1 /* 2 * Copyright 2010 Christoph Bumiller 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 */ 22 23 #include <xf86drm.h> 24 #include <nouveau_drm.h> 25 #include <nvif/class.h> 26 #include "util/u_format.h" 27 #include "util/u_format_s3tc.h" 28 #include "pipe/p_screen.h" 29 30 #include "vl/vl_decoder.h" 31 #include "vl/vl_video_buffer.h" 32 33 #include "nouveau_vp3_video.h" 34 35 #include "nvc0/nvc0_context.h" 36 #include "nvc0/nvc0_screen.h" 37 38 #include "nvc0/mme/com9097.mme.h" 39 #include "nvc0/mme/com90c0.mme.h" 40 41 #include "nv50/g80_texture.xml.h" 42 43 static boolean 44 nvc0_screen_is_format_supported(struct pipe_screen *pscreen, 45 enum pipe_format format, 46 enum pipe_texture_target target, 47 unsigned sample_count, 48 unsigned bindings) 49 { 50 const struct util_format_description *desc = util_format_description(format); 51 52 if (sample_count > 8) 53 return false; 54 if (!(0x117 & (1 << sample_count))) /* 0, 1, 2, 4 or 8 */ 55 return false; 56 57 /* Short-circuit the rest of the logic -- this is used by the state tracker 58 * to determine valid MS levels in a no-attachments scenario. 59 */ 60 if (format == PIPE_FORMAT_NONE && bindings & PIPE_BIND_RENDER_TARGET) 61 return true; 62 63 if (!util_format_is_supported(format, bindings)) 64 return false; 65 66 if ((bindings & PIPE_BIND_SAMPLER_VIEW) && (target != PIPE_BUFFER)) 67 if (util_format_get_blocksizebits(format) == 3 * 32) 68 return false; 69 70 if (bindings & PIPE_BIND_LINEAR) 71 if (util_format_is_depth_or_stencil(format) || 72 (target != PIPE_TEXTURE_1D && 73 target != PIPE_TEXTURE_2D && 74 target != PIPE_TEXTURE_RECT) || 75 sample_count > 1) 76 return false; 77 78 /* Restrict ETC2 and ASTC formats here. These are only supported on GK20A. 79 */ 80 if ((desc->layout == UTIL_FORMAT_LAYOUT_ETC || 81 desc->layout == UTIL_FORMAT_LAYOUT_ASTC) && 82 /* The claim is that this should work on GM107 but it doesn't. Need to 83 * test further and figure out if it's a nouveau issue or a HW one. 84 nouveau_screen(pscreen)->class_3d < GM107_3D_CLASS && 85 */ 86 nouveau_screen(pscreen)->class_3d != NVEA_3D_CLASS) 87 return false; 88 89 /* shared is always supported */ 90 bindings &= ~(PIPE_BIND_LINEAR | 91 PIPE_BIND_SHARED); 92 93 if (bindings & PIPE_BIND_SHADER_IMAGE && sample_count > 1 && 94 nouveau_screen(pscreen)->class_3d >= GM107_3D_CLASS) { 95 /* MS images are currently unsupported on Maxwell because they have to 96 * be handled explicitly. */ 97 return false; 98 } 99 100 return (( nvc0_format_table[format].usage | 101 nvc0_vertex_format[format].usage) & bindings) == bindings; 102 } 103 104 static int 105 nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param) 106 { 107 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d; 108 struct nouveau_device *dev = nouveau_screen(pscreen)->device; 109 110 switch (param) { 111 /* non-boolean caps */ 112 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS: 113 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS: 114 return 15; 115 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS: 116 return 12; 117 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS: 118 return 2048; 119 case PIPE_CAP_MIN_TEXEL_OFFSET: 120 return -8; 121 case PIPE_CAP_MAX_TEXEL_OFFSET: 122 return 7; 123 case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET: 124 return -32; 125 case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET: 126 return 31; 127 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE: 128 return 128 * 1024 * 1024; 129 case PIPE_CAP_GLSL_FEATURE_LEVEL: 130 return 430; 131 case PIPE_CAP_MAX_RENDER_TARGETS: 132 return 8; 133 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS: 134 return 1; 135 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS: 136 return 4; 137 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS: 138 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS: 139 return 128; 140 case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES: 141 case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS: 142 return 1024; 143 case PIPE_CAP_MAX_VERTEX_STREAMS: 144 return 4; 145 case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE: 146 return 2048; 147 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT: 148 return 256; 149 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT: 150 if (class_3d < GM107_3D_CLASS) 151 return 256; /* IMAGE bindings require alignment to 256 */ 152 return 16; 153 case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT: 154 return 16; 155 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT: 156 return NOUVEAU_MIN_BUFFER_MAP_ALIGN; 157 case PIPE_CAP_MAX_VIEWPORTS: 158 return NVC0_MAX_VIEWPORTS; 159 case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS: 160 return 4; 161 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK: 162 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50; 163 case PIPE_CAP_ENDIANNESS: 164 return PIPE_ENDIAN_LITTLE; 165 case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS: 166 return 30; 167 case PIPE_CAP_MAX_WINDOW_RECTANGLES: 168 return NVC0_MAX_WINDOW_RECTANGLES; 169 170 /* supported caps */ 171 case PIPE_CAP_TEXTURE_MIRROR_CLAMP: 172 case PIPE_CAP_TEXTURE_SWIZZLE: 173 case PIPE_CAP_TEXTURE_SHADOW_MAP: 174 case PIPE_CAP_NPOT_TEXTURES: 175 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES: 176 case PIPE_CAP_MIXED_COLOR_DEPTH_BITS: 177 case PIPE_CAP_ANISOTROPIC_FILTER: 178 case PIPE_CAP_SEAMLESS_CUBE_MAP: 179 case PIPE_CAP_CUBE_MAP_ARRAY: 180 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS: 181 case PIPE_CAP_TEXTURE_MULTISAMPLE: 182 case PIPE_CAP_TWO_SIDED_STENCIL: 183 case PIPE_CAP_DEPTH_CLIP_DISABLE: 184 case PIPE_CAP_POINT_SPRITE: 185 case PIPE_CAP_TGSI_TEXCOORD: 186 case PIPE_CAP_SM3: 187 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED: 188 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED: 189 case PIPE_CAP_VERTEX_COLOR_CLAMPED: 190 case PIPE_CAP_QUERY_TIMESTAMP: 191 case PIPE_CAP_QUERY_TIME_ELAPSED: 192 case PIPE_CAP_OCCLUSION_QUERY: 193 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME: 194 case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS: 195 case PIPE_CAP_QUERY_PIPELINE_STATISTICS: 196 case PIPE_CAP_BLEND_EQUATION_SEPARATE: 197 case PIPE_CAP_INDEP_BLEND_ENABLE: 198 case PIPE_CAP_INDEP_BLEND_FUNC: 199 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT: 200 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER: 201 case PIPE_CAP_PRIMITIVE_RESTART: 202 case PIPE_CAP_TGSI_INSTANCEID: 203 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR: 204 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS: 205 case PIPE_CAP_CONDITIONAL_RENDER: 206 case PIPE_CAP_TEXTURE_BARRIER: 207 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION: 208 case PIPE_CAP_START_INSTANCE: 209 case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT: 210 case PIPE_CAP_DRAW_INDIRECT: 211 case PIPE_CAP_USER_CONSTANT_BUFFERS: 212 case PIPE_CAP_USER_INDEX_BUFFERS: 213 case PIPE_CAP_USER_VERTEX_BUFFERS: 214 case PIPE_CAP_TEXTURE_QUERY_LOD: 215 case PIPE_CAP_SAMPLE_SHADING: 216 case PIPE_CAP_TEXTURE_GATHER_OFFSETS: 217 case PIPE_CAP_TEXTURE_GATHER_SM5: 218 case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE: 219 case PIPE_CAP_CONDITIONAL_RENDER_INVERTED: 220 case PIPE_CAP_SAMPLER_VIEW_TARGET: 221 case PIPE_CAP_CLIP_HALFZ: 222 case PIPE_CAP_POLYGON_OFFSET_CLAMP: 223 case PIPE_CAP_MULTISAMPLE_Z_RESOLVE: 224 case PIPE_CAP_TEXTURE_FLOAT_LINEAR: 225 case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR: 226 case PIPE_CAP_DEPTH_BOUNDS_TEST: 227 case PIPE_CAP_TGSI_TXQS: 228 case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS: 229 case PIPE_CAP_FORCE_PERSAMPLE_INTERP: 230 case PIPE_CAP_SHAREABLE_SHADERS: 231 case PIPE_CAP_CLEAR_TEXTURE: 232 case PIPE_CAP_DRAW_PARAMETERS: 233 case PIPE_CAP_TGSI_PACK_HALF_FLOAT: 234 case PIPE_CAP_MULTI_DRAW_INDIRECT: 235 case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS: 236 case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL: 237 case PIPE_CAP_QUERY_BUFFER_OBJECT: 238 case PIPE_CAP_INVALIDATE_BUFFER: 239 case PIPE_CAP_STRING_MARKER: 240 case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT: 241 case PIPE_CAP_CULL_DISTANCE: 242 case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES: 243 case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR: 244 case PIPE_CAP_TGSI_VOTE: 245 case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED: 246 case PIPE_CAP_TGSI_ARRAY_COMPONENTS: 247 return 1; 248 case PIPE_CAP_COMPUTE: 249 return (class_3d < GP100_3D_CLASS); 250 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE: 251 return (class_3d >= NVE4_3D_CLASS) ? 1 : 0; 252 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER: 253 return nouveau_screen(pscreen)->vram_domain & NOUVEAU_BO_VRAM ? 1 : 0; 254 case PIPE_CAP_TGSI_FS_FBFETCH: 255 return class_3d >= NVE4_3D_CLASS; /* needs testing on fermi */ 256 257 /* unsupported caps */ 258 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT: 259 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER: 260 case PIPE_CAP_SHADER_STENCIL_EXPORT: 261 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS: 262 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY: 263 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY: 264 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY: 265 case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT: 266 case PIPE_CAP_FAKE_SW_MSAA: 267 case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION: 268 case PIPE_CAP_VERTEXID_NOBASE: 269 case PIPE_CAP_RESOURCE_FROM_USER_MEMORY: 270 case PIPE_CAP_DEVICE_RESET_STATUS_QUERY: 271 case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL: 272 case PIPE_CAP_GENERATE_MIPMAP: 273 case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY: 274 case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS: 275 case PIPE_CAP_QUERY_MEMORY_INFO: 276 case PIPE_CAP_PCI_GROUP: 277 case PIPE_CAP_PCI_BUS: 278 case PIPE_CAP_PCI_DEVICE: 279 case PIPE_CAP_PCI_FUNCTION: 280 case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS: 281 case PIPE_CAP_TGSI_CAN_READ_OUTPUTS: 282 case PIPE_CAP_NATIVE_FENCE_FD: 283 case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY: 284 return 0; 285 286 case PIPE_CAP_VENDOR_ID: 287 return 0x10de; 288 case PIPE_CAP_DEVICE_ID: { 289 uint64_t device_id; 290 if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) { 291 NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n"); 292 return -1; 293 } 294 return device_id; 295 } 296 case PIPE_CAP_ACCELERATED: 297 return 1; 298 case PIPE_CAP_VIDEO_MEMORY: 299 return dev->vram_size >> 20; 300 case PIPE_CAP_UMA: 301 return 0; 302 } 303 304 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param); 305 return 0; 306 } 307 308 static int 309 nvc0_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader, 310 enum pipe_shader_cap param) 311 { 312 const uint16_t class_3d = nouveau_screen(pscreen)->class_3d; 313 314 switch (shader) { 315 case PIPE_SHADER_VERTEX: 316 case PIPE_SHADER_GEOMETRY: 317 case PIPE_SHADER_FRAGMENT: 318 case PIPE_SHADER_COMPUTE: 319 case PIPE_SHADER_TESS_CTRL: 320 case PIPE_SHADER_TESS_EVAL: 321 break; 322 default: 323 return 0; 324 } 325 326 switch (param) { 327 case PIPE_SHADER_CAP_PREFERRED_IR: 328 return PIPE_SHADER_IR_TGSI; 329 case PIPE_SHADER_CAP_SUPPORTED_IRS: 330 return 1 << PIPE_SHADER_IR_TGSI; 331 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS: 332 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS: 333 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS: 334 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS: 335 return 16384; 336 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH: 337 return 16; 338 case PIPE_SHADER_CAP_MAX_INPUTS: 339 if (shader == PIPE_SHADER_VERTEX) 340 return 32; 341 /* NOTE: These only count our slots for GENERIC varyings. 342 * The address space may be larger, but the actual hard limit seems to be 343 * less than what the address space layout permits, so don't add TEXCOORD, 344 * COLOR, etc. here. 345 */ 346 if (shader == PIPE_SHADER_FRAGMENT) 347 return 0x1f0 / 16; 348 /* Actually this counts CLIPVERTEX, which occupies the last generic slot, 349 * and excludes 0x60 per-patch inputs. 350 */ 351 return 0x200 / 16; 352 case PIPE_SHADER_CAP_MAX_OUTPUTS: 353 return 32; 354 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE: 355 return 65536; 356 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS: 357 return NVC0_MAX_PIPE_CONSTBUFS; 358 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR: 359 return shader != PIPE_SHADER_FRAGMENT; 360 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR: 361 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR: 362 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR: 363 return 1; 364 case PIPE_SHADER_CAP_MAX_PREDS: 365 return 0; 366 case PIPE_SHADER_CAP_MAX_TEMPS: 367 return NVC0_CAP_MAX_PROGRAM_TEMPS; 368 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED: 369 return 1; 370 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED: 371 return 1; 372 case PIPE_SHADER_CAP_SUBROUTINES: 373 return 1; 374 case PIPE_SHADER_CAP_INTEGERS: 375 return 1; 376 case PIPE_SHADER_CAP_DOUBLES: 377 return 1; 378 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED: 379 return 1; 380 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED: 381 return 1; 382 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED: 383 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE: 384 case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD: 385 return 0; 386 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS: 387 return NVC0_MAX_BUFFERS; 388 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS: 389 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16; 390 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS: 391 return (class_3d >= NVE4_3D_CLASS) ? 32 : 16; 392 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT: 393 return 32; 394 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES: 395 if (class_3d >= NVE4_3D_CLASS) 396 return NVC0_MAX_IMAGES; 397 if (shader == PIPE_SHADER_FRAGMENT || shader == PIPE_SHADER_COMPUTE) 398 return NVC0_MAX_IMAGES; 399 return 0; 400 default: 401 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param); 402 return 0; 403 } 404 } 405 406 static float 407 nvc0_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param) 408 { 409 switch (param) { 410 case PIPE_CAPF_MAX_LINE_WIDTH: 411 case PIPE_CAPF_MAX_LINE_WIDTH_AA: 412 return 10.0f; 413 case PIPE_CAPF_MAX_POINT_WIDTH: 414 return 63.0f; 415 case PIPE_CAPF_MAX_POINT_WIDTH_AA: 416 return 63.375f; 417 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY: 418 return 16.0f; 419 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS: 420 return 15.0f; 421 case PIPE_CAPF_GUARD_BAND_LEFT: 422 case PIPE_CAPF_GUARD_BAND_TOP: 423 return 0.0f; 424 case PIPE_CAPF_GUARD_BAND_RIGHT: 425 case PIPE_CAPF_GUARD_BAND_BOTTOM: 426 return 0.0f; /* that or infinity */ 427 } 428 429 NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param); 430 return 0.0f; 431 } 432 433 static int 434 nvc0_screen_get_compute_param(struct pipe_screen *pscreen, 435 enum pipe_shader_ir ir_type, 436 enum pipe_compute_cap param, void *data) 437 { 438 struct nvc0_screen *screen = nvc0_screen(pscreen); 439 const uint16_t obj_class = screen->compute->oclass; 440 441 #define RET(x) do { \ 442 if (data) \ 443 memcpy(data, x, sizeof(x)); \ 444 return sizeof(x); \ 445 } while (0) 446 447 switch (param) { 448 case PIPE_COMPUTE_CAP_GRID_DIMENSION: 449 RET((uint64_t []) { 3 }); 450 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE: 451 if (obj_class >= NVE4_COMPUTE_CLASS) { 452 RET(((uint64_t []) { 0x7fffffff, 65535, 65535 })); 453 } else { 454 RET(((uint64_t []) { 65535, 65535, 65535 })); 455 } 456 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE: 457 RET(((uint64_t []) { 1024, 1024, 64 })); 458 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK: 459 RET((uint64_t []) { 1024 }); 460 case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK: 461 if (obj_class >= NVE4_COMPUTE_CLASS) { 462 RET((uint64_t []) { 1024 }); 463 } else { 464 RET((uint64_t []) { 512 }); 465 } 466 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */ 467 RET((uint64_t []) { 1ULL << 40 }); 468 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */ 469 switch (obj_class) { 470 case GM200_COMPUTE_CLASS: 471 RET((uint64_t []) { 96 << 10 }); 472 break; 473 case GM107_COMPUTE_CLASS: 474 RET((uint64_t []) { 64 << 10 }); 475 break; 476 default: 477 RET((uint64_t []) { 48 << 10 }); 478 break; 479 } 480 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */ 481 RET((uint64_t []) { 512 << 10 }); 482 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */ 483 RET((uint64_t []) { 4096 }); 484 case PIPE_COMPUTE_CAP_SUBGROUP_SIZE: 485 RET((uint32_t []) { 32 }); 486 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE: 487 RET((uint64_t []) { 1ULL << 40 }); 488 case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED: 489 RET((uint32_t []) { 0 }); 490 case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS: 491 RET((uint32_t []) { screen->mp_count_compute }); 492 case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY: 493 RET((uint32_t []) { 512 }); /* FIXME: arbitrary limit */ 494 case PIPE_COMPUTE_CAP_ADDRESS_BITS: 495 RET((uint32_t []) { 64 }); 496 default: 497 return 0; 498 } 499 500 #undef RET 501 } 502 503 static void 504 nvc0_screen_destroy(struct pipe_screen *pscreen) 505 { 506 struct nvc0_screen *screen = nvc0_screen(pscreen); 507 508 if (!nouveau_drm_screen_unref(&screen->base)) 509 return; 510 511 if (screen->base.fence.current) { 512 struct nouveau_fence *current = NULL; 513 514 /* nouveau_fence_wait will create a new current fence, so wait on the 515 * _current_ one, and remove both. 516 */ 517 nouveau_fence_ref(screen->base.fence.current, ¤t); 518 nouveau_fence_wait(current, NULL); 519 nouveau_fence_ref(NULL, ¤t); 520 nouveau_fence_ref(NULL, &screen->base.fence.current); 521 } 522 if (screen->base.pushbuf) 523 screen->base.pushbuf->user_priv = NULL; 524 525 if (screen->blitter) 526 nvc0_blitter_destroy(screen); 527 if (screen->pm.prog) { 528 screen->pm.prog->code = NULL; /* hardcoded, don't FREE */ 529 nvc0_program_destroy(NULL, screen->pm.prog); 530 FREE(screen->pm.prog); 531 } 532 533 nouveau_bo_ref(NULL, &screen->text); 534 nouveau_bo_ref(NULL, &screen->uniform_bo); 535 nouveau_bo_ref(NULL, &screen->tls); 536 nouveau_bo_ref(NULL, &screen->txc); 537 nouveau_bo_ref(NULL, &screen->fence.bo); 538 nouveau_bo_ref(NULL, &screen->poly_cache); 539 540 nouveau_heap_destroy(&screen->lib_code); 541 nouveau_heap_destroy(&screen->text_heap); 542 543 FREE(screen->default_tsc); 544 FREE(screen->tic.entries); 545 546 nouveau_object_del(&screen->eng3d); 547 nouveau_object_del(&screen->eng2d); 548 nouveau_object_del(&screen->m2mf); 549 nouveau_object_del(&screen->compute); 550 nouveau_object_del(&screen->nvsw); 551 552 nouveau_screen_fini(&screen->base); 553 554 FREE(screen); 555 } 556 557 static int 558 nvc0_graph_set_macro(struct nvc0_screen *screen, uint32_t m, unsigned pos, 559 unsigned size, const uint32_t *data) 560 { 561 struct nouveau_pushbuf *push = screen->base.pushbuf; 562 563 size /= 4; 564 565 assert((pos + size) <= 0x800); 566 567 BEGIN_NVC0(push, SUBC_3D(NVC0_GRAPH_MACRO_ID), 2); 568 PUSH_DATA (push, (m - 0x3800) / 8); 569 PUSH_DATA (push, pos); 570 BEGIN_1IC0(push, SUBC_3D(NVC0_GRAPH_MACRO_UPLOAD_POS), size + 1); 571 PUSH_DATA (push, pos); 572 PUSH_DATAp(push, data, size); 573 574 return pos + size; 575 } 576 577 static void 578 nvc0_magic_3d_init(struct nouveau_pushbuf *push, uint16_t obj_class) 579 { 580 BEGIN_NVC0(push, SUBC_3D(0x10cc), 1); 581 PUSH_DATA (push, 0xff); 582 BEGIN_NVC0(push, SUBC_3D(0x10e0), 2); 583 PUSH_DATA (push, 0xff); 584 PUSH_DATA (push, 0xff); 585 BEGIN_NVC0(push, SUBC_3D(0x10ec), 2); 586 PUSH_DATA (push, 0xff); 587 PUSH_DATA (push, 0xff); 588 BEGIN_NVC0(push, SUBC_3D(0x074c), 1); 589 PUSH_DATA (push, 0x3f); 590 591 BEGIN_NVC0(push, SUBC_3D(0x16a8), 1); 592 PUSH_DATA (push, (3 << 16) | 3); 593 BEGIN_NVC0(push, SUBC_3D(0x1794), 1); 594 PUSH_DATA (push, (2 << 16) | 2); 595 596 if (obj_class < GM107_3D_CLASS) { 597 BEGIN_NVC0(push, SUBC_3D(0x12ac), 1); 598 PUSH_DATA (push, 0); 599 } 600 BEGIN_NVC0(push, SUBC_3D(0x0218), 1); 601 PUSH_DATA (push, 0x10); 602 BEGIN_NVC0(push, SUBC_3D(0x10fc), 1); 603 PUSH_DATA (push, 0x10); 604 BEGIN_NVC0(push, SUBC_3D(0x1290), 1); 605 PUSH_DATA (push, 0x10); 606 BEGIN_NVC0(push, SUBC_3D(0x12d8), 2); 607 PUSH_DATA (push, 0x10); 608 PUSH_DATA (push, 0x10); 609 BEGIN_NVC0(push, SUBC_3D(0x1140), 1); 610 PUSH_DATA (push, 0x10); 611 BEGIN_NVC0(push, SUBC_3D(0x1610), 1); 612 PUSH_DATA (push, 0xe); 613 614 BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_GEN_MODE), 1); 615 PUSH_DATA (push, NVC0_3D_VERTEX_ID_GEN_MODE_DRAW_ARRAYS_ADD_START); 616 BEGIN_NVC0(push, SUBC_3D(0x030c), 1); 617 PUSH_DATA (push, 0); 618 BEGIN_NVC0(push, SUBC_3D(0x0300), 1); 619 PUSH_DATA (push, 3); 620 621 BEGIN_NVC0(push, SUBC_3D(0x02d0), 1); 622 PUSH_DATA (push, 0x3fffff); 623 BEGIN_NVC0(push, SUBC_3D(0x0fdc), 1); 624 PUSH_DATA (push, 1); 625 BEGIN_NVC0(push, SUBC_3D(0x19c0), 1); 626 PUSH_DATA (push, 1); 627 628 if (obj_class < GM107_3D_CLASS) { 629 BEGIN_NVC0(push, SUBC_3D(0x075c), 1); 630 PUSH_DATA (push, 3); 631 632 if (obj_class >= NVE4_3D_CLASS) { 633 BEGIN_NVC0(push, SUBC_3D(0x07fc), 1); 634 PUSH_DATA (push, 1); 635 } 636 } 637 638 /* TODO: find out what software methods 0x1528, 0x1280 and (on nve4) 0x02dc 639 * are supposed to do */ 640 } 641 642 static void 643 nvc0_screen_fence_emit(struct pipe_screen *pscreen, u32 *sequence) 644 { 645 struct nvc0_screen *screen = nvc0_screen(pscreen); 646 struct nouveau_pushbuf *push = screen->base.pushbuf; 647 648 /* we need to do it after possible flush in MARK_RING */ 649 *sequence = ++screen->base.fence.sequence; 650 651 assert(PUSH_AVAIL(push) + push->rsvd_kick >= 5); 652 PUSH_DATA (push, NVC0_FIFO_PKHDR_SQ(NVC0_3D(QUERY_ADDRESS_HIGH), 4)); 653 PUSH_DATAh(push, screen->fence.bo->offset); 654 PUSH_DATA (push, screen->fence.bo->offset); 655 PUSH_DATA (push, *sequence); 656 PUSH_DATA (push, NVC0_3D_QUERY_GET_FENCE | NVC0_3D_QUERY_GET_SHORT | 657 (0xf << NVC0_3D_QUERY_GET_UNIT__SHIFT)); 658 } 659 660 static u32 661 nvc0_screen_fence_update(struct pipe_screen *pscreen) 662 { 663 struct nvc0_screen *screen = nvc0_screen(pscreen); 664 return screen->fence.map[0]; 665 } 666 667 static int 668 nvc0_screen_init_compute(struct nvc0_screen *screen) 669 { 670 screen->base.base.get_compute_param = nvc0_screen_get_compute_param; 671 672 switch (screen->base.device->chipset & ~0xf) { 673 case 0xc0: 674 case 0xd0: 675 return nvc0_screen_compute_setup(screen, screen->base.pushbuf); 676 case 0xe0: 677 case 0xf0: 678 case 0x100: 679 case 0x110: 680 case 0x120: 681 return nve4_screen_compute_setup(screen, screen->base.pushbuf); 682 case 0x130: 683 return 0; 684 default: 685 return -1; 686 } 687 } 688 689 static int 690 nvc0_screen_resize_tls_area(struct nvc0_screen *screen, 691 uint32_t lpos, uint32_t lneg, uint32_t cstack) 692 { 693 struct nouveau_bo *bo = NULL; 694 int ret; 695 uint64_t size = (lpos + lneg) * 32 + cstack; 696 697 if (size >= (1 << 20)) { 698 NOUVEAU_ERR("requested TLS size too large: 0x%"PRIx64"\n", size); 699 return -1; 700 } 701 702 size *= (screen->base.device->chipset >= 0xe0) ? 64 : 48; /* max warps */ 703 size = align(size, 0x8000); 704 size *= screen->mp_count; 705 706 size = align(size, 1 << 17); 707 708 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base), 1 << 17, size, 709 NULL, &bo); 710 if (ret) 711 return ret; 712 nouveau_bo_ref(NULL, &screen->tls); 713 screen->tls = bo; 714 return 0; 715 } 716 717 int 718 nvc0_screen_resize_text_area(struct nvc0_screen *screen, uint64_t size) 719 { 720 struct nouveau_pushbuf *push = screen->base.pushbuf; 721 struct nouveau_bo *bo; 722 int ret; 723 724 ret = nouveau_bo_new(screen->base.device, NV_VRAM_DOMAIN(&screen->base), 725 1 << 17, size, NULL, &bo); 726 if (ret) 727 return ret; 728 729 nouveau_bo_ref(NULL, &screen->text); 730 screen->text = bo; 731 732 nouveau_heap_destroy(&screen->lib_code); 733 nouveau_heap_destroy(&screen->text_heap); 734 735 /* XXX: getting a page fault at the end of the code buffer every few 736 * launches, don't use the last 256 bytes to work around them - prefetch ? 737 */ 738 nouveau_heap_init(&screen->text_heap, 0, size - 0x100); 739 740 /* update the code segment setup */ 741 BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2); 742 PUSH_DATAh(push, screen->text->offset); 743 PUSH_DATA (push, screen->text->offset); 744 if (screen->compute) { 745 BEGIN_NVC0(push, NVC0_CP(CODE_ADDRESS_HIGH), 2); 746 PUSH_DATAh(push, screen->text->offset); 747 PUSH_DATA (push, screen->text->offset); 748 } 749 750 return 0; 751 } 752 753 #define FAIL_SCREEN_INIT(str, err) \ 754 do { \ 755 NOUVEAU_ERR(str, err); \ 756 goto fail; \ 757 } while(0) 758 759 struct nouveau_screen * 760 nvc0_screen_create(struct nouveau_device *dev) 761 { 762 struct nvc0_screen *screen; 763 struct pipe_screen *pscreen; 764 struct nouveau_object *chan; 765 struct nouveau_pushbuf *push; 766 uint64_t value; 767 uint32_t obj_class; 768 uint32_t flags; 769 int ret; 770 unsigned i; 771 772 switch (dev->chipset & ~0xf) { 773 case 0xc0: 774 case 0xd0: 775 case 0xe0: 776 case 0xf0: 777 case 0x100: 778 case 0x110: 779 case 0x120: 780 case 0x130: 781 break; 782 default: 783 return NULL; 784 } 785 786 screen = CALLOC_STRUCT(nvc0_screen); 787 if (!screen) 788 return NULL; 789 pscreen = &screen->base.base; 790 pscreen->destroy = nvc0_screen_destroy; 791 792 ret = nouveau_screen_init(&screen->base, dev); 793 if (ret) 794 FAIL_SCREEN_INIT("Base screen init failed: %d\n", ret); 795 chan = screen->base.channel; 796 push = screen->base.pushbuf; 797 push->user_priv = screen; 798 push->rsvd_kick = 5; 799 800 screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER | 801 PIPE_BIND_SHADER_BUFFER | 802 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER | 803 PIPE_BIND_COMMAND_ARGS_BUFFER | PIPE_BIND_QUERY_BUFFER; 804 screen->base.sysmem_bindings |= 805 PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER; 806 807 if (screen->base.vram_domain & NOUVEAU_BO_GART) { 808 screen->base.sysmem_bindings |= screen->base.vidmem_bindings; 809 screen->base.vidmem_bindings = 0; 810 } 811 812 pscreen->context_create = nvc0_create; 813 pscreen->is_format_supported = nvc0_screen_is_format_supported; 814 pscreen->get_param = nvc0_screen_get_param; 815 pscreen->get_shader_param = nvc0_screen_get_shader_param; 816 pscreen->get_paramf = nvc0_screen_get_paramf; 817 pscreen->get_driver_query_info = nvc0_screen_get_driver_query_info; 818 pscreen->get_driver_query_group_info = nvc0_screen_get_driver_query_group_info; 819 820 nvc0_screen_init_resource_functions(pscreen); 821 822 screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param; 823 screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported; 824 825 flags = NOUVEAU_BO_GART | NOUVEAU_BO_MAP; 826 if (screen->base.drm->version >= 0x01000202) 827 flags |= NOUVEAU_BO_COHERENT; 828 829 ret = nouveau_bo_new(dev, flags, 0, 4096, NULL, &screen->fence.bo); 830 if (ret) 831 FAIL_SCREEN_INIT("Error allocating fence BO: %d\n", ret); 832 nouveau_bo_map(screen->fence.bo, 0, NULL); 833 screen->fence.map = screen->fence.bo->map; 834 screen->base.fence.emit = nvc0_screen_fence_emit; 835 screen->base.fence.update = nvc0_screen_fence_update; 836 837 838 ret = nouveau_object_new(chan, (dev->chipset < 0xe0) ? 0x1f906e : 0x906e, 839 NVIF_CLASS_SW_GF100, NULL, 0, &screen->nvsw); 840 if (ret) 841 FAIL_SCREEN_INIT("Error creating SW object: %d\n", ret); 842 843 BEGIN_NVC0(push, SUBC_SW(NV01_SUBCHAN_OBJECT), 1); 844 PUSH_DATA (push, screen->nvsw->handle); 845 846 switch (dev->chipset & ~0xf) { 847 case 0x130: 848 case 0x120: 849 case 0x110: 850 case 0x100: 851 case 0xf0: 852 obj_class = NVF0_P2MF_CLASS; 853 break; 854 case 0xe0: 855 obj_class = NVE4_P2MF_CLASS; 856 break; 857 default: 858 obj_class = NVC0_M2MF_CLASS; 859 break; 860 } 861 ret = nouveau_object_new(chan, 0xbeef323f, obj_class, NULL, 0, 862 &screen->m2mf); 863 if (ret) 864 FAIL_SCREEN_INIT("Error allocating PGRAPH context for M2MF: %d\n", ret); 865 866 BEGIN_NVC0(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1); 867 PUSH_DATA (push, screen->m2mf->oclass); 868 if (screen->m2mf->oclass == NVE4_P2MF_CLASS) { 869 BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1); 870 PUSH_DATA (push, 0xa0b5); 871 } 872 873 ret = nouveau_object_new(chan, 0xbeef902d, NVC0_2D_CLASS, NULL, 0, 874 &screen->eng2d); 875 if (ret) 876 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 2D: %d\n", ret); 877 878 BEGIN_NVC0(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1); 879 PUSH_DATA (push, screen->eng2d->oclass); 880 BEGIN_NVC0(push, SUBC_2D(NVC0_2D_SINGLE_GPC), 1); 881 PUSH_DATA (push, 0); 882 BEGIN_NVC0(push, NVC0_2D(OPERATION), 1); 883 PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY); 884 BEGIN_NVC0(push, NVC0_2D(CLIP_ENABLE), 1); 885 PUSH_DATA (push, 0); 886 BEGIN_NVC0(push, NVC0_2D(COLOR_KEY_ENABLE), 1); 887 PUSH_DATA (push, 0); 888 BEGIN_NVC0(push, SUBC_2D(0x0884), 1); 889 PUSH_DATA (push, 0x3f); 890 BEGIN_NVC0(push, SUBC_2D(0x0888), 1); 891 PUSH_DATA (push, 1); 892 BEGIN_NVC0(push, NVC0_2D(COND_MODE), 1); 893 PUSH_DATA (push, NV50_2D_COND_MODE_ALWAYS); 894 895 BEGIN_NVC0(push, SUBC_2D(NVC0_GRAPH_NOTIFY_ADDRESS_HIGH), 2); 896 PUSH_DATAh(push, screen->fence.bo->offset + 16); 897 PUSH_DATA (push, screen->fence.bo->offset + 16); 898 899 switch (dev->chipset & ~0xf) { 900 case 0x130: 901 obj_class = GP100_3D_CLASS; 902 break; 903 case 0x120: 904 obj_class = GM200_3D_CLASS; 905 break; 906 case 0x110: 907 obj_class = GM107_3D_CLASS; 908 break; 909 case 0x100: 910 case 0xf0: 911 obj_class = NVF0_3D_CLASS; 912 break; 913 case 0xe0: 914 switch (dev->chipset) { 915 case 0xea: 916 obj_class = NVEA_3D_CLASS; 917 break; 918 default: 919 obj_class = NVE4_3D_CLASS; 920 break; 921 } 922 break; 923 case 0xd0: 924 obj_class = NVC8_3D_CLASS; 925 break; 926 case 0xc0: 927 default: 928 switch (dev->chipset) { 929 case 0xc8: 930 obj_class = NVC8_3D_CLASS; 931 break; 932 case 0xc1: 933 obj_class = NVC1_3D_CLASS; 934 break; 935 default: 936 obj_class = NVC0_3D_CLASS; 937 break; 938 } 939 break; 940 } 941 ret = nouveau_object_new(chan, 0xbeef003d, obj_class, NULL, 0, 942 &screen->eng3d); 943 if (ret) 944 FAIL_SCREEN_INIT("Error allocating PGRAPH context for 3D: %d\n", ret); 945 screen->base.class_3d = obj_class; 946 947 BEGIN_NVC0(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1); 948 PUSH_DATA (push, screen->eng3d->oclass); 949 950 BEGIN_NVC0(push, NVC0_3D(COND_MODE), 1); 951 PUSH_DATA (push, NVC0_3D_COND_MODE_ALWAYS); 952 953 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", true)) { 954 /* kill shaders after about 1 second (at 100 MHz) */ 955 BEGIN_NVC0(push, NVC0_3D(WATCHDOG_TIMER), 1); 956 PUSH_DATA (push, 0x17); 957 } 958 959 IMMED_NVC0(push, NVC0_3D(ZETA_COMP_ENABLE), 960 screen->base.drm->version >= 0x01000101); 961 BEGIN_NVC0(push, NVC0_3D(RT_COMP_ENABLE(0)), 8); 962 for (i = 0; i < 8; ++i) 963 PUSH_DATA(push, screen->base.drm->version >= 0x01000101); 964 965 BEGIN_NVC0(push, NVC0_3D(RT_CONTROL), 1); 966 PUSH_DATA (push, 1); 967 968 BEGIN_NVC0(push, NVC0_3D(CSAA_ENABLE), 1); 969 PUSH_DATA (push, 0); 970 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_ENABLE), 1); 971 PUSH_DATA (push, 0); 972 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_MODE), 1); 973 PUSH_DATA (push, NVC0_3D_MULTISAMPLE_MODE_MS1); 974 BEGIN_NVC0(push, NVC0_3D(MULTISAMPLE_CTRL), 1); 975 PUSH_DATA (push, 0); 976 BEGIN_NVC0(push, NVC0_3D(LINE_WIDTH_SEPARATE), 1); 977 PUSH_DATA (push, 1); 978 BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_WITH_DRAW_ARRAYS), 1); 979 PUSH_DATA (push, 1); 980 BEGIN_NVC0(push, NVC0_3D(BLEND_SEPARATE_ALPHA), 1); 981 PUSH_DATA (push, 1); 982 BEGIN_NVC0(push, NVC0_3D(BLEND_ENABLE_COMMON), 1); 983 PUSH_DATA (push, 0); 984 BEGIN_NVC0(push, NVC0_3D(SHADE_MODEL), 1); 985 PUSH_DATA (push, NVC0_3D_SHADE_MODEL_SMOOTH); 986 if (screen->eng3d->oclass < NVE4_3D_CLASS) { 987 IMMED_NVC0(push, NVC0_3D(TEX_MISC), 0); 988 } else { 989 BEGIN_NVC0(push, NVE4_3D(TEX_CB_INDEX), 1); 990 PUSH_DATA (push, 15); 991 } 992 BEGIN_NVC0(push, NVC0_3D(CALL_LIMIT_LOG), 1); 993 PUSH_DATA (push, 8); /* 128 */ 994 BEGIN_NVC0(push, NVC0_3D(ZCULL_STATCTRS_ENABLE), 1); 995 PUSH_DATA (push, 1); 996 if (screen->eng3d->oclass >= NVC1_3D_CLASS) { 997 BEGIN_NVC0(push, NVC0_3D(CACHE_SPLIT), 1); 998 PUSH_DATA (push, NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1); 999 } 1000 1001 nvc0_magic_3d_init(push, screen->eng3d->oclass); 1002 1003 ret = nvc0_screen_resize_text_area(screen, 1 << 19); 1004 if (ret) 1005 FAIL_SCREEN_INIT("Error allocating TEXT area: %d\n", ret); 1006 1007 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 12, 7 << 16, NULL, 1008 &screen->uniform_bo); 1009 if (ret) 1010 FAIL_SCREEN_INIT("Error allocating uniform BO: %d\n", ret); 1011 1012 PUSH_REFN (push, screen->uniform_bo, NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_WR); 1013 1014 /* return { 0.0, 0.0, 0.0, 0.0 } for out-of-bounds vtxbuf access */ 1015 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3); 1016 PUSH_DATA (push, 256); 1017 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO); 1018 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO); 1019 BEGIN_1IC0(push, NVC0_3D(CB_POS), 5); 1020 PUSH_DATA (push, 0); 1021 PUSH_DATAf(push, 0.0f); 1022 PUSH_DATAf(push, 0.0f); 1023 PUSH_DATAf(push, 0.0f); 1024 PUSH_DATAf(push, 0.0f); 1025 BEGIN_NVC0(push, NVC0_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2); 1026 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO); 1027 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_RUNOUT_INFO); 1028 1029 if (screen->base.drm->version >= 0x01000101) { 1030 ret = nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value); 1031 if (ret) 1032 FAIL_SCREEN_INIT("NOUVEAU_GETPARAM_GRAPH_UNITS failed: %d\n", ret); 1033 } else { 1034 if (dev->chipset >= 0xe0 && dev->chipset < 0xf0) 1035 value = (8 << 8) | 4; 1036 else 1037 value = (16 << 8) | 4; 1038 } 1039 screen->gpc_count = value & 0x000000ff; 1040 screen->mp_count = value >> 8; 1041 screen->mp_count_compute = screen->mp_count; 1042 1043 ret = nvc0_screen_resize_tls_area(screen, 128 * 16, 0, 0x200); 1044 if (ret) 1045 FAIL_SCREEN_INIT("Error allocating TLS area: %d\n", ret); 1046 1047 BEGIN_NVC0(push, NVC0_3D(TEMP_ADDRESS_HIGH), 4); 1048 PUSH_DATAh(push, screen->tls->offset); 1049 PUSH_DATA (push, screen->tls->offset); 1050 PUSH_DATA (push, screen->tls->size >> 32); 1051 PUSH_DATA (push, screen->tls->size); 1052 BEGIN_NVC0(push, NVC0_3D(WARP_TEMP_ALLOC), 1); 1053 PUSH_DATA (push, 0); 1054 /* Reduce likelihood of collision with real buffers by placing the hole at 1055 * the top of the 4G area. This will have to be dealt with for real 1056 * eventually by blocking off that area from the VM. 1057 */ 1058 BEGIN_NVC0(push, NVC0_3D(LOCAL_BASE), 1); 1059 PUSH_DATA (push, 0xff << 24); 1060 1061 if (screen->eng3d->oclass < GM107_3D_CLASS) { 1062 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 20, NULL, 1063 &screen->poly_cache); 1064 if (ret) 1065 FAIL_SCREEN_INIT("Error allocating poly cache BO: %d\n", ret); 1066 1067 BEGIN_NVC0(push, NVC0_3D(VERTEX_QUARANTINE_ADDRESS_HIGH), 3); 1068 PUSH_DATAh(push, screen->poly_cache->offset); 1069 PUSH_DATA (push, screen->poly_cache->offset); 1070 PUSH_DATA (push, 3); 1071 } 1072 1073 ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 1 << 17, 1 << 17, NULL, 1074 &screen->txc); 1075 if (ret) 1076 FAIL_SCREEN_INIT("Error allocating txc BO: %d\n", ret); 1077 1078 BEGIN_NVC0(push, NVC0_3D(TIC_ADDRESS_HIGH), 3); 1079 PUSH_DATAh(push, screen->txc->offset); 1080 PUSH_DATA (push, screen->txc->offset); 1081 PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1); 1082 if (screen->eng3d->oclass >= GM107_3D_CLASS) { 1083 screen->tic.maxwell = true; 1084 if (screen->eng3d->oclass == GM107_3D_CLASS) { 1085 screen->tic.maxwell = 1086 debug_get_bool_option("NOUVEAU_MAXWELL_TIC", true); 1087 IMMED_NVC0(push, SUBC_3D(0x0f10), screen->tic.maxwell); 1088 } 1089 } 1090 1091 BEGIN_NVC0(push, NVC0_3D(TSC_ADDRESS_HIGH), 3); 1092 PUSH_DATAh(push, screen->txc->offset + 65536); 1093 PUSH_DATA (push, screen->txc->offset + 65536); 1094 PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1); 1095 1096 BEGIN_NVC0(push, NVC0_3D(SCREEN_Y_CONTROL), 1); 1097 PUSH_DATA (push, 0); 1098 BEGIN_NVC0(push, NVC0_3D(WINDOW_OFFSET_X), 2); 1099 PUSH_DATA (push, 0); 1100 PUSH_DATA (push, 0); 1101 BEGIN_NVC0(push, NVC0_3D(ZCULL_REGION), 1); /* deactivate ZCULL */ 1102 PUSH_DATA (push, 0x3f); 1103 1104 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_MODE), 1); 1105 PUSH_DATA (push, NVC0_3D_CLIP_RECTS_MODE_INSIDE_ANY); 1106 BEGIN_NVC0(push, NVC0_3D(CLIP_RECT_HORIZ(0)), 8 * 2); 1107 for (i = 0; i < 8 * 2; ++i) 1108 PUSH_DATA(push, 0); 1109 BEGIN_NVC0(push, NVC0_3D(CLIP_RECTS_EN), 1); 1110 PUSH_DATA (push, 0); 1111 BEGIN_NVC0(push, NVC0_3D(CLIPID_ENABLE), 1); 1112 PUSH_DATA (push, 0); 1113 1114 /* neither scissors, viewport nor stencil mask should affect clears */ 1115 BEGIN_NVC0(push, NVC0_3D(CLEAR_FLAGS), 1); 1116 PUSH_DATA (push, 0); 1117 1118 BEGIN_NVC0(push, NVC0_3D(VIEWPORT_TRANSFORM_EN), 1); 1119 PUSH_DATA (push, 1); 1120 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) { 1121 BEGIN_NVC0(push, NVC0_3D(DEPTH_RANGE_NEAR(i)), 2); 1122 PUSH_DATAf(push, 0.0f); 1123 PUSH_DATAf(push, 1.0f); 1124 } 1125 BEGIN_NVC0(push, NVC0_3D(VIEW_VOLUME_CLIP_CTRL), 1); 1126 PUSH_DATA (push, NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1); 1127 1128 /* We use scissors instead of exact view volume clipping, 1129 * so they're always enabled. 1130 */ 1131 for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) { 1132 BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(i)), 3); 1133 PUSH_DATA (push, 1); 1134 PUSH_DATA (push, 8192 << 16); 1135 PUSH_DATA (push, 8192 << 16); 1136 } 1137 1138 #define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n); 1139 1140 i = 0; 1141 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_PER_INSTANCE, mme9097_per_instance_bf); 1142 MK_MACRO(NVC0_3D_MACRO_BLEND_ENABLES, mme9097_blend_enables); 1143 MK_MACRO(NVC0_3D_MACRO_VERTEX_ARRAY_SELECT, mme9097_vertex_array_select); 1144 MK_MACRO(NVC0_3D_MACRO_TEP_SELECT, mme9097_tep_select); 1145 MK_MACRO(NVC0_3D_MACRO_GP_SELECT, mme9097_gp_select); 1146 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_FRONT, mme9097_poly_mode_front); 1147 MK_MACRO(NVC0_3D_MACRO_POLYGON_MODE_BACK, mme9097_poly_mode_back); 1148 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT, mme9097_draw_arrays_indirect); 1149 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT, mme9097_draw_elts_indirect); 1150 MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT, mme9097_draw_arrays_indirect_count); 1151 MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT, mme9097_draw_elts_indirect_count); 1152 MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE, mme9097_query_buffer_write); 1153 MK_MACRO(NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT, mme90c0_launch_grid_indirect); 1154 1155 BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1); 1156 PUSH_DATA (push, 1); 1157 BEGIN_NVC0(push, NVC0_3D(RT_SEPARATE_FRAG_DATA), 1); 1158 PUSH_DATA (push, 1); 1159 BEGIN_NVC0(push, NVC0_3D(MACRO_GP_SELECT), 1); 1160 PUSH_DATA (push, 0x40); 1161 BEGIN_NVC0(push, NVC0_3D(LAYER), 1); 1162 PUSH_DATA (push, 0); 1163 BEGIN_NVC0(push, NVC0_3D(MACRO_TEP_SELECT), 1); 1164 PUSH_DATA (push, 0x30); 1165 BEGIN_NVC0(push, NVC0_3D(PATCH_VERTICES), 1); 1166 PUSH_DATA (push, 3); 1167 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(2)), 1); 1168 PUSH_DATA (push, 0x20); 1169 BEGIN_NVC0(push, NVC0_3D(SP_SELECT(0)), 1); 1170 PUSH_DATA (push, 0x00); 1171 screen->save_state.patch_vertices = 3; 1172 1173 BEGIN_NVC0(push, NVC0_3D(POINT_COORD_REPLACE), 1); 1174 PUSH_DATA (push, 0); 1175 BEGIN_NVC0(push, NVC0_3D(POINT_RASTER_RULES), 1); 1176 PUSH_DATA (push, NVC0_3D_POINT_RASTER_RULES_OGL); 1177 1178 IMMED_NVC0(push, NVC0_3D(EDGEFLAG), 1); 1179 1180 if (nvc0_screen_init_compute(screen)) 1181 goto fail; 1182 1183 /* XXX: Compute and 3D are somehow aliased on Fermi. */ 1184 for (i = 0; i < 5; ++i) { 1185 /* TIC and TSC entries for each unit (nve4+ only) */ 1186 /* auxiliary constants (6 user clip planes, base instance id) */ 1187 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3); 1188 PUSH_DATA (push, NVC0_CB_AUX_SIZE); 1189 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(i)); 1190 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(i)); 1191 BEGIN_NVC0(push, NVC0_3D(CB_BIND(i)), 1); 1192 PUSH_DATA (push, (15 << 4) | 1); 1193 if (screen->eng3d->oclass >= NVE4_3D_CLASS) { 1194 unsigned j; 1195 BEGIN_1IC0(push, NVC0_3D(CB_POS), 9); 1196 PUSH_DATA (push, NVC0_CB_AUX_UNK_INFO); 1197 for (j = 0; j < 8; ++j) 1198 PUSH_DATA(push, j); 1199 } else { 1200 BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1); 1201 PUSH_DATA (push, 0x54); 1202 } 1203 1204 /* MS sample coordinate offsets: these do not work with _ALT modes ! */ 1205 BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 2 * 8); 1206 PUSH_DATA (push, NVC0_CB_AUX_MS_INFO); 1207 PUSH_DATA (push, 0); /* 0 */ 1208 PUSH_DATA (push, 0); 1209 PUSH_DATA (push, 1); /* 1 */ 1210 PUSH_DATA (push, 0); 1211 PUSH_DATA (push, 0); /* 2 */ 1212 PUSH_DATA (push, 1); 1213 PUSH_DATA (push, 1); /* 3 */ 1214 PUSH_DATA (push, 1); 1215 PUSH_DATA (push, 2); /* 4 */ 1216 PUSH_DATA (push, 0); 1217 PUSH_DATA (push, 3); /* 5 */ 1218 PUSH_DATA (push, 0); 1219 PUSH_DATA (push, 2); /* 6 */ 1220 PUSH_DATA (push, 1); 1221 PUSH_DATA (push, 3); /* 7 */ 1222 PUSH_DATA (push, 1); 1223 } 1224 BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1); 1225 PUSH_DATA (push, 0); 1226 1227 PUSH_KICK (push); 1228 1229 screen->tic.entries = CALLOC(4096, sizeof(void *)); 1230 screen->tsc.entries = screen->tic.entries + 2048; 1231 1232 if (!nvc0_blitter_create(screen)) 1233 goto fail; 1234 1235 screen->default_tsc = CALLOC_STRUCT(nv50_tsc_entry); 1236 screen->default_tsc->tsc[0] = G80_TSC_0_SRGB_CONVERSION; 1237 1238 nouveau_fence_new(&screen->base, &screen->base.fence.current); 1239 1240 return &screen->base; 1241 1242 fail: 1243 screen->base.base.context_create = NULL; 1244 return &screen->base; 1245 } 1246 1247 int 1248 nvc0_screen_tic_alloc(struct nvc0_screen *screen, void *entry) 1249 { 1250 int i = screen->tic.next; 1251 1252 while (screen->tic.lock[i / 32] & (1 << (i % 32))) 1253 i = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1); 1254 1255 screen->tic.next = (i + 1) & (NVC0_TIC_MAX_ENTRIES - 1); 1256 1257 if (screen->tic.entries[i]) 1258 nv50_tic_entry(screen->tic.entries[i])->id = -1; 1259 1260 screen->tic.entries[i] = entry; 1261 return i; 1262 } 1263 1264 int 1265 nvc0_screen_tsc_alloc(struct nvc0_screen *screen, void *entry) 1266 { 1267 int i = screen->tsc.next; 1268 1269 while (screen->tsc.lock[i / 32] & (1 << (i % 32))) 1270 i = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1); 1271 1272 screen->tsc.next = (i + 1) & (NVC0_TSC_MAX_ENTRIES - 1); 1273 1274 if (screen->tsc.entries[i]) 1275 nv50_tsc_entry(screen->tsc.entries[i])->id = -1; 1276 1277 screen->tsc.entries[i] = entry; 1278 return i; 1279 } 1280