• 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 #include "compiler/nir/nir.h"
31 
32 #include "nv50/nv50_context.h"
33 #include "nv50/nv50_screen.h"
34 
35 #include "nouveau_vp3_video.h"
36 
37 #include "nv_object.xml.h"
38 
39 /* affected by LOCAL_WARPS_LOG_ALLOC / LOCAL_WARPS_NO_CLAMP */
40 #define LOCAL_WARPS_ALLOC 32
41 /* affected by STACK_WARPS_LOG_ALLOC / STACK_WARPS_NO_CLAMP */
42 #define STACK_WARPS_ALLOC 32
43 
44 #define THREADS_IN_WARP 32
45 
46 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)47 nv50_screen_is_format_supported(struct pipe_screen *pscreen,
48                                 enum pipe_format format,
49                                 enum pipe_texture_target target,
50                                 unsigned sample_count,
51                                 unsigned storage_sample_count,
52                                 unsigned bindings)
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    if (sample_count == 8 && util_format_get_blocksizebits(format) >= 128)
59       return false;
60 
61    if (MAX2(1, sample_count) != MAX2(1, storage_sample_count))
62       return false;
63 
64    /* Short-circuit the rest of the logic -- this is used by the gallium frontend
65     * to determine valid MS levels in a no-attachments scenario.
66     */
67    if (format == PIPE_FORMAT_NONE && bindings & PIPE_BIND_RENDER_TARGET)
68       return true;
69 
70    switch (format) {
71    case PIPE_FORMAT_Z16_UNORM:
72       if (nv50_screen(pscreen)->tesla->oclass < NVA0_3D_CLASS)
73          return false;
74       break;
75    default:
76       break;
77    }
78 
79    if (bindings & PIPE_BIND_LINEAR)
80       if (util_format_is_depth_or_stencil(format) ||
81           (target != PIPE_TEXTURE_1D &&
82            target != PIPE_TEXTURE_2D &&
83            target != PIPE_TEXTURE_RECT) ||
84           sample_count > 1)
85          return false;
86 
87    /* shared is always supported */
88    bindings &= ~(PIPE_BIND_LINEAR |
89                  PIPE_BIND_SHARED);
90 
91    if (bindings & PIPE_BIND_INDEX_BUFFER) {
92       if (format != PIPE_FORMAT_R8_UINT &&
93           format != PIPE_FORMAT_R16_UINT &&
94           format != PIPE_FORMAT_R32_UINT)
95          return false;
96       bindings &= ~PIPE_BIND_INDEX_BUFFER;
97    }
98 
99    return (( nv50_format_table[format].usage |
100             nv50_vertex_format[format].usage) & bindings) == bindings;
101 }
102 
103 static int
nv50_screen_get_param(struct pipe_screen * pscreen,enum pipe_cap param)104 nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
105 {
106    const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
107    struct nouveau_device *dev = nouveau_screen(pscreen)->device;
108    static bool debug_cap_printed[PIPE_CAP_LAST] = {};
109 
110    switch (param) {
111    /* non-boolean caps */
112    case PIPE_CAP_MAX_TEXTURE_2D_SIZE:
113       return 8192;
114    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
115       return 12;
116    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
117       return 14;
118    case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
119       return 512;
120    case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
121    case PIPE_CAP_MIN_TEXEL_OFFSET:
122       return -8;
123    case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
124    case PIPE_CAP_MAX_TEXEL_OFFSET:
125       return 7;
126    case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
127       return 128 * 1024 * 1024;
128    case PIPE_CAP_GLSL_FEATURE_LEVEL:
129       return 330;
130    case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY:
131       return 330;
132    case PIPE_CAP_ESSL_FEATURE_LEVEL:
133       return class_3d >= NVA3_3D_CLASS ? 310 : 300;
134    case PIPE_CAP_MAX_RENDER_TARGETS:
135       return 8;
136    case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
137       return 1;
138    case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
139       return NV50_MAX_GLOBALS - 1;
140    case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
141    case PIPE_CAP_RASTERIZER_SUBPIXEL_BITS:
142       return 8;
143    case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
144       return 4;
145    case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
146       return 64;
147    case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
148       return 4;
149    case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
150    case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
151       return 1024;
152    case PIPE_CAP_MAX_VERTEX_STREAMS:
153       return 1;
154    case PIPE_CAP_MAX_GS_INVOCATIONS:
155       return 0;
156    case PIPE_CAP_MAX_SHADER_BUFFER_SIZE:
157       return 1 << 27;
158    case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
159       return 2048;
160    case PIPE_CAP_MAX_VERTEX_ELEMENT_SRC_OFFSET:
161       return 2047;
162    case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
163       return 256;
164    case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
165       return 16; /* 256 for binding as RT, but that's not possible in GL */
166    case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
167       return 256; /* the access limit is aligned to 256 */
168    case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
169       return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
170    case PIPE_CAP_MAX_VIEWPORTS:
171       return NV50_MAX_VIEWPORTS;
172    case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
173       return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
174    case PIPE_CAP_ENDIANNESS:
175       return PIPE_ENDIAN_LITTLE;
176    case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
177       return (class_3d >= NVA3_3D_CLASS) ? 4 : 0;
178    case PIPE_CAP_MAX_WINDOW_RECTANGLES:
179       return NV50_MAX_WINDOW_RECTANGLES;
180    case PIPE_CAP_MAX_TEXTURE_UPLOAD_MEMORY_BUDGET:
181       return 16 * 1024 * 1024;
182    case PIPE_CAP_MAX_VARYINGS:
183       return 15;
184    case PIPE_CAP_MAX_VERTEX_BUFFERS:
185       return 16;
186    case PIPE_CAP_GL_BEGIN_END_BUFFER_SIZE:
187       return 512 * 1024; /* TODO: Investigate tuning this */
188    case PIPE_CAP_MAX_TEXTURE_MB:
189       return 0; /* TODO: use 1/2 of VRAM for this? */
190 
191    case PIPE_CAP_SUPPORTED_PRIM_MODES_WITH_RESTART:
192    case PIPE_CAP_SUPPORTED_PRIM_MODES:
193       return BITFIELD_MASK(PIPE_PRIM_MAX);
194 
195    /* supported caps */
196    case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
197    case PIPE_CAP_TEXTURE_MIRROR_CLAMP_TO_EDGE:
198    case PIPE_CAP_TEXTURE_SWIZZLE:
199    case PIPE_CAP_TEXTURE_SHADOW_MAP:
200    case PIPE_CAP_NPOT_TEXTURES:
201    case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
202    case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
203    case PIPE_CAP_ANISOTROPIC_FILTER:
204    case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
205    case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
206    case PIPE_CAP_DEPTH_CLIP_DISABLE:
207    case PIPE_CAP_POINT_SPRITE:
208    case PIPE_CAP_FRAGMENT_SHADER_TEXTURE_LOD:
209    case PIPE_CAP_FRAGMENT_SHADER_DERIVATIVES:
210    case PIPE_CAP_VERTEX_SHADER_SATURATE:
211    case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
212    case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
213    case PIPE_CAP_VERTEX_COLOR_CLAMPED:
214    case PIPE_CAP_QUERY_TIMESTAMP:
215    case PIPE_CAP_QUERY_TIME_ELAPSED:
216    case PIPE_CAP_OCCLUSION_QUERY:
217    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
218    case PIPE_CAP_INDEP_BLEND_ENABLE:
219    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
220    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
221    case PIPE_CAP_PRIMITIVE_RESTART:
222    case PIPE_CAP_PRIMITIVE_RESTART_FIXED_INDEX:
223    case PIPE_CAP_TGSI_INSTANCEID:
224    case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
225    case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
226    case PIPE_CAP_CONDITIONAL_RENDER:
227    case PIPE_CAP_TEXTURE_BARRIER:
228    case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
229    case PIPE_CAP_START_INSTANCE:
230    case PIPE_CAP_USER_VERTEX_BUFFERS:
231    case PIPE_CAP_TEXTURE_MULTISAMPLE:
232    case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
233    case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
234    case PIPE_CAP_SAMPLER_VIEW_TARGET:
235    case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
236    case PIPE_CAP_CLIP_HALFZ:
237    case PIPE_CAP_POLYGON_OFFSET_CLAMP:
238    case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
239    case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
240    case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
241    case PIPE_CAP_DEPTH_BOUNDS_TEST:
242    case PIPE_CAP_TGSI_TXQS:
243    case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
244    case PIPE_CAP_CLEAR_TEXTURE:
245    case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
246    case PIPE_CAP_INVALIDATE_BUFFER:
247    case PIPE_CAP_STRING_MARKER:
248    case PIPE_CAP_CULL_DISTANCE:
249    case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
250    case PIPE_CAP_TGSI_MUL_ZERO_WINS:
251    case PIPE_CAP_TGSI_TEX_TXF_LZ:
252    case PIPE_CAP_TGSI_CLOCK:
253    case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
254    case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
255    case PIPE_CAP_DEST_SURFACE_SRGB_CONTROL:
256    case PIPE_CAP_TGSI_DIV:
257    case PIPE_CAP_PREFER_IMM_ARRAYS_AS_CONSTBUF:
258    case PIPE_CAP_FLATSHADE:
259    case PIPE_CAP_ALPHA_TEST:
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       return 1;
268    case PIPE_CAP_SEAMLESS_CUBE_MAP:
269       return 1; /* class_3d >= NVA0_3D_CLASS; */
270    /* supported on nva0+ */
271    case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
272       return class_3d >= NVA0_3D_CLASS;
273    /* supported on nva3+ */
274    case PIPE_CAP_CUBE_MAP_ARRAY:
275    case PIPE_CAP_INDEP_BLEND_FUNC:
276    case PIPE_CAP_TEXTURE_QUERY_LOD:
277    case PIPE_CAP_SAMPLE_SHADING:
278    case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
279       return class_3d >= NVA3_3D_CLASS;
280 
281    /* unsupported caps */
282    case PIPE_CAP_EMULATE_NONFIXED_PRIMITIVE_RESTART:
283    case PIPE_CAP_DEPTH_CLIP_DISABLE_SEPARATE:
284    case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
285    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
286    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
287    case PIPE_CAP_SHADER_STENCIL_EXPORT:
288    case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
289    case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
290    case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
291    case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
292    case PIPE_CAP_TGSI_TEXCOORD:
293    case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
294    case PIPE_CAP_TEXTURE_GATHER_SM5:
295    case PIPE_CAP_FAKE_SW_MSAA:
296    case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
297    case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
298    case PIPE_CAP_DRAW_INDIRECT:
299    case PIPE_CAP_MULTI_DRAW_INDIRECT:
300    case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
301    case PIPE_CAP_VERTEXID_NOBASE:
302    case PIPE_CAP_MULTISAMPLE_Z_RESOLVE: /* potentially supported on some hw */
303    case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
304    case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
305    case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
306    case PIPE_CAP_DRAW_PARAMETERS:
307    case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
308    case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
309    case PIPE_CAP_TGSI_FS_POINT_IS_SYSVAL:
310    case PIPE_CAP_GENERATE_MIPMAP:
311    case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
312    case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
313    case PIPE_CAP_QUERY_BUFFER_OBJECT:
314    case PIPE_CAP_QUERY_MEMORY_INFO:
315    case PIPE_CAP_PCI_GROUP:
316    case PIPE_CAP_PCI_BUS:
317    case PIPE_CAP_PCI_DEVICE:
318    case PIPE_CAP_PCI_FUNCTION:
319    case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
320    case PIPE_CAP_TGSI_VOTE:
321    case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
322    case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
323    case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
324    case PIPE_CAP_NATIVE_FENCE_FD:
325    case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
326    case PIPE_CAP_FBFETCH:
327    case PIPE_CAP_DOUBLES:
328    case PIPE_CAP_INT64:
329    case PIPE_CAP_INT64_DIVMOD:
330    case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
331    case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
332    case PIPE_CAP_TGSI_BALLOT:
333    case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
334    case PIPE_CAP_POST_DEPTH_COVERAGE:
335    case PIPE_CAP_BINDLESS_TEXTURE:
336    case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
337    case PIPE_CAP_QUERY_SO_OVERFLOW:
338    case PIPE_CAP_MEMOBJ:
339    case PIPE_CAP_LOAD_CONSTBUF:
340    case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS:
341    case PIPE_CAP_TILE_RASTER_ORDER:
342    case PIPE_CAP_FRAMEBUFFER_MSAA_CONSTRAINTS:
343    case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
344    case PIPE_CAP_CONTEXT_PRIORITY_MASK:
345    case PIPE_CAP_FENCE_SIGNAL:
346    case PIPE_CAP_CONSTBUF0_FLAGS:
347    case PIPE_CAP_PACKED_UNIFORMS:
348    case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_TRIANGLES:
349    case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_POINTS_LINES:
350    case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES:
351    case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_POINTS_LINES:
352    case PIPE_CAP_CONSERVATIVE_RASTER_POST_DEPTH_COVERAGE:
353    case PIPE_CAP_MAX_CONSERVATIVE_RASTER_SUBPIXEL_PRECISION_BIAS:
354    case PIPE_CAP_PROGRAMMABLE_SAMPLE_LOCATIONS:
355    case PIPE_CAP_MAX_COMBINED_SHADER_BUFFERS:
356    case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTERS:
357    case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTER_BUFFERS:
358    case PIPE_CAP_SURFACE_SAMPLE_COUNT:
359    case PIPE_CAP_TGSI_ATOMFADD:
360    case PIPE_CAP_QUERY_PIPELINE_STATISTICS_SINGLE:
361    case PIPE_CAP_RGB_OVERRIDE_DST_ALPHA_BLEND:
362    case PIPE_CAP_GLSL_TESS_LEVELS_AS_INPUTS:
363    case PIPE_CAP_NIR_COMPACT_ARRAYS:
364    case PIPE_CAP_IMAGE_LOAD_FORMATTED:
365    case PIPE_CAP_COMPUTE_SHADER_DERIVATIVES:
366    case PIPE_CAP_ATOMIC_FLOAT_MINMAX:
367    case PIPE_CAP_CONSERVATIVE_RASTER_INNER_COVERAGE:
368    case PIPE_CAP_FRAGMENT_SHADER_INTERLOCK:
369    case PIPE_CAP_CS_DERIVED_SYSTEM_VALUES_SUPPORTED:
370    case PIPE_CAP_FBFETCH_COHERENT:
371    case PIPE_CAP_TGSI_SKIP_SHRINK_IO_ARRAYS:
372    case PIPE_CAP_TGSI_ATOMINC_WRAP:
373    case PIPE_CAP_DEMOTE_TO_HELPER_INVOCATION:
374    case PIPE_CAP_TGSI_TG4_COMPONENT_IN_SWIZZLE:
375    case PIPE_CAP_OPENCL_INTEGER_FUNCTIONS:
376    case PIPE_CAP_INTEGER_MULTIPLY_32X16: /* could be done */
377    case PIPE_CAP_FRONTEND_NOOP:
378    case PIPE_CAP_GL_SPIRV:
379    case PIPE_CAP_SHADER_SAMPLES_IDENTICAL:
380    case PIPE_CAP_TEXTURE_SHADOW_LOD:
381    case PIPE_CAP_VIEWPORT_TRANSFORM_LOWERED:
382    case PIPE_CAP_PSIZ_CLAMPED:
383    case PIPE_CAP_VIEWPORT_SWIZZLE:
384    case PIPE_CAP_VIEWPORT_MASK:
385    case PIPE_CAP_TEXTURE_BUFFER_SAMPLER:
386    case PIPE_CAP_PREFER_REAL_BUFFER_IN_CONSTBUF0:
387    case PIPE_CAP_MAP_UNSYNCHRONIZED_THREAD_SAFE: /* when we fix MT stuff */
388    case PIPE_CAP_ALPHA_TO_COVERAGE_DITHER_CONTROL:
389    case PIPE_CAP_SHADER_ATOMIC_INT64:
390    case PIPE_CAP_GLSL_ZERO_INIT:
391    case PIPE_CAP_BLEND_EQUATION_ADVANCED:
392    case PIPE_CAP_NO_CLIP_ON_COPY_TEX:
393    case PIPE_CAP_DEVICE_PROTECTED_CONTENT:
394    case PIPE_CAP_NIR_IMAGES_AS_DEREF:
395       return 0;
396 
397    case PIPE_CAP_VENDOR_ID:
398       return 0x10de;
399    case PIPE_CAP_DEVICE_ID: {
400       uint64_t device_id;
401       if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
402          NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
403          return -1;
404       }
405       return device_id;
406    }
407    case PIPE_CAP_ACCELERATED:
408       return 1;
409    case PIPE_CAP_VIDEO_MEMORY:
410       return dev->vram_size >> 20;
411    case PIPE_CAP_UMA:
412       return 0;
413 
414    default:
415       if (!debug_cap_printed[param]) {
416          debug_printf("%s: unhandled cap %d\n", __func__, param);
417          debug_cap_printed[param] = true;
418       }
419       FALLTHROUGH;
420    /* caps where we want the default value */
421    case PIPE_CAP_DMABUF:
422    case PIPE_CAP_THROTTLE:
423       return u_pipe_screen_get_param_defaults(pscreen, param);
424    }
425 }
426 
427 static int
nv50_screen_get_shader_param(struct pipe_screen * pscreen,enum pipe_shader_type shader,enum pipe_shader_cap param)428 nv50_screen_get_shader_param(struct pipe_screen *pscreen,
429                              enum pipe_shader_type shader,
430                              enum pipe_shader_cap param)
431 {
432    const struct nouveau_screen *screen = nouveau_screen(pscreen);
433 
434    switch (shader) {
435    case PIPE_SHADER_VERTEX:
436    case PIPE_SHADER_GEOMETRY:
437    case PIPE_SHADER_FRAGMENT:
438    case PIPE_SHADER_COMPUTE:
439       break;
440    default:
441       return 0;
442    }
443 
444    switch (param) {
445    case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
446    case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
447    case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
448    case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
449       return 16384;
450    case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
451       return 4;
452    case PIPE_SHADER_CAP_MAX_INPUTS:
453       if (shader == PIPE_SHADER_VERTEX)
454          return 32;
455       return 15;
456    case PIPE_SHADER_CAP_MAX_OUTPUTS:
457       return 16;
458    case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
459       return 65536;
460    case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
461       return NV50_MAX_PIPE_CONSTBUFS;
462    case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
463       return shader != PIPE_SHADER_FRAGMENT;
464    case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
465    case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
466    case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
467       return 1;
468    case PIPE_SHADER_CAP_MAX_TEMPS:
469       return nv50_screen(pscreen)->max_tls_space / ONE_TEMP_SIZE;
470    case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
471       return 1;
472    case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
473       return 1;
474    case PIPE_SHADER_CAP_INT64_ATOMICS:
475    case PIPE_SHADER_CAP_FP16:
476    case PIPE_SHADER_CAP_FP16_DERIVATIVES:
477    case PIPE_SHADER_CAP_FP16_CONST_BUFFERS:
478    case PIPE_SHADER_CAP_INT16:
479    case PIPE_SHADER_CAP_GLSL_16BIT_CONSTS:
480    case PIPE_SHADER_CAP_SUBROUTINES:
481       return 0; /* please inline, or provide function declarations */
482    case PIPE_SHADER_CAP_INTEGERS:
483       return 1;
484    case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
485       return 1;
486    case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
487       /* The chip could handle more sampler views than samplers */
488    case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
489       return MIN2(16, PIPE_MAX_SAMPLERS);
490    case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
491       return shader == PIPE_SHADER_COMPUTE ? NV50_MAX_GLOBALS - 1 : 0;
492    case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
493       return shader == PIPE_SHADER_COMPUTE ? NV50_MAX_GLOBALS - 1 : 0;
494    case PIPE_SHADER_CAP_PREFERRED_IR:
495       return screen->prefer_nir ? PIPE_SHADER_IR_NIR : PIPE_SHADER_IR_TGSI;
496    case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
497       return 32;
498    case PIPE_SHADER_CAP_SUPPORTED_IRS:
499       return (1 << PIPE_SHADER_IR_TGSI) | (1 << PIPE_SHADER_IR_NIR);
500    case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
501    case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
502    case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
503    case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
504    case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
505    case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
506    case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
507    case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
508       return 0;
509    default:
510       NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
511       return 0;
512    }
513 }
514 
515 static float
nv50_screen_get_paramf(struct pipe_screen * pscreen,enum pipe_capf param)516 nv50_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
517 {
518    switch (param) {
519    case PIPE_CAPF_MAX_LINE_WIDTH:
520    case PIPE_CAPF_MAX_LINE_WIDTH_AA:
521       return 10.0f;
522    case PIPE_CAPF_MAX_POINT_WIDTH:
523    case PIPE_CAPF_MAX_POINT_WIDTH_AA:
524       return 64.0f;
525    case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
526       return 16.0f;
527    case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
528       return 15.0f;
529    case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE:
530    case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE:
531    case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY:
532       return 0.0f;
533    }
534 
535    NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
536    return 0.0f;
537 }
538 
539 static int
nv50_screen_get_compute_param(struct pipe_screen * pscreen,enum pipe_shader_ir ir_type,enum pipe_compute_cap param,void * data)540 nv50_screen_get_compute_param(struct pipe_screen *pscreen,
541                               enum pipe_shader_ir ir_type,
542                               enum pipe_compute_cap param, void *data)
543 {
544    struct nv50_screen *screen = nv50_screen(pscreen);
545 
546 #define RET(x) do {                  \
547    if (data)                         \
548       memcpy(data, x, sizeof(x));    \
549    return sizeof(x);                 \
550 } while (0)
551 
552    switch (param) {
553    case PIPE_COMPUTE_CAP_GRID_DIMENSION:
554       RET((uint64_t []) { 3 });
555    case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
556       RET(((uint64_t []) { 65535, 65535, 65535 }));
557    case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
558       RET(((uint64_t []) { 512, 512, 64 }));
559    case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
560       RET((uint64_t []) { 512 });
561    case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g0-15[] */
562       RET((uint64_t []) { 1ULL << 32 });
563    case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
564       RET((uint64_t []) { 16 << 10 });
565    case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
566       RET((uint64_t []) { 16 << 10 });
567    case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
568       RET((uint64_t []) { 4096 });
569    case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
570       RET((uint32_t []) { 32 });
571    case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
572       RET((uint64_t []) { 1ULL << 40 });
573    case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
574       RET((uint32_t []) { 0 });
575    case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
576       RET((uint32_t []) { screen->mp_count });
577    case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
578       RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */
579    case PIPE_COMPUTE_CAP_ADDRESS_BITS:
580       RET((uint32_t []) { 32 });
581    case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
582       RET((uint64_t []) { 0 });
583    default:
584       return 0;
585    }
586 
587 #undef RET
588 }
589 
590 static void
nv50_screen_destroy(struct pipe_screen * pscreen)591 nv50_screen_destroy(struct pipe_screen *pscreen)
592 {
593    struct nv50_screen *screen = nv50_screen(pscreen);
594 
595    if (!nouveau_drm_screen_unref(&screen->base))
596       return;
597 
598    nouveau_fence_cleanup(&screen->base);
599 
600    if (screen->base.pushbuf)
601       screen->base.pushbuf->user_priv = NULL;
602 
603    if (screen->blitter)
604       nv50_blitter_destroy(screen);
605    if (screen->pm.prog) {
606       screen->pm.prog->code = NULL; /* hardcoded, don't FREE */
607       nv50_program_destroy(NULL, screen->pm.prog);
608       FREE(screen->pm.prog);
609    }
610 
611    nouveau_bo_ref(NULL, &screen->code);
612    nouveau_bo_ref(NULL, &screen->tls_bo);
613    nouveau_bo_ref(NULL, &screen->stack_bo);
614    nouveau_bo_ref(NULL, &screen->txc);
615    nouveau_bo_ref(NULL, &screen->uniforms);
616    nouveau_bo_ref(NULL, &screen->fence.bo);
617 
618    nouveau_heap_destroy(&screen->vp_code_heap);
619    nouveau_heap_destroy(&screen->gp_code_heap);
620    nouveau_heap_destroy(&screen->fp_code_heap);
621 
622    FREE(screen->tic.entries);
623 
624    nouveau_object_del(&screen->tesla);
625    nouveau_object_del(&screen->eng2d);
626    nouveau_object_del(&screen->m2mf);
627    nouveau_object_del(&screen->compute);
628    nouveau_object_del(&screen->sync);
629 
630    nouveau_screen_fini(&screen->base);
631 
632    FREE(screen);
633 }
634 
635 static void
nv50_screen_fence_emit(struct pipe_screen * pscreen,u32 * sequence)636 nv50_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence)
637 {
638    struct nv50_screen *screen = nv50_screen(pscreen);
639    struct nouveau_pushbuf *push = screen->base.pushbuf;
640 
641    /* we need to do it after possible flush in MARK_RING */
642    *sequence = ++screen->base.fence.sequence;
643 
644    assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5);
645    PUSH_DATA (push, NV50_FIFO_PKHDR(NV50_3D(QUERY_ADDRESS_HIGH), 4));
646    PUSH_DATAh(push, screen->fence.bo->offset);
647    PUSH_DATA (push, screen->fence.bo->offset);
648    PUSH_DATA (push, *sequence);
649    PUSH_DATA (push, NV50_3D_QUERY_GET_MODE_WRITE_UNK0 |
650                     NV50_3D_QUERY_GET_UNK4 |
651                     NV50_3D_QUERY_GET_UNIT_CROP |
652                     NV50_3D_QUERY_GET_TYPE_QUERY |
653                     NV50_3D_QUERY_GET_QUERY_SELECT_ZERO |
654                     NV50_3D_QUERY_GET_SHORT);
655 }
656 
657 static u32
nv50_screen_fence_update(struct pipe_screen * pscreen)658 nv50_screen_fence_update(struct pipe_screen *pscreen)
659 {
660    return nv50_screen(pscreen)->fence.map[0];
661 }
662 
663 static void
nv50_screen_init_hwctx(struct nv50_screen * screen)664 nv50_screen_init_hwctx(struct nv50_screen *screen)
665 {
666    struct nouveau_pushbuf *push = screen->base.pushbuf;
667    struct nv04_fifo *fifo;
668    unsigned i;
669 
670    fifo = (struct nv04_fifo *)screen->base.channel->data;
671 
672    BEGIN_NV04(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
673    PUSH_DATA (push, screen->m2mf->handle);
674    BEGIN_NV04(push, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY), 3);
675    PUSH_DATA (push, screen->sync->handle);
676    PUSH_DATA (push, fifo->vram);
677    PUSH_DATA (push, fifo->vram);
678 
679    BEGIN_NV04(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
680    PUSH_DATA (push, screen->eng2d->handle);
681    BEGIN_NV04(push, NV50_2D(DMA_NOTIFY), 4);
682    PUSH_DATA (push, screen->sync->handle);
683    PUSH_DATA (push, fifo->vram);
684    PUSH_DATA (push, fifo->vram);
685    PUSH_DATA (push, fifo->vram);
686    BEGIN_NV04(push, NV50_2D(OPERATION), 1);
687    PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
688    BEGIN_NV04(push, NV50_2D(CLIP_ENABLE), 1);
689    PUSH_DATA (push, 0);
690    BEGIN_NV04(push, NV50_2D(COLOR_KEY_ENABLE), 1);
691    PUSH_DATA (push, 0);
692    BEGIN_NV04(push, NV50_2D(SET_PIXELS_FROM_MEMORY_SAFE_OVERLAP), 1);
693    PUSH_DATA (push, 1);
694    BEGIN_NV04(push, NV50_2D(COND_MODE), 1);
695    PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS);
696 
697    BEGIN_NV04(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
698    PUSH_DATA (push, screen->tesla->handle);
699 
700    BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
701    PUSH_DATA (push, NV50_3D_COND_MODE_ALWAYS);
702 
703    BEGIN_NV04(push, NV50_3D(DMA_NOTIFY), 1);
704    PUSH_DATA (push, screen->sync->handle);
705    BEGIN_NV04(push, NV50_3D(DMA_ZETA), 11);
706    for (i = 0; i < 11; ++i)
707       PUSH_DATA(push, fifo->vram);
708    BEGIN_NV04(push, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN);
709    for (i = 0; i < NV50_3D_DMA_COLOR__LEN; ++i)
710       PUSH_DATA(push, fifo->vram);
711 
712    BEGIN_NV04(push, NV50_3D(REG_MODE), 1);
713    PUSH_DATA (push, NV50_3D_REG_MODE_STRIPED);
714    BEGIN_NV04(push, NV50_3D(UNK1400_LANES), 1);
715    PUSH_DATA (push, 0xf);
716 
717    if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) {
718       BEGIN_NV04(push, NV50_3D(WATCHDOG_TIMER), 1);
719       PUSH_DATA (push, 0x18);
720    }
721 
722    BEGIN_NV04(push, NV50_3D(ZETA_COMP_ENABLE), 1);
723    PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
724 
725    BEGIN_NV04(push, NV50_3D(RT_COMP_ENABLE(0)), 8);
726    for (i = 0; i < 8; ++i)
727       PUSH_DATA(push, screen->base.drm->version >= 0x01000101);
728 
729    BEGIN_NV04(push, NV50_3D(RT_CONTROL), 1);
730    PUSH_DATA (push, 1);
731 
732    BEGIN_NV04(push, NV50_3D(CSAA_ENABLE), 1);
733    PUSH_DATA (push, 0);
734    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_ENABLE), 1);
735    PUSH_DATA (push, 0);
736    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_MODE), 1);
737    PUSH_DATA (push, NV50_3D_MULTISAMPLE_MODE_MS1);
738    BEGIN_NV04(push, NV50_3D(MULTISAMPLE_CTRL), 1);
739    PUSH_DATA (push, 0);
740    BEGIN_NV04(push, NV50_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1);
741    PUSH_DATA (push, 1);
742    BEGIN_NV04(push, NV50_3D(BLEND_SEPARATE_ALPHA), 1);
743    PUSH_DATA (push, 1);
744 
745    if (screen->tesla->oclass >= NVA0_3D_CLASS) {
746       BEGIN_NV04(push, SUBC_3D(NVA0_3D_TEX_MISC), 1);
747       PUSH_DATA (push, 0);
748    }
749 
750    BEGIN_NV04(push, NV50_3D(SCREEN_Y_CONTROL), 1);
751    PUSH_DATA (push, 0);
752    BEGIN_NV04(push, NV50_3D(WINDOW_OFFSET_X), 2);
753    PUSH_DATA (push, 0);
754    PUSH_DATA (push, 0);
755    BEGIN_NV04(push, NV50_3D(ZCULL_REGION), 1);
756    PUSH_DATA (push, 0x3f);
757 
758    BEGIN_NV04(push, NV50_3D(VP_ADDRESS_HIGH), 2);
759    PUSH_DATAh(push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
760    PUSH_DATA (push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
761 
762    BEGIN_NV04(push, NV50_3D(FP_ADDRESS_HIGH), 2);
763    PUSH_DATAh(push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
764    PUSH_DATA (push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
765 
766    BEGIN_NV04(push, NV50_3D(GP_ADDRESS_HIGH), 2);
767    PUSH_DATAh(push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
768    PUSH_DATA (push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
769 
770    BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
771    PUSH_DATAh(push, screen->tls_bo->offset);
772    PUSH_DATA (push, screen->tls_bo->offset);
773    PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
774 
775    BEGIN_NV04(push, NV50_3D(STACK_ADDRESS_HIGH), 3);
776    PUSH_DATAh(push, screen->stack_bo->offset);
777    PUSH_DATA (push, screen->stack_bo->offset);
778    PUSH_DATA (push, 4);
779 
780    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
781    PUSH_DATAh(push, screen->uniforms->offset + (0 << 16));
782    PUSH_DATA (push, screen->uniforms->offset + (0 << 16));
783    PUSH_DATA (push, (NV50_CB_PVP << 16) | 0x0000);
784 
785    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
786    PUSH_DATAh(push, screen->uniforms->offset + (1 << 16));
787    PUSH_DATA (push, screen->uniforms->offset + (1 << 16));
788    PUSH_DATA (push, (NV50_CB_PGP << 16) | 0x0000);
789 
790    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
791    PUSH_DATAh(push, screen->uniforms->offset + (2 << 16));
792    PUSH_DATA (push, screen->uniforms->offset + (2 << 16));
793    PUSH_DATA (push, (NV50_CB_PFP << 16) | 0x0000);
794 
795    BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
796    PUSH_DATAh(push, screen->uniforms->offset + (4 << 16));
797    PUSH_DATA (push, screen->uniforms->offset + (4 << 16));
798    PUSH_DATA (push, (NV50_CB_AUX << 16) | (NV50_CB_AUX_SIZE & 0xffff));
799 
800    BEGIN_NI04(push, NV50_3D(SET_PROGRAM_CB), 3);
801    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf01);
802    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf21);
803    PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf31);
804 
805    /* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
806    BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
807    PUSH_DATA (push, (NV50_CB_AUX_RUNOUT_OFFSET << (8 - 2)) | NV50_CB_AUX);
808    BEGIN_NI04(push, NV50_3D(CB_DATA(0)), 4);
809    PUSH_DATAf(push, 0.0f);
810    PUSH_DATAf(push, 0.0f);
811    PUSH_DATAf(push, 0.0f);
812    PUSH_DATAf(push, 0.0f);
813    BEGIN_NV04(push, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
814    PUSH_DATAh(push, screen->uniforms->offset + (4 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
815    PUSH_DATA (push, screen->uniforms->offset + (4 << 16) + NV50_CB_AUX_RUNOUT_OFFSET);
816 
817    /* set the membar offset */
818    BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
819    PUSH_DATA (push, (NV50_CB_AUX_MEMBAR_OFFSET << (8 - 2)) | NV50_CB_AUX);
820    BEGIN_NI04(push, NV50_3D(CB_DATA(0)), 1);
821    PUSH_DATA (push, screen->uniforms->offset + (4 << 16) + NV50_CB_AUX_MEMBAR_OFFSET);
822 
823    nv50_upload_ms_info(push);
824 
825    /* max TIC (bits 4:8) & TSC bindings, per program type */
826    for (i = 0; i < NV50_MAX_3D_SHADER_STAGES; ++i) {
827       BEGIN_NV04(push, NV50_3D(TEX_LIMITS(i)), 1);
828       PUSH_DATA (push, 0x54);
829    }
830 
831    BEGIN_NV04(push, NV50_3D(TIC_ADDRESS_HIGH), 3);
832    PUSH_DATAh(push, screen->txc->offset);
833    PUSH_DATA (push, screen->txc->offset);
834    PUSH_DATA (push, NV50_TIC_MAX_ENTRIES - 1);
835 
836    BEGIN_NV04(push, NV50_3D(TSC_ADDRESS_HIGH), 3);
837    PUSH_DATAh(push, screen->txc->offset + 65536);
838    PUSH_DATA (push, screen->txc->offset + 65536);
839    PUSH_DATA (push, NV50_TSC_MAX_ENTRIES - 1);
840 
841    BEGIN_NV04(push, NV50_3D(LINKED_TSC), 1);
842    PUSH_DATA (push, 0);
843 
844    BEGIN_NV04(push, NV50_3D(CLIP_RECTS_EN), 1);
845    PUSH_DATA (push, 0);
846    BEGIN_NV04(push, NV50_3D(CLIP_RECTS_MODE), 1);
847    PUSH_DATA (push, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY);
848    BEGIN_NV04(push, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
849    for (i = 0; i < 8 * 2; ++i)
850       PUSH_DATA(push, 0);
851    BEGIN_NV04(push, NV50_3D(CLIPID_ENABLE), 1);
852    PUSH_DATA (push, 0);
853 
854    BEGIN_NV04(push, NV50_3D(VIEWPORT_TRANSFORM_EN), 1);
855    PUSH_DATA (push, 1);
856    for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
857       BEGIN_NV04(push, NV50_3D(DEPTH_RANGE_NEAR(i)), 2);
858       PUSH_DATAf(push, 0.0f);
859       PUSH_DATAf(push, 1.0f);
860       BEGIN_NV04(push, NV50_3D(VIEWPORT_HORIZ(i)), 2);
861       PUSH_DATA (push, 8192 << 16);
862       PUSH_DATA (push, 8192 << 16);
863    }
864 
865    BEGIN_NV04(push, NV50_3D(VIEW_VOLUME_CLIP_CTRL), 1);
866 #ifdef NV50_SCISSORS_CLIPPING
867    PUSH_DATA (push, 0x0000);
868 #else
869    PUSH_DATA (push, 0x1080);
870 #endif
871 
872    BEGIN_NV04(push, NV50_3D(CLEAR_FLAGS), 1);
873    PUSH_DATA (push, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT);
874 
875    /* We use scissors instead of exact view volume clipping,
876     * so they're always enabled.
877     */
878    for (i = 0; i < NV50_MAX_VIEWPORTS; i++) {
879       BEGIN_NV04(push, NV50_3D(SCISSOR_ENABLE(i)), 3);
880       PUSH_DATA (push, 1);
881       PUSH_DATA (push, 8192 << 16);
882       PUSH_DATA (push, 8192 << 16);
883    }
884 
885    BEGIN_NV04(push, NV50_3D(RASTERIZE_ENABLE), 1);
886    PUSH_DATA (push, 1);
887    BEGIN_NV04(push, NV50_3D(POINT_RASTER_RULES), 1);
888    PUSH_DATA (push, NV50_3D_POINT_RASTER_RULES_OGL);
889    BEGIN_NV04(push, NV50_3D(FRAG_COLOR_CLAMP_EN), 1);
890    PUSH_DATA (push, 0x11111111);
891    BEGIN_NV04(push, NV50_3D(EDGEFLAG), 1);
892    PUSH_DATA (push, 1);
893 
894    BEGIN_NV04(push, NV50_3D(VB_ELEMENT_BASE), 1);
895    PUSH_DATA (push, 0);
896    if (screen->base.class_3d >= NV84_3D_CLASS) {
897       BEGIN_NV04(push, NV84_3D(VERTEX_ID_BASE), 1);
898       PUSH_DATA (push, 0);
899    }
900 
901    BEGIN_NV04(push, NV50_3D(UNK0FDC), 1);
902    PUSH_DATA (push, 1);
903    BEGIN_NV04(push, NV50_3D(UNK19C0), 1);
904    PUSH_DATA (push, 1);
905 
906    PUSH_KICK (push);
907 }
908 
nv50_tls_alloc(struct nv50_screen * screen,unsigned tls_space,uint64_t * tls_size)909 static int nv50_tls_alloc(struct nv50_screen *screen, unsigned tls_space,
910       uint64_t *tls_size)
911 {
912    struct nouveau_device *dev = screen->base.device;
913    int ret;
914 
915    screen->cur_tls_space = util_next_power_of_two(tls_space / ONE_TEMP_SIZE) *
916          ONE_TEMP_SIZE;
917    if (nouveau_mesa_debug)
918       debug_printf("allocating space for %u temps\n",
919             util_next_power_of_two(tls_space / ONE_TEMP_SIZE));
920    *tls_size = screen->cur_tls_space * util_next_power_of_two(screen->TPs) *
921          screen->MPsInTP * LOCAL_WARPS_ALLOC * THREADS_IN_WARP;
922 
923    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
924                         *tls_size, NULL, &screen->tls_bo);
925    if (ret) {
926       NOUVEAU_ERR("Failed to allocate local bo: %d\n", ret);
927       return ret;
928    }
929 
930    return 0;
931 }
932 
nv50_tls_realloc(struct nv50_screen * screen,unsigned tls_space)933 int nv50_tls_realloc(struct nv50_screen *screen, unsigned tls_space)
934 {
935    struct nouveau_pushbuf *push = screen->base.pushbuf;
936    int ret;
937    uint64_t tls_size;
938 
939    if (tls_space < screen->cur_tls_space)
940       return 0;
941    if (tls_space > screen->max_tls_space) {
942       /* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
943        * LOCAL_WARPS_NO_CLAMP) */
944       NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
945             (unsigned)(tls_space / ONE_TEMP_SIZE),
946             (unsigned)(screen->max_tls_space / ONE_TEMP_SIZE));
947       return -ENOMEM;
948    }
949 
950    nouveau_bo_ref(NULL, &screen->tls_bo);
951    ret = nv50_tls_alloc(screen, tls_space, &tls_size);
952    if (ret)
953       return ret;
954 
955    BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
956    PUSH_DATAh(push, screen->tls_bo->offset);
957    PUSH_DATA (push, screen->tls_bo->offset);
958    PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
959 
960    return 1;
961 }
962 
963 static const nir_shader_compiler_options nir_options = {
964    .fuse_ffma16 = false, /* nir doesn't track mad vs fma */
965    .fuse_ffma32 = false, /* nir doesn't track mad vs fma */
966    .fuse_ffma64 = false, /* nir doesn't track mad vs fma */
967    .lower_flrp32 = true,
968    .lower_flrp64 = true,
969    .lower_fpow = false,
970    .lower_uadd_carry = true,
971    .lower_usub_borrow = true,
972    .lower_ffract = true,
973    .lower_pack_half_2x16 = true,
974    .lower_pack_unorm_2x16 = true,
975    .lower_pack_snorm_2x16 = true,
976    .lower_pack_unorm_4x8 = true,
977    .lower_pack_snorm_4x8 = true,
978    .lower_unpack_half_2x16 = true,
979    .lower_unpack_unorm_2x16 = true,
980    .lower_unpack_snorm_2x16 = true,
981    .lower_unpack_unorm_4x8 = true,
982    .lower_unpack_snorm_4x8 = true,
983    .lower_extract_byte = true,
984    .lower_extract_word = true,
985    .lower_insert_byte = true,
986    .lower_insert_word = true,
987    .lower_all_io_to_temps = false,
988    .lower_cs_local_index_from_id = true,
989    .lower_rotate = true,
990    .lower_to_scalar = true,
991    .use_interpolated_input_intrinsics = true,
992    .max_unroll_iterations = 32,
993 };
994 
995 static const void *
nv50_screen_get_compiler_options(struct pipe_screen * pscreen,enum pipe_shader_ir ir,enum pipe_shader_type shader)996 nv50_screen_get_compiler_options(struct pipe_screen *pscreen,
997                                  enum pipe_shader_ir ir,
998                                  enum pipe_shader_type shader)
999 {
1000    if (ir == PIPE_SHADER_IR_NIR)
1001       return &nir_options;
1002    return NULL;
1003 }
1004 
1005 struct nouveau_screen *
nv50_screen_create(struct nouveau_device * dev)1006 nv50_screen_create(struct nouveau_device *dev)
1007 {
1008    struct nv50_screen *screen;
1009    struct pipe_screen *pscreen;
1010    struct nouveau_object *chan;
1011    uint64_t value;
1012    uint32_t tesla_class;
1013    unsigned stack_size;
1014    int ret;
1015 
1016    screen = CALLOC_STRUCT(nv50_screen);
1017    if (!screen)
1018       return NULL;
1019    pscreen = &screen->base.base;
1020    pscreen->destroy = nv50_screen_destroy;
1021 
1022    ret = nouveau_screen_init(&screen->base, dev);
1023    if (ret) {
1024       NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret);
1025       goto fail;
1026    }
1027 
1028    /* TODO: Prevent FIFO prefetch before transfer of index buffers and
1029     *  admit them to VRAM.
1030     */
1031    screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
1032       PIPE_BIND_VERTEX_BUFFER;
1033    screen->base.sysmem_bindings |=
1034       PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
1035 
1036    screen->base.pushbuf->user_priv = screen;
1037    screen->base.pushbuf->rsvd_kick = 5;
1038 
1039    chan = screen->base.channel;
1040 
1041    pscreen->context_create = nv50_create;
1042    pscreen->is_format_supported = nv50_screen_is_format_supported;
1043    pscreen->get_param = nv50_screen_get_param;
1044    pscreen->get_shader_param = nv50_screen_get_shader_param;
1045    pscreen->get_paramf = nv50_screen_get_paramf;
1046    pscreen->get_compute_param = nv50_screen_get_compute_param;
1047    pscreen->get_driver_query_info = nv50_screen_get_driver_query_info;
1048    pscreen->get_driver_query_group_info = nv50_screen_get_driver_query_group_info;
1049 
1050    /* nir stuff */
1051    pscreen->get_compiler_options = nv50_screen_get_compiler_options;
1052 
1053    nv50_screen_init_resource_functions(pscreen);
1054 
1055    if (screen->base.device->chipset < 0x84 ||
1056        debug_get_bool_option("NOUVEAU_PMPEG", false)) {
1057       /* PMPEG */
1058       nouveau_screen_init_vdec(&screen->base);
1059    } else if (screen->base.device->chipset < 0x98 ||
1060               screen->base.device->chipset == 0xa0) {
1061       /* VP2 */
1062       screen->base.base.get_video_param = nv84_screen_get_video_param;
1063       screen->base.base.is_video_format_supported = nv84_screen_video_supported;
1064    } else {
1065       /* VP3/4 */
1066       screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
1067       screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
1068    }
1069 
1070    ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096,
1071                         NULL, &screen->fence.bo);
1072    if (ret) {
1073       NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret);
1074       goto fail;
1075    }
1076 
1077    nouveau_bo_map(screen->fence.bo, 0, NULL);
1078    screen->fence.map = screen->fence.bo->map;
1079    screen->base.fence.emit = nv50_screen_fence_emit;
1080    screen->base.fence.update = nv50_screen_fence_update;
1081 
1082    ret = nouveau_object_new(chan, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS,
1083                             &(struct nv04_notify){ .length = 32 },
1084                             sizeof(struct nv04_notify), &screen->sync);
1085    if (ret) {
1086       NOUVEAU_ERR("Failed to allocate notifier: %d\n", ret);
1087       goto fail;
1088    }
1089 
1090    ret = nouveau_object_new(chan, 0xbeef5039, NV50_M2MF_CLASS,
1091                             NULL, 0, &screen->m2mf);
1092    if (ret) {
1093       NOUVEAU_ERR("Failed to allocate PGRAPH context for M2MF: %d\n", ret);
1094       goto fail;
1095    }
1096 
1097    ret = nouveau_object_new(chan, 0xbeef502d, NV50_2D_CLASS,
1098                             NULL, 0, &screen->eng2d);
1099    if (ret) {
1100       NOUVEAU_ERR("Failed to allocate PGRAPH context for 2D: %d\n", ret);
1101       goto fail;
1102    }
1103 
1104    switch (dev->chipset & 0xf0) {
1105    case 0x50:
1106       tesla_class = NV50_3D_CLASS;
1107       break;
1108    case 0x80:
1109    case 0x90:
1110       tesla_class = NV84_3D_CLASS;
1111       break;
1112    case 0xa0:
1113       switch (dev->chipset) {
1114       case 0xa0:
1115       case 0xaa:
1116       case 0xac:
1117          tesla_class = NVA0_3D_CLASS;
1118          break;
1119       case 0xaf:
1120          tesla_class = NVAF_3D_CLASS;
1121          break;
1122       default:
1123          tesla_class = NVA3_3D_CLASS;
1124          break;
1125       }
1126       break;
1127    default:
1128       NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", dev->chipset);
1129       goto fail;
1130    }
1131    screen->base.class_3d = tesla_class;
1132 
1133    ret = nouveau_object_new(chan, 0xbeef5097, tesla_class,
1134                             NULL, 0, &screen->tesla);
1135    if (ret) {
1136       NOUVEAU_ERR("Failed to allocate PGRAPH context for 3D: %d\n", ret);
1137       goto fail;
1138    }
1139 
1140    /* This over-allocates by a page. The GP, which would execute at the end of
1141     * the last page, would trigger faults. The going theory is that it
1142     * prefetches up to a certain amount.
1143     */
1144    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
1145                         (3 << NV50_CODE_BO_SIZE_LOG2) + 0x1000,
1146                         NULL, &screen->code);
1147    if (ret) {
1148       NOUVEAU_ERR("Failed to allocate code bo: %d\n", ret);
1149       goto fail;
1150    }
1151 
1152    nouveau_heap_init(&screen->vp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
1153    nouveau_heap_init(&screen->gp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
1154    nouveau_heap_init(&screen->fp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
1155 
1156    nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
1157 
1158    screen->TPs = util_bitcount(value & 0xffff);
1159    screen->MPsInTP = util_bitcount(value & 0x0f000000);
1160 
1161    screen->mp_count = screen->TPs * screen->MPsInTP;
1162 
1163    stack_size = util_next_power_of_two(screen->TPs) * screen->MPsInTP *
1164          STACK_WARPS_ALLOC * 64 * 8;
1165 
1166    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, stack_size, NULL,
1167                         &screen->stack_bo);
1168    if (ret) {
1169       NOUVEAU_ERR("Failed to allocate stack bo: %d\n", ret);
1170       goto fail;
1171    }
1172 
1173    uint64_t size_of_one_temp = util_next_power_of_two(screen->TPs) *
1174          screen->MPsInTP * LOCAL_WARPS_ALLOC *  THREADS_IN_WARP *
1175          ONE_TEMP_SIZE;
1176    screen->max_tls_space = dev->vram_size / size_of_one_temp * ONE_TEMP_SIZE;
1177    screen->max_tls_space /= 2; /* half of vram */
1178 
1179    /* hw can address max 64 KiB */
1180    screen->max_tls_space = MIN2(screen->max_tls_space, 64 << 10);
1181 
1182    uint64_t tls_size;
1183    unsigned tls_space = 4/*temps*/ * ONE_TEMP_SIZE;
1184    ret = nv50_tls_alloc(screen, tls_space, &tls_size);
1185    if (ret)
1186       goto fail;
1187 
1188    if (nouveau_mesa_debug)
1189       debug_printf("TPs = %u, MPsInTP = %u, VRAM = %"PRIu64" MiB, tls_size = %"PRIu64" KiB\n",
1190             screen->TPs, screen->MPsInTP, dev->vram_size >> 20, tls_size >> 10);
1191 
1192    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 5 << 16, NULL,
1193                         &screen->uniforms);
1194    if (ret) {
1195       NOUVEAU_ERR("Failed to allocate uniforms bo: %d\n", ret);
1196       goto fail;
1197    }
1198 
1199    ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 3 << 16, NULL,
1200                         &screen->txc);
1201    if (ret) {
1202       NOUVEAU_ERR("Failed to allocate TIC/TSC bo: %d\n", ret);
1203       goto fail;
1204    }
1205 
1206    screen->tic.entries = CALLOC(4096, sizeof(void *));
1207    screen->tsc.entries = screen->tic.entries + 2048;
1208 
1209    if (!nv50_blitter_create(screen))
1210       goto fail;
1211 
1212    nv50_screen_init_hwctx(screen);
1213 
1214    ret = nv50_screen_compute_setup(screen, screen->base.pushbuf);
1215    if (ret) {
1216       NOUVEAU_ERR("Failed to init compute context: %d\n", ret);
1217       goto fail;
1218    }
1219 
1220    nouveau_fence_new(&screen->base, &screen->base.fence.current);
1221 
1222    return &screen->base;
1223 
1224 fail:
1225    screen->base.base.context_create = NULL;
1226    return &screen->base;
1227 }
1228 
1229 int
nv50_screen_tic_alloc(struct nv50_screen * screen,void * entry)1230 nv50_screen_tic_alloc(struct nv50_screen *screen, void *entry)
1231 {
1232    int i = screen->tic.next;
1233 
1234    while (screen->tic.lock[i / 32] & (1 << (i % 32)))
1235       i = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
1236 
1237    screen->tic.next = (i + 1) & (NV50_TIC_MAX_ENTRIES - 1);
1238 
1239    if (screen->tic.entries[i])
1240       nv50_tic_entry(screen->tic.entries[i])->id = -1;
1241 
1242    screen->tic.entries[i] = entry;
1243    return i;
1244 }
1245 
1246 int
nv50_screen_tsc_alloc(struct nv50_screen * screen,void * entry)1247 nv50_screen_tsc_alloc(struct nv50_screen *screen, void *entry)
1248 {
1249    int i = screen->tsc.next;
1250 
1251    while (screen->tsc.lock[i / 32] & (1 << (i % 32)))
1252       i = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
1253 
1254    screen->tsc.next = (i + 1) & (NV50_TSC_MAX_ENTRIES - 1);
1255 
1256    if (screen->tsc.entries[i])
1257       nv50_tsc_entry(screen->tsc.entries[i])->id = -1;
1258 
1259    screen->tsc.entries[i] = entry;
1260    return i;
1261 }
1262