• 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 <errno.h>
24 #include <xf86drm.h>
25 #include <nouveau_drm.h>
26 #include "util/format/u_format.h"
27 #include "util/format/u_format_s3tc.h"
28 #include "util/u_screen.h"
29 #include "pipe/p_screen.h"
30 
31 #include "nv50_ir_driver.h"
32 
33 #include "nv50/nv50_context.h"
34 #include "nv50/nv50_screen.h"
35 
36 #include "nouveau_vp3_video.h"
37 
38 #include "nv_object.xml.h"
39 
40 /* affected by LOCAL_WARPS_LOG_ALLOC / LOCAL_WARPS_NO_CLAMP */
41 #define LOCAL_WARPS_ALLOC 32
42 /* affected by STACK_WARPS_LOG_ALLOC / STACK_WARPS_NO_CLAMP */
43 #define STACK_WARPS_ALLOC 32
44 
45 #define THREADS_IN_WARP 32
46 
47 static bool
nv50_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)48 nv50_screen_is_format_supported(struct pipe_screen *pscreen,
49                                 enum pipe_format format,
50                                 enum pipe_texture_target target,
51                                 unsigned sample_count,
52                                 unsigned storage_sample_count,
53                                 unsigned bindings)
54 {
55    if (sample_count > 8)
56       return false;
57    if (!(0x117 & (1 << sample_count))) /* 0, 1, 2, 4 or 8 */
58       return false;
59    if (sample_count == 8 && util_format_get_blocksizebits(format) >= 128)
60       return false;
61 
62    if (MAX2(1, sample_count) != MAX2(1, storage_sample_count))
63       return false;
64 
65    /* Short-circuit the rest of the logic -- this is used by the gallium frontend
66     * to determine valid MS levels in a no-attachments scenario.
67     */
68    if (format == PIPE_FORMAT_NONE && bindings & PIPE_BIND_RENDER_TARGET)
69       return true;
70 
71    switch (format) {
72    case PIPE_FORMAT_Z16_UNORM:
73       if (nv50_screen(pscreen)->tesla->oclass < NVA0_3D_CLASS)
74          return false;
75       break;
76    default:
77       break;
78    }
79 
80    if (bindings & PIPE_BIND_LINEAR)
81       if (util_format_is_depth_or_stencil(format) ||
82           (target != PIPE_TEXTURE_1D &&
83            target != PIPE_TEXTURE_2D &&
84            target != PIPE_TEXTURE_RECT) ||
85           sample_count > 1)
86          return false;
87 
88    /* shared is always supported */
89    bindings &= ~(PIPE_BIND_LINEAR |
90                  PIPE_BIND_SHARED);
91 
92    if (bindings & PIPE_BIND_INDEX_BUFFER) {
93       if (format != PIPE_FORMAT_R8_UINT &&
94           format != PIPE_FORMAT_R16_UINT &&
95           format != PIPE_FORMAT_R32_UINT)
96          return false;
97       bindings &= ~PIPE_BIND_INDEX_BUFFER;
98    }
99 
100    return (( nv50_format_table[format].usage |
101             nv50_vertex_format[format].usage) & bindings) == bindings;
102 }
103 
104 static int
nv50_screen_get_param(struct pipe_screen * pscreen,enum pipe_cap param)105 nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
106 {
107    struct nouveau_screen *screen = nouveau_screen(pscreen);
108    const uint16_t class_3d = screen->class_3d;
109    struct nouveau_device *dev = screen->device;
110    static bool debug_cap_printed[PIPE_CAP_LAST] = {};
111 
112    switch (param) {
113    /* non-boolean caps */
114    case PIPE_CAP_MAX_TEXTURE_2D_SIZE:
115       return 8192;
116    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
117       return 12;
118    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
119       return 14;
120    case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
121       return 512;
122    case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
123    case PIPE_CAP_MIN_TEXEL_OFFSET:
124       return -8;
125    case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
126    case PIPE_CAP_MAX_TEXEL_OFFSET:
127       return 7;
128    case PIPE_CAP_MAX_TEXEL_BUFFER_ELEMENTS_UINT:
129       return 128 * 1024 * 1024;
130    case PIPE_CAP_GLSL_FEATURE_LEVEL:
131       return 330;
132    case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY:
133       return 330;
134    case PIPE_CAP_ESSL_FEATURE_LEVEL:
135       return class_3d >= NVA3_3D_CLASS ? 310 : 300;
136    case PIPE_CAP_MAX_RENDER_TARGETS:
137       return 8;
138    case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
139       return 1;
140    case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
141       return NV50_MAX_GLOBALS - 1;
142    case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
143    case PIPE_CAP_RASTERIZER_SUBPIXEL_BITS:
144       return 8;
145    case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
146       return 4;
147    case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
148       return 64;
149    case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
150       return 4;
151    case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
152    case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
153       return 1024;
154    case PIPE_CAP_MAX_VERTEX_STREAMS:
155       return 1;
156    case PIPE_CAP_MAX_GS_INVOCATIONS:
157       return 0;
158    case PIPE_CAP_MAX_SHADER_BUFFER_SIZE_UINT:
159       return 1 << 27;
160    case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
161       return 2048;
162    case PIPE_CAP_MAX_VERTEX_ELEMENT_SRC_OFFSET:
163       return 2047;
164    case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
165       return 256;
166    case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
167       return 16; /* 256 for binding as RT, but that's not possible in GL */
168    case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
169       return 256; /* the access limit is aligned to 256 */
170    case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
171       return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
172    case PIPE_CAP_MAX_VIEWPORTS:
173       return NV50_MAX_VIEWPORTS;
174    case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
175       return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
176    case PIPE_CAP_ENDIANNESS:
177       return PIPE_ENDIAN_LITTLE;
178    case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
179       return (class_3d >= NVA3_3D_CLASS) ? 4 : 0;
180    case PIPE_CAP_MAX_WINDOW_RECTANGLES:
181       return NV50_MAX_WINDOW_RECTANGLES;
182    case PIPE_CAP_MAX_TEXTURE_UPLOAD_MEMORY_BUDGET:
183       return 16 * 1024 * 1024;
184    case PIPE_CAP_MAX_VARYINGS:
185       return 15;
186    case PIPE_CAP_MAX_VERTEX_BUFFERS:
187       return 16;
188    case PIPE_CAP_GL_BEGIN_END_BUFFER_SIZE:
189       return 512 * 1024; /* TODO: Investigate tuning this */
190    case PIPE_CAP_MAX_TEXTURE_MB:
191       return 0; /* TODO: use 1/2 of VRAM for this? */
192 
193    case PIPE_CAP_SUPPORTED_PRIM_MODES_WITH_RESTART:
194    case PIPE_CAP_SUPPORTED_PRIM_MODES:
195       return BITFIELD_MASK(PIPE_PRIM_MAX);
196 
197    /* supported caps */
198    case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
199    case PIPE_CAP_TEXTURE_MIRROR_CLAMP_TO_EDGE:
200    case PIPE_CAP_TEXTURE_SWIZZLE:
201    case PIPE_CAP_TEXTURE_SHADOW_MAP:
202    case PIPE_CAP_NPOT_TEXTURES:
203    case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
204    case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
205    case PIPE_CAP_ANISOTROPIC_FILTER:
206    case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
207    case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
208    case PIPE_CAP_DEPTH_CLIP_DISABLE:
209    case PIPE_CAP_POINT_SPRITE:
210    case PIPE_CAP_FRAGMENT_SHADER_TEXTURE_LOD:
211    case PIPE_CAP_FRAGMENT_SHADER_DERIVATIVES:
212    case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
213    case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
214    case PIPE_CAP_VERTEX_COLOR_CLAMPED:
215    case PIPE_CAP_QUERY_TIMESTAMP:
216    case PIPE_CAP_QUERY_TIME_ELAPSED:
217    case PIPE_CAP_OCCLUSION_QUERY:
218    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
219    case PIPE_CAP_INDEP_BLEND_ENABLE:
220    case PIPE_CAP_FS_COORD_ORIGIN_UPPER_LEFT:
221    case PIPE_CAP_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
222    case PIPE_CAP_POINT_COORD_ORIGIN_UPPER_LEFT:
223    case PIPE_CAP_PRIMITIVE_RESTART:
224    case PIPE_CAP_PRIMITIVE_RESTART_FIXED_INDEX:
225    case PIPE_CAP_VS_INSTANCEID:
226    case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
227    case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
228    case PIPE_CAP_CONDITIONAL_RENDER:
229    case PIPE_CAP_TEXTURE_BARRIER:
230    case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
231    case PIPE_CAP_START_INSTANCE:
232    case PIPE_CAP_USER_VERTEX_BUFFERS:
233    case PIPE_CAP_TEXTURE_MULTISAMPLE:
234    case PIPE_CAP_FS_FINE_DERIVATIVE:
235    case PIPE_CAP_SAMPLER_VIEW_TARGET:
236    case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
237    case PIPE_CAP_CLIP_HALFZ:
238    case PIPE_CAP_POLYGON_OFFSET_CLAMP:
239    case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
240    case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
241    case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
242    case PIPE_CAP_DEPTH_BOUNDS_TEST:
243    case PIPE_CAP_TEXTURE_QUERY_SAMPLES:
244    case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
245    case PIPE_CAP_CLEAR_TEXTURE:
246    case PIPE_CAP_FS_FACE_IS_INTEGER_SYSVAL:
247    case PIPE_CAP_INVALIDATE_BUFFER:
248    case PIPE_CAP_STRING_MARKER:
249    case PIPE_CAP_CULL_DISTANCE:
250    case PIPE_CAP_SHADER_ARRAY_COMPONENTS:
251    case PIPE_CAP_LEGACY_MATH_RULES:
252    case PIPE_CAP_TGSI_TEX_TXF_LZ:
253    case PIPE_CAP_SHADER_CLOCK:
254    case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
255    case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
256    case PIPE_CAP_DEST_SURFACE_SRGB_CONTROL:
257    case PIPE_CAP_TGSI_DIV:
258    case PIPE_CAP_PREFER_IMM_ARRAYS_AS_CONSTBUF:
259    case PIPE_CAP_FLATSHADE:
260    case PIPE_CAP_POINT_SIZE_FIXED:
261    case PIPE_CAP_TWO_SIDED_COLOR:
262    case PIPE_CAP_CLIP_PLANES:
263    case PIPE_CAP_PACKED_STREAM_OUTPUT:
264    case PIPE_CAP_CLEAR_SCISSORED:
265    case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
266    case PIPE_CAP_COMPUTE:
267    case PIPE_CAP_GL_CLAMP:
268    case PIPE_CAP_TEXRECT:
269    case PIPE_CAP_ALLOW_DYNAMIC_VAO_FASTPATH:
270    case PIPE_CAP_SHAREABLE_SHADERS:
271    case PIPE_CAP_PREFER_BACK_BUFFER_REUSE:
272       return 1;
273 
274    case PIPE_CAP_ALPHA_TEST:
275       /* nvc0 has fixed function alpha test support, but nv50 doesn't.  The TGSI
276        * backend emits the conditional discard code against a driver-uploaded
277        * uniform, but with NIR we can have the st emit it for us.
278        */
279       return class_3d >= NVC0_3D_CLASS || !screen->prefer_nir;
280 
281    case PIPE_CAP_TEXTURE_TRANSFER_MODES:
282       return PIPE_TEXTURE_TRANSFER_BLIT;
283    case PIPE_CAP_SEAMLESS_CUBE_MAP:
284       return 1; /* class_3d >= NVA0_3D_CLASS; */
285    /* supported on nva0+ */
286    case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
287       return class_3d >= NVA0_3D_CLASS;
288    /* supported on nva3+ */
289    case PIPE_CAP_CUBE_MAP_ARRAY:
290    case PIPE_CAP_INDEP_BLEND_FUNC:
291    case PIPE_CAP_TEXTURE_QUERY_LOD:
292    case PIPE_CAP_SAMPLE_SHADING:
293    case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
294       return class_3d >= NVA3_3D_CLASS;
295 
296    /* unsupported caps */
297    case PIPE_CAP_EMULATE_NONFIXED_PRIMITIVE_RESTART:
298    case PIPE_CAP_DEPTH_CLIP_DISABLE_SEPARATE:
299    case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
300    case PIPE_CAP_FS_COORD_ORIGIN_LOWER_LEFT:
301    case PIPE_CAP_FS_COORD_PIXEL_CENTER_INTEGER:
302    case PIPE_CAP_SHADER_STENCIL_EXPORT:
303    case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
304    case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
305    case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
306    case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
307    case PIPE_CAP_VERTEX_ATTRIB_ELEMENT_ALIGNED_ONLY:
308    case PIPE_CAP_TGSI_TEXCOORD:
309    case PIPE_CAP_VS_LAYER_VIEWPORT:
310    case PIPE_CAP_TEXTURE_GATHER_SM5:
311    case PIPE_CAP_FAKE_SW_MSAA:
312    case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
313    case PIPE_CAP_VS_WINDOW_SPACE_POSITION:
314    case PIPE_CAP_DRAW_INDIRECT:
315    case PIPE_CAP_MULTI_DRAW_INDIRECT:
316    case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
317    case PIPE_CAP_VERTEXID_NOBASE:
318    case PIPE_CAP_MULTISAMPLE_Z_RESOLVE: /* potentially supported on some hw */
319    case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
320    case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
321    case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
322    case PIPE_CAP_DRAW_PARAMETERS:
323    case PIPE_CAP_SHADER_PACK_HALF_FLOAT:
324    case PIPE_CAP_FS_POSITION_IS_SYSVAL:
325    case PIPE_CAP_FS_POINT_IS_SYSVAL:
326    case PIPE_CAP_GENERATE_MIPMAP:
327    case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
328    case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
329    case PIPE_CAP_QUERY_BUFFER_OBJECT:
330    case PIPE_CAP_QUERY_MEMORY_INFO:
331    case PIPE_CAP_PCI_GROUP:
332    case PIPE_CAP_PCI_BUS:
333    case PIPE_CAP_PCI_DEVICE:
334    case PIPE_CAP_PCI_FUNCTION:
335    case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
336    case PIPE_CAP_SHADER_GROUP_VOTE:
337    case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
338    case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
339    case PIPE_CAP_SHADER_CAN_READ_OUTPUTS:
340    case PIPE_CAP_NATIVE_FENCE_FD:
341    case PIPE_CAP_FBFETCH:
342    case PIPE_CAP_DOUBLES:
343    case PIPE_CAP_INT64:
344    case PIPE_CAP_INT64_DIVMOD:
345    case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
346    case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
347    case PIPE_CAP_SHADER_BALLOT:
348    case PIPE_CAP_TES_LAYER_VIEWPORT:
349    case PIPE_CAP_POST_DEPTH_COVERAGE:
350    case PIPE_CAP_BINDLESS_TEXTURE:
351    case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
352    case PIPE_CAP_QUERY_SO_OVERFLOW:
353    case PIPE_CAP_MEMOBJ:
354    case PIPE_CAP_LOAD_CONSTBUF:
355    case PIPE_CAP_TILE_RASTER_ORDER:
356    case PIPE_CAP_FRAMEBUFFER_MSAA_CONSTRAINTS:
357    case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
358    case PIPE_CAP_CONTEXT_PRIORITY_MASK:
359    case PIPE_CAP_FENCE_SIGNAL:
360    case PIPE_CAP_CONSTBUF0_FLAGS:
361    case PIPE_CAP_PACKED_UNIFORMS:
362    case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_TRIANGLES:
363    case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_POINTS_LINES:
364    case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES:
365    case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_POINTS_LINES:
366    case PIPE_CAP_CONSERVATIVE_RASTER_POST_DEPTH_COVERAGE:
367    case PIPE_CAP_MAX_CONSERVATIVE_RASTER_SUBPIXEL_PRECISION_BIAS:
368    case PIPE_CAP_PROGRAMMABLE_SAMPLE_LOCATIONS:
369    case PIPE_CAP_MAX_COMBINED_SHADER_BUFFERS:
370    case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTERS:
371    case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTER_BUFFERS:
372    case PIPE_CAP_SURFACE_SAMPLE_COUNT:
373    case PIPE_CAP_IMAGE_ATOMIC_FLOAT_ADD:
374    case PIPE_CAP_QUERY_PIPELINE_STATISTICS_SINGLE:
375    case PIPE_CAP_RGB_OVERRIDE_DST_ALPHA_BLEND:
376    case PIPE_CAP_GLSL_TESS_LEVELS_AS_INPUTS:
377    case PIPE_CAP_NIR_COMPACT_ARRAYS:
378    case PIPE_CAP_IMAGE_LOAD_FORMATTED:
379    case PIPE_CAP_IMAGE_STORE_FORMATTED:
380    case PIPE_CAP_COMPUTE_SHADER_DERIVATIVES:
381    case PIPE_CAP_ATOMIC_FLOAT_MINMAX:
382    case PIPE_CAP_CONSERVATIVE_RASTER_INNER_COVERAGE:
383    case PIPE_CAP_FRAGMENT_SHADER_INTERLOCK:
384    case PIPE_CAP_CS_DERIVED_SYSTEM_VALUES_SUPPORTED:
385    case PIPE_CAP_FBFETCH_COHERENT:
386    case PIPE_CAP_IMAGE_ATOMIC_INC_WRAP:
387    case PIPE_CAP_DEMOTE_TO_HELPER_INVOCATION:
388    case PIPE_CAP_TGSI_TG4_COMPONENT_IN_SWIZZLE:
389    case PIPE_CAP_OPENCL_INTEGER_FUNCTIONS:
390    case PIPE_CAP_INTEGER_MULTIPLY_32X16: /* could be done */
391    case PIPE_CAP_FRONTEND_NOOP:
392    case PIPE_CAP_GL_SPIRV:
393    case PIPE_CAP_SHADER_SAMPLES_IDENTICAL:
394    case PIPE_CAP_TEXTURE_SHADOW_LOD:
395    case PIPE_CAP_VIEWPORT_TRANSFORM_LOWERED:
396    case PIPE_CAP_PSIZ_CLAMPED:
397    case PIPE_CAP_VIEWPORT_SWIZZLE:
398    case PIPE_CAP_VIEWPORT_MASK:
399    case PIPE_CAP_TEXTURE_BUFFER_SAMPLER:
400    case PIPE_CAP_PREFER_REAL_BUFFER_IN_CONSTBUF0:
401    case PIPE_CAP_MAP_UNSYNCHRONIZED_THREAD_SAFE: /* when we fix MT stuff */
402    case PIPE_CAP_ALPHA_TO_COVERAGE_DITHER_CONTROL:
403    case PIPE_CAP_SHADER_ATOMIC_INT64:
404    case PIPE_CAP_GLSL_ZERO_INIT:
405    case PIPE_CAP_BLEND_EQUATION_ADVANCED:
406    case PIPE_CAP_NO_CLIP_ON_COPY_TEX:
407    case PIPE_CAP_DEVICE_PROTECTED_CONTENT:
408    case PIPE_CAP_NIR_IMAGES_AS_DEREF:
409    case PIPE_CAP_SAMPLER_REDUCTION_MINMAX:
410    case PIPE_CAP_SAMPLER_REDUCTION_MINMAX_ARB:
411    case PIPE_CAP_DRAW_VERTEX_STATE:
412    case PIPE_CAP_PREFER_POT_ALIGNED_VARYINGS:
413    case PIPE_CAP_MAX_SPARSE_TEXTURE_SIZE:
414    case PIPE_CAP_MAX_SPARSE_3D_TEXTURE_SIZE:
415    case PIPE_CAP_MAX_SPARSE_ARRAY_TEXTURE_LAYERS:
416    case PIPE_CAP_SPARSE_TEXTURE_FULL_ARRAY_CUBE_MIPMAPS:
417    case PIPE_CAP_QUERY_SPARSE_TEXTURE_RESIDENCY:
418    case PIPE_CAP_CLAMP_SPARSE_TEXTURE_LOD:
419    case PIPE_CAP_HARDWARE_GL_SELECT:
420       return 0;
421 
422    case PIPE_CAP_VENDOR_ID:
423       return 0x10de;
424    case PIPE_CAP_DEVICE_ID: {
425       uint64_t device_id;
426       if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
427          NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
428          return -1;
429       }
430       return device_id;
431    }
432    case PIPE_CAP_ACCELERATED:
433       return 1;
434    case PIPE_CAP_VIDEO_MEMORY:
435       return dev->vram_size >> 20;
436    case PIPE_CAP_UMA:
437       return 0;
438 
439    default:
440       if (!debug_cap_printed[param]) {
441          debug_printf("%s: unhandled cap %d\n", __func__, param);
442          debug_cap_printed[param] = true;
443       }
444       FALLTHROUGH;
445    /* caps where we want the default value */
446    case PIPE_CAP_DMABUF:
447    case PIPE_CAP_THROTTLE:
448       return u_pipe_screen_get_param_defaults(pscreen, param);
449    }
450 }
451 
452 static int
nv50_screen_get_shader_param(struct pipe_screen * pscreen,enum pipe_shader_type shader,enum pipe_shader_cap param)453 nv50_screen_get_shader_param(struct pipe_screen *pscreen,
454                              enum pipe_shader_type shader,
455                              enum pipe_shader_cap param)
456 {
457    const struct nouveau_screen *screen = nouveau_screen(pscreen);
458 
459    switch (shader) {
460    case PIPE_SHADER_VERTEX:
461    case PIPE_SHADER_GEOMETRY:
462    case PIPE_SHADER_FRAGMENT:
463    case PIPE_SHADER_COMPUTE:
464       break;
465    default:
466       return 0;
467    }
468 
469    switch (param) {
470    case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
471    case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
472    case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
473    case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
474       return 16384;
475    case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
476       return 4;
477    case PIPE_SHADER_CAP_MAX_INPUTS:
478       if (shader == PIPE_SHADER_VERTEX)
479          return 32;
480       return 15;
481    case PIPE_SHADER_CAP_MAX_OUTPUTS:
482       return 16;
483    case PIPE_SHADER_CAP_MAX_CONST_BUFFER0_SIZE:
484       return 65536;
485    case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
486       return NV50_MAX_PIPE_CONSTBUFS;
487    case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
488       return shader != PIPE_SHADER_FRAGMENT;
489    case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
490    case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
491    case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
492       return 1;
493    case PIPE_SHADER_CAP_MAX_TEMPS:
494       return nv50_screen(pscreen)->max_tls_space / ONE_TEMP_SIZE;
495    case PIPE_SHADER_CAP_CONT_SUPPORTED:
496       return 1;
497    case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
498       return 1;
499    case PIPE_SHADER_CAP_INT64_ATOMICS:
500    case PIPE_SHADER_CAP_FP16:
501    case PIPE_SHADER_CAP_FP16_DERIVATIVES:
502    case PIPE_SHADER_CAP_FP16_CONST_BUFFERS:
503    case PIPE_SHADER_CAP_INT16:
504    case PIPE_SHADER_CAP_GLSL_16BIT_CONSTS:
505    case PIPE_SHADER_CAP_SUBROUTINES:
506       return 0; /* please inline, or provide function declarations */
507    case PIPE_SHADER_CAP_INTEGERS:
508       return 1;
509    case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
510       /* The chip could handle more sampler views than samplers */
511    case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
512       return MIN2(16, PIPE_MAX_SAMPLERS);
513    case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
514       return shader == PIPE_SHADER_COMPUTE ? NV50_MAX_GLOBALS - 1 : 0;
515    case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
516       return shader == PIPE_SHADER_COMPUTE ? NV50_MAX_GLOBALS - 1 : 0;
517    case PIPE_SHADER_CAP_PREFERRED_IR:
518       return screen->prefer_nir ? PIPE_SHADER_IR_NIR : PIPE_SHADER_IR_TGSI;
519    case PIPE_SHADER_CAP_SUPPORTED_IRS:
520       return (1 << PIPE_SHADER_IR_TGSI) | (1 << PIPE_SHADER_IR_NIR);
521    case PIPE_SHADER_CAP_DROUND_SUPPORTED:
522    case PIPE_SHADER_CAP_DFRACEXP_DLDEXP_SUPPORTED:
523    case PIPE_SHADER_CAP_LDEXP_SUPPORTED:
524    case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
525    case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
526    case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
527       return 0;
528    default:
529       NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
530       return 0;
531    }
532 }
533 
534 static float
nv50_screen_get_paramf(struct pipe_screen * pscreen,enum pipe_capf param)535 nv50_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
536 {
537    switch (param) {
538    case PIPE_CAPF_MIN_LINE_WIDTH:
539    case PIPE_CAPF_MIN_LINE_WIDTH_AA:
540    case PIPE_CAPF_MIN_POINT_SIZE:
541    case PIPE_CAPF_MIN_POINT_SIZE_AA:
542       return 1;
543    case PIPE_CAPF_POINT_SIZE_GRANULARITY:
544    case PIPE_CAPF_LINE_WIDTH_GRANULARITY:
545       return 0.1;
546    case PIPE_CAPF_MAX_LINE_WIDTH:
547    case PIPE_CAPF_MAX_LINE_WIDTH_AA:
548       return 10.0f;
549    case PIPE_CAPF_MAX_POINT_SIZE:
550    case PIPE_CAPF_MAX_POINT_SIZE_AA:
551       return 64.0f;
552    case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
553       return 16.0f;
554    case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
555       return 15.0f;
556    case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE:
557    case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE:
558    case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY:
559       return 0.0f;
560    }
561 
562    NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
563    return 0.0f;
564 }
565 
566 static int
nv50_screen_get_compute_param(struct pipe_screen * pscreen,enum pipe_shader_ir ir_type,enum pipe_compute_cap param,void * data)567 nv50_screen_get_compute_param(struct pipe_screen *pscreen,
568                               enum pipe_shader_ir ir_type,
569                               enum pipe_compute_cap param, void *data)
570 {
571    struct nv50_screen *screen = nv50_screen(pscreen);
572 
573 #define RET(x) do {                  \
574    if (data)                         \
575       memcpy(data, x, sizeof(x));    \
576    return sizeof(x);                 \
577 } while (0)
578 
579    switch (param) {
580    case PIPE_COMPUTE_CAP_GRID_DIMENSION:
581       RET((uint64_t []) { 3 });
582    case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
583       RET(((uint64_t []) { 65535, 65535, 65535 }));
584    case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
585       RET(((uint64_t []) { 512, 512, 64 }));
586    case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
587       RET((uint64_t []) { 512 });
588    case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g0-15[] */
589       RET((uint64_t []) { 1ULL << 32 });
590    case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
591       RET((uint64_t []) { 16 << 10 });
592    case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
593       RET((uint64_t []) { 16 << 10 });
594    case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
595       RET((uint64_t []) { 4096 });
596    case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
597       RET((uint32_t []) { 32 });
598    case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
599       RET((uint64_t []) { 1ULL << 40 });
600    case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
601       RET((uint32_t []) { 0 });
602    case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
603       RET((uint32_t []) { screen->mp_count });
604    case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
605       RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
606    case PIPE_COMPUTE_CAP_ADDRESS_BITS:
607       RET((uint32_t []) { 32 });
608    case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
609       RET((uint64_t []) { 0 });
610    default:
611       return 0;
612    }
613 
614 #undef RET
615 }
616 
617 static void
nv50_screen_destroy(struct pipe_screen * pscreen)618 nv50_screen_destroy(struct pipe_screen *pscreen)
619 {
620    struct nv50_screen *screen = nv50_screen(pscreen);
621 
622    if (!nouveau_drm_screen_unref(&screen->base))
623       return;
624 
625    nouveau_fence_cleanup(&screen->base);
626 
627    if (screen->base.pushbuf)
628       screen->base.pushbuf->user_priv = NULL;
629 
630    if (screen->blitter)
631       nv50_blitter_destroy(screen);
632    if (screen->pm.prog) {
633       screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
634       nv50_program_destroy(NULL, screen->pm.prog);
635       FREE(screen->pm.prog);
636    }
637 
638    nouveau_bo_ref(NULL, &screen->code);
639    nouveau_bo_ref(NULL, &screen->tls_bo);
640    nouveau_bo_ref(NULL, &screen->stack_bo);
641    nouveau_bo_ref(NULL, &screen->txc);
642    nouveau_bo_ref(NULL, &screen->uniforms);
643    nouveau_bo_ref(NULL, &screen->fence.bo);
644 
645    nouveau_heap_destroy(&screen->vp_code_heap);
646    nouveau_heap_destroy(&screen->gp_code_heap);
647    nouveau_heap_destroy(&screen->fp_code_heap);
648 
649    FREE(screen->tic.entries);
650 
651    nouveau_object_del(&screen->tesla);
652    nouveau_object_del(&screen->eng2d);
653    nouveau_object_del(&screen->m2mf);
654    nouveau_object_del(&screen->compute);
655    nouveau_object_del(&screen->sync);
656 
657    nouveau_screen_fini(&screen->base);
658 
659    FREE(screen);
660 }
661 
662 static void
nv50_screen_fence_emit(struct pipe_screen * pscreen,u32 * sequence)663 nv50_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
664 {
665    struct nv50_screen *screen = nv50_screen(pscreen);
666    struct nouveau_pushbuf *push = screen->base.pushbuf;
667 
668    /* we need to do it after possible flush in MARK_RING */
669    *sequence = ++screen->base.fence.sequence;
670 
671    assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
672    PUSH_DATA (push, NV50_FIFO_PKHDR(NV50_3D(QUERY_ADDRESS_HIGH), 4));
673    PUSH_DATAh(push, screen->fence.bo->offset);
674    PUSH_DATA (push, screen->fence.bo->offset);
675    PUSH_DATA (push, *sequence);
676    PUSH_DATA (push, NV50_3D_QUERY_GET_MODE_WRITE_UNK0 |
677                     NV50_3D_QUERY_GET_UNK4 |
678                     NV50_3D_QUERY_GET_UNIT_CROP |
679                     NV50_3D_QUERY_GET_TYPE_QUERY |
680                     NV50_3D_QUERY_GET_QUERY_SELECT_ZERO |
681                     NV50_3D_QUERY_GET_SHORT);
682 }
683 
684 static u32
nv50_screen_fence_update(struct pipe_screen * pscreen)685 nv50_screen_fence_update(struct pipe_screen *pscreen)
686 {
687    return nv50_screen(pscreen)->fence.map[0];
688 }
689 
690 static void
nv50_screen_init_hwctx(struct nv50_screen * screen)691 nv50_screen_init_hwctx(struct nv50_screen *screen)
692 {
693    struct nouveau_pushbuf *push = screen->base.pushbuf;
694    struct nv04_fifo *fifo;
695    unsigned i;
696 
697    fifo = (struct nv04_fifo *)screen->base.channel->data;
698 
699    BEGIN_NV04(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
700    PUSH_DATA (push, screen->m2mf->handle);
701    BEGIN_NV04(push, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY), 3);
702    PUSH_DATA (push, screen->sync->handle);
703    PUSH_DATA (push, fifo->vram);
704    PUSH_DATA (push, fifo->vram);
705 
706    BEGIN_NV04(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
707    PUSH_DATA (push, screen->eng2d->handle);
708    BEGIN_NV04(push, NV50_2D(DMA_NOTIFY), 4);
709    PUSH_DATA (push, screen->sync->handle);
710    PUSH_DATA (push, fifo->vram);
711    PUSH_DATA (push, fifo->vram);
712    PUSH_DATA (push, fifo->vram);
713    BEGIN_NV04(push, NV50_2D(OPERATION), 1);
714    PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
715    BEGIN_NV04(push, NV50_2D(CLIP_ENABLE), 1);
716    PUSH_DATA (push, 0);
717    BEGIN_NV04(push, NV50_2D(COLOR_KEY_ENABLE), 1);
718    PUSH_DATA (push, 0);
719    BEGIN_NV04(push, NV50_2D(SET_PIXELS_FROM_MEMORY_SAFE_OVERLAP), 1);
720    PUSH_DATA (push, 1);
721    BEGIN_NV04(push, NV50_2D(COND_MODE), 1);
722    PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
723 
724    BEGIN_NV04(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
725    PUSH_DATA (push, screen->tesla->handle);
726 
727    BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
728    PUSH_DATA (push, NV50_3D_COND_MODE_ALWAYS);
729 
730    BEGIN_NV04(push, NV50_3D(DMA_NOTIFY), 1);
731    PUSH_DATA (push, screen->sync->handle);
732    BEGIN_NV04(push, NV50_3D(DMA_ZETA), 11);
733    for (i = 0; i < 11; ++i)
734       PUSH_DATA(push, fifo->vram);
735    BEGIN_NV04(push, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN);
736    for (i = 0; i < NV50_3D_DMA_COLOR__LEN; ++i)
737       PUSH_DATA(push, fifo->vram);
738 
739    BEGIN_NV04(push, NV50_3D(REG_MODE), 1);
740    PUSH_DATA (push, NV50_3D_REG_MODE_STRIPED);
741    BEGIN_NV04(push, NV50_3D(UNK1400_LANES), 1);
742    PUSH_DATA (push, 0xf);
743 
744    if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
745       BEGIN_NV04(push, NV50_3D(WATCHDOG_TIMER), 1);
746       PUSH_DATA (push, 0x18);
747    }
748 
749    BEGIN_NV04(push, NV50_3D(ZETA_COMP_ENABLE), 1);
750    PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
751 
752    BEGIN_NV04(push, NV50_3D(RT_COMP_ENABLE(0)), 8);
753    for (i = 0; i < 8; ++i)
754       PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
755 
756    BEGIN_NV04(push, NV50_3D(RT_CONTROL), 1);
757    PUSH_DATA (push, 1);
758 
759    BEGIN_NV04(push, NV50_3D(CSAA_ENABLE), 1);
760    PUSH_DATA (push, 0);
761    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_ENABLE), 1);
762    PUSH_DATA (push, 0);
763    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_MODE), 1);
764    PUSH_DATA (push, NV50_3D_MULTISAMPLE_MODE_MS1);
765    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_CTRL), 1);
766    PUSH_DATA (push, 0);
767    BEGIN_NV04(push, NV50_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
768    PUSH_DATA (push, 1);
769    BEGIN_NV04(push, NV50_3D(BLEND_SEPARATE_ALPHA), 1);
770    PUSH_DATA (push, 1);
771 
772    if (screen->tesla->oclass >= NVA0_3D_CLASS) {
773       BEGIN_NV04(push, SUBC_3D(NVA0_3D_TEX_MISC), 1);
774       PUSH_DATA (push, 0);
775    }
776 
777    BEGIN_NV04(push, NV50_3D(SCREEN_Y_CONTROL), 1);
778    PUSH_DATA (push, 0);
779    BEGIN_NV04(push, NV50_3D(WINDOW_OFFSET_X), 2);
780    PUSH_DATA (push, 0);
781    PUSH_DATA (push, 0);
782    BEGIN_NV04(push, NV50_3D(ZCULL_REGION), 1);
783    PUSH_DATA (push, 0x3f);
784 
785    BEGIN_NV04(push, NV50_3D(VP_ADDRESS_HIGH), 2);
786    PUSH_DATAh(push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
787    PUSH_DATA (push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
788 
789    BEGIN_NV04(push, NV50_3D(FP_ADDRESS_HIGH), 2);
790    PUSH_DATAh(push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
791    PUSH_DATA (push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
792 
793    BEGIN_NV04(push, NV50_3D(GP_ADDRESS_HIGH), 2);
794    PUSH_DATAh(push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
795    PUSH_DATA (push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
796 
797    BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
798    PUSH_DATAh(push, screen->tls_bo->offset);
799    PUSH_DATA (push, screen->tls_bo->offset);
800    PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
801 
802    BEGIN_NV04(push, NV50_3D(STACK_ADDRESS_HIGH), 3);
803    PUSH_DATAh(push, screen->stack_bo->offset);
804    PUSH_DATA (push, screen->stack_bo->offset);
805    PUSH_DATA (push, 4);
806 
807    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
808    PUSH_DATAh(push, screen->uniforms->offset + (0 << 16));
809    PUSH_DATA (push, screen->uniforms->offset + (0 << 16));
810    PUSH_DATA (push, (NV50_CB_PVP << 16) | 0x0000);
811 
812    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
813    PUSH_DATAh(push, screen->uniforms->offset + (1 << 16));
814    PUSH_DATA (push, screen->uniforms->offset + (1 << 16));
815    PUSH_DATA (push, (NV50_CB_PGP << 16) | 0x0000);
816 
817    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
818    PUSH_DATAh(push, screen->uniforms->offset + (2 << 16));
819    PUSH_DATA (push, screen->uniforms->offset + (2 << 16));
820    PUSH_DATA (push, (NV50_CB_PFP << 16) | 0x0000);
821 
822    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
823    PUSH_DATAh(push, screen->uniforms->offset + (4 << 16));
824    PUSH_DATA (push, screen->uniforms->offset + (4 << 16));
825    PUSH_DATA (push, (NV50_CB_AUX << 16) | (NV50_CB_AUX_SIZE & 0xffff));
826 
827    BEGIN_NI04(push, NV50_3D(SET_PROGRAM_CB), 3);
828    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf01);
829    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf21);
830    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf31);
831 
832    /* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
833    BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
834    PUSH_DATA (push, (NV50_CB_AUX_RUNOUT_OFFSET << (8 - 2)) | NV50_CB_AUX);
835    BEGIN_NI04(push, NV50_3D(CB_DATA(0)), 4);
836    PUSH_DATAf(push, 0.0f);
837    PUSH_DATAf(push, 0.0f);
838    PUSH_DATAf(push, 0.0f);
839    PUSH_DATAf(push, 0.0f);
840    BEGIN_NV04(push, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
841    PUSH_DATAh(push, screen->uniforms->offset + (4 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
842    PUSH_DATA (push, screen->uniforms->offset + (4 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
843 
844    /* set the membar offset */
845    BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
846    PUSH_DATA (push, (NV50_CB_AUX_MEMBAR_OFFSET << (8 - 2)) | NV50_CB_AUX);
847    BEGIN_NI04(push, NV50_3D(CB_DATA(0)), 1);
848    PUSH_DATA (push, screen->uniforms->offset + (4 << 16) + NV50_CB_AUX_MEMBAR_OFFSET);
849 
850    nv50_upload_ms_info(push);
851 
852    /* max TIC (bits 4:8) & TSC bindings, per program type */
853    for (i = 0; i < NV50_MAX_3D_SHADER_STAGES; ++i) {
854       BEGIN_NV04(push, NV50_3D(TEX_LIMITS(i)), 1);
855       PUSH_DATA (push, 0x54);
856    }
857 
858    BEGIN_NV04(push, NV50_3D(TIC_ADDRESS_HIGH), 3);
859    PUSH_DATAh(push, screen->txc->offset);
860    PUSH_DATA (push, screen->txc->offset);
861    PUSH_DATA (push, NV50_TIC_MAX_ENTRIES - 1);
862 
863    BEGIN_NV04(push, NV50_3D(TSC_ADDRESS_HIGH), 3);
864    PUSH_DATAh(push, screen->txc->offset + 65536);
865    PUSH_DATA (push, screen->txc->offset + 65536);
866    PUSH_DATA (push, NV50_TSC_MAX_ENTRIES - 1);
867 
868    BEGIN_NV04(push, NV50_3D(LINKED_TSC), 1);
869    PUSH_DATA (push, 0);
870 
871    BEGIN_NV04(push, NV50_3D(CLIP_RECTS_EN), 1);
872    PUSH_DATA (push, 0);
873    BEGIN_NV04(push, NV50_3D(CLIP_RECTS_MODE), 1);
874    PUSH_DATA (push, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY);
875    BEGIN_NV04(push, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
876    for (i = 0; i < 8 * 2; ++i)
877       PUSH_DATA(push, 0);
878    BEGIN_NV04(push, NV50_3D(CLIPID_ENABLE), 1);
879    PUSH_DATA (push, 0);
880 
881    BEGIN_NV04(push, NV50_3D(VIEWPORT_TRANSFORM_EN), 1);
882    PUSH_DATA (push, 1);
883    for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
884       BEGIN_NV04(push, NV50_3D(DEPTH_RANGE_NEAR(i)), 2);
885       PUSH_DATAf(push, 0.0f);
886       PUSH_DATAf(push, 1.0f);
887       BEGIN_NV04(push, NV50_3D(VIEWPORT_HORIZ(i)), 2);
888       PUSH_DATA (push, 8192 << 16);
889       PUSH_DATA (push, 8192 << 16);
890    }
891 
892    BEGIN_NV04(push, NV50_3D(VIEW_VOLUME_CLIP_CTRL), 1);
893 #ifdef NV50_SCISSORS_CLIPPING
894    PUSH_DATA (push, 0x0000);
895 #else
896    PUSH_DATA (push, 0x1080);
897 #endif
898 
899    BEGIN_NV04(push, NV50_3D(CLEAR_FLAGS), 1);
900    PUSH_DATA (push, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT);
901 
902    /* We use scissors instead of exact view volume clipping,
903     * so they're always enabled.
904     */
905    for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
906       BEGIN_NV04(push, NV50_3D(SCISSOR_ENABLE(i)), 3);
907       PUSH_DATA (push, 1);
908       PUSH_DATA (push, 8192 << 16);
909       PUSH_DATA (push, 8192 << 16);
910    }
911 
912    BEGIN_NV04(push, NV50_3D(RASTERIZE_ENABLE), 1);
913    PUSH_DATA (push, 1);
914    BEGIN_NV04(push, NV50_3D(POINT_RASTER_RULES), 1);
915    PUSH_DATA (push, NV50_3D_POINT_RASTER_RULES_OGL);
916    BEGIN_NV04(push, NV50_3D(FRAG_COLOR_CLAMP_EN), 1);
917    PUSH_DATA (push, 0x11111111);
918    BEGIN_NV04(push, NV50_3D(EDGEFLAG), 1);
919    PUSH_DATA (push, 1);
920 
921    BEGIN_NV04(push, NV50_3D(VB_ELEMENT_BASE), 1);
922    PUSH_DATA (push, 0);
923    if (screen->base.class_3d >= NV84_3D_CLASS) {
924       BEGIN_NV04(push, NV84_3D(VERTEX_ID_BASE), 1);
925       PUSH_DATA (push, 0);
926    }
927 
928    BEGIN_NV04(push, NV50_3D(UNK0FDC), 1);
929    PUSH_DATA (push, 1);
930    BEGIN_NV04(push, NV50_3D(UNK19C0), 1);
931    PUSH_DATA (push, 1);
932 
933    PUSH_KICK (push);
934 }
935 
nv50_tls_alloc(struct nv50_screen * screen,unsigned tls_space,uint64_t * tls_size)936 static int nv50_tls_alloc(struct nv50_screen *screen, unsigned tls_space,
937       uint64_t *tls_size)
938 {
939    struct nouveau_device *dev = screen->base.device;
940    int ret;
941 
942    assert(tls_space % ONE_TEMP_SIZE == 0);
943    screen->cur_tls_space = util_next_power_of_two(tls_space / ONE_TEMP_SIZE) *
944          ONE_TEMP_SIZE;
945    if (nouveau_mesa_debug)
946       debug_printf("allocating space for %u temps\n",
947             util_next_power_of_two(tls_space / ONE_TEMP_SIZE));
948    *tls_size = screen->cur_tls_space * util_next_power_of_two(screen->TPs) *
949          screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP;
950 
951    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
952                         *tls_size, NULL, &screen->tls_bo);
953    if (ret) {
954       NOUVEAU_ERR("Failed to allocate local bo: %d\n", ret);
955       return ret;
956    }
957 
958    return 0;
959 }
960 
nv50_tls_realloc(struct nv50_screen * screen,unsigned tls_space)961 int nv50_tls_realloc(struct nv50_screen *screen, unsigned tls_space)
962 {
963    struct nouveau_pushbuf *push = screen->base.pushbuf;
964    int ret;
965    uint64_t tls_size;
966 
967    if (tls_space < screen->cur_tls_space)
968       return 0;
969    if (tls_space > screen->max_tls_space) {
970       /* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
971        * LOCAL_WARPS_NO_CLAMP) */
972       NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
973             (unsigned)(tls_space / ONE_TEMP_SIZE),
974             (unsigned)(screen->max_tls_space / ONE_TEMP_SIZE));
975       return -ENOMEM;
976    }
977 
978    nouveau_bo_ref(NULL, &screen->tls_bo);
979    ret = nv50_tls_alloc(screen, tls_space, &tls_size);
980    if (ret)
981       return ret;
982 
983    BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
984    PUSH_DATAh(push, screen->tls_bo->offset);
985    PUSH_DATA (push, screen->tls_bo->offset);
986    PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
987 
988    return 1;
989 }
990 
991 static const void *
nv50_screen_get_compiler_options(struct pipe_screen * pscreen,enum pipe_shader_ir ir,enum pipe_shader_type shader)992 nv50_screen_get_compiler_options(struct pipe_screen *pscreen,
993                                  enum pipe_shader_ir ir,
994                                  enum pipe_shader_type shader)
995 {
996    if (ir == PIPE_SHADER_IR_NIR)
997       return nv50_ir_nir_shader_compiler_options(NVISA_G80_CHIPSET, shader);
998    return NULL;
999 }
1000 
1001 struct nouveau_screen *
nv50_screen_create(struct nouveau_device * dev)1002 nv50_screen_create(struct nouveau_device *dev)
1003 {
1004    struct nv50_screen *screen;
1005    struct pipe_screen *pscreen;
1006    struct nouveau_object *chan;
1007    uint64_t value;
1008    uint32_t tesla_class;
1009    unsigned stack_size;
1010    int ret;
1011 
1012    screen = CALLOC_STRUCT(nv50_screen);
1013    if (!screen)
1014       return NULL;
1015    pscreen = &screen->base.base;
1016    pscreen->destroy = nv50_screen_destroy;
1017 
1018    ret = nouveau_screen_init(&screen->base, dev);
1019    if (ret) {
1020       NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret);
1021       goto fail;
1022    }
1023 
1024    /* TODO: Prevent FIFO prefetch before transfer of index buffers and
1025     *  admit them to VRAM.
1026     */
1027    screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
1028       PIPE_BIND_VERTEX_BUFFER;
1029    screen->base.sysmem_bindings |=
1030       PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
1031 
1032    screen->base.pushbuf->user_priv = screen;
1033    screen->base.pushbuf->rsvd_kick = 5;
1034 
1035    chan = screen->base.channel;
1036 
1037    pscreen->context_create = nv50_create;
1038    pscreen->is_format_supported = nv50_screen_is_format_supported;
1039    pscreen->get_param = nv50_screen_get_param;
1040    pscreen->get_shader_param = nv50_screen_get_shader_param;
1041    pscreen->get_paramf = nv50_screen_get_paramf;
1042    pscreen->get_compute_param = nv50_screen_get_compute_param;
1043    pscreen->get_driver_query_info = nv50_screen_get_driver_query_info;
1044    pscreen->get_driver_query_group_info = nv50_screen_get_driver_query_group_info;
1045 
1046    /* nir stuff */
1047    pscreen->get_compiler_options = nv50_screen_get_compiler_options;
1048 
1049    nv50_screen_init_resource_functions(pscreen);
1050 
1051    if (screen->base.device->chipset < 0x84 ||
1052        debug_get_bool_option("NOUVEAU_PMPEG", false)) {
1053       /* PMPEG */
1054       nouveau_screen_init_vdec(&screen->base);
1055    } else if (screen->base.device->chipset < 0x98 ||
1056               screen->base.device->chipset == 0xa0) {
1057       /* VP2 */
1058       screen->base.base.get_video_param = nv84_screen_get_video_param;
1059       screen->base.base.is_video_format_supported = nv84_screen_video_supported;
1060    } else {
1061       /* VP3/4 */
1062       screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
1063       screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
1064    }
1065 
1066    ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096,
1067                         NULL, &screen->fence.bo);
1068    if (ret) {
1069       NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret);
1070       goto fail;
1071    }
1072 
1073    nouveau_bo_map(screen->fence.bo, 0, NULL);
1074    screen->fence.map = screen->fence.bo->map;
1075    screen->base.fence.emit = nv50_screen_fence_emit;
1076    screen->base.fence.update = nv50_screen_fence_update;
1077 
1078    ret = nouveau_object_new(chan, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS,
1079                             &(struct nv04_notify){ .length = 32 },
1080                             sizeof(struct nv04_notify), &screen->sync);
1081    if (ret) {
1082       NOUVEAU_ERR("Failed to allocate notifier: %d\n", ret);
1083       goto fail;
1084    }
1085 
1086    ret = nouveau_object_new(chan, 0xbeef5039, NV50_M2MF_CLASS,
1087                             NULL, 0, &screen->m2mf);
1088    if (ret) {
1089       NOUVEAU_ERR("Failed to allocate PGRAPH context for M2MF: %d\n", ret);
1090       goto fail;
1091    }
1092 
1093    ret = nouveau_object_new(chan, 0xbeef502d, NV50_2D_CLASS,
1094                             NULL, 0, &screen->eng2d);
1095    if (ret) {
1096       NOUVEAU_ERR("Failed to allocate PGRAPH context for 2D: %d\n", ret);
1097       goto fail;
1098    }
1099 
1100    switch (dev->chipset & 0xf0) {
1101    case 0x50:
1102       tesla_class = NV50_3D_CLASS;
1103       break;
1104    case 0x80:
1105    case 0x90:
1106       tesla_class = NV84_3D_CLASS;
1107       break;
1108    case 0xa0:
1109       switch (dev->chipset) {
1110       case 0xa0:
1111       case 0xaa:
1112       case 0xac:
1113          tesla_class = NVA0_3D_CLASS;
1114          break;
1115       case 0xaf:
1116          tesla_class = NVAF_3D_CLASS;
1117          break;
1118       default:
1119          tesla_class = NVA3_3D_CLASS;
1120          break;
1121       }
1122       break;
1123    default:
1124       NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", dev->chipset);
1125       goto fail;
1126    }
1127    screen->base.class_3d = tesla_class;
1128 
1129    ret = nouveau_object_new(chan, 0xbeef5097, tesla_class,
1130                             NULL, 0, &screen->tesla);
1131    if (ret) {
1132       NOUVEAU_ERR("Failed to allocate PGRAPH context for 3D: %d\n", ret);
1133       goto fail;
1134    }
1135 
1136    /* This over-allocates by a page. The GP, which would execute at the end of
1137     * the last page, would trigger faults. The going theory is that it
1138     * prefetches up to a certain amount.
1139     */
1140    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
1141                         (3 << NV50_CODE_BO_SIZE_LOG2) + 0x1000,
1142                         NULL, &screen->code);
1143    if (ret) {
1144       NOUVEAU_ERR("Failed to allocate code bo: %d\n", ret);
1145       goto fail;
1146    }
1147 
1148    nouveau_heap_init(&screen->vp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
1149    nouveau_heap_init(&screen->gp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
1150    nouveau_heap_init(&screen->fp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
1151 
1152    nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
1153 
1154    screen->TPs = util_bitcount(value & 0xffff);
1155    screen->MPsInTP = util_bitcount(value & 0x0f000000);
1156 
1157    screen->mp_count = screen->TPs * screen->MPsInTP;
1158 
1159    stack_size = util_next_power_of_two(screen->TPs) * screen->MPsInTP *
1160          STACK_WARPS_ALLOC * 64 * 8;
1161 
1162    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, stack_size, NULL,
1163                         &screen->stack_bo);
1164    if (ret) {
1165       NOUVEAU_ERR("Failed to allocate stack bo: %d\n", ret);
1166       goto fail;
1167    }
1168 
1169    uint64_t size_of_one_temp = util_next_power_of_two(screen->TPs) *
1170          screen->MPsInTP * LOCAL_WARPS_ALLOC *  THREADS_IN_WARP *
1171          ONE_TEMP_SIZE;
1172    screen->max_tls_space = dev->vram_size / size_of_one_temp * ONE_TEMP_SIZE;
1173    screen->max_tls_space /= 2; /* half of vram */
1174 
1175    /* hw can address max 64 KiB */
1176    screen->max_tls_space = MIN2(screen->max_tls_space, 64 << 10);
1177 
1178    uint64_t tls_size;
1179    unsigned tls_space = 4/*temps*/ * ONE_TEMP_SIZE;
1180    ret = nv50_tls_alloc(screen, tls_space, &tls_size);
1181    if (ret)
1182       goto fail;
1183 
1184    if (nouveau_mesa_debug)
1185       debug_printf("TPs = %u, MPsInTP = %u, VRAM = %"PRIu64" MiB, tls_size = %"PRIu64" KiB\n",
1186             screen->TPs, screen->MPsInTP, dev->vram_size >> 20, tls_size >> 10);
1187 
1188    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 5 << 16, NULL,
1189                         &screen->uniforms);
1190    if (ret) {
1191       NOUVEAU_ERR("Failed to allocate uniforms bo: %d\n", ret);
1192       goto fail;
1193    }
1194 
1195    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 3 << 16, NULL,
1196                         &screen->txc);
1197    if (ret) {
1198       NOUVEAU_ERR("Failed to allocate TIC/TSC bo: %d\n", ret);
1199       goto fail;
1200    }
1201 
1202    screen->tic.entries = CALLOC(4096, sizeof(void *));
1203    screen->tsc.entries = screen->tic.entries + 2048;
1204 
1205    if (!nv50_blitter_create(screen))
1206       goto fail;
1207 
1208    nv50_screen_init_hwctx(screen);
1209 
1210    ret = nv50_screen_compute_setup(screen, screen->base.pushbuf);
1211    if (ret) {
1212       NOUVEAU_ERR("Failed to init compute context: %d\n", ret);
1213       goto fail;
1214    }
1215 
1216    nouveau_fence_new(&screen->base, &screen->base.fence.current);
1217 
1218    return &screen->base;
1219 
1220 fail:
1221    screen->base.base.context_create = NULL;
1222    return &screen->base;
1223 }
1224 
1225 int
nv50_screen_tic_alloc(struct nv50_screen * screen,void * entry)1226 nv50_screen_tic_alloc(struct nv50_screen *screen, void *entry)
1227 {
1228    int i = screen->tic.next;
1229 
1230    while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1231       i = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
1232 
1233    screen->tic.next = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
1234 
1235    if (screen->tic.entries[i])
1236       nv50_tic_entry(screen->tic.entries[i])->id = -1;
1237 
1238    screen->tic.entries[i] = entry;
1239    return i;
1240 }
1241 
1242 int
nv50_screen_tsc_alloc(struct nv50_screen * screen,void * entry)1243 nv50_screen_tsc_alloc(struct nv50_screen *screen, void *entry)
1244 {
1245    int i = screen->tsc.next;
1246 
1247    while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1248       i = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
1249 
1250    screen->tsc.next = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
1251 
1252    if (screen->tsc.entries[i])
1253       nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1254 
1255    screen->tsc.entries[i] = entry;
1256    return i;
1257 }
1258