anv: Implement VK_EXT_custom_border_color
[mesa.git] / src / intel / vulkan / anv_device.c
1 /*
2 * Copyright © 2015 Intel Corporation
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 (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #include <assert.h>
25 #include <stdbool.h>
26 #include <string.h>
27 #include <sys/mman.h>
28 #include <sys/sysinfo.h>
29 #include <unistd.h>
30 #include <fcntl.h>
31 #include <xf86drm.h>
32 #include "drm-uapi/drm_fourcc.h"
33
34 #include "anv_private.h"
35 #include "util/debug.h"
36 #include "util/build_id.h"
37 #include "util/disk_cache.h"
38 #include "util/mesa-sha1.h"
39 #include "util/os_file.h"
40 #include "util/u_atomic.h"
41 #include "util/u_string.h"
42 #include "util/xmlpool.h"
43 #include "git_sha1.h"
44 #include "vk_util.h"
45 #include "common/gen_aux_map.h"
46 #include "common/gen_defines.h"
47 #include "compiler/glsl_types.h"
48
49 #include "genxml/gen7_pack.h"
50
51 static const char anv_dri_options_xml[] =
52 DRI_CONF_BEGIN
53 DRI_CONF_SECTION_PERFORMANCE
54 DRI_CONF_VK_X11_OVERRIDE_MIN_IMAGE_COUNT(0)
55 DRI_CONF_VK_X11_STRICT_IMAGE_COUNT("false")
56 DRI_CONF_SECTION_END
57
58 DRI_CONF_SECTION_DEBUG
59 DRI_CONF_ALWAYS_FLUSH_CACHE("false")
60 DRI_CONF_VK_WSI_FORCE_BGRA8_UNORM_FIRST("false")
61 DRI_CONF_SECTION_END
62 DRI_CONF_END;
63
64 /* This is probably far to big but it reflects the max size used for messages
65 * in OpenGLs KHR_debug.
66 */
67 #define MAX_DEBUG_MESSAGE_LENGTH 4096
68
69 static void
70 compiler_debug_log(void *data, const char *fmt, ...)
71 {
72 char str[MAX_DEBUG_MESSAGE_LENGTH];
73 struct anv_device *device = (struct anv_device *)data;
74 struct anv_instance *instance = device->physical->instance;
75
76 if (list_is_empty(&instance->debug_report_callbacks.callbacks))
77 return;
78
79 va_list args;
80 va_start(args, fmt);
81 (void) vsnprintf(str, MAX_DEBUG_MESSAGE_LENGTH, fmt, args);
82 va_end(args);
83
84 vk_debug_report(&instance->debug_report_callbacks,
85 VK_DEBUG_REPORT_DEBUG_BIT_EXT,
86 VK_DEBUG_REPORT_OBJECT_TYPE_UNKNOWN_EXT,
87 0, 0, 0, "anv", str);
88 }
89
90 static void
91 compiler_perf_log(void *data, const char *fmt, ...)
92 {
93 va_list args;
94 va_start(args, fmt);
95
96 if (unlikely(INTEL_DEBUG & DEBUG_PERF))
97 intel_logd_v(fmt, args);
98
99 va_end(args);
100 }
101
102 static uint64_t
103 anv_compute_heap_size(int fd, uint64_t gtt_size)
104 {
105 /* Query the total ram from the system */
106 struct sysinfo info;
107 sysinfo(&info);
108
109 uint64_t total_ram = (uint64_t)info.totalram * (uint64_t)info.mem_unit;
110
111 /* We don't want to burn too much ram with the GPU. If the user has 4GiB
112 * or less, we use at most half. If they have more than 4GiB, we use 3/4.
113 */
114 uint64_t available_ram;
115 if (total_ram <= 4ull * 1024ull * 1024ull * 1024ull)
116 available_ram = total_ram / 2;
117 else
118 available_ram = total_ram * 3 / 4;
119
120 /* We also want to leave some padding for things we allocate in the driver,
121 * so don't go over 3/4 of the GTT either.
122 */
123 uint64_t available_gtt = gtt_size * 3 / 4;
124
125 return MIN2(available_ram, available_gtt);
126 }
127
128 static VkResult
129 anv_physical_device_init_heaps(struct anv_physical_device *device, int fd)
130 {
131 if (anv_gem_get_context_param(fd, 0, I915_CONTEXT_PARAM_GTT_SIZE,
132 &device->gtt_size) == -1) {
133 /* If, for whatever reason, we can't actually get the GTT size from the
134 * kernel (too old?) fall back to the aperture size.
135 */
136 anv_perf_warn(NULL, NULL,
137 "Failed to get I915_CONTEXT_PARAM_GTT_SIZE: %m");
138
139 if (anv_gem_get_aperture(fd, &device->gtt_size) == -1) {
140 return vk_errorfi(device->instance, NULL,
141 VK_ERROR_INITIALIZATION_FAILED,
142 "failed to get aperture size: %m");
143 }
144 }
145
146 /* We only allow 48-bit addresses with softpin because knowing the actual
147 * address is required for the vertex cache flush workaround.
148 */
149 device->supports_48bit_addresses = (device->info.gen >= 8) &&
150 device->has_softpin &&
151 device->gtt_size > (4ULL << 30 /* GiB */);
152
153 uint64_t heap_size = anv_compute_heap_size(fd, device->gtt_size);
154
155 if (heap_size > (2ull << 30) && !device->supports_48bit_addresses) {
156 /* When running with an overridden PCI ID, we may get a GTT size from
157 * the kernel that is greater than 2 GiB but the execbuf check for 48bit
158 * address support can still fail. Just clamp the address space size to
159 * 2 GiB if we don't have 48-bit support.
160 */
161 intel_logw("%s:%d: The kernel reported a GTT size larger than 2 GiB but "
162 "not support for 48-bit addresses",
163 __FILE__, __LINE__);
164 heap_size = 2ull << 30;
165 }
166
167 device->memory.heap_count = 1;
168 device->memory.heaps[0] = (struct anv_memory_heap) {
169 .size = heap_size,
170 .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
171 };
172
173 uint32_t type_count = 0;
174 for (uint32_t heap = 0; heap < device->memory.heap_count; heap++) {
175 if (device->info.has_llc) {
176 /* Big core GPUs share LLC with the CPU and thus one memory type can be
177 * both cached and coherent at the same time.
178 */
179 device->memory.types[type_count++] = (struct anv_memory_type) {
180 .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
181 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
182 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
183 VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
184 .heapIndex = heap,
185 };
186 } else {
187 /* The spec requires that we expose a host-visible, coherent memory
188 * type, but Atom GPUs don't share LLC. Thus we offer two memory types
189 * to give the application a choice between cached, but not coherent and
190 * coherent but uncached (WC though).
191 */
192 device->memory.types[type_count++] = (struct anv_memory_type) {
193 .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
194 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
195 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
196 .heapIndex = heap,
197 };
198 device->memory.types[type_count++] = (struct anv_memory_type) {
199 .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
200 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
201 VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
202 .heapIndex = heap,
203 };
204 }
205 }
206 device->memory.type_count = type_count;
207
208 return VK_SUCCESS;
209 }
210
211 static VkResult
212 anv_physical_device_init_uuids(struct anv_physical_device *device)
213 {
214 const struct build_id_note *note =
215 build_id_find_nhdr_for_addr(anv_physical_device_init_uuids);
216 if (!note) {
217 return vk_errorfi(device->instance, NULL,
218 VK_ERROR_INITIALIZATION_FAILED,
219 "Failed to find build-id");
220 }
221
222 unsigned build_id_len = build_id_length(note);
223 if (build_id_len < 20) {
224 return vk_errorfi(device->instance, NULL,
225 VK_ERROR_INITIALIZATION_FAILED,
226 "build-id too short. It needs to be a SHA");
227 }
228
229 memcpy(device->driver_build_sha1, build_id_data(note), 20);
230
231 struct mesa_sha1 sha1_ctx;
232 uint8_t sha1[20];
233 STATIC_ASSERT(VK_UUID_SIZE <= sizeof(sha1));
234
235 /* The pipeline cache UUID is used for determining when a pipeline cache is
236 * invalid. It needs both a driver build and the PCI ID of the device.
237 */
238 _mesa_sha1_init(&sha1_ctx);
239 _mesa_sha1_update(&sha1_ctx, build_id_data(note), build_id_len);
240 _mesa_sha1_update(&sha1_ctx, &device->info.chipset_id,
241 sizeof(device->info.chipset_id));
242 _mesa_sha1_update(&sha1_ctx, &device->always_use_bindless,
243 sizeof(device->always_use_bindless));
244 _mesa_sha1_update(&sha1_ctx, &device->has_a64_buffer_access,
245 sizeof(device->has_a64_buffer_access));
246 _mesa_sha1_update(&sha1_ctx, &device->has_bindless_images,
247 sizeof(device->has_bindless_images));
248 _mesa_sha1_update(&sha1_ctx, &device->has_bindless_samplers,
249 sizeof(device->has_bindless_samplers));
250 _mesa_sha1_final(&sha1_ctx, sha1);
251 memcpy(device->pipeline_cache_uuid, sha1, VK_UUID_SIZE);
252
253 /* The driver UUID is used for determining sharability of images and memory
254 * between two Vulkan instances in separate processes. People who want to
255 * share memory need to also check the device UUID (below) so all this
256 * needs to be is the build-id.
257 */
258 memcpy(device->driver_uuid, build_id_data(note), VK_UUID_SIZE);
259
260 /* The device UUID uniquely identifies the given device within the machine.
261 * Since we never have more than one device, this doesn't need to be a real
262 * UUID. However, on the off-chance that someone tries to use this to
263 * cache pre-tiled images or something of the like, we use the PCI ID and
264 * some bits of ISL info to ensure that this is safe.
265 */
266 _mesa_sha1_init(&sha1_ctx);
267 _mesa_sha1_update(&sha1_ctx, &device->info.chipset_id,
268 sizeof(device->info.chipset_id));
269 _mesa_sha1_update(&sha1_ctx, &device->isl_dev.has_bit6_swizzling,
270 sizeof(device->isl_dev.has_bit6_swizzling));
271 _mesa_sha1_final(&sha1_ctx, sha1);
272 memcpy(device->device_uuid, sha1, VK_UUID_SIZE);
273
274 return VK_SUCCESS;
275 }
276
277 static void
278 anv_physical_device_init_disk_cache(struct anv_physical_device *device)
279 {
280 #ifdef ENABLE_SHADER_CACHE
281 char renderer[10];
282 ASSERTED int len = snprintf(renderer, sizeof(renderer), "anv_%04x",
283 device->info.chipset_id);
284 assert(len == sizeof(renderer) - 2);
285
286 char timestamp[41];
287 _mesa_sha1_format(timestamp, device->driver_build_sha1);
288
289 const uint64_t driver_flags =
290 brw_get_compiler_config_value(device->compiler);
291 device->disk_cache = disk_cache_create(renderer, timestamp, driver_flags);
292 #else
293 device->disk_cache = NULL;
294 #endif
295 }
296
297 static void
298 anv_physical_device_free_disk_cache(struct anv_physical_device *device)
299 {
300 #ifdef ENABLE_SHADER_CACHE
301 if (device->disk_cache)
302 disk_cache_destroy(device->disk_cache);
303 #else
304 assert(device->disk_cache == NULL);
305 #endif
306 }
307
308 static uint64_t
309 get_available_system_memory()
310 {
311 char *meminfo = os_read_file("/proc/meminfo", NULL);
312 if (!meminfo)
313 return 0;
314
315 char *str = strstr(meminfo, "MemAvailable:");
316 if (!str) {
317 free(meminfo);
318 return 0;
319 }
320
321 uint64_t kb_mem_available;
322 if (sscanf(str, "MemAvailable: %" PRIx64, &kb_mem_available) == 1) {
323 free(meminfo);
324 return kb_mem_available << 10;
325 }
326
327 free(meminfo);
328 return 0;
329 }
330
331 static VkResult
332 anv_physical_device_try_create(struct anv_instance *instance,
333 drmDevicePtr drm_device,
334 struct anv_physical_device **device_out)
335 {
336 const char *primary_path = drm_device->nodes[DRM_NODE_PRIMARY];
337 const char *path = drm_device->nodes[DRM_NODE_RENDER];
338 VkResult result;
339 int fd;
340 int master_fd = -1;
341
342 brw_process_intel_debug_variable();
343
344 fd = open(path, O_RDWR | O_CLOEXEC);
345 if (fd < 0)
346 return vk_error(VK_ERROR_INCOMPATIBLE_DRIVER);
347
348 struct gen_device_info devinfo;
349 if (!gen_get_device_info_from_fd(fd, &devinfo)) {
350 result = vk_error(VK_ERROR_INCOMPATIBLE_DRIVER);
351 goto fail_fd;
352 }
353
354 const char *device_name = gen_get_device_name(devinfo.chipset_id);
355
356 if (devinfo.is_haswell) {
357 intel_logw("Haswell Vulkan support is incomplete");
358 } else if (devinfo.gen == 7 && !devinfo.is_baytrail) {
359 intel_logw("Ivy Bridge Vulkan support is incomplete");
360 } else if (devinfo.gen == 7 && devinfo.is_baytrail) {
361 intel_logw("Bay Trail Vulkan support is incomplete");
362 } else if (devinfo.gen >= 8 && devinfo.gen <= 11) {
363 /* Gen8-11 fully supported */
364 } else if (devinfo.gen == 12) {
365 intel_logw("Vulkan is not yet fully supported on gen12");
366 } else {
367 result = vk_errorfi(instance, NULL, VK_ERROR_INCOMPATIBLE_DRIVER,
368 "Vulkan not yet supported on %s", device_name);
369 goto fail_fd;
370 }
371
372 struct anv_physical_device *device =
373 vk_alloc(&instance->alloc, sizeof(*device), 8,
374 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
375 if (device == NULL) {
376 result = vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
377 goto fail_fd;
378 }
379
380 vk_object_base_init(NULL, &device->base, VK_OBJECT_TYPE_PHYSICAL_DEVICE);
381 device->instance = instance;
382
383 assert(strlen(path) < ARRAY_SIZE(device->path));
384 snprintf(device->path, ARRAY_SIZE(device->path), "%s", path);
385
386 device->info = devinfo;
387 device->name = device_name;
388
389 device->no_hw = device->info.no_hw;
390 if (getenv("INTEL_NO_HW") != NULL)
391 device->no_hw = true;
392
393 device->pci_info.domain = drm_device->businfo.pci->domain;
394 device->pci_info.bus = drm_device->businfo.pci->bus;
395 device->pci_info.device = drm_device->businfo.pci->dev;
396 device->pci_info.function = drm_device->businfo.pci->func;
397
398 device->cmd_parser_version = -1;
399 if (device->info.gen == 7) {
400 device->cmd_parser_version =
401 anv_gem_get_param(fd, I915_PARAM_CMD_PARSER_VERSION);
402 if (device->cmd_parser_version == -1) {
403 result = vk_errorfi(device->instance, NULL,
404 VK_ERROR_INITIALIZATION_FAILED,
405 "failed to get command parser version");
406 goto fail_alloc;
407 }
408 }
409
410 if (!anv_gem_get_param(fd, I915_PARAM_HAS_WAIT_TIMEOUT)) {
411 result = vk_errorfi(device->instance, NULL,
412 VK_ERROR_INITIALIZATION_FAILED,
413 "kernel missing gem wait");
414 goto fail_alloc;
415 }
416
417 if (!anv_gem_get_param(fd, I915_PARAM_HAS_EXECBUF2)) {
418 result = vk_errorfi(device->instance, NULL,
419 VK_ERROR_INITIALIZATION_FAILED,
420 "kernel missing execbuf2");
421 goto fail_alloc;
422 }
423
424 if (!device->info.has_llc &&
425 anv_gem_get_param(fd, I915_PARAM_MMAP_VERSION) < 1) {
426 result = vk_errorfi(device->instance, NULL,
427 VK_ERROR_INITIALIZATION_FAILED,
428 "kernel missing wc mmap");
429 goto fail_alloc;
430 }
431
432 device->has_softpin = anv_gem_get_param(fd, I915_PARAM_HAS_EXEC_SOFTPIN);
433 device->has_exec_async = anv_gem_get_param(fd, I915_PARAM_HAS_EXEC_ASYNC);
434 device->has_exec_capture = anv_gem_get_param(fd, I915_PARAM_HAS_EXEC_CAPTURE);
435 device->has_exec_fence = anv_gem_get_param(fd, I915_PARAM_HAS_EXEC_FENCE);
436 device->has_syncobj = anv_gem_get_param(fd, I915_PARAM_HAS_EXEC_FENCE_ARRAY);
437 device->has_syncobj_wait = device->has_syncobj &&
438 anv_gem_supports_syncobj_wait(fd);
439 device->has_context_priority = anv_gem_has_context_priority(fd);
440
441 result = anv_physical_device_init_heaps(device, fd);
442 if (result != VK_SUCCESS)
443 goto fail_alloc;
444
445 device->use_softpin = device->has_softpin &&
446 device->supports_48bit_addresses;
447
448 device->has_context_isolation =
449 anv_gem_get_param(fd, I915_PARAM_HAS_CONTEXT_ISOLATION);
450
451 device->always_use_bindless =
452 env_var_as_boolean("ANV_ALWAYS_BINDLESS", false);
453
454 /* We first got the A64 messages on broadwell and we can only use them if
455 * we can pass addresses directly into the shader which requires softpin.
456 */
457 device->has_a64_buffer_access = device->info.gen >= 8 &&
458 device->use_softpin;
459
460 /* We first get bindless image access on Skylake and we can only really do
461 * it if we don't have any relocations so we need softpin.
462 */
463 device->has_bindless_images = device->info.gen >= 9 &&
464 device->use_softpin;
465
466 /* We've had bindless samplers since Ivy Bridge (forever in Vulkan terms)
467 * because it's just a matter of setting the sampler address in the sample
468 * message header. However, we've not bothered to wire it up for vec4 so
469 * we leave it disabled on gen7.
470 */
471 device->has_bindless_samplers = device->info.gen >= 8;
472
473 device->has_implicit_ccs = device->info.has_aux_map;
474
475 device->has_mem_available = get_available_system_memory() != 0;
476
477 device->always_flush_cache =
478 driQueryOptionb(&instance->dri_options, "always_flush_cache");
479
480 device->has_mmap_offset =
481 anv_gem_get_param(fd, I915_PARAM_MMAP_GTT_VERSION) >= 4;
482
483 /* GENs prior to 8 do not support EU/Subslice info */
484 if (device->info.gen >= 8) {
485 device->subslice_total = anv_gem_get_param(fd, I915_PARAM_SUBSLICE_TOTAL);
486 device->eu_total = anv_gem_get_param(fd, I915_PARAM_EU_TOTAL);
487
488 /* Without this information, we cannot get the right Braswell
489 * brandstrings, and we have to use conservative numbers for GPGPU on
490 * many platforms, but otherwise, things will just work.
491 */
492 if (device->subslice_total < 1 || device->eu_total < 1) {
493 intel_logw("Kernel 4.1 required to properly query GPU properties");
494 }
495 } else if (device->info.gen == 7) {
496 device->subslice_total = 1 << (device->info.gt - 1);
497 }
498
499 if (device->info.is_cherryview &&
500 device->subslice_total > 0 && device->eu_total > 0) {
501 /* Logical CS threads = EUs per subslice * num threads per EU */
502 uint32_t max_cs_threads =
503 device->eu_total / device->subslice_total * device->info.num_thread_per_eu;
504
505 /* Fuse configurations may give more threads than expected, never less. */
506 if (max_cs_threads > device->info.max_cs_threads)
507 device->info.max_cs_threads = max_cs_threads;
508 }
509
510 device->compiler = brw_compiler_create(NULL, &device->info);
511 if (device->compiler == NULL) {
512 result = vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
513 goto fail_alloc;
514 }
515 device->compiler->shader_debug_log = compiler_debug_log;
516 device->compiler->shader_perf_log = compiler_perf_log;
517 device->compiler->supports_pull_constants = false;
518 device->compiler->constant_buffer_0_is_relative =
519 device->info.gen < 8 || !device->has_context_isolation;
520 device->compiler->supports_shader_constants = true;
521 device->compiler->compact_params = false;
522
523 /* Broadwell PRM says:
524 *
525 * "Before Gen8, there was a historical configuration control field to
526 * swizzle address bit[6] for in X/Y tiling modes. This was set in three
527 * different places: TILECTL[1:0], ARB_MODE[5:4], and
528 * DISP_ARB_CTL[14:13].
529 *
530 * For Gen8 and subsequent generations, the swizzle fields are all
531 * reserved, and the CPU's memory controller performs all address
532 * swizzling modifications."
533 */
534 bool swizzled =
535 device->info.gen < 8 && anv_gem_get_bit6_swizzle(fd, I915_TILING_X);
536
537 isl_device_init(&device->isl_dev, &device->info, swizzled);
538
539 result = anv_physical_device_init_uuids(device);
540 if (result != VK_SUCCESS)
541 goto fail_compiler;
542
543 anv_physical_device_init_disk_cache(device);
544
545 if (instance->enabled_extensions.KHR_display) {
546 master_fd = open(primary_path, O_RDWR | O_CLOEXEC);
547 if (master_fd >= 0) {
548 /* prod the device with a GETPARAM call which will fail if
549 * we don't have permission to even render on this device
550 */
551 if (anv_gem_get_param(master_fd, I915_PARAM_CHIPSET_ID) == 0) {
552 close(master_fd);
553 master_fd = -1;
554 }
555 }
556 }
557 device->master_fd = master_fd;
558
559 result = anv_init_wsi(device);
560 if (result != VK_SUCCESS)
561 goto fail_disk_cache;
562
563 device->perf = anv_get_perf(&device->info, fd);
564
565 anv_physical_device_get_supported_extensions(device,
566 &device->supported_extensions);
567
568
569 device->local_fd = fd;
570
571 *device_out = device;
572
573 return VK_SUCCESS;
574
575 fail_disk_cache:
576 anv_physical_device_free_disk_cache(device);
577 fail_compiler:
578 ralloc_free(device->compiler);
579 fail_alloc:
580 vk_free(&instance->alloc, device);
581 fail_fd:
582 close(fd);
583 if (master_fd != -1)
584 close(master_fd);
585 return result;
586 }
587
588 static void
589 anv_physical_device_destroy(struct anv_physical_device *device)
590 {
591 anv_finish_wsi(device);
592 anv_physical_device_free_disk_cache(device);
593 ralloc_free(device->compiler);
594 ralloc_free(device->perf);
595 close(device->local_fd);
596 if (device->master_fd >= 0)
597 close(device->master_fd);
598 vk_object_base_finish(&device->base);
599 vk_free(&device->instance->alloc, device);
600 }
601
602 static void *
603 default_alloc_func(void *pUserData, size_t size, size_t align,
604 VkSystemAllocationScope allocationScope)
605 {
606 return malloc(size);
607 }
608
609 static void *
610 default_realloc_func(void *pUserData, void *pOriginal, size_t size,
611 size_t align, VkSystemAllocationScope allocationScope)
612 {
613 return realloc(pOriginal, size);
614 }
615
616 static void
617 default_free_func(void *pUserData, void *pMemory)
618 {
619 free(pMemory);
620 }
621
622 static const VkAllocationCallbacks default_alloc = {
623 .pUserData = NULL,
624 .pfnAllocation = default_alloc_func,
625 .pfnReallocation = default_realloc_func,
626 .pfnFree = default_free_func,
627 };
628
629 VkResult anv_EnumerateInstanceExtensionProperties(
630 const char* pLayerName,
631 uint32_t* pPropertyCount,
632 VkExtensionProperties* pProperties)
633 {
634 VK_OUTARRAY_MAKE(out, pProperties, pPropertyCount);
635
636 for (int i = 0; i < ANV_INSTANCE_EXTENSION_COUNT; i++) {
637 if (anv_instance_extensions_supported.extensions[i]) {
638 vk_outarray_append(&out, prop) {
639 *prop = anv_instance_extensions[i];
640 }
641 }
642 }
643
644 return vk_outarray_status(&out);
645 }
646
647 VkResult anv_CreateInstance(
648 const VkInstanceCreateInfo* pCreateInfo,
649 const VkAllocationCallbacks* pAllocator,
650 VkInstance* pInstance)
651 {
652 struct anv_instance *instance;
653 VkResult result;
654
655 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO);
656
657 struct anv_instance_extension_table enabled_extensions = {};
658 for (uint32_t i = 0; i < pCreateInfo->enabledExtensionCount; i++) {
659 int idx;
660 for (idx = 0; idx < ANV_INSTANCE_EXTENSION_COUNT; idx++) {
661 if (strcmp(pCreateInfo->ppEnabledExtensionNames[i],
662 anv_instance_extensions[idx].extensionName) == 0)
663 break;
664 }
665
666 if (idx >= ANV_INSTANCE_EXTENSION_COUNT)
667 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);
668
669 if (!anv_instance_extensions_supported.extensions[idx])
670 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);
671
672 enabled_extensions.extensions[idx] = true;
673 }
674
675 instance = vk_alloc2(&default_alloc, pAllocator, sizeof(*instance), 8,
676 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
677 if (!instance)
678 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
679
680 vk_object_base_init(NULL, &instance->base, VK_OBJECT_TYPE_INSTANCE);
681
682 if (pAllocator)
683 instance->alloc = *pAllocator;
684 else
685 instance->alloc = default_alloc;
686
687 instance->app_info = (struct anv_app_info) { .api_version = 0 };
688 if (pCreateInfo->pApplicationInfo) {
689 const VkApplicationInfo *app = pCreateInfo->pApplicationInfo;
690
691 instance->app_info.app_name =
692 vk_strdup(&instance->alloc, app->pApplicationName,
693 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
694 instance->app_info.app_version = app->applicationVersion;
695
696 instance->app_info.engine_name =
697 vk_strdup(&instance->alloc, app->pEngineName,
698 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
699 instance->app_info.engine_version = app->engineVersion;
700
701 instance->app_info.api_version = app->apiVersion;
702 }
703
704 if (instance->app_info.api_version == 0)
705 instance->app_info.api_version = VK_API_VERSION_1_0;
706
707 instance->enabled_extensions = enabled_extensions;
708
709 for (unsigned i = 0; i < ARRAY_SIZE(instance->dispatch.entrypoints); i++) {
710 /* Vulkan requires that entrypoints for extensions which have not been
711 * enabled must not be advertised.
712 */
713 if (!anv_instance_entrypoint_is_enabled(i, instance->app_info.api_version,
714 &instance->enabled_extensions)) {
715 instance->dispatch.entrypoints[i] = NULL;
716 } else {
717 instance->dispatch.entrypoints[i] =
718 anv_instance_dispatch_table.entrypoints[i];
719 }
720 }
721
722 for (unsigned i = 0; i < ARRAY_SIZE(instance->physical_device_dispatch.entrypoints); i++) {
723 /* Vulkan requires that entrypoints for extensions which have not been
724 * enabled must not be advertised.
725 */
726 if (!anv_physical_device_entrypoint_is_enabled(i, instance->app_info.api_version,
727 &instance->enabled_extensions)) {
728 instance->physical_device_dispatch.entrypoints[i] = NULL;
729 } else {
730 instance->physical_device_dispatch.entrypoints[i] =
731 anv_physical_device_dispatch_table.entrypoints[i];
732 }
733 }
734
735 for (unsigned i = 0; i < ARRAY_SIZE(instance->device_dispatch.entrypoints); i++) {
736 /* Vulkan requires that entrypoints for extensions which have not been
737 * enabled must not be advertised.
738 */
739 if (!anv_device_entrypoint_is_enabled(i, instance->app_info.api_version,
740 &instance->enabled_extensions, NULL)) {
741 instance->device_dispatch.entrypoints[i] = NULL;
742 } else {
743 instance->device_dispatch.entrypoints[i] =
744 anv_device_dispatch_table.entrypoints[i];
745 }
746 }
747
748 instance->physical_devices_enumerated = false;
749 list_inithead(&instance->physical_devices);
750
751 result = vk_debug_report_instance_init(&instance->debug_report_callbacks);
752 if (result != VK_SUCCESS) {
753 vk_free2(&default_alloc, pAllocator, instance);
754 return vk_error(result);
755 }
756
757 instance->pipeline_cache_enabled =
758 env_var_as_boolean("ANV_ENABLE_PIPELINE_CACHE", true);
759
760 glsl_type_singleton_init_or_ref();
761
762 VG(VALGRIND_CREATE_MEMPOOL(instance, 0, false));
763
764 driParseOptionInfo(&instance->available_dri_options, anv_dri_options_xml);
765 driParseConfigFiles(&instance->dri_options, &instance->available_dri_options,
766 0, "anv", NULL,
767 instance->app_info.engine_name,
768 instance->app_info.engine_version);
769
770 *pInstance = anv_instance_to_handle(instance);
771
772 return VK_SUCCESS;
773 }
774
775 void anv_DestroyInstance(
776 VkInstance _instance,
777 const VkAllocationCallbacks* pAllocator)
778 {
779 ANV_FROM_HANDLE(anv_instance, instance, _instance);
780
781 if (!instance)
782 return;
783
784 list_for_each_entry_safe(struct anv_physical_device, pdevice,
785 &instance->physical_devices, link)
786 anv_physical_device_destroy(pdevice);
787
788 vk_free(&instance->alloc, (char *)instance->app_info.app_name);
789 vk_free(&instance->alloc, (char *)instance->app_info.engine_name);
790
791 VG(VALGRIND_DESTROY_MEMPOOL(instance));
792
793 vk_debug_report_instance_destroy(&instance->debug_report_callbacks);
794
795 glsl_type_singleton_decref();
796
797 driDestroyOptionCache(&instance->dri_options);
798 driDestroyOptionInfo(&instance->available_dri_options);
799
800 vk_object_base_finish(&instance->base);
801 vk_free(&instance->alloc, instance);
802 }
803
804 static VkResult
805 anv_enumerate_physical_devices(struct anv_instance *instance)
806 {
807 if (instance->physical_devices_enumerated)
808 return VK_SUCCESS;
809
810 instance->physical_devices_enumerated = true;
811
812 /* TODO: Check for more devices ? */
813 drmDevicePtr devices[8];
814 int max_devices;
815
816 max_devices = drmGetDevices2(0, devices, ARRAY_SIZE(devices));
817 if (max_devices < 1)
818 return VK_SUCCESS;
819
820 VkResult result = VK_SUCCESS;
821 for (unsigned i = 0; i < (unsigned)max_devices; i++) {
822 if (devices[i]->available_nodes & 1 << DRM_NODE_RENDER &&
823 devices[i]->bustype == DRM_BUS_PCI &&
824 devices[i]->deviceinfo.pci->vendor_id == 0x8086) {
825
826 struct anv_physical_device *pdevice;
827 result = anv_physical_device_try_create(instance, devices[i],
828 &pdevice);
829 /* Incompatible DRM device, skip. */
830 if (result == VK_ERROR_INCOMPATIBLE_DRIVER) {
831 result = VK_SUCCESS;
832 continue;
833 }
834
835 /* Error creating the physical device, report the error. */
836 if (result != VK_SUCCESS)
837 break;
838
839 list_addtail(&pdevice->link, &instance->physical_devices);
840 }
841 }
842 drmFreeDevices(devices, max_devices);
843
844 /* If we successfully enumerated any devices, call it success */
845 return result;
846 }
847
848 VkResult anv_EnumeratePhysicalDevices(
849 VkInstance _instance,
850 uint32_t* pPhysicalDeviceCount,
851 VkPhysicalDevice* pPhysicalDevices)
852 {
853 ANV_FROM_HANDLE(anv_instance, instance, _instance);
854 VK_OUTARRAY_MAKE(out, pPhysicalDevices, pPhysicalDeviceCount);
855
856 VkResult result = anv_enumerate_physical_devices(instance);
857 if (result != VK_SUCCESS)
858 return result;
859
860 list_for_each_entry(struct anv_physical_device, pdevice,
861 &instance->physical_devices, link) {
862 vk_outarray_append(&out, i) {
863 *i = anv_physical_device_to_handle(pdevice);
864 }
865 }
866
867 return vk_outarray_status(&out);
868 }
869
870 VkResult anv_EnumeratePhysicalDeviceGroups(
871 VkInstance _instance,
872 uint32_t* pPhysicalDeviceGroupCount,
873 VkPhysicalDeviceGroupProperties* pPhysicalDeviceGroupProperties)
874 {
875 ANV_FROM_HANDLE(anv_instance, instance, _instance);
876 VK_OUTARRAY_MAKE(out, pPhysicalDeviceGroupProperties,
877 pPhysicalDeviceGroupCount);
878
879 VkResult result = anv_enumerate_physical_devices(instance);
880 if (result != VK_SUCCESS)
881 return result;
882
883 list_for_each_entry(struct anv_physical_device, pdevice,
884 &instance->physical_devices, link) {
885 vk_outarray_append(&out, p) {
886 p->physicalDeviceCount = 1;
887 memset(p->physicalDevices, 0, sizeof(p->physicalDevices));
888 p->physicalDevices[0] = anv_physical_device_to_handle(pdevice);
889 p->subsetAllocation = false;
890
891 vk_foreach_struct(ext, p->pNext)
892 anv_debug_ignored_stype(ext->sType);
893 }
894 }
895
896 return vk_outarray_status(&out);
897 }
898
899 void anv_GetPhysicalDeviceFeatures(
900 VkPhysicalDevice physicalDevice,
901 VkPhysicalDeviceFeatures* pFeatures)
902 {
903 ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
904
905 *pFeatures = (VkPhysicalDeviceFeatures) {
906 .robustBufferAccess = true,
907 .fullDrawIndexUint32 = true,
908 .imageCubeArray = true,
909 .independentBlend = true,
910 .geometryShader = true,
911 .tessellationShader = true,
912 .sampleRateShading = true,
913 .dualSrcBlend = true,
914 .logicOp = true,
915 .multiDrawIndirect = true,
916 .drawIndirectFirstInstance = true,
917 .depthClamp = true,
918 .depthBiasClamp = true,
919 .fillModeNonSolid = true,
920 .depthBounds = pdevice->info.gen >= 12,
921 .wideLines = true,
922 .largePoints = true,
923 .alphaToOne = true,
924 .multiViewport = true,
925 .samplerAnisotropy = true,
926 .textureCompressionETC2 = pdevice->info.gen >= 8 ||
927 pdevice->info.is_baytrail,
928 .textureCompressionASTC_LDR = pdevice->info.gen >= 9, /* FINISHME CHV */
929 .textureCompressionBC = true,
930 .occlusionQueryPrecise = true,
931 .pipelineStatisticsQuery = true,
932 .fragmentStoresAndAtomics = true,
933 .shaderTessellationAndGeometryPointSize = true,
934 .shaderImageGatherExtended = true,
935 .shaderStorageImageExtendedFormats = true,
936 .shaderStorageImageMultisample = false,
937 .shaderStorageImageReadWithoutFormat = false,
938 .shaderStorageImageWriteWithoutFormat = true,
939 .shaderUniformBufferArrayDynamicIndexing = true,
940 .shaderSampledImageArrayDynamicIndexing = true,
941 .shaderStorageBufferArrayDynamicIndexing = true,
942 .shaderStorageImageArrayDynamicIndexing = true,
943 .shaderClipDistance = true,
944 .shaderCullDistance = true,
945 .shaderFloat64 = pdevice->info.gen >= 8 &&
946 pdevice->info.has_64bit_float,
947 .shaderInt64 = pdevice->info.gen >= 8 &&
948 pdevice->info.has_64bit_int,
949 .shaderInt16 = pdevice->info.gen >= 8,
950 .shaderResourceMinLod = pdevice->info.gen >= 9,
951 .variableMultisampleRate = true,
952 .inheritedQueries = true,
953 };
954
955 /* We can't do image stores in vec4 shaders */
956 pFeatures->vertexPipelineStoresAndAtomics =
957 pdevice->compiler->scalar_stage[MESA_SHADER_VERTEX] &&
958 pdevice->compiler->scalar_stage[MESA_SHADER_GEOMETRY];
959
960 struct anv_app_info *app_info = &pdevice->instance->app_info;
961
962 /* The new DOOM and Wolfenstein games require depthBounds without
963 * checking for it. They seem to run fine without it so just claim it's
964 * there and accept the consequences.
965 */
966 if (app_info->engine_name && strcmp(app_info->engine_name, "idTech") == 0)
967 pFeatures->depthBounds = true;
968 }
969
970 static void
971 anv_get_physical_device_features_1_1(struct anv_physical_device *pdevice,
972 VkPhysicalDeviceVulkan11Features *f)
973 {
974 assert(f->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES);
975
976 f->storageBuffer16BitAccess = pdevice->info.gen >= 8;
977 f->uniformAndStorageBuffer16BitAccess = pdevice->info.gen >= 8;
978 f->storagePushConstant16 = pdevice->info.gen >= 8;
979 f->storageInputOutput16 = false;
980 f->multiview = true;
981 f->multiviewGeometryShader = true;
982 f->multiviewTessellationShader = true;
983 f->variablePointersStorageBuffer = true;
984 f->variablePointers = true;
985 f->protectedMemory = false;
986 f->samplerYcbcrConversion = true;
987 f->shaderDrawParameters = true;
988 }
989
990 static void
991 anv_get_physical_device_features_1_2(struct anv_physical_device *pdevice,
992 VkPhysicalDeviceVulkan12Features *f)
993 {
994 assert(f->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES);
995
996 f->samplerMirrorClampToEdge = true;
997 f->drawIndirectCount = true;
998 f->storageBuffer8BitAccess = pdevice->info.gen >= 8;
999 f->uniformAndStorageBuffer8BitAccess = pdevice->info.gen >= 8;
1000 f->storagePushConstant8 = pdevice->info.gen >= 8;
1001 f->shaderBufferInt64Atomics = pdevice->info.gen >= 9 &&
1002 pdevice->use_softpin;
1003 f->shaderSharedInt64Atomics = false;
1004 f->shaderFloat16 = pdevice->info.gen >= 8;
1005 f->shaderInt8 = pdevice->info.gen >= 8;
1006
1007 bool descIndexing = pdevice->has_a64_buffer_access &&
1008 pdevice->has_bindless_images;
1009 f->descriptorIndexing = descIndexing;
1010 f->shaderInputAttachmentArrayDynamicIndexing = false;
1011 f->shaderUniformTexelBufferArrayDynamicIndexing = descIndexing;
1012 f->shaderStorageTexelBufferArrayDynamicIndexing = descIndexing;
1013 f->shaderUniformBufferArrayNonUniformIndexing = false;
1014 f->shaderSampledImageArrayNonUniformIndexing = descIndexing;
1015 f->shaderStorageBufferArrayNonUniformIndexing = descIndexing;
1016 f->shaderStorageImageArrayNonUniformIndexing = descIndexing;
1017 f->shaderInputAttachmentArrayNonUniformIndexing = false;
1018 f->shaderUniformTexelBufferArrayNonUniformIndexing = descIndexing;
1019 f->shaderStorageTexelBufferArrayNonUniformIndexing = descIndexing;
1020 f->descriptorBindingUniformBufferUpdateAfterBind = false;
1021 f->descriptorBindingSampledImageUpdateAfterBind = descIndexing;
1022 f->descriptorBindingStorageImageUpdateAfterBind = descIndexing;
1023 f->descriptorBindingStorageBufferUpdateAfterBind = descIndexing;
1024 f->descriptorBindingUniformTexelBufferUpdateAfterBind = descIndexing;
1025 f->descriptorBindingStorageTexelBufferUpdateAfterBind = descIndexing;
1026 f->descriptorBindingUpdateUnusedWhilePending = descIndexing;
1027 f->descriptorBindingPartiallyBound = descIndexing;
1028 f->descriptorBindingVariableDescriptorCount = false;
1029 f->runtimeDescriptorArray = descIndexing;
1030
1031 f->samplerFilterMinmax = pdevice->info.gen >= 9;
1032 f->scalarBlockLayout = true;
1033 f->imagelessFramebuffer = true;
1034 f->uniformBufferStandardLayout = true;
1035 f->shaderSubgroupExtendedTypes = true;
1036 f->separateDepthStencilLayouts = true;
1037 f->hostQueryReset = true;
1038 f->timelineSemaphore = true;
1039 f->bufferDeviceAddress = pdevice->has_a64_buffer_access;
1040 f->bufferDeviceAddressCaptureReplay = pdevice->has_a64_buffer_access;
1041 f->bufferDeviceAddressMultiDevice = false;
1042 f->vulkanMemoryModel = true;
1043 f->vulkanMemoryModelDeviceScope = true;
1044 f->vulkanMemoryModelAvailabilityVisibilityChains = true;
1045 f->shaderOutputViewportIndex = true;
1046 f->shaderOutputLayer = true;
1047 f->subgroupBroadcastDynamicId = true;
1048 }
1049
1050 void anv_GetPhysicalDeviceFeatures2(
1051 VkPhysicalDevice physicalDevice,
1052 VkPhysicalDeviceFeatures2* pFeatures)
1053 {
1054 ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
1055 anv_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features);
1056
1057 VkPhysicalDeviceVulkan11Features core_1_1 = {
1058 .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES,
1059 };
1060 anv_get_physical_device_features_1_1(pdevice, &core_1_1);
1061
1062 VkPhysicalDeviceVulkan12Features core_1_2 = {
1063 .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES,
1064 };
1065 anv_get_physical_device_features_1_2(pdevice, &core_1_2);
1066
1067 #define CORE_FEATURE(major, minor, feature) \
1068 features->feature = core_##major##_##minor.feature
1069
1070
1071 vk_foreach_struct(ext, pFeatures->pNext) {
1072 switch (ext->sType) {
1073 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_8BIT_STORAGE_FEATURES_KHR: {
1074 VkPhysicalDevice8BitStorageFeaturesKHR *features =
1075 (VkPhysicalDevice8BitStorageFeaturesKHR *)ext;
1076 CORE_FEATURE(1, 2, storageBuffer8BitAccess);
1077 CORE_FEATURE(1, 2, uniformAndStorageBuffer8BitAccess);
1078 CORE_FEATURE(1, 2, storagePushConstant8);
1079 break;
1080 }
1081
1082 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES: {
1083 VkPhysicalDevice16BitStorageFeatures *features =
1084 (VkPhysicalDevice16BitStorageFeatures *)ext;
1085 CORE_FEATURE(1, 1, storageBuffer16BitAccess);
1086 CORE_FEATURE(1, 1, uniformAndStorageBuffer16BitAccess);
1087 CORE_FEATURE(1, 1, storagePushConstant16);
1088 CORE_FEATURE(1, 1, storageInputOutput16);
1089 break;
1090 }
1091
1092 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES_EXT: {
1093 VkPhysicalDeviceBufferDeviceAddressFeaturesEXT *features = (void *)ext;
1094 features->bufferDeviceAddress = pdevice->has_a64_buffer_access;
1095 features->bufferDeviceAddressCaptureReplay = false;
1096 features->bufferDeviceAddressMultiDevice = false;
1097 break;
1098 }
1099
1100 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES_KHR: {
1101 VkPhysicalDeviceBufferDeviceAddressFeaturesKHR *features = (void *)ext;
1102 CORE_FEATURE(1, 2, bufferDeviceAddress);
1103 CORE_FEATURE(1, 2, bufferDeviceAddressCaptureReplay);
1104 CORE_FEATURE(1, 2, bufferDeviceAddressMultiDevice);
1105 break;
1106 }
1107
1108 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_COMPUTE_SHADER_DERIVATIVES_FEATURES_NV: {
1109 VkPhysicalDeviceComputeShaderDerivativesFeaturesNV *features =
1110 (VkPhysicalDeviceComputeShaderDerivativesFeaturesNV *)ext;
1111 features->computeDerivativeGroupQuads = true;
1112 features->computeDerivativeGroupLinear = true;
1113 break;
1114 }
1115
1116 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONDITIONAL_RENDERING_FEATURES_EXT: {
1117 VkPhysicalDeviceConditionalRenderingFeaturesEXT *features =
1118 (VkPhysicalDeviceConditionalRenderingFeaturesEXT*)ext;
1119 features->conditionalRendering = pdevice->info.gen >= 8 ||
1120 pdevice->info.is_haswell;
1121 features->inheritedConditionalRendering = pdevice->info.gen >= 8 ||
1122 pdevice->info.is_haswell;
1123 break;
1124 }
1125
1126 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT: {
1127 VkPhysicalDeviceCustomBorderColorFeaturesEXT *features =
1128 (VkPhysicalDeviceCustomBorderColorFeaturesEXT *)ext;
1129 features->customBorderColors = pdevice->info.gen >= 8;
1130 features->customBorderColorWithoutFormat = pdevice->info.gen >= 8;
1131 break;
1132 }
1133
1134 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_CLIP_ENABLE_FEATURES_EXT: {
1135 VkPhysicalDeviceDepthClipEnableFeaturesEXT *features =
1136 (VkPhysicalDeviceDepthClipEnableFeaturesEXT *)ext;
1137 features->depthClipEnable = true;
1138 break;
1139 }
1140
1141 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR: {
1142 VkPhysicalDeviceFloat16Int8FeaturesKHR *features = (void *)ext;
1143 CORE_FEATURE(1, 2, shaderFloat16);
1144 CORE_FEATURE(1, 2, shaderInt8);
1145 break;
1146 }
1147
1148 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FRAGMENT_SHADER_INTERLOCK_FEATURES_EXT: {
1149 VkPhysicalDeviceFragmentShaderInterlockFeaturesEXT *features =
1150 (VkPhysicalDeviceFragmentShaderInterlockFeaturesEXT *)ext;
1151 features->fragmentShaderSampleInterlock = pdevice->info.gen >= 9;
1152 features->fragmentShaderPixelInterlock = pdevice->info.gen >= 9;
1153 features->fragmentShaderShadingRateInterlock = false;
1154 break;
1155 }
1156
1157 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_HOST_QUERY_RESET_FEATURES_EXT: {
1158 VkPhysicalDeviceHostQueryResetFeaturesEXT *features =
1159 (VkPhysicalDeviceHostQueryResetFeaturesEXT *)ext;
1160 CORE_FEATURE(1, 2, hostQueryReset);
1161 break;
1162 }
1163
1164 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT: {
1165 VkPhysicalDeviceDescriptorIndexingFeaturesEXT *features =
1166 (VkPhysicalDeviceDescriptorIndexingFeaturesEXT *)ext;
1167 CORE_FEATURE(1, 2, shaderInputAttachmentArrayDynamicIndexing);
1168 CORE_FEATURE(1, 2, shaderUniformTexelBufferArrayDynamicIndexing);
1169 CORE_FEATURE(1, 2, shaderStorageTexelBufferArrayDynamicIndexing);
1170 CORE_FEATURE(1, 2, shaderUniformBufferArrayNonUniformIndexing);
1171 CORE_FEATURE(1, 2, shaderSampledImageArrayNonUniformIndexing);
1172 CORE_FEATURE(1, 2, shaderStorageBufferArrayNonUniformIndexing);
1173 CORE_FEATURE(1, 2, shaderStorageImageArrayNonUniformIndexing);
1174 CORE_FEATURE(1, 2, shaderInputAttachmentArrayNonUniformIndexing);
1175 CORE_FEATURE(1, 2, shaderUniformTexelBufferArrayNonUniformIndexing);
1176 CORE_FEATURE(1, 2, shaderStorageTexelBufferArrayNonUniformIndexing);
1177 CORE_FEATURE(1, 2, descriptorBindingUniformBufferUpdateAfterBind);
1178 CORE_FEATURE(1, 2, descriptorBindingSampledImageUpdateAfterBind);
1179 CORE_FEATURE(1, 2, descriptorBindingStorageImageUpdateAfterBind);
1180 CORE_FEATURE(1, 2, descriptorBindingStorageBufferUpdateAfterBind);
1181 CORE_FEATURE(1, 2, descriptorBindingUniformTexelBufferUpdateAfterBind);
1182 CORE_FEATURE(1, 2, descriptorBindingStorageTexelBufferUpdateAfterBind);
1183 CORE_FEATURE(1, 2, descriptorBindingUpdateUnusedWhilePending);
1184 CORE_FEATURE(1, 2, descriptorBindingPartiallyBound);
1185 CORE_FEATURE(1, 2, descriptorBindingVariableDescriptorCount);
1186 CORE_FEATURE(1, 2, runtimeDescriptorArray);
1187 break;
1188 }
1189
1190 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INDEX_TYPE_UINT8_FEATURES_EXT: {
1191 VkPhysicalDeviceIndexTypeUint8FeaturesEXT *features =
1192 (VkPhysicalDeviceIndexTypeUint8FeaturesEXT *)ext;
1193 features->indexTypeUint8 = true;
1194 break;
1195 }
1196
1197 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INLINE_UNIFORM_BLOCK_FEATURES_EXT: {
1198 VkPhysicalDeviceInlineUniformBlockFeaturesEXT *features =
1199 (VkPhysicalDeviceInlineUniformBlockFeaturesEXT *)ext;
1200 features->inlineUniformBlock = true;
1201 features->descriptorBindingInlineUniformBlockUpdateAfterBind = true;
1202 break;
1203 }
1204
1205 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_LINE_RASTERIZATION_FEATURES_EXT: {
1206 VkPhysicalDeviceLineRasterizationFeaturesEXT *features =
1207 (VkPhysicalDeviceLineRasterizationFeaturesEXT *)ext;
1208 features->rectangularLines = true;
1209 features->bresenhamLines = true;
1210 /* Support for Smooth lines with MSAA was removed on gen11. From the
1211 * BSpec section "Multisample ModesState" table for "AA Line Support
1212 * Requirements":
1213 *
1214 * GEN10:BUG:######## NUM_MULTISAMPLES == 1
1215 *
1216 * Fortunately, this isn't a case most people care about.
1217 */
1218 features->smoothLines = pdevice->info.gen < 10;
1219 features->stippledRectangularLines = false;
1220 features->stippledBresenhamLines = true;
1221 features->stippledSmoothLines = false;
1222 break;
1223 }
1224
1225 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES: {
1226 VkPhysicalDeviceMultiviewFeatures *features =
1227 (VkPhysicalDeviceMultiviewFeatures *)ext;
1228 CORE_FEATURE(1, 1, multiview);
1229 CORE_FEATURE(1, 1, multiviewGeometryShader);
1230 CORE_FEATURE(1, 1, multiviewTessellationShader);
1231 break;
1232 }
1233
1234 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_IMAGELESS_FRAMEBUFFER_FEATURES_KHR: {
1235 VkPhysicalDeviceImagelessFramebufferFeaturesKHR *features =
1236 (VkPhysicalDeviceImagelessFramebufferFeaturesKHR *)ext;
1237 CORE_FEATURE(1, 2, imagelessFramebuffer);
1238 break;
1239 }
1240
1241 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PIPELINE_EXECUTABLE_PROPERTIES_FEATURES_KHR: {
1242 VkPhysicalDevicePipelineExecutablePropertiesFeaturesKHR *features =
1243 (VkPhysicalDevicePipelineExecutablePropertiesFeaturesKHR *)ext;
1244 features->pipelineExecutableInfo = true;
1245 break;
1246 }
1247
1248 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRIVATE_DATA_FEATURES_EXT: {
1249 VkPhysicalDevicePrivateDataFeaturesEXT *features = (void *)ext;
1250 features->privateData = true;
1251 break;
1252 }
1253
1254 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES: {
1255 VkPhysicalDeviceProtectedMemoryFeatures *features = (void *)ext;
1256 CORE_FEATURE(1, 1, protectedMemory);
1257 break;
1258 }
1259
1260 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_FEATURES_EXT: {
1261 VkPhysicalDeviceRobustness2FeaturesEXT *features = (void *)ext;
1262 features->robustBufferAccess2 = true;
1263 features->robustImageAccess2 = true;
1264 features->nullDescriptor = true;
1265 break;
1266 }
1267
1268 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES: {
1269 VkPhysicalDeviceSamplerYcbcrConversionFeatures *features =
1270 (VkPhysicalDeviceSamplerYcbcrConversionFeatures *) ext;
1271 CORE_FEATURE(1, 1, samplerYcbcrConversion);
1272 break;
1273 }
1274
1275 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SCALAR_BLOCK_LAYOUT_FEATURES_EXT: {
1276 VkPhysicalDeviceScalarBlockLayoutFeaturesEXT *features =
1277 (VkPhysicalDeviceScalarBlockLayoutFeaturesEXT *)ext;
1278 CORE_FEATURE(1, 2, scalarBlockLayout);
1279 break;
1280 }
1281
1282 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SEPARATE_DEPTH_STENCIL_LAYOUTS_FEATURES_KHR: {
1283 VkPhysicalDeviceSeparateDepthStencilLayoutsFeaturesKHR *features =
1284 (VkPhysicalDeviceSeparateDepthStencilLayoutsFeaturesKHR *)ext;
1285 CORE_FEATURE(1, 2, separateDepthStencilLayouts);
1286 break;
1287 }
1288
1289 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_INT64_FEATURES_KHR: {
1290 VkPhysicalDeviceShaderAtomicInt64FeaturesKHR *features = (void *)ext;
1291 CORE_FEATURE(1, 2, shaderBufferInt64Atomics);
1292 CORE_FEATURE(1, 2, shaderSharedInt64Atomics);
1293 break;
1294 }
1295
1296 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DEMOTE_TO_HELPER_INVOCATION_FEATURES_EXT: {
1297 VkPhysicalDeviceShaderDemoteToHelperInvocationFeaturesEXT *features = (void *)ext;
1298 features->shaderDemoteToHelperInvocation = true;
1299 break;
1300 }
1301
1302 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_CLOCK_FEATURES_KHR: {
1303 VkPhysicalDeviceShaderClockFeaturesKHR *features =
1304 (VkPhysicalDeviceShaderClockFeaturesKHR *)ext;
1305 features->shaderSubgroupClock = true;
1306 features->shaderDeviceClock = false;
1307 break;
1308 }
1309
1310 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETERS_FEATURES: {
1311 VkPhysicalDeviceShaderDrawParametersFeatures *features = (void *)ext;
1312 CORE_FEATURE(1, 1, shaderDrawParameters);
1313 break;
1314 }
1315
1316 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_SUBGROUP_EXTENDED_TYPES_FEATURES_KHR: {
1317 VkPhysicalDeviceShaderSubgroupExtendedTypesFeaturesKHR *features =
1318 (VkPhysicalDeviceShaderSubgroupExtendedTypesFeaturesKHR *)ext;
1319 CORE_FEATURE(1, 2, shaderSubgroupExtendedTypes);
1320 break;
1321 }
1322
1323 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_FEATURES_EXT: {
1324 VkPhysicalDeviceSubgroupSizeControlFeaturesEXT *features =
1325 (VkPhysicalDeviceSubgroupSizeControlFeaturesEXT *)ext;
1326 features->subgroupSizeControl = true;
1327 features->computeFullSubgroups = true;
1328 break;
1329 }
1330
1331 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TEXEL_BUFFER_ALIGNMENT_FEATURES_EXT: {
1332 VkPhysicalDeviceTexelBufferAlignmentFeaturesEXT *features =
1333 (VkPhysicalDeviceTexelBufferAlignmentFeaturesEXT *)ext;
1334 features->texelBufferAlignment = true;
1335 break;
1336 }
1337
1338 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES_KHR: {
1339 VkPhysicalDeviceTimelineSemaphoreFeaturesKHR *features =
1340 (VkPhysicalDeviceTimelineSemaphoreFeaturesKHR *) ext;
1341 CORE_FEATURE(1, 2, timelineSemaphore);
1342 break;
1343 }
1344
1345 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTERS_FEATURES: {
1346 VkPhysicalDeviceVariablePointersFeatures *features = (void *)ext;
1347 CORE_FEATURE(1, 1, variablePointersStorageBuffer);
1348 CORE_FEATURE(1, 1, variablePointers);
1349 break;
1350 }
1351
1352 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT: {
1353 VkPhysicalDeviceTransformFeedbackFeaturesEXT *features =
1354 (VkPhysicalDeviceTransformFeedbackFeaturesEXT *)ext;
1355 features->transformFeedback = true;
1356 features->geometryStreams = true;
1357 break;
1358 }
1359
1360 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_UNIFORM_BUFFER_STANDARD_LAYOUT_FEATURES_KHR: {
1361 VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR *features =
1362 (VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR *)ext;
1363 CORE_FEATURE(1, 2, uniformBufferStandardLayout);
1364 break;
1365 }
1366
1367 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_FEATURES_EXT: {
1368 VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT *features =
1369 (VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT *)ext;
1370 features->vertexAttributeInstanceRateDivisor = true;
1371 features->vertexAttributeInstanceRateZeroDivisor = true;
1372 break;
1373 }
1374
1375 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES:
1376 anv_get_physical_device_features_1_1(pdevice, (void *)ext);
1377 break;
1378
1379 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES:
1380 anv_get_physical_device_features_1_2(pdevice, (void *)ext);
1381 break;
1382
1383 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_MEMORY_MODEL_FEATURES_KHR: {
1384 VkPhysicalDeviceVulkanMemoryModelFeaturesKHR *features = (void *)ext;
1385 CORE_FEATURE(1, 2, vulkanMemoryModel);
1386 CORE_FEATURE(1, 2, vulkanMemoryModelDeviceScope);
1387 CORE_FEATURE(1, 2, vulkanMemoryModelAvailabilityVisibilityChains);
1388 break;
1389 }
1390
1391 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_YCBCR_IMAGE_ARRAYS_FEATURES_EXT: {
1392 VkPhysicalDeviceYcbcrImageArraysFeaturesEXT *features =
1393 (VkPhysicalDeviceYcbcrImageArraysFeaturesEXT *)ext;
1394 features->ycbcrImageArrays = true;
1395 break;
1396 }
1397
1398 default:
1399 anv_debug_ignored_stype(ext->sType);
1400 break;
1401 }
1402 }
1403
1404 #undef CORE_FEATURE
1405 }
1406
1407 #define MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS 64
1408
1409 #define MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS 64
1410 #define MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS 256
1411
1412 #define MAX_CUSTOM_BORDER_COLORS 4096
1413
1414 void anv_GetPhysicalDeviceProperties(
1415 VkPhysicalDevice physicalDevice,
1416 VkPhysicalDeviceProperties* pProperties)
1417 {
1418 ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
1419 const struct gen_device_info *devinfo = &pdevice->info;
1420
1421 /* See assertions made when programming the buffer surface state. */
1422 const uint32_t max_raw_buffer_sz = devinfo->gen >= 7 ?
1423 (1ul << 30) : (1ul << 27);
1424
1425 const uint32_t max_ssbos = pdevice->has_a64_buffer_access ? UINT16_MAX : 64;
1426 const uint32_t max_textures =
1427 pdevice->has_bindless_images ? UINT16_MAX : 128;
1428 const uint32_t max_samplers =
1429 pdevice->has_bindless_samplers ? UINT16_MAX :
1430 (devinfo->gen >= 8 || devinfo->is_haswell) ? 128 : 16;
1431 const uint32_t max_images =
1432 pdevice->has_bindless_images ? UINT16_MAX : MAX_IMAGES;
1433
1434 /* If we can use bindless for everything, claim a high per-stage limit,
1435 * otherwise use the binding table size, minus the slots reserved for
1436 * render targets and one slot for the descriptor buffer. */
1437 const uint32_t max_per_stage =
1438 pdevice->has_bindless_images && pdevice->has_a64_buffer_access
1439 ? UINT32_MAX : MAX_BINDING_TABLE_SIZE - MAX_RTS - 1;
1440
1441 /* Limit max_threads to 64 for the GPGPU_WALKER command */
1442 const uint32_t max_workgroup_size = 32 * MIN2(64, devinfo->max_cs_threads);
1443
1444 VkSampleCountFlags sample_counts =
1445 isl_device_get_sample_counts(&pdevice->isl_dev);
1446
1447
1448 VkPhysicalDeviceLimits limits = {
1449 .maxImageDimension1D = (1 << 14),
1450 .maxImageDimension2D = (1 << 14),
1451 .maxImageDimension3D = (1 << 11),
1452 .maxImageDimensionCube = (1 << 14),
1453 .maxImageArrayLayers = (1 << 11),
1454 .maxTexelBufferElements = 128 * 1024 * 1024,
1455 .maxUniformBufferRange = (1ul << 27),
1456 .maxStorageBufferRange = max_raw_buffer_sz,
1457 .maxPushConstantsSize = MAX_PUSH_CONSTANTS_SIZE,
1458 .maxMemoryAllocationCount = UINT32_MAX,
1459 .maxSamplerAllocationCount = 64 * 1024,
1460 .bufferImageGranularity = 64, /* A cache line */
1461 .sparseAddressSpaceSize = 0,
1462 .maxBoundDescriptorSets = MAX_SETS,
1463 .maxPerStageDescriptorSamplers = max_samplers,
1464 .maxPerStageDescriptorUniformBuffers = MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS,
1465 .maxPerStageDescriptorStorageBuffers = max_ssbos,
1466 .maxPerStageDescriptorSampledImages = max_textures,
1467 .maxPerStageDescriptorStorageImages = max_images,
1468 .maxPerStageDescriptorInputAttachments = MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS,
1469 .maxPerStageResources = max_per_stage,
1470 .maxDescriptorSetSamplers = 6 * max_samplers, /* number of stages * maxPerStageDescriptorSamplers */
1471 .maxDescriptorSetUniformBuffers = 6 * MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS, /* number of stages * maxPerStageDescriptorUniformBuffers */
1472 .maxDescriptorSetUniformBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2,
1473 .maxDescriptorSetStorageBuffers = 6 * max_ssbos, /* number of stages * maxPerStageDescriptorStorageBuffers */
1474 .maxDescriptorSetStorageBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2,
1475 .maxDescriptorSetSampledImages = 6 * max_textures, /* number of stages * maxPerStageDescriptorSampledImages */
1476 .maxDescriptorSetStorageImages = 6 * max_images, /* number of stages * maxPerStageDescriptorStorageImages */
1477 .maxDescriptorSetInputAttachments = MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS,
1478 .maxVertexInputAttributes = MAX_VBS,
1479 .maxVertexInputBindings = MAX_VBS,
1480 .maxVertexInputAttributeOffset = 2047,
1481 .maxVertexInputBindingStride = 2048,
1482 .maxVertexOutputComponents = 128,
1483 .maxTessellationGenerationLevel = 64,
1484 .maxTessellationPatchSize = 32,
1485 .maxTessellationControlPerVertexInputComponents = 128,
1486 .maxTessellationControlPerVertexOutputComponents = 128,
1487 .maxTessellationControlPerPatchOutputComponents = 128,
1488 .maxTessellationControlTotalOutputComponents = 2048,
1489 .maxTessellationEvaluationInputComponents = 128,
1490 .maxTessellationEvaluationOutputComponents = 128,
1491 .maxGeometryShaderInvocations = 32,
1492 .maxGeometryInputComponents = 64,
1493 .maxGeometryOutputComponents = 128,
1494 .maxGeometryOutputVertices = 256,
1495 .maxGeometryTotalOutputComponents = 1024,
1496 .maxFragmentInputComponents = 116, /* 128 components - (PSIZ, CLIP_DIST0, CLIP_DIST1) */
1497 .maxFragmentOutputAttachments = 8,
1498 .maxFragmentDualSrcAttachments = 1,
1499 .maxFragmentCombinedOutputResources = 8,
1500 .maxComputeSharedMemorySize = 64 * 1024,
1501 .maxComputeWorkGroupCount = { 65535, 65535, 65535 },
1502 .maxComputeWorkGroupInvocations = max_workgroup_size,
1503 .maxComputeWorkGroupSize = {
1504 max_workgroup_size,
1505 max_workgroup_size,
1506 max_workgroup_size,
1507 },
1508 .subPixelPrecisionBits = 8,
1509 .subTexelPrecisionBits = 8,
1510 .mipmapPrecisionBits = 8,
1511 .maxDrawIndexedIndexValue = UINT32_MAX,
1512 .maxDrawIndirectCount = UINT32_MAX,
1513 .maxSamplerLodBias = 16,
1514 .maxSamplerAnisotropy = 16,
1515 .maxViewports = MAX_VIEWPORTS,
1516 .maxViewportDimensions = { (1 << 14), (1 << 14) },
1517 .viewportBoundsRange = { INT16_MIN, INT16_MAX },
1518 .viewportSubPixelBits = 13, /* We take a float? */
1519 .minMemoryMapAlignment = 4096, /* A page */
1520 /* The dataport requires texel alignment so we need to assume a worst
1521 * case of R32G32B32A32 which is 16 bytes.
1522 */
1523 .minTexelBufferOffsetAlignment = 16,
1524 .minUniformBufferOffsetAlignment = ANV_UBO_ALIGNMENT,
1525 .minStorageBufferOffsetAlignment = 4,
1526 .minTexelOffset = -8,
1527 .maxTexelOffset = 7,
1528 .minTexelGatherOffset = -32,
1529 .maxTexelGatherOffset = 31,
1530 .minInterpolationOffset = -0.5,
1531 .maxInterpolationOffset = 0.4375,
1532 .subPixelInterpolationOffsetBits = 4,
1533 .maxFramebufferWidth = (1 << 14),
1534 .maxFramebufferHeight = (1 << 14),
1535 .maxFramebufferLayers = (1 << 11),
1536 .framebufferColorSampleCounts = sample_counts,
1537 .framebufferDepthSampleCounts = sample_counts,
1538 .framebufferStencilSampleCounts = sample_counts,
1539 .framebufferNoAttachmentsSampleCounts = sample_counts,
1540 .maxColorAttachments = MAX_RTS,
1541 .sampledImageColorSampleCounts = sample_counts,
1542 .sampledImageIntegerSampleCounts = sample_counts,
1543 .sampledImageDepthSampleCounts = sample_counts,
1544 .sampledImageStencilSampleCounts = sample_counts,
1545 .storageImageSampleCounts = VK_SAMPLE_COUNT_1_BIT,
1546 .maxSampleMaskWords = 1,
1547 .timestampComputeAndGraphics = true,
1548 .timestampPeriod = 1000000000.0 / devinfo->timestamp_frequency,
1549 .maxClipDistances = 8,
1550 .maxCullDistances = 8,
1551 .maxCombinedClipAndCullDistances = 8,
1552 .discreteQueuePriorities = 2,
1553 .pointSizeRange = { 0.125, 255.875 },
1554 .lineWidthRange = {
1555 0.0,
1556 (devinfo->gen >= 9 || devinfo->is_cherryview) ?
1557 2047.9921875 : 7.9921875,
1558 },
1559 .pointSizeGranularity = (1.0 / 8.0),
1560 .lineWidthGranularity = (1.0 / 128.0),
1561 .strictLines = false,
1562 .standardSampleLocations = true,
1563 .optimalBufferCopyOffsetAlignment = 128,
1564 .optimalBufferCopyRowPitchAlignment = 128,
1565 .nonCoherentAtomSize = 64,
1566 };
1567
1568 *pProperties = (VkPhysicalDeviceProperties) {
1569 .apiVersion = anv_physical_device_api_version(pdevice),
1570 .driverVersion = vk_get_driver_version(),
1571 .vendorID = 0x8086,
1572 .deviceID = pdevice->info.chipset_id,
1573 .deviceType = VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU,
1574 .limits = limits,
1575 .sparseProperties = {0}, /* Broadwell doesn't do sparse. */
1576 };
1577
1578 snprintf(pProperties->deviceName, sizeof(pProperties->deviceName),
1579 "%s", pdevice->name);
1580 memcpy(pProperties->pipelineCacheUUID,
1581 pdevice->pipeline_cache_uuid, VK_UUID_SIZE);
1582 }
1583
1584 static void
1585 anv_get_physical_device_properties_1_1(struct anv_physical_device *pdevice,
1586 VkPhysicalDeviceVulkan11Properties *p)
1587 {
1588 assert(p->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_PROPERTIES);
1589
1590 memcpy(p->deviceUUID, pdevice->device_uuid, VK_UUID_SIZE);
1591 memcpy(p->driverUUID, pdevice->driver_uuid, VK_UUID_SIZE);
1592 memset(p->deviceLUID, 0, VK_LUID_SIZE);
1593 p->deviceNodeMask = 0;
1594 p->deviceLUIDValid = false;
1595
1596 p->subgroupSize = BRW_SUBGROUP_SIZE;
1597 VkShaderStageFlags scalar_stages = 0;
1598 for (unsigned stage = 0; stage < MESA_SHADER_STAGES; stage++) {
1599 if (pdevice->compiler->scalar_stage[stage])
1600 scalar_stages |= mesa_to_vk_shader_stage(stage);
1601 }
1602 p->subgroupSupportedStages = scalar_stages;
1603 p->subgroupSupportedOperations = VK_SUBGROUP_FEATURE_BASIC_BIT |
1604 VK_SUBGROUP_FEATURE_VOTE_BIT |
1605 VK_SUBGROUP_FEATURE_BALLOT_BIT |
1606 VK_SUBGROUP_FEATURE_SHUFFLE_BIT |
1607 VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT |
1608 VK_SUBGROUP_FEATURE_QUAD_BIT;
1609 if (pdevice->info.gen >= 8) {
1610 /* TODO: There's no technical reason why these can't be made to
1611 * work on gen7 but they don't at the moment so it's best to leave
1612 * the feature disabled than enabled and broken.
1613 */
1614 p->subgroupSupportedOperations |= VK_SUBGROUP_FEATURE_ARITHMETIC_BIT |
1615 VK_SUBGROUP_FEATURE_CLUSTERED_BIT;
1616 }
1617 p->subgroupQuadOperationsInAllStages = pdevice->info.gen >= 8;
1618
1619 p->pointClippingBehavior = VK_POINT_CLIPPING_BEHAVIOR_USER_CLIP_PLANES_ONLY;
1620 p->maxMultiviewViewCount = 16;
1621 p->maxMultiviewInstanceIndex = UINT32_MAX / 16;
1622 p->protectedNoFault = false;
1623 /* This value doesn't matter for us today as our per-stage descriptors are
1624 * the real limit.
1625 */
1626 p->maxPerSetDescriptors = 1024;
1627 p->maxMemoryAllocationSize = MAX_MEMORY_ALLOCATION_SIZE;
1628 }
1629
1630 static void
1631 anv_get_physical_device_properties_1_2(struct anv_physical_device *pdevice,
1632 VkPhysicalDeviceVulkan12Properties *p)
1633 {
1634 assert(p->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_PROPERTIES);
1635
1636 p->driverID = VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA_KHR;
1637 memset(p->driverName, 0, sizeof(p->driverName));
1638 snprintf(p->driverName, VK_MAX_DRIVER_NAME_SIZE_KHR,
1639 "Intel open-source Mesa driver");
1640 memset(p->driverInfo, 0, sizeof(p->driverInfo));
1641 snprintf(p->driverInfo, VK_MAX_DRIVER_INFO_SIZE_KHR,
1642 "Mesa " PACKAGE_VERSION MESA_GIT_SHA1);
1643 p->conformanceVersion = (VkConformanceVersionKHR) {
1644 .major = 1,
1645 .minor = 2,
1646 .subminor = 0,
1647 .patch = 0,
1648 };
1649
1650 p->denormBehaviorIndependence =
1651 VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_ALL_KHR;
1652 p->roundingModeIndependence =
1653 VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_NONE_KHR;
1654
1655 /* Broadwell does not support HF denorms and there are restrictions
1656 * other gens. According to Kabylake's PRM:
1657 *
1658 * "math - Extended Math Function
1659 * [...]
1660 * Restriction : Half-float denorms are always retained."
1661 */
1662 p->shaderDenormFlushToZeroFloat16 = false;
1663 p->shaderDenormPreserveFloat16 = pdevice->info.gen > 8;
1664 p->shaderRoundingModeRTEFloat16 = true;
1665 p->shaderRoundingModeRTZFloat16 = true;
1666 p->shaderSignedZeroInfNanPreserveFloat16 = true;
1667
1668 p->shaderDenormFlushToZeroFloat32 = true;
1669 p->shaderDenormPreserveFloat32 = true;
1670 p->shaderRoundingModeRTEFloat32 = true;
1671 p->shaderRoundingModeRTZFloat32 = true;
1672 p->shaderSignedZeroInfNanPreserveFloat32 = true;
1673
1674 p->shaderDenormFlushToZeroFloat64 = true;
1675 p->shaderDenormPreserveFloat64 = true;
1676 p->shaderRoundingModeRTEFloat64 = true;
1677 p->shaderRoundingModeRTZFloat64 = true;
1678 p->shaderSignedZeroInfNanPreserveFloat64 = true;
1679
1680 /* It's a bit hard to exactly map our implementation to the limits
1681 * described here. The bindless surface handle in the extended
1682 * message descriptors is 20 bits and it's an index into the table of
1683 * RENDER_SURFACE_STATE structs that starts at bindless surface base
1684 * address. Given that most things consume two surface states per
1685 * view (general/sampled for textures and write-only/read-write for
1686 * images), we claim 2^19 things.
1687 *
1688 * For SSBOs, we just use A64 messages so there is no real limit
1689 * there beyond the limit on the total size of a descriptor set.
1690 */
1691 const unsigned max_bindless_views = 1 << 19;
1692 p->maxUpdateAfterBindDescriptorsInAllPools = max_bindless_views;
1693 p->shaderUniformBufferArrayNonUniformIndexingNative = false;
1694 p->shaderSampledImageArrayNonUniformIndexingNative = false;
1695 p->shaderStorageBufferArrayNonUniformIndexingNative = true;
1696 p->shaderStorageImageArrayNonUniformIndexingNative = false;
1697 p->shaderInputAttachmentArrayNonUniformIndexingNative = false;
1698 p->robustBufferAccessUpdateAfterBind = true;
1699 p->quadDivergentImplicitLod = false;
1700 p->maxPerStageDescriptorUpdateAfterBindSamplers = max_bindless_views;
1701 p->maxPerStageDescriptorUpdateAfterBindUniformBuffers = MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS;
1702 p->maxPerStageDescriptorUpdateAfterBindStorageBuffers = UINT32_MAX;
1703 p->maxPerStageDescriptorUpdateAfterBindSampledImages = max_bindless_views;
1704 p->maxPerStageDescriptorUpdateAfterBindStorageImages = max_bindless_views;
1705 p->maxPerStageDescriptorUpdateAfterBindInputAttachments = MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS;
1706 p->maxPerStageUpdateAfterBindResources = UINT32_MAX;
1707 p->maxDescriptorSetUpdateAfterBindSamplers = max_bindless_views;
1708 p->maxDescriptorSetUpdateAfterBindUniformBuffers = 6 * MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS;
1709 p->maxDescriptorSetUpdateAfterBindUniformBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2;
1710 p->maxDescriptorSetUpdateAfterBindStorageBuffers = UINT32_MAX;
1711 p->maxDescriptorSetUpdateAfterBindStorageBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2;
1712 p->maxDescriptorSetUpdateAfterBindSampledImages = max_bindless_views;
1713 p->maxDescriptorSetUpdateAfterBindStorageImages = max_bindless_views;
1714 p->maxDescriptorSetUpdateAfterBindInputAttachments = MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS;
1715
1716 /* We support all of the depth resolve modes */
1717 p->supportedDepthResolveModes = VK_RESOLVE_MODE_SAMPLE_ZERO_BIT_KHR |
1718 VK_RESOLVE_MODE_AVERAGE_BIT_KHR |
1719 VK_RESOLVE_MODE_MIN_BIT_KHR |
1720 VK_RESOLVE_MODE_MAX_BIT_KHR;
1721 /* Average doesn't make sense for stencil so we don't support that */
1722 p->supportedStencilResolveModes = VK_RESOLVE_MODE_SAMPLE_ZERO_BIT_KHR;
1723 if (pdevice->info.gen >= 8) {
1724 /* The advanced stencil resolve modes currently require stencil
1725 * sampling be supported by the hardware.
1726 */
1727 p->supportedStencilResolveModes |= VK_RESOLVE_MODE_MIN_BIT_KHR |
1728 VK_RESOLVE_MODE_MAX_BIT_KHR;
1729 }
1730 p->independentResolveNone = true;
1731 p->independentResolve = true;
1732
1733 p->filterMinmaxSingleComponentFormats = pdevice->info.gen >= 9;
1734 p->filterMinmaxImageComponentMapping = pdevice->info.gen >= 9;
1735
1736 p->maxTimelineSemaphoreValueDifference = UINT64_MAX;
1737
1738 p->framebufferIntegerColorSampleCounts =
1739 isl_device_get_sample_counts(&pdevice->isl_dev);
1740 }
1741
1742 void anv_GetPhysicalDeviceProperties2(
1743 VkPhysicalDevice physicalDevice,
1744 VkPhysicalDeviceProperties2* pProperties)
1745 {
1746 ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
1747
1748 anv_GetPhysicalDeviceProperties(physicalDevice, &pProperties->properties);
1749
1750 VkPhysicalDeviceVulkan11Properties core_1_1 = {
1751 .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_PROPERTIES,
1752 };
1753 anv_get_physical_device_properties_1_1(pdevice, &core_1_1);
1754
1755 VkPhysicalDeviceVulkan12Properties core_1_2 = {
1756 .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_PROPERTIES,
1757 };
1758 anv_get_physical_device_properties_1_2(pdevice, &core_1_2);
1759
1760 #define CORE_RENAMED_PROPERTY(major, minor, ext_property, core_property) \
1761 memcpy(&properties->ext_property, &core_##major##_##minor.core_property, \
1762 sizeof(core_##major##_##minor.core_property))
1763
1764 #define CORE_PROPERTY(major, minor, property) \
1765 CORE_RENAMED_PROPERTY(major, minor, property, property)
1766
1767 vk_foreach_struct(ext, pProperties->pNext) {
1768 switch (ext->sType) {
1769 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_PROPERTIES_EXT: {
1770 VkPhysicalDeviceCustomBorderColorPropertiesEXT *properties =
1771 (VkPhysicalDeviceCustomBorderColorPropertiesEXT *)ext;
1772 properties->maxCustomBorderColorSamplers = MAX_CUSTOM_BORDER_COLORS;
1773 break;
1774 }
1775
1776 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_STENCIL_RESOLVE_PROPERTIES_KHR: {
1777 VkPhysicalDeviceDepthStencilResolvePropertiesKHR *properties =
1778 (VkPhysicalDeviceDepthStencilResolvePropertiesKHR *)ext;
1779 CORE_PROPERTY(1, 2, supportedDepthResolveModes);
1780 CORE_PROPERTY(1, 2, supportedStencilResolveModes);
1781 CORE_PROPERTY(1, 2, independentResolveNone);
1782 CORE_PROPERTY(1, 2, independentResolve);
1783 break;
1784 }
1785
1786 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_PROPERTIES_EXT: {
1787 VkPhysicalDeviceDescriptorIndexingPropertiesEXT *properties =
1788 (VkPhysicalDeviceDescriptorIndexingPropertiesEXT *)ext;
1789 CORE_PROPERTY(1, 2, maxUpdateAfterBindDescriptorsInAllPools);
1790 CORE_PROPERTY(1, 2, shaderUniformBufferArrayNonUniformIndexingNative);
1791 CORE_PROPERTY(1, 2, shaderSampledImageArrayNonUniformIndexingNative);
1792 CORE_PROPERTY(1, 2, shaderStorageBufferArrayNonUniformIndexingNative);
1793 CORE_PROPERTY(1, 2, shaderStorageImageArrayNonUniformIndexingNative);
1794 CORE_PROPERTY(1, 2, shaderInputAttachmentArrayNonUniformIndexingNative);
1795 CORE_PROPERTY(1, 2, robustBufferAccessUpdateAfterBind);
1796 CORE_PROPERTY(1, 2, quadDivergentImplicitLod);
1797 CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindSamplers);
1798 CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindUniformBuffers);
1799 CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindStorageBuffers);
1800 CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindSampledImages);
1801 CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindStorageImages);
1802 CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindInputAttachments);
1803 CORE_PROPERTY(1, 2, maxPerStageUpdateAfterBindResources);
1804 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindSamplers);
1805 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindUniformBuffers);
1806 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindUniformBuffersDynamic);
1807 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindStorageBuffers);
1808 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindStorageBuffersDynamic);
1809 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindSampledImages);
1810 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindStorageImages);
1811 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindInputAttachments);
1812 break;
1813 }
1814
1815 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES_KHR: {
1816 VkPhysicalDeviceDriverPropertiesKHR *properties =
1817 (VkPhysicalDeviceDriverPropertiesKHR *) ext;
1818 CORE_PROPERTY(1, 2, driverID);
1819 CORE_PROPERTY(1, 2, driverName);
1820 CORE_PROPERTY(1, 2, driverInfo);
1821 CORE_PROPERTY(1, 2, conformanceVersion);
1822 break;
1823 }
1824
1825 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_MEMORY_HOST_PROPERTIES_EXT: {
1826 VkPhysicalDeviceExternalMemoryHostPropertiesEXT *props =
1827 (VkPhysicalDeviceExternalMemoryHostPropertiesEXT *) ext;
1828 /* Userptr needs page aligned memory. */
1829 props->minImportedHostPointerAlignment = 4096;
1830 break;
1831 }
1832
1833 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES: {
1834 VkPhysicalDeviceIDProperties *properties =
1835 (VkPhysicalDeviceIDProperties *)ext;
1836 CORE_PROPERTY(1, 1, deviceUUID);
1837 CORE_PROPERTY(1, 1, driverUUID);
1838 CORE_PROPERTY(1, 1, deviceLUID);
1839 CORE_PROPERTY(1, 1, deviceLUIDValid);
1840 break;
1841 }
1842
1843 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INLINE_UNIFORM_BLOCK_PROPERTIES_EXT: {
1844 VkPhysicalDeviceInlineUniformBlockPropertiesEXT *props =
1845 (VkPhysicalDeviceInlineUniformBlockPropertiesEXT *)ext;
1846 props->maxInlineUniformBlockSize = MAX_INLINE_UNIFORM_BLOCK_SIZE;
1847 props->maxPerStageDescriptorInlineUniformBlocks =
1848 MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
1849 props->maxPerStageDescriptorUpdateAfterBindInlineUniformBlocks =
1850 MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
1851 props->maxDescriptorSetInlineUniformBlocks =
1852 MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
1853 props->maxDescriptorSetUpdateAfterBindInlineUniformBlocks =
1854 MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
1855 break;
1856 }
1857
1858 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_LINE_RASTERIZATION_PROPERTIES_EXT: {
1859 VkPhysicalDeviceLineRasterizationPropertiesEXT *props =
1860 (VkPhysicalDeviceLineRasterizationPropertiesEXT *)ext;
1861 /* In the Skylake PRM Vol. 7, subsection titled "GIQ (Diamond)
1862 * Sampling Rules - Legacy Mode", it says the following:
1863 *
1864 * "Note that the device divides a pixel into a 16x16 array of
1865 * subpixels, referenced by their upper left corners."
1866 *
1867 * This is the only known reference in the PRMs to the subpixel
1868 * precision of line rasterization and a "16x16 array of subpixels"
1869 * implies 4 subpixel precision bits. Empirical testing has shown
1870 * that 4 subpixel precision bits applies to all line rasterization
1871 * types.
1872 */
1873 props->lineSubPixelPrecisionBits = 4;
1874 break;
1875 }
1876
1877 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES: {
1878 VkPhysicalDeviceMaintenance3Properties *properties =
1879 (VkPhysicalDeviceMaintenance3Properties *)ext;
1880 /* This value doesn't matter for us today as our per-stage
1881 * descriptors are the real limit.
1882 */
1883 CORE_PROPERTY(1, 1, maxPerSetDescriptors);
1884 CORE_PROPERTY(1, 1, maxMemoryAllocationSize);
1885 break;
1886 }
1887
1888 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_PROPERTIES: {
1889 VkPhysicalDeviceMultiviewProperties *properties =
1890 (VkPhysicalDeviceMultiviewProperties *)ext;
1891 CORE_PROPERTY(1, 1, maxMultiviewViewCount);
1892 CORE_PROPERTY(1, 1, maxMultiviewInstanceIndex);
1893 break;
1894 }
1895
1896 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PCI_BUS_INFO_PROPERTIES_EXT: {
1897 VkPhysicalDevicePCIBusInfoPropertiesEXT *properties =
1898 (VkPhysicalDevicePCIBusInfoPropertiesEXT *)ext;
1899 properties->pciDomain = pdevice->pci_info.domain;
1900 properties->pciBus = pdevice->pci_info.bus;
1901 properties->pciDevice = pdevice->pci_info.device;
1902 properties->pciFunction = pdevice->pci_info.function;
1903 break;
1904 }
1905
1906 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_POINT_CLIPPING_PROPERTIES: {
1907 VkPhysicalDevicePointClippingProperties *properties =
1908 (VkPhysicalDevicePointClippingProperties *) ext;
1909 CORE_PROPERTY(1, 1, pointClippingBehavior);
1910 break;
1911 }
1912
1913 #pragma GCC diagnostic push
1914 #pragma GCC diagnostic ignored "-Wswitch"
1915 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRESENTATION_PROPERTIES_ANDROID: {
1916 VkPhysicalDevicePresentationPropertiesANDROID *props =
1917 (VkPhysicalDevicePresentationPropertiesANDROID *)ext;
1918 props->sharedImage = VK_FALSE;
1919 break;
1920 }
1921 #pragma GCC diagnostic pop
1922
1923 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_PROPERTIES: {
1924 VkPhysicalDeviceProtectedMemoryProperties *properties =
1925 (VkPhysicalDeviceProtectedMemoryProperties *)ext;
1926 CORE_PROPERTY(1, 1, protectedNoFault);
1927 break;
1928 }
1929
1930 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR: {
1931 VkPhysicalDevicePushDescriptorPropertiesKHR *properties =
1932 (VkPhysicalDevicePushDescriptorPropertiesKHR *) ext;
1933 properties->maxPushDescriptors = MAX_PUSH_DESCRIPTORS;
1934 break;
1935 }
1936
1937 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_PROPERTIES_EXT: {
1938 VkPhysicalDeviceRobustness2PropertiesEXT *properties = (void *)ext;
1939 properties->robustStorageBufferAccessSizeAlignment =
1940 ANV_SSBO_BOUNDS_CHECK_ALIGNMENT;
1941 properties->robustUniformBufferAccessSizeAlignment =
1942 ANV_UBO_ALIGNMENT;
1943 break;
1944 }
1945
1946 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_FILTER_MINMAX_PROPERTIES_EXT: {
1947 VkPhysicalDeviceSamplerFilterMinmaxPropertiesEXT *properties =
1948 (VkPhysicalDeviceSamplerFilterMinmaxPropertiesEXT *)ext;
1949 CORE_PROPERTY(1, 2, filterMinmaxImageComponentMapping);
1950 CORE_PROPERTY(1, 2, filterMinmaxSingleComponentFormats);
1951 break;
1952 }
1953
1954 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES: {
1955 VkPhysicalDeviceSubgroupProperties *properties = (void *)ext;
1956 CORE_PROPERTY(1, 1, subgroupSize);
1957 CORE_RENAMED_PROPERTY(1, 1, supportedStages,
1958 subgroupSupportedStages);
1959 CORE_RENAMED_PROPERTY(1, 1, supportedOperations,
1960 subgroupSupportedOperations);
1961 CORE_RENAMED_PROPERTY(1, 1, quadOperationsInAllStages,
1962 subgroupQuadOperationsInAllStages);
1963 break;
1964 }
1965
1966 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_PROPERTIES_EXT: {
1967 VkPhysicalDeviceSubgroupSizeControlPropertiesEXT *props =
1968 (VkPhysicalDeviceSubgroupSizeControlPropertiesEXT *)ext;
1969 STATIC_ASSERT(8 <= BRW_SUBGROUP_SIZE && BRW_SUBGROUP_SIZE <= 32);
1970 props->minSubgroupSize = 8;
1971 props->maxSubgroupSize = 32;
1972 props->maxComputeWorkgroupSubgroups = pdevice->info.max_cs_threads;
1973 props->requiredSubgroupSizeStages = VK_SHADER_STAGE_COMPUTE_BIT;
1974 break;
1975 }
1976 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT_CONTROLS_PROPERTIES_KHR : {
1977 VkPhysicalDeviceFloatControlsPropertiesKHR *properties = (void *)ext;
1978 CORE_PROPERTY(1, 2, denormBehaviorIndependence);
1979 CORE_PROPERTY(1, 2, roundingModeIndependence);
1980 CORE_PROPERTY(1, 2, shaderDenormFlushToZeroFloat16);
1981 CORE_PROPERTY(1, 2, shaderDenormPreserveFloat16);
1982 CORE_PROPERTY(1, 2, shaderRoundingModeRTEFloat16);
1983 CORE_PROPERTY(1, 2, shaderRoundingModeRTZFloat16);
1984 CORE_PROPERTY(1, 2, shaderSignedZeroInfNanPreserveFloat16);
1985 CORE_PROPERTY(1, 2, shaderDenormFlushToZeroFloat32);
1986 CORE_PROPERTY(1, 2, shaderDenormPreserveFloat32);
1987 CORE_PROPERTY(1, 2, shaderRoundingModeRTEFloat32);
1988 CORE_PROPERTY(1, 2, shaderRoundingModeRTZFloat32);
1989 CORE_PROPERTY(1, 2, shaderSignedZeroInfNanPreserveFloat32);
1990 CORE_PROPERTY(1, 2, shaderDenormFlushToZeroFloat64);
1991 CORE_PROPERTY(1, 2, shaderDenormPreserveFloat64);
1992 CORE_PROPERTY(1, 2, shaderRoundingModeRTEFloat64);
1993 CORE_PROPERTY(1, 2, shaderRoundingModeRTZFloat64);
1994 CORE_PROPERTY(1, 2, shaderSignedZeroInfNanPreserveFloat64);
1995 break;
1996 }
1997
1998 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TEXEL_BUFFER_ALIGNMENT_PROPERTIES_EXT: {
1999 VkPhysicalDeviceTexelBufferAlignmentPropertiesEXT *props =
2000 (VkPhysicalDeviceTexelBufferAlignmentPropertiesEXT *)ext;
2001
2002 /* From the SKL PRM Vol. 2d, docs for RENDER_SURFACE_STATE::Surface
2003 * Base Address:
2004 *
2005 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field
2006 * specifies the base address of the first element of the surface,
2007 * computed in software by adding the surface base address to the
2008 * byte offset of the element in the buffer. The base address must
2009 * be aligned to element size."
2010 *
2011 * The typed dataport messages require that things be texel aligned.
2012 * Otherwise, we may just load/store the wrong data or, in the worst
2013 * case, there may be hangs.
2014 */
2015 props->storageTexelBufferOffsetAlignmentBytes = 16;
2016 props->storageTexelBufferOffsetSingleTexelAlignment = true;
2017
2018 /* The sampler, however, is much more forgiving and it can handle
2019 * arbitrary byte alignment for linear and buffer surfaces. It's
2020 * hard to find a good PRM citation for this but years of empirical
2021 * experience demonstrate that this is true.
2022 */
2023 props->uniformTexelBufferOffsetAlignmentBytes = 1;
2024 props->uniformTexelBufferOffsetSingleTexelAlignment = false;
2025 break;
2026 }
2027
2028 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_PROPERTIES_KHR: {
2029 VkPhysicalDeviceTimelineSemaphorePropertiesKHR *properties =
2030 (VkPhysicalDeviceTimelineSemaphorePropertiesKHR *) ext;
2031 CORE_PROPERTY(1, 2, maxTimelineSemaphoreValueDifference);
2032 break;
2033 }
2034
2035 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT: {
2036 VkPhysicalDeviceTransformFeedbackPropertiesEXT *props =
2037 (VkPhysicalDeviceTransformFeedbackPropertiesEXT *)ext;
2038
2039 props->maxTransformFeedbackStreams = MAX_XFB_STREAMS;
2040 props->maxTransformFeedbackBuffers = MAX_XFB_BUFFERS;
2041 props->maxTransformFeedbackBufferSize = (1ull << 32);
2042 props->maxTransformFeedbackStreamDataSize = 128 * 4;
2043 props->maxTransformFeedbackBufferDataSize = 128 * 4;
2044 props->maxTransformFeedbackBufferDataStride = 2048;
2045 props->transformFeedbackQueries = true;
2046 props->transformFeedbackStreamsLinesTriangles = false;
2047 props->transformFeedbackRasterizationStreamSelect = false;
2048 props->transformFeedbackDraw = true;
2049 break;
2050 }
2051
2052 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT: {
2053 VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT *props =
2054 (VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT *)ext;
2055 /* We have to restrict this a bit for multiview */
2056 props->maxVertexAttribDivisor = UINT32_MAX / 16;
2057 break;
2058 }
2059
2060 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_PROPERTIES:
2061 anv_get_physical_device_properties_1_1(pdevice, (void *)ext);
2062 break;
2063
2064 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_PROPERTIES:
2065 anv_get_physical_device_properties_1_2(pdevice, (void *)ext);
2066 break;
2067
2068 default:
2069 anv_debug_ignored_stype(ext->sType);
2070 break;
2071 }
2072 }
2073
2074 #undef CORE_RENAMED_PROPERTY
2075 #undef CORE_PROPERTY
2076 }
2077
2078 /* We support exactly one queue family. */
2079 static const VkQueueFamilyProperties
2080 anv_queue_family_properties = {
2081 .queueFlags = VK_QUEUE_GRAPHICS_BIT |
2082 VK_QUEUE_COMPUTE_BIT |
2083 VK_QUEUE_TRANSFER_BIT,
2084 .queueCount = 1,
2085 .timestampValidBits = 36, /* XXX: Real value here */
2086 .minImageTransferGranularity = { 1, 1, 1 },
2087 };
2088
2089 void anv_GetPhysicalDeviceQueueFamilyProperties(
2090 VkPhysicalDevice physicalDevice,
2091 uint32_t* pCount,
2092 VkQueueFamilyProperties* pQueueFamilyProperties)
2093 {
2094 VK_OUTARRAY_MAKE(out, pQueueFamilyProperties, pCount);
2095
2096 vk_outarray_append(&out, p) {
2097 *p = anv_queue_family_properties;
2098 }
2099 }
2100
2101 void anv_GetPhysicalDeviceQueueFamilyProperties2(
2102 VkPhysicalDevice physicalDevice,
2103 uint32_t* pQueueFamilyPropertyCount,
2104 VkQueueFamilyProperties2* pQueueFamilyProperties)
2105 {
2106
2107 VK_OUTARRAY_MAKE(out, pQueueFamilyProperties, pQueueFamilyPropertyCount);
2108
2109 vk_outarray_append(&out, p) {
2110 p->queueFamilyProperties = anv_queue_family_properties;
2111
2112 vk_foreach_struct(s, p->pNext) {
2113 anv_debug_ignored_stype(s->sType);
2114 }
2115 }
2116 }
2117
2118 void anv_GetPhysicalDeviceMemoryProperties(
2119 VkPhysicalDevice physicalDevice,
2120 VkPhysicalDeviceMemoryProperties* pMemoryProperties)
2121 {
2122 ANV_FROM_HANDLE(anv_physical_device, physical_device, physicalDevice);
2123
2124 pMemoryProperties->memoryTypeCount = physical_device->memory.type_count;
2125 for (uint32_t i = 0; i < physical_device->memory.type_count; i++) {
2126 pMemoryProperties->memoryTypes[i] = (VkMemoryType) {
2127 .propertyFlags = physical_device->memory.types[i].propertyFlags,
2128 .heapIndex = physical_device->memory.types[i].heapIndex,
2129 };
2130 }
2131
2132 pMemoryProperties->memoryHeapCount = physical_device->memory.heap_count;
2133 for (uint32_t i = 0; i < physical_device->memory.heap_count; i++) {
2134 pMemoryProperties->memoryHeaps[i] = (VkMemoryHeap) {
2135 .size = physical_device->memory.heaps[i].size,
2136 .flags = physical_device->memory.heaps[i].flags,
2137 };
2138 }
2139 }
2140
2141 static void
2142 anv_get_memory_budget(VkPhysicalDevice physicalDevice,
2143 VkPhysicalDeviceMemoryBudgetPropertiesEXT *memoryBudget)
2144 {
2145 ANV_FROM_HANDLE(anv_physical_device, device, physicalDevice);
2146 uint64_t sys_available = get_available_system_memory();
2147 assert(sys_available > 0);
2148
2149 VkDeviceSize total_heaps_size = 0;
2150 for (size_t i = 0; i < device->memory.heap_count; i++)
2151 total_heaps_size += device->memory.heaps[i].size;
2152
2153 for (size_t i = 0; i < device->memory.heap_count; i++) {
2154 VkDeviceSize heap_size = device->memory.heaps[i].size;
2155 VkDeviceSize heap_used = device->memory.heaps[i].used;
2156 VkDeviceSize heap_budget;
2157
2158 double heap_proportion = (double) heap_size / total_heaps_size;
2159 VkDeviceSize sys_available_prop = sys_available * heap_proportion;
2160
2161 /*
2162 * Let's not incite the app to starve the system: report at most 90% of
2163 * available system memory.
2164 */
2165 uint64_t heap_available = sys_available_prop * 9 / 10;
2166 heap_budget = MIN2(heap_size, heap_used + heap_available);
2167
2168 /*
2169 * Round down to the nearest MB
2170 */
2171 heap_budget &= ~((1ull << 20) - 1);
2172
2173 /*
2174 * The heapBudget value must be non-zero for array elements less than
2175 * VkPhysicalDeviceMemoryProperties::memoryHeapCount. The heapBudget
2176 * value must be less than or equal to VkMemoryHeap::size for each heap.
2177 */
2178 assert(0 < heap_budget && heap_budget <= heap_size);
2179
2180 memoryBudget->heapUsage[i] = heap_used;
2181 memoryBudget->heapBudget[i] = heap_budget;
2182 }
2183
2184 /* The heapBudget and heapUsage values must be zero for array elements
2185 * greater than or equal to VkPhysicalDeviceMemoryProperties::memoryHeapCount
2186 */
2187 for (uint32_t i = device->memory.heap_count; i < VK_MAX_MEMORY_HEAPS; i++) {
2188 memoryBudget->heapBudget[i] = 0;
2189 memoryBudget->heapUsage[i] = 0;
2190 }
2191 }
2192
2193 void anv_GetPhysicalDeviceMemoryProperties2(
2194 VkPhysicalDevice physicalDevice,
2195 VkPhysicalDeviceMemoryProperties2* pMemoryProperties)
2196 {
2197 anv_GetPhysicalDeviceMemoryProperties(physicalDevice,
2198 &pMemoryProperties->memoryProperties);
2199
2200 vk_foreach_struct(ext, pMemoryProperties->pNext) {
2201 switch (ext->sType) {
2202 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_BUDGET_PROPERTIES_EXT:
2203 anv_get_memory_budget(physicalDevice, (void*)ext);
2204 break;
2205 default:
2206 anv_debug_ignored_stype(ext->sType);
2207 break;
2208 }
2209 }
2210 }
2211
2212 void
2213 anv_GetDeviceGroupPeerMemoryFeatures(
2214 VkDevice device,
2215 uint32_t heapIndex,
2216 uint32_t localDeviceIndex,
2217 uint32_t remoteDeviceIndex,
2218 VkPeerMemoryFeatureFlags* pPeerMemoryFeatures)
2219 {
2220 assert(localDeviceIndex == 0 && remoteDeviceIndex == 0);
2221 *pPeerMemoryFeatures = VK_PEER_MEMORY_FEATURE_COPY_SRC_BIT |
2222 VK_PEER_MEMORY_FEATURE_COPY_DST_BIT |
2223 VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT |
2224 VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT;
2225 }
2226
2227 PFN_vkVoidFunction anv_GetInstanceProcAddr(
2228 VkInstance _instance,
2229 const char* pName)
2230 {
2231 ANV_FROM_HANDLE(anv_instance, instance, _instance);
2232
2233 /* The Vulkan 1.0 spec for vkGetInstanceProcAddr has a table of exactly
2234 * when we have to return valid function pointers, NULL, or it's left
2235 * undefined. See the table for exact details.
2236 */
2237 if (pName == NULL)
2238 return NULL;
2239
2240 #define LOOKUP_ANV_ENTRYPOINT(entrypoint) \
2241 if (strcmp(pName, "vk" #entrypoint) == 0) \
2242 return (PFN_vkVoidFunction)anv_##entrypoint
2243
2244 LOOKUP_ANV_ENTRYPOINT(EnumerateInstanceExtensionProperties);
2245 LOOKUP_ANV_ENTRYPOINT(EnumerateInstanceLayerProperties);
2246 LOOKUP_ANV_ENTRYPOINT(EnumerateInstanceVersion);
2247 LOOKUP_ANV_ENTRYPOINT(CreateInstance);
2248
2249 /* GetInstanceProcAddr() can also be called with a NULL instance.
2250 * See https://gitlab.khronos.org/vulkan/vulkan/issues/2057
2251 */
2252 LOOKUP_ANV_ENTRYPOINT(GetInstanceProcAddr);
2253
2254 #undef LOOKUP_ANV_ENTRYPOINT
2255
2256 if (instance == NULL)
2257 return NULL;
2258
2259 int idx = anv_get_instance_entrypoint_index(pName);
2260 if (idx >= 0)
2261 return instance->dispatch.entrypoints[idx];
2262
2263 idx = anv_get_physical_device_entrypoint_index(pName);
2264 if (idx >= 0)
2265 return instance->physical_device_dispatch.entrypoints[idx];
2266
2267 idx = anv_get_device_entrypoint_index(pName);
2268 if (idx >= 0)
2269 return instance->device_dispatch.entrypoints[idx];
2270
2271 return NULL;
2272 }
2273
2274 /* With version 1+ of the loader interface the ICD should expose
2275 * vk_icdGetInstanceProcAddr to work around certain LD_PRELOAD issues seen in apps.
2276 */
2277 PUBLIC
2278 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL vk_icdGetInstanceProcAddr(
2279 VkInstance instance,
2280 const char* pName);
2281
2282 PUBLIC
2283 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL vk_icdGetInstanceProcAddr(
2284 VkInstance instance,
2285 const char* pName)
2286 {
2287 return anv_GetInstanceProcAddr(instance, pName);
2288 }
2289
2290 PFN_vkVoidFunction anv_GetDeviceProcAddr(
2291 VkDevice _device,
2292 const char* pName)
2293 {
2294 ANV_FROM_HANDLE(anv_device, device, _device);
2295
2296 if (!device || !pName)
2297 return NULL;
2298
2299 int idx = anv_get_device_entrypoint_index(pName);
2300 if (idx < 0)
2301 return NULL;
2302
2303 return device->dispatch.entrypoints[idx];
2304 }
2305
2306 /* With version 4+ of the loader interface the ICD should expose
2307 * vk_icdGetPhysicalDeviceProcAddr()
2308 */
2309 PUBLIC
2310 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL vk_icdGetPhysicalDeviceProcAddr(
2311 VkInstance _instance,
2312 const char* pName);
2313
2314 PFN_vkVoidFunction vk_icdGetPhysicalDeviceProcAddr(
2315 VkInstance _instance,
2316 const char* pName)
2317 {
2318 ANV_FROM_HANDLE(anv_instance, instance, _instance);
2319
2320 if (!pName || !instance)
2321 return NULL;
2322
2323 int idx = anv_get_physical_device_entrypoint_index(pName);
2324 if (idx < 0)
2325 return NULL;
2326
2327 return instance->physical_device_dispatch.entrypoints[idx];
2328 }
2329
2330
2331 VkResult
2332 anv_CreateDebugReportCallbackEXT(VkInstance _instance,
2333 const VkDebugReportCallbackCreateInfoEXT* pCreateInfo,
2334 const VkAllocationCallbacks* pAllocator,
2335 VkDebugReportCallbackEXT* pCallback)
2336 {
2337 ANV_FROM_HANDLE(anv_instance, instance, _instance);
2338 return vk_create_debug_report_callback(&instance->debug_report_callbacks,
2339 pCreateInfo, pAllocator, &instance->alloc,
2340 pCallback);
2341 }
2342
2343 void
2344 anv_DestroyDebugReportCallbackEXT(VkInstance _instance,
2345 VkDebugReportCallbackEXT _callback,
2346 const VkAllocationCallbacks* pAllocator)
2347 {
2348 ANV_FROM_HANDLE(anv_instance, instance, _instance);
2349 vk_destroy_debug_report_callback(&instance->debug_report_callbacks,
2350 _callback, pAllocator, &instance->alloc);
2351 }
2352
2353 void
2354 anv_DebugReportMessageEXT(VkInstance _instance,
2355 VkDebugReportFlagsEXT flags,
2356 VkDebugReportObjectTypeEXT objectType,
2357 uint64_t object,
2358 size_t location,
2359 int32_t messageCode,
2360 const char* pLayerPrefix,
2361 const char* pMessage)
2362 {
2363 ANV_FROM_HANDLE(anv_instance, instance, _instance);
2364 vk_debug_report(&instance->debug_report_callbacks, flags, objectType,
2365 object, location, messageCode, pLayerPrefix, pMessage);
2366 }
2367
2368 static struct anv_state
2369 anv_state_pool_emit_data(struct anv_state_pool *pool, size_t size, size_t align, const void *p)
2370 {
2371 struct anv_state state;
2372
2373 state = anv_state_pool_alloc(pool, size, align);
2374 memcpy(state.map, p, size);
2375
2376 return state;
2377 }
2378
2379 static void
2380 anv_device_init_border_colors(struct anv_device *device)
2381 {
2382 if (device->info.is_haswell) {
2383 static const struct hsw_border_color border_colors[] = {
2384 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK] = { .float32 = { 0.0, 0.0, 0.0, 0.0 } },
2385 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK] = { .float32 = { 0.0, 0.0, 0.0, 1.0 } },
2386 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE] = { .float32 = { 1.0, 1.0, 1.0, 1.0 } },
2387 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK] = { .uint32 = { 0, 0, 0, 0 } },
2388 [VK_BORDER_COLOR_INT_OPAQUE_BLACK] = { .uint32 = { 0, 0, 0, 1 } },
2389 [VK_BORDER_COLOR_INT_OPAQUE_WHITE] = { .uint32 = { 1, 1, 1, 1 } },
2390 };
2391
2392 device->border_colors =
2393 anv_state_pool_emit_data(&device->dynamic_state_pool,
2394 sizeof(border_colors), 512, border_colors);
2395 } else {
2396 static const struct gen8_border_color border_colors[] = {
2397 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK] = { .float32 = { 0.0, 0.0, 0.0, 0.0 } },
2398 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK] = { .float32 = { 0.0, 0.0, 0.0, 1.0 } },
2399 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE] = { .float32 = { 1.0, 1.0, 1.0, 1.0 } },
2400 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK] = { .uint32 = { 0, 0, 0, 0 } },
2401 [VK_BORDER_COLOR_INT_OPAQUE_BLACK] = { .uint32 = { 0, 0, 0, 1 } },
2402 [VK_BORDER_COLOR_INT_OPAQUE_WHITE] = { .uint32 = { 1, 1, 1, 1 } },
2403 };
2404
2405 device->border_colors =
2406 anv_state_pool_emit_data(&device->dynamic_state_pool,
2407 sizeof(border_colors), 64, border_colors);
2408 }
2409 }
2410
2411 static VkResult
2412 anv_device_init_trivial_batch(struct anv_device *device)
2413 {
2414 VkResult result = anv_device_alloc_bo(device, 4096,
2415 ANV_BO_ALLOC_MAPPED,
2416 0 /* explicit_address */,
2417 &device->trivial_batch_bo);
2418 if (result != VK_SUCCESS)
2419 return result;
2420
2421 struct anv_batch batch = {
2422 .start = device->trivial_batch_bo->map,
2423 .next = device->trivial_batch_bo->map,
2424 .end = device->trivial_batch_bo->map + 4096,
2425 };
2426
2427 anv_batch_emit(&batch, GEN7_MI_BATCH_BUFFER_END, bbe);
2428 anv_batch_emit(&batch, GEN7_MI_NOOP, noop);
2429
2430 if (!device->info.has_llc)
2431 gen_clflush_range(batch.start, batch.next - batch.start);
2432
2433 return VK_SUCCESS;
2434 }
2435
2436 VkResult anv_EnumerateDeviceExtensionProperties(
2437 VkPhysicalDevice physicalDevice,
2438 const char* pLayerName,
2439 uint32_t* pPropertyCount,
2440 VkExtensionProperties* pProperties)
2441 {
2442 ANV_FROM_HANDLE(anv_physical_device, device, physicalDevice);
2443 VK_OUTARRAY_MAKE(out, pProperties, pPropertyCount);
2444
2445 for (int i = 0; i < ANV_DEVICE_EXTENSION_COUNT; i++) {
2446 if (device->supported_extensions.extensions[i]) {
2447 vk_outarray_append(&out, prop) {
2448 *prop = anv_device_extensions[i];
2449 }
2450 }
2451 }
2452
2453 return vk_outarray_status(&out);
2454 }
2455
2456 static void
2457 anv_device_init_dispatch(struct anv_device *device)
2458 {
2459 const struct anv_instance *instance = device->physical->instance;
2460
2461 const struct anv_device_dispatch_table *genX_table;
2462 switch (device->info.gen) {
2463 case 12:
2464 genX_table = &gen12_device_dispatch_table;
2465 break;
2466 case 11:
2467 genX_table = &gen11_device_dispatch_table;
2468 break;
2469 case 10:
2470 genX_table = &gen10_device_dispatch_table;
2471 break;
2472 case 9:
2473 genX_table = &gen9_device_dispatch_table;
2474 break;
2475 case 8:
2476 genX_table = &gen8_device_dispatch_table;
2477 break;
2478 case 7:
2479 if (device->info.is_haswell)
2480 genX_table = &gen75_device_dispatch_table;
2481 else
2482 genX_table = &gen7_device_dispatch_table;
2483 break;
2484 default:
2485 unreachable("unsupported gen\n");
2486 }
2487
2488 for (unsigned i = 0; i < ARRAY_SIZE(device->dispatch.entrypoints); i++) {
2489 /* Vulkan requires that entrypoints for extensions which have not been
2490 * enabled must not be advertised.
2491 */
2492 if (!anv_device_entrypoint_is_enabled(i, instance->app_info.api_version,
2493 &instance->enabled_extensions,
2494 &device->enabled_extensions)) {
2495 device->dispatch.entrypoints[i] = NULL;
2496 } else if (genX_table->entrypoints[i]) {
2497 device->dispatch.entrypoints[i] = genX_table->entrypoints[i];
2498 } else {
2499 device->dispatch.entrypoints[i] =
2500 anv_device_dispatch_table.entrypoints[i];
2501 }
2502 }
2503 }
2504
2505 static int
2506 vk_priority_to_gen(int priority)
2507 {
2508 switch (priority) {
2509 case VK_QUEUE_GLOBAL_PRIORITY_LOW_EXT:
2510 return GEN_CONTEXT_LOW_PRIORITY;
2511 case VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_EXT:
2512 return GEN_CONTEXT_MEDIUM_PRIORITY;
2513 case VK_QUEUE_GLOBAL_PRIORITY_HIGH_EXT:
2514 return GEN_CONTEXT_HIGH_PRIORITY;
2515 case VK_QUEUE_GLOBAL_PRIORITY_REALTIME_EXT:
2516 return GEN_CONTEXT_REALTIME_PRIORITY;
2517 default:
2518 unreachable("Invalid priority");
2519 }
2520 }
2521
2522 static VkResult
2523 anv_device_init_hiz_clear_value_bo(struct anv_device *device)
2524 {
2525 VkResult result = anv_device_alloc_bo(device, 4096,
2526 ANV_BO_ALLOC_MAPPED,
2527 0 /* explicit_address */,
2528 &device->hiz_clear_bo);
2529 if (result != VK_SUCCESS)
2530 return result;
2531
2532 union isl_color_value hiz_clear = { .u32 = { 0, } };
2533 hiz_clear.f32[0] = ANV_HZ_FC_VAL;
2534
2535 memcpy(device->hiz_clear_bo->map, hiz_clear.u32, sizeof(hiz_clear.u32));
2536
2537 if (!device->info.has_llc)
2538 gen_clflush_range(device->hiz_clear_bo->map, sizeof(hiz_clear.u32));
2539
2540 return VK_SUCCESS;
2541 }
2542
2543 static bool
2544 get_bo_from_pool(struct gen_batch_decode_bo *ret,
2545 struct anv_block_pool *pool,
2546 uint64_t address)
2547 {
2548 anv_block_pool_foreach_bo(bo, pool) {
2549 uint64_t bo_address = gen_48b_address(bo->offset);
2550 if (address >= bo_address && address < (bo_address + bo->size)) {
2551 *ret = (struct gen_batch_decode_bo) {
2552 .addr = bo_address,
2553 .size = bo->size,
2554 .map = bo->map,
2555 };
2556 return true;
2557 }
2558 }
2559 return false;
2560 }
2561
2562 /* Finding a buffer for batch decoding */
2563 static struct gen_batch_decode_bo
2564 decode_get_bo(void *v_batch, bool ppgtt, uint64_t address)
2565 {
2566 struct anv_device *device = v_batch;
2567 struct gen_batch_decode_bo ret_bo = {};
2568
2569 assert(ppgtt);
2570
2571 if (get_bo_from_pool(&ret_bo, &device->dynamic_state_pool.block_pool, address))
2572 return ret_bo;
2573 if (get_bo_from_pool(&ret_bo, &device->instruction_state_pool.block_pool, address))
2574 return ret_bo;
2575 if (get_bo_from_pool(&ret_bo, &device->binding_table_pool.block_pool, address))
2576 return ret_bo;
2577 if (get_bo_from_pool(&ret_bo, &device->surface_state_pool.block_pool, address))
2578 return ret_bo;
2579
2580 if (!device->cmd_buffer_being_decoded)
2581 return (struct gen_batch_decode_bo) { };
2582
2583 struct anv_batch_bo **bo;
2584
2585 u_vector_foreach(bo, &device->cmd_buffer_being_decoded->seen_bbos) {
2586 /* The decoder zeroes out the top 16 bits, so we need to as well */
2587 uint64_t bo_address = (*bo)->bo->offset & (~0ull >> 16);
2588
2589 if (address >= bo_address && address < bo_address + (*bo)->bo->size) {
2590 return (struct gen_batch_decode_bo) {
2591 .addr = bo_address,
2592 .size = (*bo)->bo->size,
2593 .map = (*bo)->bo->map,
2594 };
2595 }
2596 }
2597
2598 return (struct gen_batch_decode_bo) { };
2599 }
2600
2601 struct gen_aux_map_buffer {
2602 struct gen_buffer base;
2603 struct anv_state state;
2604 };
2605
2606 static struct gen_buffer *
2607 gen_aux_map_buffer_alloc(void *driver_ctx, uint32_t size)
2608 {
2609 struct gen_aux_map_buffer *buf = malloc(sizeof(struct gen_aux_map_buffer));
2610 if (!buf)
2611 return NULL;
2612
2613 struct anv_device *device = (struct anv_device*)driver_ctx;
2614 assert(device->physical->supports_48bit_addresses &&
2615 device->physical->use_softpin);
2616
2617 struct anv_state_pool *pool = &device->dynamic_state_pool;
2618 buf->state = anv_state_pool_alloc(pool, size, size);
2619
2620 buf->base.gpu = pool->block_pool.bo->offset + buf->state.offset;
2621 buf->base.gpu_end = buf->base.gpu + buf->state.alloc_size;
2622 buf->base.map = buf->state.map;
2623 buf->base.driver_bo = &buf->state;
2624 return &buf->base;
2625 }
2626
2627 static void
2628 gen_aux_map_buffer_free(void *driver_ctx, struct gen_buffer *buffer)
2629 {
2630 struct gen_aux_map_buffer *buf = (struct gen_aux_map_buffer*)buffer;
2631 struct anv_device *device = (struct anv_device*)driver_ctx;
2632 struct anv_state_pool *pool = &device->dynamic_state_pool;
2633 anv_state_pool_free(pool, buf->state);
2634 free(buf);
2635 }
2636
2637 static struct gen_mapped_pinned_buffer_alloc aux_map_allocator = {
2638 .alloc = gen_aux_map_buffer_alloc,
2639 .free = gen_aux_map_buffer_free,
2640 };
2641
2642 static VkResult
2643 check_physical_device_features(VkPhysicalDevice physicalDevice,
2644 const VkPhysicalDeviceFeatures *features)
2645 {
2646 VkPhysicalDeviceFeatures supported_features;
2647 anv_GetPhysicalDeviceFeatures(physicalDevice, &supported_features);
2648 VkBool32 *supported_feature = (VkBool32 *)&supported_features;
2649 VkBool32 *enabled_feature = (VkBool32 *)features;
2650 unsigned num_features = sizeof(VkPhysicalDeviceFeatures) / sizeof(VkBool32);
2651 for (uint32_t i = 0; i < num_features; i++) {
2652 if (enabled_feature[i] && !supported_feature[i])
2653 return vk_error(VK_ERROR_FEATURE_NOT_PRESENT);
2654 }
2655
2656 return VK_SUCCESS;
2657 }
2658
2659 VkResult anv_CreateDevice(
2660 VkPhysicalDevice physicalDevice,
2661 const VkDeviceCreateInfo* pCreateInfo,
2662 const VkAllocationCallbacks* pAllocator,
2663 VkDevice* pDevice)
2664 {
2665 ANV_FROM_HANDLE(anv_physical_device, physical_device, physicalDevice);
2666 VkResult result;
2667 struct anv_device *device;
2668
2669 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO);
2670
2671 struct anv_device_extension_table enabled_extensions = { };
2672 for (uint32_t i = 0; i < pCreateInfo->enabledExtensionCount; i++) {
2673 int idx;
2674 for (idx = 0; idx < ANV_DEVICE_EXTENSION_COUNT; idx++) {
2675 if (strcmp(pCreateInfo->ppEnabledExtensionNames[i],
2676 anv_device_extensions[idx].extensionName) == 0)
2677 break;
2678 }
2679
2680 if (idx >= ANV_DEVICE_EXTENSION_COUNT)
2681 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);
2682
2683 if (!physical_device->supported_extensions.extensions[idx])
2684 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);
2685
2686 enabled_extensions.extensions[idx] = true;
2687 }
2688
2689 /* Check enabled features */
2690 bool robust_buffer_access = false;
2691 if (pCreateInfo->pEnabledFeatures) {
2692 result = check_physical_device_features(physicalDevice,
2693 pCreateInfo->pEnabledFeatures);
2694 if (result != VK_SUCCESS)
2695 return result;
2696
2697 if (pCreateInfo->pEnabledFeatures->robustBufferAccess)
2698 robust_buffer_access = true;
2699 }
2700
2701 vk_foreach_struct_const(ext, pCreateInfo->pNext) {
2702 switch (ext->sType) {
2703 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2: {
2704 const VkPhysicalDeviceFeatures2 *features = (const void *)ext;
2705 result = check_physical_device_features(physicalDevice,
2706 &features->features);
2707 if (result != VK_SUCCESS)
2708 return result;
2709
2710 if (features->features.robustBufferAccess)
2711 robust_buffer_access = true;
2712 break;
2713 }
2714
2715 default:
2716 /* Don't warn */
2717 break;
2718 }
2719 }
2720
2721 /* Check requested queues and fail if we are requested to create any
2722 * queues with flags we don't support.
2723 */
2724 assert(pCreateInfo->queueCreateInfoCount > 0);
2725 for (uint32_t i = 0; i < pCreateInfo->queueCreateInfoCount; i++) {
2726 if (pCreateInfo->pQueueCreateInfos[i].flags != 0)
2727 return vk_error(VK_ERROR_INITIALIZATION_FAILED);
2728 }
2729
2730 /* Check if client specified queue priority. */
2731 const VkDeviceQueueGlobalPriorityCreateInfoEXT *queue_priority =
2732 vk_find_struct_const(pCreateInfo->pQueueCreateInfos[0].pNext,
2733 DEVICE_QUEUE_GLOBAL_PRIORITY_CREATE_INFO_EXT);
2734
2735 VkQueueGlobalPriorityEXT priority =
2736 queue_priority ? queue_priority->globalPriority :
2737 VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_EXT;
2738
2739 device = vk_alloc2(&physical_device->instance->alloc, pAllocator,
2740 sizeof(*device), 8,
2741 VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
2742 if (!device)
2743 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2744
2745 vk_device_init(&device->vk, pCreateInfo,
2746 &physical_device->instance->alloc, pAllocator);
2747
2748 if (INTEL_DEBUG & DEBUG_BATCH) {
2749 const unsigned decode_flags =
2750 GEN_BATCH_DECODE_FULL |
2751 ((INTEL_DEBUG & DEBUG_COLOR) ? GEN_BATCH_DECODE_IN_COLOR : 0) |
2752 GEN_BATCH_DECODE_OFFSETS |
2753 GEN_BATCH_DECODE_FLOATS;
2754
2755 gen_batch_decode_ctx_init(&device->decoder_ctx,
2756 &physical_device->info,
2757 stderr, decode_flags, NULL,
2758 decode_get_bo, NULL, device);
2759 }
2760
2761 device->physical = physical_device;
2762 device->no_hw = physical_device->no_hw;
2763 device->_lost = false;
2764
2765 /* XXX(chadv): Can we dup() physicalDevice->fd here? */
2766 device->fd = open(physical_device->path, O_RDWR | O_CLOEXEC);
2767 if (device->fd == -1) {
2768 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2769 goto fail_device;
2770 }
2771
2772 device->context_id = anv_gem_create_context(device);
2773 if (device->context_id == -1) {
2774 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2775 goto fail_fd;
2776 }
2777
2778 result = anv_queue_init(device, &device->queue);
2779 if (result != VK_SUCCESS)
2780 goto fail_context_id;
2781
2782 if (physical_device->use_softpin) {
2783 if (pthread_mutex_init(&device->vma_mutex, NULL) != 0) {
2784 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2785 goto fail_queue;
2786 }
2787
2788 /* keep the page with address zero out of the allocator */
2789 util_vma_heap_init(&device->vma_lo,
2790 LOW_HEAP_MIN_ADDRESS, LOW_HEAP_SIZE);
2791
2792 util_vma_heap_init(&device->vma_cva, CLIENT_VISIBLE_HEAP_MIN_ADDRESS,
2793 CLIENT_VISIBLE_HEAP_SIZE);
2794
2795 /* Leave the last 4GiB out of the high vma range, so that no state
2796 * base address + size can overflow 48 bits. For more information see
2797 * the comment about Wa32bitGeneralStateOffset in anv_allocator.c
2798 */
2799 util_vma_heap_init(&device->vma_hi, HIGH_HEAP_MIN_ADDRESS,
2800 physical_device->gtt_size - (1ull << 32) -
2801 HIGH_HEAP_MIN_ADDRESS);
2802 }
2803
2804 list_inithead(&device->memory_objects);
2805
2806 /* As per spec, the driver implementation may deny requests to acquire
2807 * a priority above the default priority (MEDIUM) if the caller does not
2808 * have sufficient privileges. In this scenario VK_ERROR_NOT_PERMITTED_EXT
2809 * is returned.
2810 */
2811 if (physical_device->has_context_priority) {
2812 int err = anv_gem_set_context_param(device->fd, device->context_id,
2813 I915_CONTEXT_PARAM_PRIORITY,
2814 vk_priority_to_gen(priority));
2815 if (err != 0 && priority > VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_EXT) {
2816 result = vk_error(VK_ERROR_NOT_PERMITTED_EXT);
2817 goto fail_vmas;
2818 }
2819 }
2820
2821 device->info = physical_device->info;
2822 device->isl_dev = physical_device->isl_dev;
2823
2824 /* On Broadwell and later, we can use batch chaining to more efficiently
2825 * implement growing command buffers. Prior to Haswell, the kernel
2826 * command parser gets in the way and we have to fall back to growing
2827 * the batch.
2828 */
2829 device->can_chain_batches = device->info.gen >= 8;
2830
2831 device->robust_buffer_access = robust_buffer_access;
2832 device->enabled_extensions = enabled_extensions;
2833
2834 anv_device_init_dispatch(device);
2835
2836 if (pthread_mutex_init(&device->mutex, NULL) != 0) {
2837 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2838 goto fail_queue;
2839 }
2840
2841 pthread_condattr_t condattr;
2842 if (pthread_condattr_init(&condattr) != 0) {
2843 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2844 goto fail_mutex;
2845 }
2846 if (pthread_condattr_setclock(&condattr, CLOCK_MONOTONIC) != 0) {
2847 pthread_condattr_destroy(&condattr);
2848 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2849 goto fail_mutex;
2850 }
2851 if (pthread_cond_init(&device->queue_submit, &condattr) != 0) {
2852 pthread_condattr_destroy(&condattr);
2853 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2854 goto fail_mutex;
2855 }
2856 pthread_condattr_destroy(&condattr);
2857
2858 result = anv_bo_cache_init(&device->bo_cache);
2859 if (result != VK_SUCCESS)
2860 goto fail_queue_cond;
2861
2862 anv_bo_pool_init(&device->batch_bo_pool, device);
2863
2864 result = anv_state_pool_init(&device->dynamic_state_pool, device,
2865 DYNAMIC_STATE_POOL_MIN_ADDRESS, 0, 16384);
2866 if (result != VK_SUCCESS)
2867 goto fail_batch_bo_pool;
2868
2869 if (device->info.gen >= 8) {
2870 /* The border color pointer is limited to 24 bits, so we need to make
2871 * sure that any such color used at any point in the program doesn't
2872 * exceed that limit.
2873 * We achieve that by reserving all the custom border colors we support
2874 * right off the bat, so they are close to the base address.
2875 */
2876 anv_state_reserved_pool_init(&device->custom_border_colors,
2877 &device->dynamic_state_pool,
2878 sizeof(struct gen8_border_color),
2879 MAX_CUSTOM_BORDER_COLORS, 64);
2880 }
2881
2882 result = anv_state_pool_init(&device->instruction_state_pool, device,
2883 INSTRUCTION_STATE_POOL_MIN_ADDRESS, 0, 16384);
2884 if (result != VK_SUCCESS)
2885 goto fail_dynamic_state_pool;
2886
2887 result = anv_state_pool_init(&device->surface_state_pool, device,
2888 SURFACE_STATE_POOL_MIN_ADDRESS, 0, 4096);
2889 if (result != VK_SUCCESS)
2890 goto fail_instruction_state_pool;
2891
2892 if (physical_device->use_softpin) {
2893 int64_t bt_pool_offset = (int64_t)BINDING_TABLE_POOL_MIN_ADDRESS -
2894 (int64_t)SURFACE_STATE_POOL_MIN_ADDRESS;
2895 assert(INT32_MIN < bt_pool_offset && bt_pool_offset < 0);
2896 result = anv_state_pool_init(&device->binding_table_pool, device,
2897 SURFACE_STATE_POOL_MIN_ADDRESS,
2898 bt_pool_offset, 4096);
2899 if (result != VK_SUCCESS)
2900 goto fail_surface_state_pool;
2901 }
2902
2903 if (device->info.gen >= 12) {
2904 device->aux_map_ctx = gen_aux_map_init(device, &aux_map_allocator,
2905 &physical_device->info);
2906 if (!device->aux_map_ctx)
2907 goto fail_binding_table_pool;
2908 }
2909
2910 result = anv_device_alloc_bo(device, 4096, 0 /* flags */,
2911 0 /* explicit_address */,
2912 &device->workaround_bo);
2913 if (result != VK_SUCCESS)
2914 goto fail_surface_aux_map_pool;
2915
2916 result = anv_device_init_trivial_batch(device);
2917 if (result != VK_SUCCESS)
2918 goto fail_workaround_bo;
2919
2920 /* Allocate a null surface state at surface state offset 0. This makes
2921 * NULL descriptor handling trivial because we can just memset structures
2922 * to zero and they have a valid descriptor.
2923 */
2924 device->null_surface_state =
2925 anv_state_pool_alloc(&device->surface_state_pool,
2926 device->isl_dev.ss.size,
2927 device->isl_dev.ss.align);
2928 isl_null_fill_state(&device->isl_dev, device->null_surface_state.map,
2929 isl_extent3d(1, 1, 1) /* This shouldn't matter */);
2930 assert(device->null_surface_state.offset == 0);
2931
2932 if (device->info.gen >= 10) {
2933 result = anv_device_init_hiz_clear_value_bo(device);
2934 if (result != VK_SUCCESS)
2935 goto fail_trivial_batch_bo;
2936 }
2937
2938 anv_scratch_pool_init(device, &device->scratch_pool);
2939
2940 switch (device->info.gen) {
2941 case 7:
2942 if (!device->info.is_haswell)
2943 result = gen7_init_device_state(device);
2944 else
2945 result = gen75_init_device_state(device);
2946 break;
2947 case 8:
2948 result = gen8_init_device_state(device);
2949 break;
2950 case 9:
2951 result = gen9_init_device_state(device);
2952 break;
2953 case 10:
2954 result = gen10_init_device_state(device);
2955 break;
2956 case 11:
2957 result = gen11_init_device_state(device);
2958 break;
2959 case 12:
2960 result = gen12_init_device_state(device);
2961 break;
2962 default:
2963 /* Shouldn't get here as we don't create physical devices for any other
2964 * gens. */
2965 unreachable("unhandled gen");
2966 }
2967 if (result != VK_SUCCESS)
2968 goto fail_workaround_bo;
2969
2970 anv_pipeline_cache_init(&device->default_pipeline_cache, device, true);
2971
2972 anv_device_init_blorp(device);
2973
2974 anv_device_init_border_colors(device);
2975
2976 anv_device_perf_init(device);
2977
2978 *pDevice = anv_device_to_handle(device);
2979
2980 return VK_SUCCESS;
2981
2982 fail_workaround_bo:
2983 anv_scratch_pool_finish(device, &device->scratch_pool);
2984 if (device->info.gen >= 10)
2985 anv_device_release_bo(device, device->hiz_clear_bo);
2986 anv_device_release_bo(device, device->workaround_bo);
2987 fail_trivial_batch_bo:
2988 anv_device_release_bo(device, device->trivial_batch_bo);
2989 fail_surface_aux_map_pool:
2990 if (device->info.gen >= 12) {
2991 gen_aux_map_finish(device->aux_map_ctx);
2992 device->aux_map_ctx = NULL;
2993 }
2994 fail_binding_table_pool:
2995 if (physical_device->use_softpin)
2996 anv_state_pool_finish(&device->binding_table_pool);
2997 fail_surface_state_pool:
2998 anv_state_pool_finish(&device->surface_state_pool);
2999 fail_instruction_state_pool:
3000 anv_state_pool_finish(&device->instruction_state_pool);
3001 fail_dynamic_state_pool:
3002 if (device->info.gen >= 8)
3003 anv_state_reserved_pool_finish(&device->custom_border_colors);
3004 anv_state_pool_finish(&device->dynamic_state_pool);
3005 fail_batch_bo_pool:
3006 anv_bo_pool_finish(&device->batch_bo_pool);
3007 anv_bo_cache_finish(&device->bo_cache);
3008 fail_queue_cond:
3009 pthread_cond_destroy(&device->queue_submit);
3010 fail_mutex:
3011 pthread_mutex_destroy(&device->mutex);
3012 fail_vmas:
3013 if (physical_device->use_softpin) {
3014 util_vma_heap_finish(&device->vma_hi);
3015 util_vma_heap_finish(&device->vma_cva);
3016 util_vma_heap_finish(&device->vma_lo);
3017 }
3018 fail_queue:
3019 anv_queue_finish(&device->queue);
3020 fail_context_id:
3021 anv_gem_destroy_context(device, device->context_id);
3022 fail_fd:
3023 close(device->fd);
3024 fail_device:
3025 vk_free(&device->vk.alloc, device);
3026
3027 return result;
3028 }
3029
3030 void anv_DestroyDevice(
3031 VkDevice _device,
3032 const VkAllocationCallbacks* pAllocator)
3033 {
3034 ANV_FROM_HANDLE(anv_device, device, _device);
3035
3036 if (!device)
3037 return;
3038
3039 anv_device_finish_blorp(device);
3040
3041 anv_pipeline_cache_finish(&device->default_pipeline_cache);
3042
3043 anv_queue_finish(&device->queue);
3044
3045 #ifdef HAVE_VALGRIND
3046 /* We only need to free these to prevent valgrind errors. The backing
3047 * BO will go away in a couple of lines so we don't actually leak.
3048 */
3049 if (device->info.gen >= 8)
3050 anv_state_reserved_pool_finish(&device->custom_border_colors);
3051 anv_state_pool_free(&device->dynamic_state_pool, device->border_colors);
3052 anv_state_pool_free(&device->dynamic_state_pool, device->slice_hash);
3053 #endif
3054
3055 anv_scratch_pool_finish(device, &device->scratch_pool);
3056
3057 anv_device_release_bo(device, device->workaround_bo);
3058 anv_device_release_bo(device, device->trivial_batch_bo);
3059 if (device->info.gen >= 10)
3060 anv_device_release_bo(device, device->hiz_clear_bo);
3061
3062 if (device->info.gen >= 12) {
3063 gen_aux_map_finish(device->aux_map_ctx);
3064 device->aux_map_ctx = NULL;
3065 }
3066
3067 if (device->physical->use_softpin)
3068 anv_state_pool_finish(&device->binding_table_pool);
3069 anv_state_pool_finish(&device->surface_state_pool);
3070 anv_state_pool_finish(&device->instruction_state_pool);
3071 anv_state_pool_finish(&device->dynamic_state_pool);
3072
3073 anv_bo_pool_finish(&device->batch_bo_pool);
3074
3075 anv_bo_cache_finish(&device->bo_cache);
3076
3077 if (device->physical->use_softpin) {
3078 util_vma_heap_finish(&device->vma_hi);
3079 util_vma_heap_finish(&device->vma_cva);
3080 util_vma_heap_finish(&device->vma_lo);
3081 }
3082
3083 pthread_cond_destroy(&device->queue_submit);
3084 pthread_mutex_destroy(&device->mutex);
3085
3086 anv_gem_destroy_context(device, device->context_id);
3087
3088 if (INTEL_DEBUG & DEBUG_BATCH)
3089 gen_batch_decode_ctx_finish(&device->decoder_ctx);
3090
3091 close(device->fd);
3092
3093 vk_device_finish(&device->vk);
3094 vk_free(&device->vk.alloc, device);
3095 }
3096
3097 VkResult anv_EnumerateInstanceLayerProperties(
3098 uint32_t* pPropertyCount,
3099 VkLayerProperties* pProperties)
3100 {
3101 if (pProperties == NULL) {
3102 *pPropertyCount = 0;
3103 return VK_SUCCESS;
3104 }
3105
3106 /* None supported at this time */
3107 return vk_error(VK_ERROR_LAYER_NOT_PRESENT);
3108 }
3109
3110 VkResult anv_EnumerateDeviceLayerProperties(
3111 VkPhysicalDevice physicalDevice,
3112 uint32_t* pPropertyCount,
3113 VkLayerProperties* pProperties)
3114 {
3115 if (pProperties == NULL) {
3116 *pPropertyCount = 0;
3117 return VK_SUCCESS;
3118 }
3119
3120 /* None supported at this time */
3121 return vk_error(VK_ERROR_LAYER_NOT_PRESENT);
3122 }
3123
3124 void anv_GetDeviceQueue(
3125 VkDevice _device,
3126 uint32_t queueNodeIndex,
3127 uint32_t queueIndex,
3128 VkQueue* pQueue)
3129 {
3130 const VkDeviceQueueInfo2 info = {