PUSH_DATA (push, 1);
1333    for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1334       BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(i)), 2);
1335       PUSH_DATAf(push, 0.0f);
1336       PUSH_DATAf(push, 1.0f);
1337    }
1338    BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1);
1339    PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1);
1340 
1341    /* We use scissors instead of exact view volume clipping,
1342     * so they're always enabled.
1343     */
1344    for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
1345       BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(i)), 3);
1346       PUSH_DATA (push, 1);
1347       PUSH_DATA (push, 16384 << 16);
1348       PUSH_DATA (push, 16384 << 16);
1349    }
1350 
1351    if (screen->eng3d->oclass < TU102_3D_CLASS) {
1352 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
1353 
1354       i = 0;
1355       MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, mme9097_per_instance_bf);
1356       MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, mme9097_blend_enables);
1357       MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, mme9097_vertex_array_select);
1358       MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, mme9097_tep_select);
1359       MK_MACRO(NVC0_3D_MACRO_GP_SELECT, mme9097_gp_select);
1360       MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, mme9097_poly_mode_front);
1361       MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, mme9097_poly_mode_back);
1362       MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT, mme9097_draw_arrays_indirect);
1363       MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT, mme9097_draw_elts_indirect);
1364       MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT, mme9097_draw_arrays_indirect_count);
1365       MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT, mme9097_draw_elts_indirect_count);
1366       MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE, mme9097_query_buffer_write);
1367       MK_MACRO(NVC0_3D_MACRO_CONSERVATIVE_RASTER_STATE, mme9097_conservative_raster_state);
1368       MK_MACRO(NVC0_3D_MACRO_SET_PRIV_REG, mme9097_set_priv_reg);
1369       MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER, mme9097_compute_counter);
1370       MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER_TO_QUERY, mme9097_compute_counter_to_query);
1371       MK_MACRO(NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT, mme90c0_launch_grid_indirect);
1372    } else {
1373 #undef MK_MACRO
1374 #define MK_MACRO(m, n) i = tu102_graph_set_macro(screen, m, i, sizeof(n), n);
1375 
1376       i = 0;
1377       MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, mmec597_per_instance_bf);
1378       MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, mmec597_blend_enables);
1379       MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, mmec597_vertex_array_select);
1380       MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, mmec597_tep_select);
1381       MK_MACRO(NVC0_3D_MACRO_GP_SELECT, mmec597_gp_select);
1382       MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, mmec597_poly_mode_front);
1383       MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, mmec597_poly_mode_back);
1384       MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT, mmec597_draw_arrays_indirect);
1385       MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT, mmec597_draw_elts_indirect);
1386       MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT, mmec597_draw_arrays_indirect_count);
1387       MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT, mmec597_draw_elts_indirect_count);
1388       MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE, mmec597_query_buffer_write);
1389       MK_MACRO(NVC0_3D_MACRO_CONSERVATIVE_RASTER_STATE, mmec597_conservative_raster_state);
1390       MK_MACRO(NVC0_3D_MACRO_SET_PRIV_REG, mmec597_set_priv_reg);
1391       MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER, mmec597_compute_counter);
1392       MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER_TO_QUERY, mmec597_compute_counter_to_query);
1393    }
1394 
1395    BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
1396    PUSH_DATA (push, 1);
1397    BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1);
1398    PUSH_DATA (push, 1);
1399    BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1);
1400    PUSH_DATA (push, 0x40);
1401    BEGIN_NVC0(push, NVC0_3D(LAYER), 1);
1402    PUSH_DATA (push, 0);
1403    BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1);
1404    PUSH_DATA (push, 0x30);
1405    BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1);
1406    PUSH_DATA (push, 3);
1407    BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1);
1408    PUSH_DATA (push, 0x20);
1409    BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1);
1410    PUSH_DATA (push, 0x00);
1411    screen->save_state.patch_vertices = 3;
1412 
1413    BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1);
1414    PUSH_DATA (push, 0);
1415    BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1);
1416    PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL);
1417 
1418    IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1);
1419 
1420    if (nvc0_screen_init_compute(screen))
1421       goto fail;
1422 
1423    /* XXX: Compute and 3D are somehow aliased on Fermi. */
1424    for (i = 0; i < 5; ++i) {
1425       unsigned j = 0;
1426       for (j = 0; j < 16; j++)
1427          screen->cb_bindings[i][j].size = -1;
1428 
1429       /* TIC and TSC entries for each unit (nve4+ only) */
1430       /* auxiliary constants (6 user clip planes, base instance id) */
1431       nvc0_screen_bind_cb_3d(screen, push, NULL, i, 15, NVC0_CB_AUX_SIZE,
1432                              screen->uniform_bo->offset + NVC0_CB_AUX_INFO(i));
1433       if (screen->eng3d->oclass >= NVE4_3D_CLASS) {
1434          unsigned j;
1435          BEGIN_1IC0(push, NVC0_3D(CB_POS), 9);
1436          PUSH_DATA (push, NVC0_CB_AUX_UNK_INFO);
1437          for (j = 0; j < 8; ++j)
1438             PUSH_DATA(push, j);
1439       } else {
1440          BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
1441          PUSH_DATA (push, 0x54);
1442       }
1443 
1444       /* MS sample coordinate offsets: these do not work with _ALT modes ! */
1445       BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 2 * 8);
1446       PUSH_DATA (push, NVC0_CB_AUX_MS_INFO);
1447       PUSH_DATA (push, 0); /* 0 */
1448       PUSH_DATA (push, 0);
1449       PUSH_DATA (push, 1); /* 1 */
1450       PUSH_DATA (push, 0);
1451       PUSH_DATA (push, 0); /* 2 */
1452       PUSH_DATA (push, 1);
1453       PUSH_DATA (push, 1); /* 3 */
1454       PUSH_DATA (push, 1);
1455       PUSH_DATA (push, 2); /* 4 */
1456       PUSH_DATA (push, 0);
1457       PUSH_DATA (push, 3); /* 5 */
1458       PUSH_DATA (push, 0);
1459       PUSH_DATA (push, 2); /* 6 */
1460       PUSH_DATA (push, 1);
1461       PUSH_DATA (push, 3); /* 7 */
1462       PUSH_DATA (push, 1);
1463    }
1464    BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
1465    PUSH_DATA (push, 0);
1466 
1467    /* requires Nvidia provided firmware */
1468    if (screen->eng3d->oclass >= GM200_3D_CLASS) {
1469       unsigned reg = screen->eng3d->oclass >= GV100_3D_CLASS ? 0x419ba4 : 0x419f78;
1470       BEGIN_1IC0(push, NVC0_3D(MACRO_SET_PRIV_REG), 3);
1471       PUSH_DATA (push, reg);
1472       PUSH_DATA (push, 0x00000000);
1473       PUSH_DATA (push, 0x00000008);
1474    }
1475 
1476    PUSH_KICK (push);
1477 
1478    screen->tic.entries = CALLOC(
1479          NVC0_TIC_MAX_ENTRIES + NVC0_TSC_MAX_ENTRIES + NVE4_IMG_MAX_HANDLES,
1480          sizeof(void *));
1481    screen->tsc.entries = screen->tic.entries + NVC0_TIC_MAX_ENTRIES;
1482    screen->img.entries = (void *)(screen->tsc.entries + NVC0_TSC_MAX_ENTRIES);
1483 
1484    if (!nvc0_blitter_create(screen))
1485       goto fail;
1486 
1487    nouveau_device_set_classes_for_debug(dev,
1488                                         screen->eng3d->oclass,
1489                                         screen->compute->oclass,
1490                                         screen->m2mf->oclass,
1491                                         screen->copy ? screen->copy->oclass : 0);
1492    return &screen->base;
1493 
1494 fail:
1495    screen->base.base.context_create = NULL;
1496    return &screen->base;
1497 }
1498 
1499 int
nvc0_screen_tic_alloc(struct nvc0_screen * screen,void * entry)1500 nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry)
1501 {
1502    int i = screen->tic.next;
1503 
1504    while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1505       i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1506 
1507    screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1);
1508 
1509    if (screen->tic.entries[i])
1510       nv50_tic_entry(screen->tic.entries[i])->id = -1;
1511 
1512    screen->tic.entries[i] = entry;
1513    return i;
1514 }
1515 
1516 int
nvc0_screen_tsc_alloc(struct nvc0_screen * screen,void * entry)1517 nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry)
1518 {
1519    int i = screen->tsc.next;
1520 
1521    while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1522       i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1523 
1524    screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1);
1525 
1526    if (screen->tsc.entries[i])
1527       nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1528 
1529    screen->tsc.entries[i] = entry;
1530    return i;
1531 }
1532