projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
util/os_file: extend os_read_file to return the file size
[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 device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
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 /* GENs prior to 8 do not support EU/Subslice info */
481 if (device->info.gen >= 8) {
482 device->subslice_total = anv_gem_get_param(fd, I915_PARAM_SUBSLICE_TOTAL);
483 device->eu_total = anv_gem_get_param(fd, I915_PARAM_EU_TOTAL);
484
485 /* Without this information, we cannot get the right Braswell
486 * brandstrings, and we have to use conservative numbers for GPGPU on
487 * many platforms, but otherwise, things will just work.
488 */
489 if (device->subslice_total < 1 || device->eu_total < 1) {
490 intel_logw("Kernel 4.1 required to properly query GPU properties");
491 }
492 } else if (device->info.gen == 7) {
493 device->subslice_total = 1 << (device->info.gt - 1);
494 }
495
496 if (device->info.is_cherryview &&
497 device->subslice_total > 0 && device->eu_total > 0) {
498 /* Logical CS threads = EUs per subslice * num threads per EU */
499 uint32_t max_cs_threads =
500 device->eu_total / device->subslice_total * device->info.num_thread_per_eu;
501
502 /* Fuse configurations may give more threads than expected, never less. */
503 if (max_cs_threads > device->info.max_cs_threads)
504 device->info.max_cs_threads = max_cs_threads;
505 }
506
507 device->compiler = brw_compiler_create(NULL, &device->info);
508 if (device->compiler == NULL) {
509 result = vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
510 goto fail_alloc;
511 }
512 device->compiler->shader_debug_log = compiler_debug_log;
513 device->compiler->shader_perf_log = compiler_perf_log;
514 device->compiler->supports_pull_constants = false;
515 device->compiler->constant_buffer_0_is_relative =
516 device->info.gen < 8 || !device->has_context_isolation;
517 device->compiler->supports_shader_constants = true;
518 device->compiler->compact_params = false;
519
520 /* Broadwell PRM says:
521 *
522 * "Before Gen8, there was a historical configuration control field to
523 * swizzle address bit[6] for in X/Y tiling modes. This was set in three
524 * different places: TILECTL[1:0], ARB_MODE[5:4], and
525 * DISP_ARB_CTL[14:13].
526 *
527 * For Gen8 and subsequent generations, the swizzle fields are all
528 * reserved, and the CPU's memory controller performs all address
529 * swizzling modifications."
530 */
531 bool swizzled =
532 device->info.gen < 8 && anv_gem_get_bit6_swizzle(fd, I915_TILING_X);
533
534 isl_device_init(&device->isl_dev, &device->info, swizzled);
535
536 result = anv_physical_device_init_uuids(device);
537 if (result != VK_SUCCESS)
538 goto fail_compiler;
539
540 anv_physical_device_init_disk_cache(device);
541
542 if (instance->enabled_extensions.KHR_display) {
543 master_fd = open(primary_path, O_RDWR | O_CLOEXEC);
544 if (master_fd >= 0) {
545 /* prod the device with a GETPARAM call which will fail if
546 * we don't have permission to even render on this device
547 */
548 if (anv_gem_get_param(master_fd, I915_PARAM_CHIPSET_ID) == 0) {
549 close(master_fd);
550 master_fd = -1;
551 }
552 }
553 }
554 device->master_fd = master_fd;
555
556 result = anv_init_wsi(device);
557 if (result != VK_SUCCESS)
558 goto fail_disk_cache;
559
560 device->perf = anv_get_perf(&device->info, fd);
561
562 anv_physical_device_get_supported_extensions(device,
563 &device->supported_extensions);
564
565
566 device->local_fd = fd;
567
568 *device_out = device;
569
570 return VK_SUCCESS;
571
572 fail_disk_cache:
573 anv_physical_device_free_disk_cache(device);
574 fail_compiler:
575 ralloc_free(device->compiler);
576 fail_alloc:
577 vk_free(&instance->alloc, device);
578 fail_fd:
579 close(fd);
580 if (master_fd != -1)
581 close(master_fd);
582 return result;
583 }
584
585 static void
586 anv_physical_device_destroy(struct anv_physical_device *device)
587 {
588 anv_finish_wsi(device);
589 anv_physical_device_free_disk_cache(device);
590 ralloc_free(device->compiler);
591 ralloc_free(device->perf);
592 close(device->local_fd);
593 if (device->master_fd >= 0)
594 close(device->master_fd);
595 vk_free(&device->instance->alloc, device);
596 }
597
598 static void *
599 default_alloc_func(void *pUserData, size_t size, size_t align,
600 VkSystemAllocationScope allocationScope)
601 {
602 return malloc(size);
603 }
604
605 static void *
606 default_realloc_func(void *pUserData, void *pOriginal, size_t size,
607 size_t align, VkSystemAllocationScope allocationScope)
608 {
609 return realloc(pOriginal, size);
610 }
611
612 static void
613 default_free_func(void *pUserData, void *pMemory)
614 {
615 free(pMemory);
616 }
617
618 static const VkAllocationCallbacks default_alloc = {
619 .pUserData = NULL,
620 .pfnAllocation = default_alloc_func,
621 .pfnReallocation = default_realloc_func,
622 .pfnFree = default_free_func,
623 };
624
625 VkResult anv_EnumerateInstanceExtensionProperties(
626 const char* pLayerName,
627 uint32_t* pPropertyCount,
628 VkExtensionProperties* pProperties)
629 {
630 VK_OUTARRAY_MAKE(out, pProperties, pPropertyCount);
631
632 for (int i = 0; i < ANV_INSTANCE_EXTENSION_COUNT; i++) {
633 if (anv_instance_extensions_supported.extensions[i]) {
634 vk_outarray_append(&out, prop) {
635 *prop = anv_instance_extensions[i];
636 }
637 }
638 }
639
640 return vk_outarray_status(&out);
641 }
642
643 VkResult anv_CreateInstance(
644 const VkInstanceCreateInfo* pCreateInfo,
645 const VkAllocationCallbacks* pAllocator,
646 VkInstance* pInstance)
647 {
648 struct anv_instance *instance;
649 VkResult result;
650
651 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO);
652
653 struct anv_instance_extension_table enabled_extensions = {};
654 for (uint32_t i = 0; i < pCreateInfo->enabledExtensionCount; i++) {
655 int idx;
656 for (idx = 0; idx < ANV_INSTANCE_EXTENSION_COUNT; idx++) {
657 if (strcmp(pCreateInfo->ppEnabledExtensionNames[i],
658 anv_instance_extensions[idx].extensionName) == 0)
659 break;
660 }
661
662 if (idx >= ANV_INSTANCE_EXTENSION_COUNT)
663 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);
664
665 if (!anv_instance_extensions_supported.extensions[idx])
666 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);
667
668 enabled_extensions.extensions[idx] = true;
669 }
670
671 instance = vk_alloc2(&default_alloc, pAllocator, sizeof(*instance), 8,
672 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
673 if (!instance)
674 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
675
676 instance->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
677
678 if (pAllocator)
679 instance->alloc = *pAllocator;
680 else
681 instance->alloc = default_alloc;
682
683 instance->app_info = (struct anv_app_info) { .api_version = 0 };
684 if (pCreateInfo->pApplicationInfo) {
685 const VkApplicationInfo *app = pCreateInfo->pApplicationInfo;
686
687 instance->app_info.app_name =
688 vk_strdup(&instance->alloc, app->pApplicationName,
689 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
690 instance->app_info.app_version = app->applicationVersion;
691
692 instance->app_info.engine_name =
693 vk_strdup(&instance->alloc, app->pEngineName,
694 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
695 instance->app_info.engine_version = app->engineVersion;
696
697 instance->app_info.api_version = app->apiVersion;
698 }
699
700 if (instance->app_info.api_version == 0)
701 instance->app_info.api_version = VK_API_VERSION_1_0;
702
703 instance->enabled_extensions = enabled_extensions;
704
705 for (unsigned i = 0; i < ARRAY_SIZE(instance->dispatch.entrypoints); i++) {
706 /* Vulkan requires that entrypoints for extensions which have not been
707 * enabled must not be advertised.
708 */
709 if (!anv_instance_entrypoint_is_enabled(i, instance->app_info.api_version,
710 &instance->enabled_extensions)) {
711 instance->dispatch.entrypoints[i] = NULL;
712 } else {
713 instance->dispatch.entrypoints[i] =
714 anv_instance_dispatch_table.entrypoints[i];
715 }
716 }
717
718 for (unsigned i = 0; i < ARRAY_SIZE(instance->physical_device_dispatch.entrypoints); i++) {
719 /* Vulkan requires that entrypoints for extensions which have not been
720 * enabled must not be advertised.
721 */
722 if (!anv_physical_device_entrypoint_is_enabled(i, instance->app_info.api_version,
723 &instance->enabled_extensions)) {
724 instance->physical_device_dispatch.entrypoints[i] = NULL;
725 } else {
726 instance->physical_device_dispatch.entrypoints[i] =
727 anv_physical_device_dispatch_table.entrypoints[i];
728 }
729 }
730
731 for (unsigned i = 0; i < ARRAY_SIZE(instance->device_dispatch.entrypoints); i++) {
732 /* Vulkan requires that entrypoints for extensions which have not been
733 * enabled must not be advertised.
734 */
735 if (!anv_device_entrypoint_is_enabled(i, instance->app_info.api_version,
736 &instance->enabled_extensions, NULL)) {
737 instance->device_dispatch.entrypoints[i] = NULL;
738 } else {
739 instance->device_dispatch.entrypoints[i] =
740 anv_device_dispatch_table.entrypoints[i];
741 }
742 }
743
744 instance->physical_devices_enumerated = false;
745 list_inithead(&instance->physical_devices);
746
747 result = vk_debug_report_instance_init(&instance->debug_report_callbacks);
748 if (result != VK_SUCCESS) {
749 vk_free2(&default_alloc, pAllocator, instance);
750 return vk_error(result);
751 }
752
753 instance->pipeline_cache_enabled =
754 env_var_as_boolean("ANV_ENABLE_PIPELINE_CACHE", true);
755
756 glsl_type_singleton_init_or_ref();
757
758 VG(VALGRIND_CREATE_MEMPOOL(instance, 0, false));
759
760 driParseOptionInfo(&instance->available_dri_options, anv_dri_options_xml);
761 driParseConfigFiles(&instance->dri_options, &instance->available_dri_options,
762 0, "anv", NULL,
763 instance->app_info.engine_name,
764 instance->app_info.engine_version);
765
766 *pInstance = anv_instance_to_handle(instance);
767
768 return VK_SUCCESS;
769 }
770
771 void anv_DestroyInstance(
772 VkInstance _instance,
773 const VkAllocationCallbacks* pAllocator)
774 {
775 ANV_FROM_HANDLE(anv_instance, instance, _instance);
776
777 if (!instance)
778 return;
779
780 list_for_each_entry_safe(struct anv_physical_device, pdevice,
781 &instance->physical_devices, link)
782 anv_physical_device_destroy(pdevice);
783
784 vk_free(&instance->alloc, (char *)instance->app_info.app_name);
785 vk_free(&instance->alloc, (char *)instance->app_info.engine_name);
786
787 VG(VALGRIND_DESTROY_MEMPOOL(instance));
788
789 vk_debug_report_instance_destroy(&instance->debug_report_callbacks);
790
791 glsl_type_singleton_decref();
792
793 driDestroyOptionCache(&instance->dri_options);
794 driDestroyOptionInfo(&instance->available_dri_options);
795
796 vk_free(&instance->alloc, instance);
797 }
798
799 static VkResult
800 anv_enumerate_physical_devices(struct anv_instance *instance)
801 {
802 if (instance->physical_devices_enumerated)
803 return VK_SUCCESS;
804
805 instance->physical_devices_enumerated = true;
806
807 /* TODO: Check for more devices ? */
808 drmDevicePtr devices[8];
809 int max_devices;
810
811 max_devices = drmGetDevices2(0, devices, ARRAY_SIZE(devices));
812 if (max_devices < 1)
813 return VK_SUCCESS;
814
815 VkResult result = VK_SUCCESS;
816 for (unsigned i = 0; i < (unsigned)max_devices; i++) {
817 if (devices[i]->available_nodes & 1 << DRM_NODE_RENDER &&
818 devices[i]->bustype == DRM_BUS_PCI &&
819 devices[i]->deviceinfo.pci->vendor_id == 0x8086) {
820
821 struct anv_physical_device *pdevice;
822 result = anv_physical_device_try_create(instance, devices[i],
823 &pdevice);
824 /* Incompatible DRM device, skip. */
825 if (result == VK_ERROR_INCOMPATIBLE_DRIVER) {
826 result = VK_SUCCESS;
827 continue;
828 }
829
830 /* Error creating the physical device, report the error. */
831 if (result != VK_SUCCESS)
832 break;
833
834 list_addtail(&pdevice->link, &instance->physical_devices);
835 }
836 }
837 drmFreeDevices(devices, max_devices);
838
839 /* If we successfully enumerated any devices, call it success */
840 return result;
841 }
842
843 VkResult anv_EnumeratePhysicalDevices(
844 VkInstance _instance,
845 uint32_t* pPhysicalDeviceCount,
846 VkPhysicalDevice* pPhysicalDevices)
847 {
848 ANV_FROM_HANDLE(anv_instance, instance, _instance);
849 VK_OUTARRAY_MAKE(out, pPhysicalDevices, pPhysicalDeviceCount);
850
851 VkResult result = anv_enumerate_physical_devices(instance);
852 if (result != VK_SUCCESS)
853 return result;
854
855 list_for_each_entry(struct anv_physical_device, pdevice,
856 &instance->physical_devices, link) {
857 vk_outarray_append(&out, i) {
858 *i = anv_physical_device_to_handle(pdevice);
859 }
860 }
861
862 return vk_outarray_status(&out);
863 }
864
865 VkResult anv_EnumeratePhysicalDeviceGroups(
866 VkInstance _instance,
867 uint32_t* pPhysicalDeviceGroupCount,
868 VkPhysicalDeviceGroupProperties* pPhysicalDeviceGroupProperties)
869 {
870 ANV_FROM_HANDLE(anv_instance, instance, _instance);
871 VK_OUTARRAY_MAKE(out, pPhysicalDeviceGroupProperties,
872 pPhysicalDeviceGroupCount);
873
874 VkResult result = anv_enumerate_physical_devices(instance);
875 if (result != VK_SUCCESS)
876 return result;
877
878 list_for_each_entry(struct anv_physical_device, pdevice,
879 &instance->physical_devices, link) {
880 vk_outarray_append(&out, p) {
881 p->physicalDeviceCount = 1;
882 memset(p->physicalDevices, 0, sizeof(p->physicalDevices));
883 p->physicalDevices[0] = anv_physical_device_to_handle(pdevice);
884 p->subsetAllocation = false;
885
886 vk_foreach_struct(ext, p->pNext)
887 anv_debug_ignored_stype(ext->sType);
888 }
889 }
890
891 return vk_outarray_status(&out);
892 }
893
894 void anv_GetPhysicalDeviceFeatures(
895 VkPhysicalDevice physicalDevice,
896 VkPhysicalDeviceFeatures* pFeatures)
897 {
898 ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
899
900 *pFeatures = (VkPhysicalDeviceFeatures) {
901 .robustBufferAccess = true,
902 .fullDrawIndexUint32 = true,
903 .imageCubeArray = true,
904 .independentBlend = true,
905 .geometryShader = true,
906 .tessellationShader = true,
907 .sampleRateShading = true,
908 .dualSrcBlend = true,
909 .logicOp = true,
910 .multiDrawIndirect = true,
911 .drawIndirectFirstInstance = true,
912 .depthClamp = true,
913 .depthBiasClamp = true,
914 .fillModeNonSolid = true,
915 .depthBounds = pdevice->info.gen >= 12,
916 .wideLines = true,
917 .largePoints = true,
918 .alphaToOne = true,
919 .multiViewport = true,
920 .samplerAnisotropy = true,
921 .textureCompressionETC2 = pdevice->info.gen >= 8 ||
922 pdevice->info.is_baytrail,
923 .textureCompressionASTC_LDR = pdevice->info.gen >= 9, /* FINISHME CHV */
924 .textureCompressionBC = true,
925 .occlusionQueryPrecise = true,
926 .pipelineStatisticsQuery = true,
927 .fragmentStoresAndAtomics = true,
928 .shaderTessellationAndGeometryPointSize = true,
929 .shaderImageGatherExtended = true,
930 .shaderStorageImageExtendedFormats = true,
931 .shaderStorageImageMultisample = false,
932 .shaderStorageImageReadWithoutFormat = false,
933 .shaderStorageImageWriteWithoutFormat = true,
934 .shaderUniformBufferArrayDynamicIndexing = true,
935 .shaderSampledImageArrayDynamicIndexing = true,
936 .shaderStorageBufferArrayDynamicIndexing = true,
937 .shaderStorageImageArrayDynamicIndexing = true,
938 .shaderClipDistance = true,
939 .shaderCullDistance = true,
940 .shaderFloat64 = pdevice->info.gen >= 8 &&
941 pdevice->info.has_64bit_float,
942 .shaderInt64 = pdevice->info.gen >= 8 &&
943 pdevice->info.has_64bit_int,
944 .shaderInt16 = pdevice->info.gen >= 8,
945 .shaderResourceMinLod = pdevice->info.gen >= 9,
946 .variableMultisampleRate = true,
947 .inheritedQueries = true,
948 };
949
950 /* We can't do image stores in vec4 shaders */
951 pFeatures->vertexPipelineStoresAndAtomics =
952 pdevice->compiler->scalar_stage[MESA_SHADER_VERTEX] &&
953 pdevice->compiler->scalar_stage[MESA_SHADER_GEOMETRY];
954
955 struct anv_app_info *app_info = &pdevice->instance->app_info;
956
957 /* The new DOOM and Wolfenstein games require depthBounds without
958 * checking for it. They seem to run fine without it so just claim it's
959 * there and accept the consequences.
960 */
961 if (app_info->engine_name && strcmp(app_info->engine_name, "idTech") == 0)
962 pFeatures->depthBounds = true;
963 }
964
965 static void
966 anv_get_physical_device_features_1_1(struct anv_physical_device *pdevice,
967 VkPhysicalDeviceVulkan11Features *f)
968 {
969 assert(f->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES);
970
971 f->storageBuffer16BitAccess = pdevice->info.gen >= 8;
972 f->uniformAndStorageBuffer16BitAccess = pdevice->info.gen >= 8;
973 f->storagePushConstant16 = pdevice->info.gen >= 8;
974 f->storageInputOutput16 = false;
975 f->multiview = true;
976 f->multiviewGeometryShader = true;
977 f->multiviewTessellationShader = true;
978 f->variablePointersStorageBuffer = true;
979 f->variablePointers = true;
980 f->protectedMemory = false;
981 f->samplerYcbcrConversion = true;
982 f->shaderDrawParameters = true;
983 }
984
985 static void
986 anv_get_physical_device_features_1_2(struct anv_physical_device *pdevice,
987 VkPhysicalDeviceVulkan12Features *f)
988 {
989 assert(f->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES);
990
991 f->samplerMirrorClampToEdge = true;
992 f->drawIndirectCount = true;
993 f->storageBuffer8BitAccess = pdevice->info.gen >= 8;
994 f->uniformAndStorageBuffer8BitAccess = pdevice->info.gen >= 8;
995 f->storagePushConstant8 = pdevice->info.gen >= 8;
996 f->shaderBufferInt64Atomics = pdevice->info.gen >= 9 &&
997 pdevice->use_softpin;
998 f->shaderSharedInt64Atomics = false;
999 f->shaderFloat16 = pdevice->info.gen >= 8;
1000 f->shaderInt8 = pdevice->info.gen >= 8;
1001
1002 bool descIndexing = pdevice->has_a64_buffer_access &&
1003 pdevice->has_bindless_images;
1004 f->descriptorIndexing = descIndexing;
1005 f->shaderInputAttachmentArrayDynamicIndexing = false;
1006 f->shaderUniformTexelBufferArrayDynamicIndexing = descIndexing;
1007 f->shaderStorageTexelBufferArrayDynamicIndexing = descIndexing;
1008 f->shaderUniformBufferArrayNonUniformIndexing = false;
1009 f->shaderSampledImageArrayNonUniformIndexing = descIndexing;
1010 f->shaderStorageBufferArrayNonUniformIndexing = descIndexing;
1011 f->shaderStorageImageArrayNonUniformIndexing = descIndexing;
1012 f->shaderInputAttachmentArrayNonUniformIndexing = false;
1013 f->shaderUniformTexelBufferArrayNonUniformIndexing = descIndexing;
1014 f->shaderStorageTexelBufferArrayNonUniformIndexing = descIndexing;
1015 f->descriptorBindingUniformBufferUpdateAfterBind = false;
1016 f->descriptorBindingSampledImageUpdateAfterBind = descIndexing;
1017 f->descriptorBindingStorageImageUpdateAfterBind = descIndexing;
1018 f->descriptorBindingStorageBufferUpdateAfterBind = descIndexing;
1019 f->descriptorBindingUniformTexelBufferUpdateAfterBind = descIndexing;
1020 f->descriptorBindingStorageTexelBufferUpdateAfterBind = descIndexing;
1021 f->descriptorBindingUpdateUnusedWhilePending = descIndexing;
1022 f->descriptorBindingPartiallyBound = descIndexing;
1023 f->descriptorBindingVariableDescriptorCount = false;
1024 f->runtimeDescriptorArray = descIndexing;
1025
1026 f->samplerFilterMinmax = pdevice->info.gen >= 9;
1027 f->scalarBlockLayout = true;
1028 f->imagelessFramebuffer = true;
1029 f->uniformBufferStandardLayout = true;
1030 f->shaderSubgroupExtendedTypes = true;
1031 f->separateDepthStencilLayouts = true;
1032 f->hostQueryReset = true;
1033 f->timelineSemaphore = true;
1034 f->bufferDeviceAddress = pdevice->has_a64_buffer_access;
1035 f->bufferDeviceAddressCaptureReplay = pdevice->has_a64_buffer_access;
1036 f->bufferDeviceAddressMultiDevice = false;
1037 f->vulkanMemoryModel = true;
1038 f->vulkanMemoryModelDeviceScope = true;
1039 f->vulkanMemoryModelAvailabilityVisibilityChains = true;
1040 f->shaderOutputViewportIndex = true;
1041 f->shaderOutputLayer = true;
1042 f->subgroupBroadcastDynamicId = true;
1043 }
1044
1045 void anv_GetPhysicalDeviceFeatures2(
1046 VkPhysicalDevice physicalDevice,
1047 VkPhysicalDeviceFeatures2* pFeatures)
1048 {
1049 ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
1050 anv_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features);
1051
1052 VkPhysicalDeviceVulkan11Features core_1_1 = {
1053 .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES,
1054 };
1055 anv_get_physical_device_features_1_1(pdevice, &core_1_1);
1056
1057 VkPhysicalDeviceVulkan12Features core_1_2 = {
1058 .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES,
1059 };
1060 anv_get_physical_device_features_1_2(pdevice, &core_1_2);
1061
1062 #define CORE_FEATURE(major, minor, feature) \
1063 features->feature = core_##major##_##minor.feature
1064
1065
1066 vk_foreach_struct(ext, pFeatures->pNext) {
1067 switch (ext->sType) {
1068 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_8BIT_STORAGE_FEATURES_KHR: {
1069 VkPhysicalDevice8BitStorageFeaturesKHR *features =
1070 (VkPhysicalDevice8BitStorageFeaturesKHR *)ext;
1071 CORE_FEATURE(1, 2, storageBuffer8BitAccess);
1072 CORE_FEATURE(1, 2, uniformAndStorageBuffer8BitAccess);
1073 CORE_FEATURE(1, 2, storagePushConstant8);
1074 break;
1075 }
1076
1077 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES: {
1078 VkPhysicalDevice16BitStorageFeatures *features =
1079 (VkPhysicalDevice16BitStorageFeatures *)ext;
1080 CORE_FEATURE(1, 1, storageBuffer16BitAccess);
1081 CORE_FEATURE(1, 1, uniformAndStorageBuffer16BitAccess);
1082 CORE_FEATURE(1, 1, storagePushConstant16);
1083 CORE_FEATURE(1, 1, storageInputOutput16);
1084 break;
1085 }
1086
1087 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES_EXT: {
1088 VkPhysicalDeviceBufferDeviceAddressFeaturesEXT *features = (void *)ext;
1089 features->bufferDeviceAddress = pdevice->has_a64_buffer_access;
1090 features->bufferDeviceAddressCaptureReplay = false;
1091 features->bufferDeviceAddressMultiDevice = false;
1092 break;
1093 }
1094
1095 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES_KHR: {
1096 VkPhysicalDeviceBufferDeviceAddressFeaturesKHR *features = (void *)ext;
1097 CORE_FEATURE(1, 2, bufferDeviceAddress);
1098 CORE_FEATURE(1, 2, bufferDeviceAddressCaptureReplay);
1099 CORE_FEATURE(1, 2, bufferDeviceAddressMultiDevice);
1100 break;
1101 }
1102
1103 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_COMPUTE_SHADER_DERIVATIVES_FEATURES_NV: {
1104 VkPhysicalDeviceComputeShaderDerivativesFeaturesNV *features =
1105 (VkPhysicalDeviceComputeShaderDerivativesFeaturesNV *)ext;
1106 features->computeDerivativeGroupQuads = true;
1107 features->computeDerivativeGroupLinear = true;
1108 break;
1109 }
1110
1111 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONDITIONAL_RENDERING_FEATURES_EXT: {
1112 VkPhysicalDeviceConditionalRenderingFeaturesEXT *features =
1113 (VkPhysicalDeviceConditionalRenderingFeaturesEXT*)ext;
1114 features->conditionalRendering = pdevice->info.gen >= 8 ||
1115 pdevice->info.is_haswell;
1116 features->inheritedConditionalRendering = pdevice->info.gen >= 8 ||
1117 pdevice->info.is_haswell;
1118 break;
1119 }
1120
1121 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_CLIP_ENABLE_FEATURES_EXT: {
1122 VkPhysicalDeviceDepthClipEnableFeaturesEXT *features =
1123 (VkPhysicalDeviceDepthClipEnableFeaturesEXT *)ext;
1124 features->depthClipEnable = true;
1125 break;
1126 }
1127
1128 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR: {
1129 VkPhysicalDeviceFloat16Int8FeaturesKHR *features = (void *)ext;
1130 CORE_FEATURE(1, 2, shaderFloat16);
1131 CORE_FEATURE(1, 2, shaderInt8);
1132 break;
1133 }
1134
1135 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FRAGMENT_SHADER_INTERLOCK_FEATURES_EXT: {
1136 VkPhysicalDeviceFragmentShaderInterlockFeaturesEXT *features =
1137 (VkPhysicalDeviceFragmentShaderInterlockFeaturesEXT *)ext;
1138 features->fragmentShaderSampleInterlock = pdevice->info.gen >= 9;
1139 features->fragmentShaderPixelInterlock = pdevice->info.gen >= 9;
1140 features->fragmentShaderShadingRateInterlock = false;
1141 break;
1142 }
1143
1144 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_HOST_QUERY_RESET_FEATURES_EXT: {
1145 VkPhysicalDeviceHostQueryResetFeaturesEXT *features =
1146 (VkPhysicalDeviceHostQueryResetFeaturesEXT *)ext;
1147 CORE_FEATURE(1, 2, hostQueryReset);
1148 break;
1149 }
1150
1151 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT: {
1152 VkPhysicalDeviceDescriptorIndexingFeaturesEXT *features =
1153 (VkPhysicalDeviceDescriptorIndexingFeaturesEXT *)ext;
1154 CORE_FEATURE(1, 2, shaderInputAttachmentArrayDynamicIndexing);
1155 CORE_FEATURE(1, 2, shaderUniformTexelBufferArrayDynamicIndexing);
1156 CORE_FEATURE(1, 2, shaderStorageTexelBufferArrayDynamicIndexing);
1157 CORE_FEATURE(1, 2, shaderUniformBufferArrayNonUniformIndexing);
1158 CORE_FEATURE(1, 2, shaderSampledImageArrayNonUniformIndexing);
1159 CORE_FEATURE(1, 2, shaderStorageBufferArrayNonUniformIndexing);
1160 CORE_FEATURE(1, 2, shaderStorageImageArrayNonUniformIndexing);
1161 CORE_FEATURE(1, 2, shaderInputAttachmentArrayNonUniformIndexing);
1162 CORE_FEATURE(1, 2, shaderUniformTexelBufferArrayNonUniformIndexing);
1163 CORE_FEATURE(1, 2, shaderStorageTexelBufferArrayNonUniformIndexing);
1164 CORE_FEATURE(1, 2, descriptorBindingUniformBufferUpdateAfterBind);
1165 CORE_FEATURE(1, 2, descriptorBindingSampledImageUpdateAfterBind);
1166 CORE_FEATURE(1, 2, descriptorBindingStorageImageUpdateAfterBind);
1167 CORE_FEATURE(1, 2, descriptorBindingStorageBufferUpdateAfterBind);
1168 CORE_FEATURE(1, 2, descriptorBindingUniformTexelBufferUpdateAfterBind);
1169 CORE_FEATURE(1, 2, descriptorBindingStorageTexelBufferUpdateAfterBind);
1170 CORE_FEATURE(1, 2, descriptorBindingUpdateUnusedWhilePending);
1171 CORE_FEATURE(1, 2, descriptorBindingPartiallyBound);
1172 CORE_FEATURE(1, 2, descriptorBindingVariableDescriptorCount);
1173 CORE_FEATURE(1, 2, runtimeDescriptorArray);
1174 break;
1175 }
1176
1177 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INDEX_TYPE_UINT8_FEATURES_EXT: {
1178 VkPhysicalDeviceIndexTypeUint8FeaturesEXT *features =
1179 (VkPhysicalDeviceIndexTypeUint8FeaturesEXT *)ext;
1180 features->indexTypeUint8 = true;
1181 break;
1182 }
1183
1184 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INLINE_UNIFORM_BLOCK_FEATURES_EXT: {
1185 VkPhysicalDeviceInlineUniformBlockFeaturesEXT *features =
1186 (VkPhysicalDeviceInlineUniformBlockFeaturesEXT *)ext;
1187 features->inlineUniformBlock = true;
1188 features->descriptorBindingInlineUniformBlockUpdateAfterBind = true;
1189 break;
1190 }
1191
1192 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_LINE_RASTERIZATION_FEATURES_EXT: {
1193 VkPhysicalDeviceLineRasterizationFeaturesEXT *features =
1194 (VkPhysicalDeviceLineRasterizationFeaturesEXT *)ext;
1195 features->rectangularLines = true;
1196 features->bresenhamLines = true;
1197 /* Support for Smooth lines with MSAA was removed on gen11. From the
1198 * BSpec section "Multisample ModesState" table for "AA Line Support
1199 * Requirements":
1200 *
1201 * GEN10:BUG:######## NUM_MULTISAMPLES == 1
1202 *
1203 * Fortunately, this isn't a case most people care about.
1204 */
1205 features->smoothLines = pdevice->info.gen < 10;
1206 features->stippledRectangularLines = false;
1207 features->stippledBresenhamLines = true;
1208 features->stippledSmoothLines = false;
1209 break;
1210 }
1211
1212 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES: {
1213 VkPhysicalDeviceMultiviewFeatures *features =
1214 (VkPhysicalDeviceMultiviewFeatures *)ext;
1215 CORE_FEATURE(1, 1, multiview);
1216 CORE_FEATURE(1, 1, multiviewGeometryShader);
1217 CORE_FEATURE(1, 1, multiviewTessellationShader);
1218 break;
1219 }
1220
1221 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_IMAGELESS_FRAMEBUFFER_FEATURES_KHR: {
1222 VkPhysicalDeviceImagelessFramebufferFeaturesKHR *features =
1223 (VkPhysicalDeviceImagelessFramebufferFeaturesKHR *)ext;
1224 CORE_FEATURE(1, 2, imagelessFramebuffer);
1225 break;
1226 }
1227
1228 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PIPELINE_EXECUTABLE_PROPERTIES_FEATURES_KHR: {
1229 VkPhysicalDevicePipelineExecutablePropertiesFeaturesKHR *features =
1230 (VkPhysicalDevicePipelineExecutablePropertiesFeaturesKHR *)ext;
1231 features->pipelineExecutableInfo = true;
1232 break;
1233 }
1234
1235 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES: {
1236 VkPhysicalDeviceProtectedMemoryFeatures *features = (void *)ext;
1237 CORE_FEATURE(1, 1, protectedMemory);
1238 break;
1239 }
1240
1241 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES: {
1242 VkPhysicalDeviceSamplerYcbcrConversionFeatures *features =
1243 (VkPhysicalDeviceSamplerYcbcrConversionFeatures *) ext;
1244 CORE_FEATURE(1, 1, samplerYcbcrConversion);
1245 break;
1246 }
1247
1248 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SCALAR_BLOCK_LAYOUT_FEATURES_EXT: {
1249 VkPhysicalDeviceScalarBlockLayoutFeaturesEXT *features =
1250 (VkPhysicalDeviceScalarBlockLayoutFeaturesEXT *)ext;
1251 CORE_FEATURE(1, 2, scalarBlockLayout);
1252 break;
1253 }
1254
1255 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SEPARATE_DEPTH_STENCIL_LAYOUTS_FEATURES_KHR: {
1256 VkPhysicalDeviceSeparateDepthStencilLayoutsFeaturesKHR *features =
1257 (VkPhysicalDeviceSeparateDepthStencilLayoutsFeaturesKHR *)ext;
1258 CORE_FEATURE(1, 2, separateDepthStencilLayouts);
1259 break;
1260 }
1261
1262 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_INT64_FEATURES_KHR: {
1263 VkPhysicalDeviceShaderAtomicInt64FeaturesKHR *features = (void *)ext;
1264 CORE_FEATURE(1, 2, shaderBufferInt64Atomics);
1265 CORE_FEATURE(1, 2, shaderSharedInt64Atomics);
1266 break;
1267 }
1268
1269 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DEMOTE_TO_HELPER_INVOCATION_FEATURES_EXT: {
1270 VkPhysicalDeviceShaderDemoteToHelperInvocationFeaturesEXT *features = (void *)ext;
1271 features->shaderDemoteToHelperInvocation = true;
1272 break;
1273 }
1274
1275 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_CLOCK_FEATURES_KHR: {
1276 VkPhysicalDeviceShaderClockFeaturesKHR *features =
1277 (VkPhysicalDeviceShaderClockFeaturesKHR *)ext;
1278 features->shaderSubgroupClock = true;
1279 features->shaderDeviceClock = false;
1280 break;
1281 }
1282
1283 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETERS_FEATURES: {
1284 VkPhysicalDeviceShaderDrawParametersFeatures *features = (void *)ext;
1285 CORE_FEATURE(1, 1, shaderDrawParameters);
1286 break;
1287 }
1288
1289 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_SUBGROUP_EXTENDED_TYPES_FEATURES_KHR: {
1290 VkPhysicalDeviceShaderSubgroupExtendedTypesFeaturesKHR *features =
1291 (VkPhysicalDeviceShaderSubgroupExtendedTypesFeaturesKHR *)ext;
1292 CORE_FEATURE(1, 2, shaderSubgroupExtendedTypes);
1293 break;
1294 }
1295
1296 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_FEATURES_EXT: {
1297 VkPhysicalDeviceSubgroupSizeControlFeaturesEXT *features =
1298 (VkPhysicalDeviceSubgroupSizeControlFeaturesEXT *)ext;
1299 features->subgroupSizeControl = true;
1300 features->computeFullSubgroups = true;
1301 break;
1302 }
1303
1304 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TEXEL_BUFFER_ALIGNMENT_FEATURES_EXT: {
1305 VkPhysicalDeviceTexelBufferAlignmentFeaturesEXT *features =
1306 (VkPhysicalDeviceTexelBufferAlignmentFeaturesEXT *)ext;
1307 features->texelBufferAlignment = true;
1308 break;
1309 }
1310
1311 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES_KHR: {
1312 VkPhysicalDeviceTimelineSemaphoreFeaturesKHR *features =
1313 (VkPhysicalDeviceTimelineSemaphoreFeaturesKHR *) ext;
1314 CORE_FEATURE(1, 2, timelineSemaphore);
1315 break;
1316 }
1317
1318 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTERS_FEATURES: {
1319 VkPhysicalDeviceVariablePointersFeatures *features = (void *)ext;
1320 CORE_FEATURE(1, 1, variablePointersStorageBuffer);
1321 CORE_FEATURE(1, 1, variablePointers);
1322 break;
1323 }
1324
1325 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT: {
1326 VkPhysicalDeviceTransformFeedbackFeaturesEXT *features =
1327 (VkPhysicalDeviceTransformFeedbackFeaturesEXT *)ext;
1328 features->transformFeedback = true;
1329 features->geometryStreams = true;
1330 break;
1331 }
1332
1333 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_UNIFORM_BUFFER_STANDARD_LAYOUT_FEATURES_KHR: {
1334 VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR *features =
1335 (VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR *)ext;
1336 CORE_FEATURE(1, 2, uniformBufferStandardLayout);
1337 break;
1338 }
1339
1340 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_FEATURES_EXT: {
1341 VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT *features =
1342 (VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT *)ext;
1343 features->vertexAttributeInstanceRateDivisor = true;
1344 features->vertexAttributeInstanceRateZeroDivisor = true;
1345 break;
1346 }
1347
1348 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES:
1349 anv_get_physical_device_features_1_1(pdevice, (void *)ext);
1350 break;
1351
1352 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES:
1353 anv_get_physical_device_features_1_2(pdevice, (void *)ext);
1354 break;
1355
1356 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_MEMORY_MODEL_FEATURES_KHR: {
1357 VkPhysicalDeviceVulkanMemoryModelFeaturesKHR *features = (void *)ext;
1358 CORE_FEATURE(1, 2, vulkanMemoryModel);
1359 CORE_FEATURE(1, 2, vulkanMemoryModelDeviceScope);
1360 CORE_FEATURE(1, 2, vulkanMemoryModelAvailabilityVisibilityChains);
1361 break;
1362 }
1363
1364 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_YCBCR_IMAGE_ARRAYS_FEATURES_EXT: {
1365 VkPhysicalDeviceYcbcrImageArraysFeaturesEXT *features =
1366 (VkPhysicalDeviceYcbcrImageArraysFeaturesEXT *)ext;
1367 features->ycbcrImageArrays = true;
1368 break;
1369 }
1370
1371 default:
1372 anv_debug_ignored_stype(ext->sType);
1373 break;
1374 }
1375 }
1376
1377 #undef CORE_FEATURE
1378 }
1379
1380 #define MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS 64
1381
1382 #define MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS 64
1383 #define MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS 256
1384
1385 void anv_GetPhysicalDeviceProperties(
1386 VkPhysicalDevice physicalDevice,
1387 VkPhysicalDeviceProperties* pProperties)
1388 {
1389 ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
1390 const struct gen_device_info *devinfo = &pdevice->info;
1391
1392 /* See assertions made when programming the buffer surface state. */
1393 const uint32_t max_raw_buffer_sz = devinfo->gen >= 7 ?
1394 (1ul << 30) : (1ul << 27);
1395
1396 const uint32_t max_ssbos = pdevice->has_a64_buffer_access ? UINT16_MAX : 64;
1397 const uint32_t max_textures =
1398 pdevice->has_bindless_images ? UINT16_MAX : 128;
1399 const uint32_t max_samplers =
1400 pdevice->has_bindless_samplers ? UINT16_MAX :
1401 (devinfo->gen >= 8 || devinfo->is_haswell) ? 128 : 16;
1402 const uint32_t max_images =
1403 pdevice->has_bindless_images ? UINT16_MAX : MAX_IMAGES;
1404
1405 /* If we can use bindless for everything, claim a high per-stage limit,
1406 * otherwise use the binding table size, minus the slots reserved for
1407 * render targets and one slot for the descriptor buffer. */
1408 const uint32_t max_per_stage =
1409 pdevice->has_bindless_images && pdevice->has_a64_buffer_access
1410 ? UINT32_MAX : MAX_BINDING_TABLE_SIZE - MAX_RTS - 1;
1411
1412 const uint32_t max_workgroup_size = 32 * devinfo->max_cs_threads;
1413
1414 VkSampleCountFlags sample_counts =
1415 isl_device_get_sample_counts(&pdevice->isl_dev);
1416
1417
1418 VkPhysicalDeviceLimits limits = {
1419 .maxImageDimension1D = (1 << 14),
1420 .maxImageDimension2D = (1 << 14),
1421 .maxImageDimension3D = (1 << 11),
1422 .maxImageDimensionCube = (1 << 14),
1423 .maxImageArrayLayers = (1 << 11),
1424 .maxTexelBufferElements = 128 * 1024 * 1024,
1425 .maxUniformBufferRange = (1ul << 27),
1426 .maxStorageBufferRange = max_raw_buffer_sz,
1427 .maxPushConstantsSize = MAX_PUSH_CONSTANTS_SIZE,
1428 .maxMemoryAllocationCount = UINT32_MAX,
1429 .maxSamplerAllocationCount = 64 * 1024,
1430 .bufferImageGranularity = 64, /* A cache line */
1431 .sparseAddressSpaceSize = 0,
1432 .maxBoundDescriptorSets = MAX_SETS,
1433 .maxPerStageDescriptorSamplers = max_samplers,
1434 .maxPerStageDescriptorUniformBuffers = MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS,
1435 .maxPerStageDescriptorStorageBuffers = max_ssbos,
1436 .maxPerStageDescriptorSampledImages = max_textures,
1437 .maxPerStageDescriptorStorageImages = max_images,
1438 .maxPerStageDescriptorInputAttachments = MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS,
1439 .maxPerStageResources = max_per_stage,
1440 .maxDescriptorSetSamplers = 6 * max_samplers, /* number of stages * maxPerStageDescriptorSamplers */
1441 .maxDescriptorSetUniformBuffers = 6 * MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS, /* number of stages * maxPerStageDescriptorUniformBuffers */
1442 .maxDescriptorSetUniformBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2,
1443 .maxDescriptorSetStorageBuffers = 6 * max_ssbos, /* number of stages * maxPerStageDescriptorStorageBuffers */
1444 .maxDescriptorSetStorageBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2,
1445 .maxDescriptorSetSampledImages = 6 * max_textures, /* number of stages * maxPerStageDescriptorSampledImages */
1446 .maxDescriptorSetStorageImages = 6 * max_images, /* number of stages * maxPerStageDescriptorStorageImages */
1447 .maxDescriptorSetInputAttachments = MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS,
1448 .maxVertexInputAttributes = MAX_VBS,
1449 .maxVertexInputBindings = MAX_VBS,
1450 .maxVertexInputAttributeOffset = 2047,
1451 .maxVertexInputBindingStride = 2048,
1452 .maxVertexOutputComponents = 128,
1453 .maxTessellationGenerationLevel = 64,
1454 .maxTessellationPatchSize = 32,
1455 .maxTessellationControlPerVertexInputComponents = 128,
1456 .maxTessellationControlPerVertexOutputComponents = 128,
1457 .maxTessellationControlPerPatchOutputComponents = 128,
1458 .maxTessellationControlTotalOutputComponents = 2048,
1459 .maxTessellationEvaluationInputComponents = 128,
1460 .maxTessellationEvaluationOutputComponents = 128,
1461 .maxGeometryShaderInvocations = 32,
1462 .maxGeometryInputComponents = 64,
1463 .maxGeometryOutputComponents = 128,
1464 .maxGeometryOutputVertices = 256,
1465 .maxGeometryTotalOutputComponents = 1024,
1466 .maxFragmentInputComponents = 116, /* 128 components - (PSIZ, CLIP_DIST0, CLIP_DIST1) */
1467 .maxFragmentOutputAttachments = 8,
1468 .maxFragmentDualSrcAttachments = 1,
1469 .maxFragmentCombinedOutputResources = 8,
1470 .maxComputeSharedMemorySize = 64 * 1024,
1471 .maxComputeWorkGroupCount = { 65535, 65535, 65535 },
1472 .maxComputeWorkGroupInvocations = max_workgroup_size,
1473 .maxComputeWorkGroupSize = {
1474 max_workgroup_size,
1475 max_workgroup_size,
1476 max_workgroup_size,
1477 },
1478 .subPixelPrecisionBits = 8,
1479 .subTexelPrecisionBits = 8,
1480 .mipmapPrecisionBits = 8,
1481 .maxDrawIndexedIndexValue = UINT32_MAX,
1482 .maxDrawIndirectCount = UINT32_MAX,
1483 .maxSamplerLodBias = 16,
1484 .maxSamplerAnisotropy = 16,
1485 .maxViewports = MAX_VIEWPORTS,
1486 .maxViewportDimensions = { (1 << 14), (1 << 14) },
1487 .viewportBoundsRange = { INT16_MIN, INT16_MAX },
1488 .viewportSubPixelBits = 13, /* We take a float? */
1489 .minMemoryMapAlignment = 4096, /* A page */
1490 /* The dataport requires texel alignment so we need to assume a worst
1491 * case of R32G32B32A32 which is 16 bytes.
1492 */
1493 .minTexelBufferOffsetAlignment = 16,
1494 /* We need 16 for UBO block reads to work and 32 for push UBOs */
1495 .minUniformBufferOffsetAlignment = 32,
1496 .minStorageBufferOffsetAlignment = 4,
1497 .minTexelOffset = -8,
1498 .maxTexelOffset = 7,
1499 .minTexelGatherOffset = -32,
1500 .maxTexelGatherOffset = 31,
1501 .minInterpolationOffset = -0.5,
1502 .maxInterpolationOffset = 0.4375,
1503 .subPixelInterpolationOffsetBits = 4,
1504 .maxFramebufferWidth = (1 << 14),
1505 .maxFramebufferHeight = (1 << 14),
1506 .maxFramebufferLayers = (1 << 11),
1507 .framebufferColorSampleCounts = sample_counts,
1508 .framebufferDepthSampleCounts = sample_counts,
1509 .framebufferStencilSampleCounts = sample_counts,
1510 .framebufferNoAttachmentsSampleCounts = sample_counts,
1511 .maxColorAttachments = MAX_RTS,
1512 .sampledImageColorSampleCounts = sample_counts,
1513 .sampledImageIntegerSampleCounts = sample_counts,
1514 .sampledImageDepthSampleCounts = sample_counts,
1515 .sampledImageStencilSampleCounts = sample_counts,
1516 .storageImageSampleCounts = VK_SAMPLE_COUNT_1_BIT,
1517 .maxSampleMaskWords = 1,
1518 .timestampComputeAndGraphics = true,
1519 .timestampPeriod = 1000000000.0 / devinfo->timestamp_frequency,
1520 .maxClipDistances = 8,
1521 .maxCullDistances = 8,
1522 .maxCombinedClipAndCullDistances = 8,
1523 .discreteQueuePriorities = 2,
1524 .pointSizeRange = { 0.125, 255.875 },
1525 .lineWidthRange = {
1526 0.0,
1527 (devinfo->gen >= 9 || devinfo->is_cherryview) ?
1528 2047.9921875 : 7.9921875,
1529 },
1530 .pointSizeGranularity = (1.0 / 8.0),
1531 .lineWidthGranularity = (1.0 / 128.0),
1532 .strictLines = false,
1533 .standardSampleLocations = true,
1534 .optimalBufferCopyOffsetAlignment = 128,
1535 .optimalBufferCopyRowPitchAlignment = 128,
1536 .nonCoherentAtomSize = 64,
1537 };
1538
1539 *pProperties = (VkPhysicalDeviceProperties) {
1540 .apiVersion = anv_physical_device_api_version(pdevice),
1541 .driverVersion = vk_get_driver_version(),
1542 .vendorID = 0x8086,
1543 .deviceID = pdevice->info.chipset_id,
1544 .deviceType = VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU,
1545 .limits = limits,
1546 .sparseProperties = {0}, /* Broadwell doesn't do sparse. */
1547 };
1548
1549 snprintf(pProperties->deviceName, sizeof(pProperties->deviceName),
1550 "%s", pdevice->name);
1551 memcpy(pProperties->pipelineCacheUUID,
1552 pdevice->pipeline_cache_uuid, VK_UUID_SIZE);
1553 }
1554
1555 static void
1556 anv_get_physical_device_properties_1_1(struct anv_physical_device *pdevice,
1557 VkPhysicalDeviceVulkan11Properties *p)
1558 {
1559 assert(p->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_PROPERTIES);
1560
1561 memcpy(p->deviceUUID, pdevice->device_uuid, VK_UUID_SIZE);
1562 memcpy(p->driverUUID, pdevice->driver_uuid, VK_UUID_SIZE);
1563 memset(p->deviceLUID, 0, VK_LUID_SIZE);
1564 p->deviceNodeMask = 0;
1565 p->deviceLUIDValid = false;
1566
1567 p->subgroupSize = BRW_SUBGROUP_SIZE;
1568 VkShaderStageFlags scalar_stages = 0;
1569 for (unsigned stage = 0; stage < MESA_SHADER_STAGES; stage++) {
1570 if (pdevice->compiler->scalar_stage[stage])
1571 scalar_stages |= mesa_to_vk_shader_stage(stage);
1572 }
1573 p->subgroupSupportedStages = scalar_stages;
1574 p->subgroupSupportedOperations = VK_SUBGROUP_FEATURE_BASIC_BIT |
1575 VK_SUBGROUP_FEATURE_VOTE_BIT |
1576 VK_SUBGROUP_FEATURE_BALLOT_BIT |
1577 VK_SUBGROUP_FEATURE_SHUFFLE_BIT |
1578 VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT |
1579 VK_SUBGROUP_FEATURE_QUAD_BIT;
1580 if (pdevice->info.gen >= 8) {
1581 /* TODO: There's no technical reason why these can't be made to
1582 * work on gen7 but they don't at the moment so it's best to leave
1583 * the feature disabled than enabled and broken.
1584 */
1585 p->subgroupSupportedOperations |= VK_SUBGROUP_FEATURE_ARITHMETIC_BIT |
1586 VK_SUBGROUP_FEATURE_CLUSTERED_BIT;
1587 }
1588 p->subgroupQuadOperationsInAllStages = pdevice->info.gen >= 8;
1589
1590 p->pointClippingBehavior = VK_POINT_CLIPPING_BEHAVIOR_USER_CLIP_PLANES_ONLY;
1591 p->maxMultiviewViewCount = 16;
1592 p->maxMultiviewInstanceIndex = UINT32_MAX / 16;
1593 p->protectedNoFault = false;
1594 /* This value doesn't matter for us today as our per-stage descriptors are
1595 * the real limit.
1596 */
1597 p->maxPerSetDescriptors = 1024;
1598 p->maxMemoryAllocationSize = MAX_MEMORY_ALLOCATION_SIZE;
1599 }
1600
1601 static void
1602 anv_get_physical_device_properties_1_2(struct anv_physical_device *pdevice,
1603 VkPhysicalDeviceVulkan12Properties *p)
1604 {
1605 assert(p->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_PROPERTIES);
1606
1607 p->driverID = VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA_KHR;
1608 memset(p->driverName, 0, sizeof(p->driverName));
1609 snprintf(p->driverName, VK_MAX_DRIVER_NAME_SIZE_KHR,
1610 "Intel open-source Mesa driver");
1611 memset(p->driverInfo, 0, sizeof(p->driverInfo));
1612 snprintf(p->driverInfo, VK_MAX_DRIVER_INFO_SIZE_KHR,
1613 "Mesa " PACKAGE_VERSION MESA_GIT_SHA1);
1614 p->conformanceVersion = (VkConformanceVersionKHR) {
1615 .major = 1,
1616 .minor = 2,
1617 .subminor = 0,
1618 .patch = 0,
1619 };
1620
1621 p->denormBehaviorIndependence =
1622 VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_ALL_KHR;
1623 p->roundingModeIndependence =
1624 VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_NONE_KHR;
1625
1626 /* Broadwell does not support HF denorms and there are restrictions
1627 * other gens. According to Kabylake's PRM:
1628 *
1629 * "math - Extended Math Function
1630 * [...]
1631 * Restriction : Half-float denorms are always retained."
1632 */
1633 p->shaderDenormFlushToZeroFloat16 = false;
1634 p->shaderDenormPreserveFloat16 = pdevice->info.gen > 8;
1635 p->shaderRoundingModeRTEFloat16 = true;
1636 p->shaderRoundingModeRTZFloat16 = true;
1637 p->shaderSignedZeroInfNanPreserveFloat16 = true;
1638
1639 p->shaderDenormFlushToZeroFloat32 = true;
1640 p->shaderDenormPreserveFloat32 = true;
1641 p->shaderRoundingModeRTEFloat32 = true;
1642 p->shaderRoundingModeRTZFloat32 = true;
1643 p->shaderSignedZeroInfNanPreserveFloat32 = true;
1644
1645 p->shaderDenormFlushToZeroFloat64 = true;
1646 p->shaderDenormPreserveFloat64 = true;
1647 p->shaderRoundingModeRTEFloat64 = true;
1648 p->shaderRoundingModeRTZFloat64 = true;
1649 p->shaderSignedZeroInfNanPreserveFloat64 = true;
1650
1651 /* It's a bit hard to exactly map our implementation to the limits
1652 * described here. The bindless surface handle in the extended
1653 * message descriptors is 20 bits and it's an index into the table of
1654 * RENDER_SURFACE_STATE structs that starts at bindless surface base
1655 * address. Given that most things consume two surface states per
1656 * view (general/sampled for textures and write-only/read-write for
1657 * images), we claim 2^19 things.
1658 *
1659 * For SSBOs, we just use A64 messages so there is no real limit
1660 * there beyond the limit on the total size of a descriptor set.
1661 */
1662 const unsigned max_bindless_views = 1 << 19;
1663 p->maxUpdateAfterBindDescriptorsInAllPools = max_bindless_views;
1664 p->shaderUniformBufferArrayNonUniformIndexingNative = false;
1665 p->shaderSampledImageArrayNonUniformIndexingNative = false;
1666 p->shaderStorageBufferArrayNonUniformIndexingNative = true;
1667 p->shaderStorageImageArrayNonUniformIndexingNative = false;
1668 p->shaderInputAttachmentArrayNonUniformIndexingNative = false;
1669 p->robustBufferAccessUpdateAfterBind = true;
1670 p->quadDivergentImplicitLod = false;
1671 p->maxPerStageDescriptorUpdateAfterBindSamplers = max_bindless_views;
1672 p->maxPerStageDescriptorUpdateAfterBindUniformBuffers = MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS;
1673 p->maxPerStageDescriptorUpdateAfterBindStorageBuffers = UINT32_MAX;
1674 p->maxPerStageDescriptorUpdateAfterBindSampledImages = max_bindless_views;
1675 p->maxPerStageDescriptorUpdateAfterBindStorageImages = max_bindless_views;
1676 p->maxPerStageDescriptorUpdateAfterBindInputAttachments = MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS;
1677 p->maxPerStageUpdateAfterBindResources = UINT32_MAX;
1678 p->maxDescriptorSetUpdateAfterBindSamplers = max_bindless_views;
1679 p->maxDescriptorSetUpdateAfterBindUniformBuffers = 6 * MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS;
1680 p->maxDescriptorSetUpdateAfterBindUniformBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2;
1681 p->maxDescriptorSetUpdateAfterBindStorageBuffers = UINT32_MAX;
1682 p->maxDescriptorSetUpdateAfterBindStorageBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2;
1683 p->maxDescriptorSetUpdateAfterBindSampledImages = max_bindless_views;
1684 p->maxDescriptorSetUpdateAfterBindStorageImages = max_bindless_views;
1685 p->maxDescriptorSetUpdateAfterBindInputAttachments = MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS;
1686
1687 /* We support all of the depth resolve modes */
1688 p->supportedDepthResolveModes = VK_RESOLVE_MODE_SAMPLE_ZERO_BIT_KHR |
1689 VK_RESOLVE_MODE_AVERAGE_BIT_KHR |
1690 VK_RESOLVE_MODE_MIN_BIT_KHR |
1691 VK_RESOLVE_MODE_MAX_BIT_KHR;
1692 /* Average doesn't make sense for stencil so we don't support that */
1693 p->supportedStencilResolveModes = VK_RESOLVE_MODE_SAMPLE_ZERO_BIT_KHR;
1694 if (pdevice->info.gen >= 8) {
1695 /* The advanced stencil resolve modes currently require stencil
1696 * sampling be supported by the hardware.
1697 */
1698 p->supportedStencilResolveModes |= VK_RESOLVE_MODE_MIN_BIT_KHR |
1699 VK_RESOLVE_MODE_MAX_BIT_KHR;
1700 }
1701 p->independentResolveNone = true;
1702 p->independentResolve = true;
1703
1704 p->filterMinmaxSingleComponentFormats = pdevice->info.gen >= 9;
1705 p->filterMinmaxImageComponentMapping = pdevice->info.gen >= 9;
1706
1707 p->maxTimelineSemaphoreValueDifference = UINT64_MAX;
1708
1709 p->framebufferIntegerColorSampleCounts =
1710 isl_device_get_sample_counts(&pdevice->isl_dev);
1711 }
1712
1713 void anv_GetPhysicalDeviceProperties2(
1714 VkPhysicalDevice physicalDevice,
1715 VkPhysicalDeviceProperties2* pProperties)
1716 {
1717 ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
1718
1719 anv_GetPhysicalDeviceProperties(physicalDevice, &pProperties->properties);
1720
1721 VkPhysicalDeviceVulkan11Properties core_1_1 = {
1722 .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_PROPERTIES,
1723 };
1724 anv_get_physical_device_properties_1_1(pdevice, &core_1_1);
1725
1726 VkPhysicalDeviceVulkan12Properties core_1_2 = {
1727 .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_PROPERTIES,
1728 };
1729 anv_get_physical_device_properties_1_2(pdevice, &core_1_2);
1730
1731 #define CORE_RENAMED_PROPERTY(major, minor, ext_property, core_property) \
1732 memcpy(&properties->ext_property, &core_##major##_##minor.core_property, \
1733 sizeof(core_##major##_##minor.core_property))
1734
1735 #define CORE_PROPERTY(major, minor, property) \
1736 CORE_RENAMED_PROPERTY(major, minor, property, property)
1737
1738 vk_foreach_struct(ext, pProperties->pNext) {
1739 switch (ext->sType) {
1740 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_STENCIL_RESOLVE_PROPERTIES_KHR: {
1741 VkPhysicalDeviceDepthStencilResolvePropertiesKHR *properties =
1742 (VkPhysicalDeviceDepthStencilResolvePropertiesKHR *)ext;
1743 CORE_PROPERTY(1, 2, supportedDepthResolveModes);
1744 CORE_PROPERTY(1, 2, supportedStencilResolveModes);
1745 CORE_PROPERTY(1, 2, independentResolveNone);
1746 CORE_PROPERTY(1, 2, independentResolve);
1747 break;
1748 }
1749
1750 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_PROPERTIES_EXT: {
1751 VkPhysicalDeviceDescriptorIndexingPropertiesEXT *properties =
1752 (VkPhysicalDeviceDescriptorIndexingPropertiesEXT *)ext;
1753 CORE_PROPERTY(1, 2, maxUpdateAfterBindDescriptorsInAllPools);
1754 CORE_PROPERTY(1, 2, shaderUniformBufferArrayNonUniformIndexingNative);
1755 CORE_PROPERTY(1, 2, shaderSampledImageArrayNonUniformIndexingNative);
1756 CORE_PROPERTY(1, 2, shaderStorageBufferArrayNonUniformIndexingNative);
1757 CORE_PROPERTY(1, 2, shaderStorageImageArrayNonUniformIndexingNative);
1758 CORE_PROPERTY(1, 2, shaderInputAttachmentArrayNonUniformIndexingNative);
1759 CORE_PROPERTY(1, 2, robustBufferAccessUpdateAfterBind);
1760 CORE_PROPERTY(1, 2, quadDivergentImplicitLod);
1761 CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindSamplers);
1762 CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindUniformBuffers);
1763 CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindStorageBuffers);
1764 CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindSampledImages);
1765 CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindStorageImages);
1766 CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindInputAttachments);
1767 CORE_PROPERTY(1, 2, maxPerStageUpdateAfterBindResources);
1768 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindSamplers);
1769 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindUniformBuffers);
1770 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindUniformBuffersDynamic);
1771 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindStorageBuffers);
1772 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindStorageBuffersDynamic);
1773 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindSampledImages);
1774 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindStorageImages);
1775 CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindInputAttachments);
1776 break;
1777 }
1778
1779 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES_KHR: {
1780 VkPhysicalDeviceDriverPropertiesKHR *properties =
1781 (VkPhysicalDeviceDriverPropertiesKHR *) ext;
1782 CORE_PROPERTY(1, 2, driverID);
1783 CORE_PROPERTY(1, 2, driverName);
1784 CORE_PROPERTY(1, 2, driverInfo);
1785 CORE_PROPERTY(1, 2, conformanceVersion);
1786 break;
1787 }
1788
1789 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_MEMORY_HOST_PROPERTIES_EXT: {
1790 VkPhysicalDeviceExternalMemoryHostPropertiesEXT *props =
1791 (VkPhysicalDeviceExternalMemoryHostPropertiesEXT *) ext;
1792 /* Userptr needs page aligned memory. */
1793 props->minImportedHostPointerAlignment = 4096;
1794 break;
1795 }
1796
1797 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES: {
1798 VkPhysicalDeviceIDProperties *properties =
1799 (VkPhysicalDeviceIDProperties *)ext;
1800 CORE_PROPERTY(1, 1, deviceUUID);
1801 CORE_PROPERTY(1, 1, driverUUID);
1802 CORE_PROPERTY(1, 1, deviceLUID);
1803 CORE_PROPERTY(1, 1, deviceLUIDValid);
1804 break;
1805 }
1806
1807 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INLINE_UNIFORM_BLOCK_PROPERTIES_EXT: {
1808 VkPhysicalDeviceInlineUniformBlockPropertiesEXT *props =
1809 (VkPhysicalDeviceInlineUniformBlockPropertiesEXT *)ext;
1810 props->maxInlineUniformBlockSize = MAX_INLINE_UNIFORM_BLOCK_SIZE;
1811 props->maxPerStageDescriptorInlineUniformBlocks =
1812 MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
1813 props->maxPerStageDescriptorUpdateAfterBindInlineUniformBlocks =
1814 MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
1815 props->maxDescriptorSetInlineUniformBlocks =
1816 MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
1817 props->maxDescriptorSetUpdateAfterBindInlineUniformBlocks =
1818 MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
1819 break;
1820 }
1821
1822 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_LINE_RASTERIZATION_PROPERTIES_EXT: {
1823 VkPhysicalDeviceLineRasterizationPropertiesEXT *props =
1824 (VkPhysicalDeviceLineRasterizationPropertiesEXT *)ext;
1825 /* In the Skylake PRM Vol. 7, subsection titled "GIQ (Diamond)
1826 * Sampling Rules - Legacy Mode", it says the following:
1827 *
1828 * "Note that the device divides a pixel into a 16x16 array of
1829 * subpixels, referenced by their upper left corners."
1830 *
1831 * This is the only known reference in the PRMs to the subpixel
1832 * precision of line rasterization and a "16x16 array of subpixels"
1833 * implies 4 subpixel precision bits. Empirical testing has shown
1834 * that 4 subpixel precision bits applies to all line rasterization
1835 * types.
1836 */
1837 props->lineSubPixelPrecisionBits = 4;
1838 break;
1839 }
1840
1841 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES: {
1842 VkPhysicalDeviceMaintenance3Properties *properties =
1843 (VkPhysicalDeviceMaintenance3Properties *)ext;
1844 /* This value doesn't matter for us today as our per-stage
1845 * descriptors are the real limit.
1846 */
1847 CORE_PROPERTY(1, 1, maxPerSetDescriptors);
1848 CORE_PROPERTY(1, 1, maxMemoryAllocationSize);
1849 break;
1850 }
1851
1852 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_PROPERTIES: {
1853 VkPhysicalDeviceMultiviewProperties *properties =
1854 (VkPhysicalDeviceMultiviewProperties *)ext;
1855 CORE_PROPERTY(1, 1, maxMultiviewViewCount);
1856 CORE_PROPERTY(1, 1, maxMultiviewInstanceIndex);
1857 break;
1858 }
1859
1860 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PCI_BUS_INFO_PROPERTIES_EXT: {
1861 VkPhysicalDevicePCIBusInfoPropertiesEXT *properties =
1862 (VkPhysicalDevicePCIBusInfoPropertiesEXT *)ext;
1863 properties->pciDomain = pdevice->pci_info.domain;
1864 properties->pciBus = pdevice->pci_info.bus;
1865 properties->pciDevice = pdevice->pci_info.device;
1866 properties->pciFunction = pdevice->pci_info.function;
1867 break;
1868 }
1869
1870 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_POINT_CLIPPING_PROPERTIES: {
1871 VkPhysicalDevicePointClippingProperties *properties =
1872 (VkPhysicalDevicePointClippingProperties *) ext;
1873 CORE_PROPERTY(1, 1, pointClippingBehavior);
1874 break;
1875 }
1876
1877 #pragma GCC diagnostic push
1878 #pragma GCC diagnostic ignored "-Wswitch"
1879 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRESENTATION_PROPERTIES_ANDROID: {
1880 VkPhysicalDevicePresentationPropertiesANDROID *props =
1881 (VkPhysicalDevicePresentationPropertiesANDROID *)ext;
1882 props->sharedImage = VK_FALSE;
1883 break;
1884 }
1885 #pragma GCC diagnostic pop
1886
1887 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_PROPERTIES: {
1888 VkPhysicalDeviceProtectedMemoryProperties *properties =
1889 (VkPhysicalDeviceProtectedMemoryProperties *)ext;
1890 CORE_PROPERTY(1, 1, protectedNoFault);
1891 break;
1892 }
1893
1894 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR: {
1895 VkPhysicalDevicePushDescriptorPropertiesKHR *properties =
1896 (VkPhysicalDevicePushDescriptorPropertiesKHR *) ext;
1897 properties->maxPushDescriptors = MAX_PUSH_DESCRIPTORS;
1898 break;
1899 }
1900
1901 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_FILTER_MINMAX_PROPERTIES_EXT: {
1902 VkPhysicalDeviceSamplerFilterMinmaxPropertiesEXT *properties =
1903 (VkPhysicalDeviceSamplerFilterMinmaxPropertiesEXT *)ext;
1904 CORE_PROPERTY(1, 2, filterMinmaxImageComponentMapping);
1905 CORE_PROPERTY(1, 2, filterMinmaxSingleComponentFormats);
1906 break;
1907 }
1908
1909 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES: {
1910 VkPhysicalDeviceSubgroupProperties *properties = (void *)ext;
1911 CORE_PROPERTY(1, 1, subgroupSize);
1912 CORE_RENAMED_PROPERTY(1, 1, supportedStages,
1913 subgroupSupportedStages);
1914 CORE_RENAMED_PROPERTY(1, 1, supportedOperations,
1915 subgroupSupportedOperations);
1916 CORE_RENAMED_PROPERTY(1, 1, quadOperationsInAllStages,
1917 subgroupQuadOperationsInAllStages);
1918 break;
1919 }
1920
1921 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_PROPERTIES_EXT: {
1922 VkPhysicalDeviceSubgroupSizeControlPropertiesEXT *props =
1923 (VkPhysicalDeviceSubgroupSizeControlPropertiesEXT *)ext;
1924 STATIC_ASSERT(8 <= BRW_SUBGROUP_SIZE && BRW_SUBGROUP_SIZE <= 32);
1925 props->minSubgroupSize = 8;
1926 props->maxSubgroupSize = 32;
1927 props->maxComputeWorkgroupSubgroups = pdevice->info.max_cs_threads;
1928 props->requiredSubgroupSizeStages = VK_SHADER_STAGE_COMPUTE_BIT;
1929 break;
1930 }
1931 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT_CONTROLS_PROPERTIES_KHR : {
1932 VkPhysicalDeviceFloatControlsPropertiesKHR *properties = (void *)ext;
1933 CORE_PROPERTY(1, 2, denormBehaviorIndependence);
1934 CORE_PROPERTY(1, 2, roundingModeIndependence);
1935 CORE_PROPERTY(1, 2, shaderDenormFlushToZeroFloat16);
1936 CORE_PROPERTY(1, 2, shaderDenormPreserveFloat16);
1937 CORE_PROPERTY(1, 2, shaderRoundingModeRTEFloat16);
1938 CORE_PROPERTY(1, 2, shaderRoundingModeRTZFloat16);
1939 CORE_PROPERTY(1, 2, shaderSignedZeroInfNanPreserveFloat16);
1940 CORE_PROPERTY(1, 2, shaderDenormFlushToZeroFloat32);
1941 CORE_PROPERTY(1, 2, shaderDenormPreserveFloat32);
1942 CORE_PROPERTY(1, 2, shaderRoundingModeRTEFloat32);
1943 CORE_PROPERTY(1, 2, shaderRoundingModeRTZFloat32);
1944 CORE_PROPERTY(1, 2, shaderSignedZeroInfNanPreserveFloat32);
1945 CORE_PROPERTY(1, 2, shaderDenormFlushToZeroFloat64);
1946 CORE_PROPERTY(1, 2, shaderDenormPreserveFloat64);
1947 CORE_PROPERTY(1, 2, shaderRoundingModeRTEFloat64);
1948 CORE_PROPERTY(1, 2, shaderRoundingModeRTZFloat64);
1949 CORE_PROPERTY(1, 2, shaderSignedZeroInfNanPreserveFloat64);
1950 break;
1951 }
1952
1953 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TEXEL_BUFFER_ALIGNMENT_PROPERTIES_EXT: {
1954 VkPhysicalDeviceTexelBufferAlignmentPropertiesEXT *props =
1955 (VkPhysicalDeviceTexelBufferAlignmentPropertiesEXT *)ext;
1956
1957 /* From the SKL PRM Vol. 2d, docs for RENDER_SURFACE_STATE::Surface
1958 * Base Address:
1959 *
1960 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field
1961 * specifies the base address of the first element of the surface,
1962 * computed in software by adding the surface base address to the
1963 * byte offset of the element in the buffer. The base address must
1964 * be aligned to element size."
1965 *
1966 * The typed dataport messages require that things be texel aligned.
1967 * Otherwise, we may just load/store the wrong data or, in the worst
1968 * case, there may be hangs.
1969 */
1970 props->storageTexelBufferOffsetAlignmentBytes = 16;
1971 props->storageTexelBufferOffsetSingleTexelAlignment = true;
1972
1973 /* The sampler, however, is much more forgiving and it can handle
1974 * arbitrary byte alignment for linear and buffer surfaces. It's
1975 * hard to find a good PRM citation for this but years of empirical
1976 * experience demonstrate that this is true.
1977 */
1978 props->uniformTexelBufferOffsetAlignmentBytes = 1;
1979 props->uniformTexelBufferOffsetSingleTexelAlignment = false;
1980 break;
1981 }
1982
1983 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_PROPERTIES_KHR: {
1984 VkPhysicalDeviceTimelineSemaphorePropertiesKHR *properties =
1985 (VkPhysicalDeviceTimelineSemaphorePropertiesKHR *) ext;
1986 CORE_PROPERTY(1, 2, maxTimelineSemaphoreValueDifference);
1987 break;
1988 }
1989
1990 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT: {
1991 VkPhysicalDeviceTransformFeedbackPropertiesEXT *props =
1992 (VkPhysicalDeviceTransformFeedbackPropertiesEXT *)ext;
1993
1994 props->maxTransformFeedbackStreams = MAX_XFB_STREAMS;
1995 props->maxTransformFeedbackBuffers = MAX_XFB_BUFFERS;
1996 props->maxTransformFeedbackBufferSize = (1ull << 32);
1997 props->maxTransformFeedbackStreamDataSize = 128 * 4;
1998 props->maxTransformFeedbackBufferDataSize = 128 * 4;
1999 props->maxTransformFeedbackBufferDataStride = 2048;
2000 props->transformFeedbackQueries = true;
2001 props->transformFeedbackStreamsLinesTriangles = false;
2002 props->transformFeedbackRasterizationStreamSelect = false;
2003 props->transformFeedbackDraw = true;
2004 break;
2005 }
2006
2007 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT: {
2008 VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT *props =
2009 (VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT *)ext;
2010 /* We have to restrict this a bit for multiview */
2011 props->maxVertexAttribDivisor = UINT32_MAX / 16;
2012 break;
2013 }
2014
2015 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_PROPERTIES:
2016 anv_get_physical_device_properties_1_1(pdevice, (void *)ext);
2017 break;
2018
2019 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_PROPERTIES:
2020 anv_get_physical_device_properties_1_2(pdevice, (void *)ext);
2021 break;
2022
2023 default:
2024 anv_debug_ignored_stype(ext->sType);
2025 break;
2026 }
2027 }
2028
2029 #undef CORE_RENAMED_PROPERTY
2030 #undef CORE_PROPERTY
2031 }
2032
2033 /* We support exactly one queue family. */
2034 static const VkQueueFamilyProperties
2035 anv_queue_family_properties = {
2036 .queueFlags = VK_QUEUE_GRAPHICS_BIT |
2037 VK_QUEUE_COMPUTE_BIT |
2038 VK_QUEUE_TRANSFER_BIT,
2039 .queueCount = 1,
2040 .timestampValidBits = 36, /* XXX: Real value here */
2041 .minImageTransferGranularity = { 1, 1, 1 },
2042 };
2043
2044 void anv_GetPhysicalDeviceQueueFamilyProperties(
2045 VkPhysicalDevice physicalDevice,
2046 uint32_t* pCount,
2047 VkQueueFamilyProperties* pQueueFamilyProperties)
2048 {
2049 VK_OUTARRAY_MAKE(out, pQueueFamilyProperties, pCount);
2050
2051 vk_outarray_append(&out, p) {
2052 *p = anv_queue_family_properties;
2053 }
2054 }
2055
2056 void anv_GetPhysicalDeviceQueueFamilyProperties2(
2057 VkPhysicalDevice physicalDevice,
2058 uint32_t* pQueueFamilyPropertyCount,
2059 VkQueueFamilyProperties2* pQueueFamilyProperties)
2060 {
2061
2062 VK_OUTARRAY_MAKE(out, pQueueFamilyProperties, pQueueFamilyPropertyCount);
2063
2064 vk_outarray_append(&out, p) {
2065 p->queueFamilyProperties = anv_queue_family_properties;
2066
2067 vk_foreach_struct(s, p->pNext) {
2068 anv_debug_ignored_stype(s->sType);
2069 }
2070 }
2071 }
2072
2073 void anv_GetPhysicalDeviceMemoryProperties(
2074 VkPhysicalDevice physicalDevice,
2075 VkPhysicalDeviceMemoryProperties* pMemoryProperties)
2076 {
2077 ANV_FROM_HANDLE(anv_physical_device, physical_device, physicalDevice);
2078
2079 pMemoryProperties->memoryTypeCount = physical_device->memory.type_count;
2080 for (uint32_t i = 0; i < physical_device->memory.type_count; i++) {
2081 pMemoryProperties->memoryTypes[i] = (VkMemoryType) {
2082 .propertyFlags = physical_device->memory.types[i].propertyFlags,
2083 .heapIndex = physical_device->memory.types[i].heapIndex,
2084 };
2085 }
2086
2087 pMemoryProperties->memoryHeapCount = physical_device->memory.heap_count;
2088 for (uint32_t i = 0; i < physical_device->memory.heap_count; i++) {
2089 pMemoryProperties->memoryHeaps[i] = (VkMemoryHeap) {
2090 .size = physical_device->memory.heaps[i].size,
2091 .flags = physical_device->memory.heaps[i].flags,
2092 };
2093 }
2094 }
2095
2096 static void
2097 anv_get_memory_budget(VkPhysicalDevice physicalDevice,
2098 VkPhysicalDeviceMemoryBudgetPropertiesEXT *memoryBudget)
2099 {
2100 ANV_FROM_HANDLE(anv_physical_device, device, physicalDevice);
2101 uint64_t sys_available = get_available_system_memory();
2102 assert(sys_available > 0);
2103
2104 VkDeviceSize total_heaps_size = 0;
2105 for (size_t i = 0; i < device->memory.heap_count; i++)
2106 total_heaps_size += device->memory.heaps[i].size;
2107
2108 for (size_t i = 0; i < device->memory.heap_count; i++) {
2109 VkDeviceSize heap_size = device->memory.heaps[i].size;
2110 VkDeviceSize heap_used = device->memory.heaps[i].used;
2111 VkDeviceSize heap_budget;
2112
2113 double heap_proportion = (double) heap_size / total_heaps_size;
2114 VkDeviceSize sys_available_prop = sys_available * heap_proportion;
2115
2116 /*
2117 * Let's not incite the app to starve the system: report at most 90% of
2118 * available system memory.
2119 */
2120 uint64_t heap_available = sys_available_prop * 9 / 10;
2121 heap_budget = MIN2(heap_size, heap_used + heap_available);
2122
2123 /*
2124 * Round down to the nearest MB
2125 */
2126 heap_budget &= ~((1ull << 20) - 1);
2127
2128 /*
2129 * The heapBudget value must be non-zero for array elements less than
2130 * VkPhysicalDeviceMemoryProperties::memoryHeapCount. The heapBudget
2131 * value must be less than or equal to VkMemoryHeap::size for each heap.
2132 */
2133 assert(0 < heap_budget && heap_budget <= heap_size);
2134
2135 memoryBudget->heapUsage[i] = heap_used;
2136 memoryBudget->heapBudget[i] = heap_budget;
2137 }
2138
2139 /* The heapBudget and heapUsage values must be zero for array elements
2140 * greater than or equal to VkPhysicalDeviceMemoryProperties::memoryHeapCount
2141 */
2142 for (uint32_t i = device->memory.heap_count; i < VK_MAX_MEMORY_HEAPS; i++) {
2143 memoryBudget->heapBudget[i] = 0;
2144 memoryBudget->heapUsage[i] = 0;
2145 }
2146 }
2147
2148 void anv_GetPhysicalDeviceMemoryProperties2(
2149 VkPhysicalDevice physicalDevice,
2150 VkPhysicalDeviceMemoryProperties2* pMemoryProperties)
2151 {
2152 anv_GetPhysicalDeviceMemoryProperties(physicalDevice,
2153 &pMemoryProperties->memoryProperties);
2154
2155 vk_foreach_struct(ext, pMemoryProperties->pNext) {
2156 switch (ext->sType) {
2157 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_BUDGET_PROPERTIES_EXT:
2158 anv_get_memory_budget(physicalDevice, (void*)ext);
2159 break;
2160 default:
2161 anv_debug_ignored_stype(ext->sType);
2162 break;
2163 }
2164 }
2165 }
2166
2167 void
2168 anv_GetDeviceGroupPeerMemoryFeatures(
2169 VkDevice device,
2170 uint32_t heapIndex,
2171 uint32_t localDeviceIndex,
2172 uint32_t remoteDeviceIndex,
2173 VkPeerMemoryFeatureFlags* pPeerMemoryFeatures)
2174 {
2175 assert(localDeviceIndex == 0 && remoteDeviceIndex == 0);
2176 *pPeerMemoryFeatures = VK_PEER_MEMORY_FEATURE_COPY_SRC_BIT |
2177 VK_PEER_MEMORY_FEATURE_COPY_DST_BIT |
2178 VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT |
2179 VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT;
2180 }
2181
2182 PFN_vkVoidFunction anv_GetInstanceProcAddr(
2183 VkInstance _instance,
2184 const char* pName)
2185 {
2186 ANV_FROM_HANDLE(anv_instance, instance, _instance);
2187
2188 /* The Vulkan 1.0 spec for vkGetInstanceProcAddr has a table of exactly
2189 * when we have to return valid function pointers, NULL, or it's left
2190 * undefined. See the table for exact details.
2191 */
2192 if (pName == NULL)
2193 return NULL;
2194
2195 #define LOOKUP_ANV_ENTRYPOINT(entrypoint) \
2196 if (strcmp(pName, "vk" #entrypoint) == 0) \
2197 return (PFN_vkVoidFunction)anv_##entrypoint
2198
2199 LOOKUP_ANV_ENTRYPOINT(EnumerateInstanceExtensionProperties);
2200 LOOKUP_ANV_ENTRYPOINT(EnumerateInstanceLayerProperties);
2201 LOOKUP_ANV_ENTRYPOINT(EnumerateInstanceVersion);
2202 LOOKUP_ANV_ENTRYPOINT(CreateInstance);
2203
2204 #undef LOOKUP_ANV_ENTRYPOINT
2205
2206 if (instance == NULL)
2207 return NULL;
2208
2209 int idx = anv_get_instance_entrypoint_index(pName);
2210 if (idx >= 0)
2211 return instance->dispatch.entrypoints[idx];
2212
2213 idx = anv_get_physical_device_entrypoint_index(pName);
2214 if (idx >= 0)
2215 return instance->physical_device_dispatch.entrypoints[idx];
2216
2217 idx = anv_get_device_entrypoint_index(pName);
2218 if (idx >= 0)
2219 return instance->device_dispatch.entrypoints[idx];
2220
2221 return NULL;
2222 }
2223
2224 /* With version 1+ of the loader interface the ICD should expose
2225 * vk_icdGetInstanceProcAddr to work around certain LD_PRELOAD issues seen in apps.
2226 */
2227 PUBLIC
2228 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL vk_icdGetInstanceProcAddr(
2229 VkInstance instance,
2230 const char* pName);
2231
2232 PUBLIC
2233 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL vk_icdGetInstanceProcAddr(
2234 VkInstance instance,
2235 const char* pName)
2236 {
2237 return anv_GetInstanceProcAddr(instance, pName);
2238 }
2239
2240 PFN_vkVoidFunction anv_GetDeviceProcAddr(
2241 VkDevice _device,
2242 const char* pName)
2243 {
2244 ANV_FROM_HANDLE(anv_device, device, _device);
2245
2246 if (!device || !pName)
2247 return NULL;
2248
2249 int idx = anv_get_device_entrypoint_index(pName);
2250 if (idx < 0)
2251 return NULL;
2252
2253 return device->dispatch.entrypoints[idx];
2254 }
2255
2256 /* With version 4+ of the loader interface the ICD should expose
2257 * vk_icdGetPhysicalDeviceProcAddr()
2258 */
2259 PUBLIC
2260 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL vk_icdGetPhysicalDeviceProcAddr(
2261 VkInstance _instance,
2262 const char* pName);
2263
2264 PFN_vkVoidFunction vk_icdGetPhysicalDeviceProcAddr(
2265 VkInstance _instance,
2266 const char* pName)
2267 {
2268 ANV_FROM_HANDLE(anv_instance, instance, _instance);
2269
2270 if (!pName || !instance)
2271 return NULL;
2272
2273 int idx = anv_get_physical_device_entrypoint_index(pName);
2274 if (idx < 0)
2275 return NULL;
2276
2277 return instance->physical_device_dispatch.entrypoints[idx];
2278 }
2279
2280
2281 VkResult
2282 anv_CreateDebugReportCallbackEXT(VkInstance _instance,
2283 const VkDebugReportCallbackCreateInfoEXT* pCreateInfo,
2284 const VkAllocationCallbacks* pAllocator,
2285 VkDebugReportCallbackEXT* pCallback)
2286 {
2287 ANV_FROM_HANDLE(anv_instance, instance, _instance);
2288 return vk_create_debug_report_callback(&instance->debug_report_callbacks,
2289 pCreateInfo, pAllocator, &instance->alloc,
2290 pCallback);
2291 }
2292
2293 void
2294 anv_DestroyDebugReportCallbackEXT(VkInstance _instance,
2295 VkDebugReportCallbackEXT _callback,
2296 const VkAllocationCallbacks* pAllocator)
2297 {
2298 ANV_FROM_HANDLE(anv_instance, instance, _instance);
2299 vk_destroy_debug_report_callback(&instance->debug_report_callbacks,
2300 _callback, pAllocator, &instance->alloc);
2301 }
2302
2303 void
2304 anv_DebugReportMessageEXT(VkInstance _instance,
2305 VkDebugReportFlagsEXT flags,
2306 VkDebugReportObjectTypeEXT objectType,
2307 uint64_t object,
2308 size_t location,
2309 int32_t messageCode,
2310 const char* pLayerPrefix,
2311 const char* pMessage)
2312 {
2313 ANV_FROM_HANDLE(anv_instance, instance, _instance);
2314 vk_debug_report(&instance->debug_report_callbacks, flags, objectType,
2315 object, location, messageCode, pLayerPrefix, pMessage);
2316 }
2317
2318 static struct anv_state
2319 anv_state_pool_emit_data(struct anv_state_pool *pool, size_t size, size_t align, const void *p)
2320 {
2321 struct anv_state state;
2322
2323 state = anv_state_pool_alloc(pool, size, align);
2324 memcpy(state.map, p, size);
2325
2326 return state;
2327 }
2328
2329 /* Haswell border color is a bit of a disaster. Float and unorm formats use a
2330 * straightforward 32-bit float color in the first 64 bytes. Instead of using
2331 * a nice float/integer union like Gen8+, Haswell specifies the integer border
2332 * color as a separate entry /after/ the float color. The layout of this entry
2333 * also depends on the format's bpp (with extra hacks for RG32), and overlaps.
2334 *
2335 * Since we don't know the format/bpp, we can't make any of the border colors
2336 * containing '1' work for all formats, as it would be in the wrong place for
2337 * some of them. We opt to make 32-bit integers work as this seems like the
2338 * most common option. Fortunately, transparent black works regardless, as
2339 * all zeroes is the same in every bit-size.
2340 */
2341 struct hsw_border_color {
2342 float float32[4];
2343 uint32_t _pad0[12];
2344 uint32_t uint32[4];
2345 uint32_t _pad1[108];
2346 };
2347
2348 struct gen8_border_color {
2349 union {
2350 float float32[4];
2351 uint32_t uint32[4];
2352 };
2353 /* Pad out to 64 bytes */
2354 uint32_t _pad[12];
2355 };
2356
2357 static void
2358 anv_device_init_border_colors(struct anv_device *device)
2359 {
2360 if (device->info.is_haswell) {
2361 static const struct hsw_border_color border_colors[] = {
2362 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK] = { .float32 = { 0.0, 0.0, 0.0, 0.0 } },
2363 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK] = { .float32 = { 0.0, 0.0, 0.0, 1.0 } },
2364 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE] = { .float32 = { 1.0, 1.0, 1.0, 1.0 } },
2365 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK] = { .uint32 = { 0, 0, 0, 0 } },
2366 [VK_BORDER_COLOR_INT_OPAQUE_BLACK] = { .uint32 = { 0, 0, 0, 1 } },
2367 [VK_BORDER_COLOR_INT_OPAQUE_WHITE] = { .uint32 = { 1, 1, 1, 1 } },
2368 };
2369
2370 device->border_colors =
2371 anv_state_pool_emit_data(&device->dynamic_state_pool,
2372 sizeof(border_colors), 512, border_colors);
2373 } else {
2374 static const struct gen8_border_color border_colors[] = {
2375 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK] = { .float32 = { 0.0, 0.0, 0.0, 0.0 } },
2376 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK] = { .float32 = { 0.0, 0.0, 0.0, 1.0 } },
2377 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE] = { .float32 = { 1.0, 1.0, 1.0, 1.0 } },
2378 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK] = { .uint32 = { 0, 0, 0, 0 } },
2379 [VK_BORDER_COLOR_INT_OPAQUE_BLACK] = { .uint32 = { 0, 0, 0, 1 } },
2380 [VK_BORDER_COLOR_INT_OPAQUE_WHITE] = { .uint32 = { 1, 1, 1, 1 } },
2381 };
2382
2383 device->border_colors =
2384 anv_state_pool_emit_data(&device->dynamic_state_pool,
2385 sizeof(border_colors), 64, border_colors);
2386 }
2387 }
2388
2389 static VkResult
2390 anv_device_init_trivial_batch(struct anv_device *device)
2391 {
2392 VkResult result = anv_device_alloc_bo(device, 4096,
2393 ANV_BO_ALLOC_MAPPED,
2394 0 /* explicit_address */,
2395 &device->trivial_batch_bo);
2396 if (result != VK_SUCCESS)
2397 return result;
2398
2399 struct anv_batch batch = {
2400 .start = device->trivial_batch_bo->map,
2401 .next = device->trivial_batch_bo->map,
2402 .end = device->trivial_batch_bo->map + 4096,
2403 };
2404
2405 anv_batch_emit(&batch, GEN7_MI_BATCH_BUFFER_END, bbe);
2406 anv_batch_emit(&batch, GEN7_MI_NOOP, noop);
2407
2408 if (!device->info.has_llc)
2409 gen_clflush_range(batch.start, batch.next - batch.start);
2410
2411 return VK_SUCCESS;
2412 }
2413
2414 VkResult anv_EnumerateDeviceExtensionProperties(
2415 VkPhysicalDevice physicalDevice,
2416 const char* pLayerName,
2417 uint32_t* pPropertyCount,
2418 VkExtensionProperties* pProperties)
2419 {
2420 ANV_FROM_HANDLE(anv_physical_device, device, physicalDevice);
2421 VK_OUTARRAY_MAKE(out, pProperties, pPropertyCount);
2422
2423 for (int i = 0; i < ANV_DEVICE_EXTENSION_COUNT; i++) {
2424 if (device->supported_extensions.extensions[i]) {
2425 vk_outarray_append(&out, prop) {
2426 *prop = anv_device_extensions[i];
2427 }
2428 }
2429 }
2430
2431 return vk_outarray_status(&out);
2432 }
2433
2434 static void
2435 anv_device_init_dispatch(struct anv_device *device)
2436 {
2437 const struct anv_instance *instance = device->physical->instance;
2438
2439 const struct anv_device_dispatch_table *genX_table;
2440 switch (device->info.gen) {
2441 case 12:
2442 genX_table = &gen12_device_dispatch_table;
2443 break;
2444 case 11:
2445 genX_table = &gen11_device_dispatch_table;
2446 break;
2447 case 10:
2448 genX_table = &gen10_device_dispatch_table;
2449 break;
2450 case 9:
2451 genX_table = &gen9_device_dispatch_table;
2452 break;
2453 case 8:
2454 genX_table = &gen8_device_dispatch_table;
2455 break;
2456 case 7:
2457 if (device->info.is_haswell)
2458 genX_table = &gen75_device_dispatch_table;
2459 else
2460 genX_table = &gen7_device_dispatch_table;
2461 break;
2462 default:
2463 unreachable("unsupported gen\n");
2464 }
2465
2466 for (unsigned i = 0; i < ARRAY_SIZE(device->dispatch.entrypoints); i++) {
2467 /* Vulkan requires that entrypoints for extensions which have not been
2468 * enabled must not be advertised.
2469 */
2470 if (!anv_device_entrypoint_is_enabled(i, instance->app_info.api_version,
2471 &instance->enabled_extensions,
2472 &device->enabled_extensions)) {
2473 device->dispatch.entrypoints[i] = NULL;
2474 } else if (genX_table->entrypoints[i]) {
2475 device->dispatch.entrypoints[i] = genX_table->entrypoints[i];
2476 } else {
2477 device->dispatch.entrypoints[i] =
2478 anv_device_dispatch_table.entrypoints[i];
2479 }
2480 }
2481 }
2482
2483 static int
2484 vk_priority_to_gen(int priority)
2485 {
2486 switch (priority) {
2487 case VK_QUEUE_GLOBAL_PRIORITY_LOW_EXT:
2488 return GEN_CONTEXT_LOW_PRIORITY;
2489 case VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_EXT:
2490 return GEN_CONTEXT_MEDIUM_PRIORITY;
2491 case VK_QUEUE_GLOBAL_PRIORITY_HIGH_EXT:
2492 return GEN_CONTEXT_HIGH_PRIORITY;
2493 case VK_QUEUE_GLOBAL_PRIORITY_REALTIME_EXT:
2494 return GEN_CONTEXT_REALTIME_PRIORITY;
2495 default:
2496 unreachable("Invalid priority");
2497 }
2498 }
2499
2500 static VkResult
2501 anv_device_init_hiz_clear_value_bo(struct anv_device *device)
2502 {
2503 VkResult result = anv_device_alloc_bo(device, 4096,
2504 ANV_BO_ALLOC_MAPPED,
2505 0 /* explicit_address */,
2506 &device->hiz_clear_bo);
2507 if (result != VK_SUCCESS)
2508 return result;
2509
2510 union isl_color_value hiz_clear = { .u32 = { 0, } };
2511 hiz_clear.f32[0] = ANV_HZ_FC_VAL;
2512
2513 memcpy(device->hiz_clear_bo->map, hiz_clear.u32, sizeof(hiz_clear.u32));
2514
2515 if (!device->info.has_llc)
2516 gen_clflush_range(device->hiz_clear_bo->map, sizeof(hiz_clear.u32));
2517
2518 return VK_SUCCESS;
2519 }
2520
2521 static bool
2522 get_bo_from_pool(struct gen_batch_decode_bo *ret,
2523 struct anv_block_pool *pool,
2524 uint64_t address)
2525 {
2526 anv_block_pool_foreach_bo(bo, pool) {
2527 uint64_t bo_address = gen_48b_address(bo->offset);
2528 if (address >= bo_address && address < (bo_address + bo->size)) {
2529 *ret = (struct gen_batch_decode_bo) {
2530 .addr = bo_address,
2531 .size = bo->size,
2532 .map = bo->map,
2533 };
2534 return true;
2535 }
2536 }
2537 return false;
2538 }
2539
2540 /* Finding a buffer for batch decoding */
2541 static struct gen_batch_decode_bo
2542 decode_get_bo(void *v_batch, bool ppgtt, uint64_t address)
2543 {
2544 struct anv_device *device = v_batch;
2545 struct gen_batch_decode_bo ret_bo = {};
2546
2547 assert(ppgtt);
2548
2549 if (get_bo_from_pool(&ret_bo, &device->dynamic_state_pool.block_pool, address))
2550 return ret_bo;
2551 if (get_bo_from_pool(&ret_bo, &device->instruction_state_pool.block_pool, address))
2552 return ret_bo;
2553 if (get_bo_from_pool(&ret_bo, &device->binding_table_pool.block_pool, address))
2554 return ret_bo;
2555 if (get_bo_from_pool(&ret_bo, &device->surface_state_pool.block_pool, address))
2556 return ret_bo;
2557
2558 if (!device->cmd_buffer_being_decoded)
2559 return (struct gen_batch_decode_bo) { };
2560
2561 struct anv_batch_bo **bo;
2562
2563 u_vector_foreach(bo, &device->cmd_buffer_being_decoded->seen_bbos) {
2564 /* The decoder zeroes out the top 16 bits, so we need to as well */
2565 uint64_t bo_address = (*bo)->bo->offset & (~0ull >> 16);
2566
2567 if (address >= bo_address && address < bo_address + (*bo)->bo->size) {
2568 return (struct gen_batch_decode_bo) {
2569 .addr = bo_address,
2570 .size = (*bo)->bo->size,
2571 .map = (*bo)->bo->map,
2572 };
2573 }
2574 }
2575
2576 return (struct gen_batch_decode_bo) { };
2577 }
2578
2579 struct gen_aux_map_buffer {
2580 struct gen_buffer base;
2581 struct anv_state state;
2582 };
2583
2584 static struct gen_buffer *
2585 gen_aux_map_buffer_alloc(void *driver_ctx, uint32_t size)
2586 {
2587 struct gen_aux_map_buffer *buf = malloc(sizeof(struct gen_aux_map_buffer));
2588 if (!buf)
2589 return NULL;
2590
2591 struct anv_device *device = (struct anv_device*)driver_ctx;
2592 assert(device->physical->supports_48bit_addresses &&
2593 device->physical->use_softpin);
2594
2595 struct anv_state_pool *pool = &device->dynamic_state_pool;
2596 buf->state = anv_state_pool_alloc(pool, size, size);
2597
2598 buf->base.gpu = pool->block_pool.bo->offset + buf->state.offset;
2599 buf->base.gpu_end = buf->base.gpu + buf->state.alloc_size;
2600 buf->base.map = buf->state.map;
2601 buf->base.driver_bo = &buf->state;
2602 return &buf->base;
2603 }
2604
2605 static void
2606 gen_aux_map_buffer_free(void *driver_ctx, struct gen_buffer *buffer)
2607 {
2608 struct gen_aux_map_buffer *buf = (struct gen_aux_map_buffer*)buffer;
2609 struct anv_device *device = (struct anv_device*)driver_ctx;
2610 struct anv_state_pool *pool = &device->dynamic_state_pool;
2611 anv_state_pool_free(pool, buf->state);
2612 free(buf);
2613 }
2614
2615 static struct gen_mapped_pinned_buffer_alloc aux_map_allocator = {
2616 .alloc = gen_aux_map_buffer_alloc,
2617 .free = gen_aux_map_buffer_free,
2618 };
2619
2620 static VkResult
2621 check_physical_device_features(VkPhysicalDevice physicalDevice,
2622 const VkPhysicalDeviceFeatures *features)
2623 {
2624 VkPhysicalDeviceFeatures supported_features;
2625 anv_GetPhysicalDeviceFeatures(physicalDevice, &supported_features);
2626 VkBool32 *supported_feature = (VkBool32 *)&supported_features;
2627 VkBool32 *enabled_feature = (VkBool32 *)features;
2628 unsigned num_features = sizeof(VkPhysicalDeviceFeatures) / sizeof(VkBool32);
2629 for (uint32_t i = 0; i < num_features; i++) {
2630 if (enabled_feature[i] && !supported_feature[i])
2631 return vk_error(VK_ERROR_FEATURE_NOT_PRESENT);
2632 }
2633
2634 return VK_SUCCESS;
2635 }
2636
2637 VkResult anv_CreateDevice(
2638 VkPhysicalDevice physicalDevice,
2639 const VkDeviceCreateInfo* pCreateInfo,
2640 const VkAllocationCallbacks* pAllocator,
2641 VkDevice* pDevice)
2642 {
2643 ANV_FROM_HANDLE(anv_physical_device, physical_device, physicalDevice);
2644 VkResult result;
2645 struct anv_device *device;
2646
2647 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO);
2648
2649 struct anv_device_extension_table enabled_extensions = { };
2650 for (uint32_t i = 0; i < pCreateInfo->enabledExtensionCount; i++) {
2651 int idx;
2652 for (idx = 0; idx < ANV_DEVICE_EXTENSION_COUNT; idx++) {
2653 if (strcmp(pCreateInfo->ppEnabledExtensionNames[i],
2654 anv_device_extensions[idx].extensionName) == 0)
2655 break;
2656 }
2657
2658 if (idx >= ANV_DEVICE_EXTENSION_COUNT)
2659 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);
2660
2661 if (!physical_device->supported_extensions.extensions[idx])
2662 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);
2663
2664 enabled_extensions.extensions[idx] = true;
2665 }
2666
2667 /* Check enabled features */
2668 bool robust_buffer_access = false;
2669 if (pCreateInfo->pEnabledFeatures) {
2670 result = check_physical_device_features(physicalDevice,
2671 pCreateInfo->pEnabledFeatures);
2672 if (result != VK_SUCCESS)
2673 return result;
2674
2675 if (pCreateInfo->pEnabledFeatures->robustBufferAccess)
2676 robust_buffer_access = true;
2677 }
2678
2679 vk_foreach_struct_const(ext, pCreateInfo->pNext) {
2680 switch (ext->sType) {
2681 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2: {
2682 const VkPhysicalDeviceFeatures2 *features = (const void *)ext;
2683 result = check_physical_device_features(physicalDevice,
2684 &features->features);
2685 if (result != VK_SUCCESS)
2686 return result;
2687
2688 if (features->features.robustBufferAccess)
2689 robust_buffer_access = true;
2690 break;
2691 }
2692
2693 default:
2694 /* Don't warn */
2695 break;
2696 }
2697 }
2698
2699 /* Check requested queues and fail if we are requested to create any
2700 * queues with flags we don't support.
2701 */
2702 assert(pCreateInfo->queueCreateInfoCount > 0);
2703 for (uint32_t i = 0; i < pCreateInfo->queueCreateInfoCount; i++) {
2704 if (pCreateInfo->pQueueCreateInfos[i].flags != 0)
2705 return vk_error(VK_ERROR_INITIALIZATION_FAILED);
2706 }
2707
2708 /* Check if client specified queue priority. */
2709 const VkDeviceQueueGlobalPriorityCreateInfoEXT *queue_priority =
2710 vk_find_struct_const(pCreateInfo->pQueueCreateInfos[0].pNext,
2711 DEVICE_QUEUE_GLOBAL_PRIORITY_CREATE_INFO_EXT);
2712
2713 VkQueueGlobalPriorityEXT priority =
2714 queue_priority ? queue_priority->globalPriority :
2715 VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_EXT;
2716
2717 device = vk_alloc2(&physical_device->instance->alloc, pAllocator,
2718 sizeof(*device), 8,
2719 VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
2720 if (!device)
2721 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2722
2723 if (INTEL_DEBUG & DEBUG_BATCH) {
2724 const unsigned decode_flags =
2725 GEN_BATCH_DECODE_FULL |
2726 ((INTEL_DEBUG & DEBUG_COLOR) ? GEN_BATCH_DECODE_IN_COLOR : 0) |
2727 GEN_BATCH_DECODE_OFFSETS |
2728 GEN_BATCH_DECODE_FLOATS;
2729
2730 gen_batch_decode_ctx_init(&device->decoder_ctx,
2731 &physical_device->info,
2732 stderr, decode_flags, NULL,
2733 decode_get_bo, NULL, device);
2734 }
2735
2736 device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
2737 device->physical = physical_device;
2738 device->no_hw = physical_device->no_hw;
2739 device->_lost = false;
2740
2741 if (pAllocator)
2742 device->alloc = *pAllocator;
2743 else
2744 device->alloc = physical_device->instance->alloc;
2745
2746 /* XXX(chadv): Can we dup() physicalDevice->fd here? */
2747 device->fd = open(physical_device->path, O_RDWR | O_CLOEXEC);
2748 if (device->fd == -1) {
2749 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2750 goto fail_device;
2751 }
2752
2753 device->context_id = anv_gem_create_context(device);
2754 if (device->context_id == -1) {
2755 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2756 goto fail_fd;
2757 }
2758
2759 result = anv_queue_init(device, &device->queue);
2760 if (result != VK_SUCCESS)
2761 goto fail_context_id;
2762
2763 if (physical_device->use_softpin) {
2764 if (pthread_mutex_init(&device->vma_mutex, NULL) != 0) {
2765 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2766 goto fail_queue;
2767 }
2768
2769 /* keep the page with address zero out of the allocator */
2770 util_vma_heap_init(&device->vma_lo,
2771 LOW_HEAP_MIN_ADDRESS, LOW_HEAP_SIZE);
2772
2773 util_vma_heap_init(&device->vma_cva, CLIENT_VISIBLE_HEAP_MIN_ADDRESS,
2774 CLIENT_VISIBLE_HEAP_SIZE);
2775
2776 /* Leave the last 4GiB out of the high vma range, so that no state
2777 * base address + size can overflow 48 bits. For more information see
2778 * the comment about Wa32bitGeneralStateOffset in anv_allocator.c
2779 */
2780 util_vma_heap_init(&device->vma_hi, HIGH_HEAP_MIN_ADDRESS,
2781 physical_device->gtt_size - (1ull << 32) -
2782 HIGH_HEAP_MIN_ADDRESS);
2783 }
2784
2785 list_inithead(&device->memory_objects);
2786
2787 /* As per spec, the driver implementation may deny requests to acquire
2788 * a priority above the default priority (MEDIUM) if the caller does not
2789 * have sufficient privileges. In this scenario VK_ERROR_NOT_PERMITTED_EXT
2790 * is returned.
2791 */
2792 if (physical_device->has_context_priority) {
2793 int err = anv_gem_set_context_param(device->fd, device->context_id,
2794 I915_CONTEXT_PARAM_PRIORITY,
2795 vk_priority_to_gen(priority));
2796 if (err != 0 && priority > VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_EXT) {
2797 result = vk_error(VK_ERROR_NOT_PERMITTED_EXT);
2798 goto fail_vmas;
2799 }
2800 }
2801
2802 device->info = physical_device->info;
2803 device->isl_dev = physical_device->isl_dev;
2804
2805 /* On Broadwell and later, we can use batch chaining to more efficiently
2806 * implement growing command buffers. Prior to Haswell, the kernel
2807 * command parser gets in the way and we have to fall back to growing
2808 * the batch.
2809 */
2810 device->can_chain_batches = device->info.gen >= 8;
2811
2812 device->robust_buffer_access = robust_buffer_access;
2813 device->enabled_extensions = enabled_extensions;
2814
2815 anv_device_init_dispatch(device);
2816
2817 if (pthread_mutex_init(&device->mutex, NULL) != 0) {
2818 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2819 goto fail_queue;
2820 }
2821
2822 pthread_condattr_t condattr;
2823 if (pthread_condattr_init(&condattr) != 0) {
2824 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2825 goto fail_mutex;
2826 }
2827 if (pthread_condattr_setclock(&condattr, CLOCK_MONOTONIC) != 0) {
2828 pthread_condattr_destroy(&condattr);
2829 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2830 goto fail_mutex;
2831 }
2832 if (pthread_cond_init(&device->queue_submit, &condattr) != 0) {
2833 pthread_condattr_destroy(&condattr);
2834 result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
2835 goto fail_mutex;
2836 }
2837 pthread_condattr_destroy(&condattr);
2838
2839 result = anv_bo_cache_init(&device->bo_cache);
2840 if (result != VK_SUCCESS)
2841 goto fail_queue_cond;
2842
2843 anv_bo_pool_init(&device->batch_bo_pool, device);
2844
2845 result = anv_state_pool_init(&device->dynamic_state_pool, device,
2846 DYNAMIC_STATE_POOL_MIN_ADDRESS, 16384);
2847 if (result != VK_SUCCESS)
2848 goto fail_batch_bo_pool;
2849
2850 result = anv_state_pool_init(&device->instruction_state_pool, device,
2851 INSTRUCTION_STATE_POOL_MIN_ADDRESS, 16384);
2852 if (result != VK_SUCCESS)
2853 goto fail_dynamic_state_pool;
2854
2855 result = anv_state_pool_init(&device->surface_state_pool, device,
2856 SURFACE_STATE_POOL_MIN_ADDRESS, 4096);
2857 if (result != VK_SUCCESS)
2858 goto fail_instruction_state_pool;
2859
2860 if (physical_device->use_softpin) {
2861 result = anv_state_pool_init(&device->binding_table_pool, device,
2862 BINDING_TABLE_POOL_MIN_ADDRESS, 4096);
2863 if (result != VK_SUCCESS)
2864 goto fail_surface_state_pool;
2865 }
2866
2867 if (device->info.gen >= 12) {
2868 device->aux_map_ctx = gen_aux_map_init(device, &aux_map_allocator,
2869 &physical_device->info);
2870 if (!device->aux_map_ctx)
2871 goto fail_binding_table_pool;
2872 }
2873
2874 result = anv_device_alloc_bo(device, 4096, 0 /* flags */,
2875 0 /* explicit_address */,
2876 &device->workaround_bo);
2877 if (result != VK_SUCCESS)
2878 goto fail_surface_aux_map_pool;
2879
2880 result = anv_device_init_trivial_batch(device);
2881 if (result != VK_SUCCESS)
2882 goto fail_workaround_bo;
2883
2884 if (device->info.gen >= 10) {
2885 result = anv_device_init_hiz_clear_value_bo(device);
2886 if (result != VK_SUCCESS)
2887 goto fail_trivial_batch_bo;
2888 }
2889
2890 anv_scratch_pool_init(device, &device->scratch_pool);
2891
2892 switch (device->info.gen) {
2893 case 7:
2894 if (!device->info.is_haswell)
2895 result = gen7_init_device_state(device);
2896 else
2897 result = gen75_init_device_state(device);
2898 break;
2899 case 8:
2900 result = gen8_init_device_state(device);
2901 break;
2902 case 9:
2903 result = gen9_init_device_state(device);
2904 break;
2905 case 10:
2906 result = gen10_init_device_state(device);
2907 break;
2908 case 11:
2909 result = gen11_init_device_state(device);
2910 break;
2911 case 12:
2912 result = gen12_init_device_state(device);
2913 break;
2914 default:
2915 /* Shouldn't get here as we don't create physical devices for any other
2916 * gens. */
2917 unreachable("unhandled gen");
2918 }
2919 if (result != VK_SUCCESS)
2920 goto fail_workaround_bo;
2921
2922 anv_pipeline_cache_init(&device->default_pipeline_cache, device, true);
2923
2924 anv_device_init_blorp(device);
2925
2926 anv_device_init_border_colors(device);
2927
2928 anv_device_perf_init(device);
2929
2930 *pDevice = anv_device_to_handle(device);
2931
2932 return VK_SUCCESS;
2933
2934 fail_workaround_bo:
2935 anv_scratch_pool_finish(device, &device->scratch_pool);
2936 if (device->info.gen >= 10)
2937 anv_device_release_bo(device, device->hiz_clear_bo);
2938 anv_device_release_bo(device, device->workaround_bo);
2939 fail_trivial_batch_bo:
2940 anv_device_release_bo(device, device->trivial_batch_bo);
2941 fail_surface_aux_map_pool:
2942 if (device->info.gen >= 12) {
2943 gen_aux_map_finish(device->aux_map_ctx);
2944 device->aux_map_ctx = NULL;
2945 }
2946 fail_binding_table_pool:
2947 if (physical_device->use_softpin)
2948 anv_state_pool_finish(&device->binding_table_pool);
2949 fail_surface_state_pool:
2950 anv_state_pool_finish(&device->surface_state_pool);
2951 fail_instruction_state_pool:
2952 anv_state_pool_finish(&device->instruction_state_pool);
2953 fail_dynamic_state_pool:
2954 anv_state_pool_finish(&device->dynamic_state_pool);
2955 fail_batch_bo_pool:
2956 anv_bo_pool_finish(&device->batch_bo_pool);
2957 anv_bo_cache_finish(&device->bo_cache);
2958 fail_queue_cond:
2959 pthread_cond_destroy(&device->queue_submit);
2960 fail_mutex:
2961 pthread_mutex_destroy(&device->mutex);
2962 fail_vmas:
2963 if (physical_device->use_softpin) {
2964 util_vma_heap_finish(&device->vma_hi);
2965 util_vma_heap_finish(&device->vma_cva);
2966 util_vma_heap_finish(&device->vma_lo);
2967 }
2968 fail_queue:
2969 anv_queue_finish(&device->queue);
2970 fail_context_id:
2971 anv_gem_destroy_context(device, device->context_id);
2972 fail_fd:
2973 close(device->fd);
2974 fail_device:
2975 vk_free(&device->alloc, device);
2976
2977 return result;
2978 }
2979
2980 void anv_DestroyDevice(
2981 VkDevice _device,
2982 const VkAllocationCallbacks* pAllocator)
2983 {
2984 ANV_FROM_HANDLE(anv_device, device, _device);
2985
2986 if (!device)
2987 return;
2988
2989 anv_device_finish_blorp(device);
2990
2991 anv_pipeline_cache_finish(&device->default_pipeline_cache);
2992
2993 anv_queue_finish(&device->queue);
2994
2995 #ifdef HAVE_VALGRIND
2996 /* We only need to free these to prevent valgrind errors. The backing
2997 * BO will go away in a couple of lines so we don't actually leak.
2998 */
2999 anv_state_pool_free(&device->dynamic_state_pool, device->border_colors);
3000 anv_state_pool_free(&device->dynamic_state_pool, device->slice_hash);
3001 #endif
3002
3003 anv_scratch_pool_finish(device, &device->scratch_pool);
3004
3005 anv_device_release_bo(device, device->workaround_bo);
3006 anv_device_release_bo(device, device->trivial_batch_bo);
3007 if (device->info.gen >= 10)
3008 anv_device_release_bo(device, device->hiz_clear_bo);
3009
3010 if (device->info.gen >= 12) {
3011 gen_aux_map_finish(device->aux_map_ctx);
3012 device->aux_map_ctx = NULL;
3013 }
3014
3015 if (device->physical->use_softpin)
3016 anv_state_pool_finish(&device->binding_table_pool);
3017 anv_state_pool_finish(&device->surface_state_pool);
3018 anv_state_pool_finish(&device->instruction_state_pool);
3019 anv_state_pool_finish(&device->dynamic_state_pool);
3020
3021 anv_bo_pool_finish(&device->batch_bo_pool);
3022
3023 anv_bo_cache_finish(&device->bo_cache);
3024
3025 if (device->physical->use_softpin) {
3026 util_vma_heap_finish(&device->vma_hi);
3027 util_vma_heap_finish(&device->vma_cva);
3028 util_vma_heap_finish(&device->vma_lo);
3029 }
3030
3031 pthread_cond_destroy(&device->queue_submit);
3032 pthread_mutex_destroy(&device->mutex);
3033
3034 anv_gem_destroy_context(device, device->context_id);
3035
3036 if (INTEL_DEBUG & DEBUG_BATCH)
3037 gen_batch_decode_ctx_finish(&device->decoder_ctx);
3038
3039 close(device->fd);
3040
3041 vk_free(&device->alloc, device);
3042 }
3043
3044 VkResult anv_EnumerateInstanceLayerProperties(
3045 uint32_t* pPropertyCount,
3046 VkLayerProperties* pProperties)
3047 {
3048 if (pProperties == NULL) {
3049 *pPropertyCount = 0;
3050 return VK_SUCCESS;
3051 }
3052
3053 /* None supported at this time */
3054 return vk_error(VK_ERROR_LAYER_NOT_PRESENT);
3055 }
3056
3057 VkResult anv_EnumerateDeviceLayerProperties(
3058 VkPhysicalDevice physicalDevice,
3059 uint32_t* pPropertyCount,
3060 VkLayerProperties* pProperties)
3061 {
3062 if (pProperties == NULL) {
3063 *pPropertyCount = 0;
3064 return VK_SUCCESS;
3065 }
3066
3067 /* None supported at this time */
3068 return vk_error(VK_ERROR_LAYER_NOT_PRESENT);
3069 }
3070
3071 void anv_GetDeviceQueue(
3072 VkDevice _device,
3073 uint32_t queueNodeIndex,
3074 uint32_t queueIndex,
3075 VkQueue* pQueue)
3076 {
3077 const VkDeviceQueueInfo2 info = {
3078 .sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_INFO_2,
3079 .pNext = NULL,
3080 .flags = 0,
3081 .queueFamilyIndex = queueNodeIndex,
3082 .queueIndex = queueIndex,
3083 };
3084
3085 anv_GetDeviceQueue2(_device, &info, pQueue);
3086 }
3087
3088 void anv_GetDeviceQueue2(
3089 VkDevice _device,
3090 const VkDeviceQueueInfo2* pQueueInfo,
3091 VkQueue* pQueue)
3092 {
3093 ANV_FROM_HANDLE(anv_device, device, _device);
3094
3095 assert(pQueueInfo->queueIndex == 0);
3096
3097 if (pQueueInfo->flags == device->queue.flags)
3098 *pQueue = anv_queue_to_handle(&device->queue);
3099 else
3100 *pQueue = NULL;
3101 }
3102
3103 VkResult
3104 _anv_device_set_lost(struct anv_device *device,
3105 const char *file, int line,
3106 const char *msg, ...)
3107 {
3108 VkResult err;
3109 va_list ap;
3110
3111 p_atomic_inc(&device->_lost);
3112
3113 va_start(ap, msg);
3114 err = __vk_errorv(device->physical->instance, device,
3115 VK_DEBUG_REPORT_OBJECT_TYPE_DEVICE_EXT,
3116 VK_ERROR_DEVICE_LOST, file, line, msg, ap);
3117 va_end(ap);
3118
3119 if (env_var_as_boolean("ANV_ABORT_ON_DEVICE_LOSS", false))
3120 abort();
3121
3122 return err;
3123 }
3124
3125 VkResult
3126 _anv_queue_set_lost(struct anv_queue *queue,
3127 const char *file, int line,
3128 const char *msg, ...)
3129 {
3130 VkResult err;
3131 va_list ap;
3132
3133 p_atomic_inc(&queue->device->_lost);
3134
3135 va_start(ap, msg);
3136 err = __vk_errorv(queue->device->physical->instance, queue->device,
3137 VK_DEBUG_REPORT_OBJECT_TYPE_DEVICE_EXT,
3138 VK_ERROR_DEVICE_LOST, file, line, msg, ap);
3139 va_end(ap);
3140
3141 if (env_var_as_boolean("ANV_ABORT_ON_DEVICE_LOSS", false))
3142 abort();
3143
3144 return err;
3145 }
3146
3147 VkResult
3148 anv_device_query_status(struct anv_device *device)
3149 {
3150 /* This isn't likely as most of the callers of this function already check
3151 * for it. However, it doesn't hurt to check and it potentially lets us
3152 * avoid an ioctl.
3153 */
3154 if (anv_device_is_lost(device))
3155 return VK_ERROR_DEVICE_LOST;
3156
3157 uint32_t active, pending;
3158 int ret = anv_gem_gpu_get_reset_stats(device, &active, &pending);
3159 if (ret == -1) {
3160 /* We don't know the real error. */
3161 return anv_device_set_lost(device, "get_reset_stats failed: %m");
3162 }
3163
3164 if (active) {
3165 return anv_device_set_lost(device, "GPU hung on one of our command buffers");
3166 } else if (pending) {
3167 return anv_device_set_lost(device, "GPU hung with commands in-flight");
3168 }
3169
3170 return VK_SUCCESS;
3171 }
3172
3173 VkResult
3174 anv_device_bo_busy(struct anv_device *device, struct anv_bo *bo)
3175 {
3176 /* Note: This only returns whether or not the BO is in use by an i915 GPU.
3177 * Other usages of the BO (such as on different hardware) will not be
3178 * flagged as "busy" by this ioctl. Use with care.
3179 */
3180 int ret = anv_gem_busy(device, bo->gem_handle);
3181 if (ret == 1) {
3182 return VK_NOT_READY;
3183 } else if (ret == -1) {
3184 /* We don't know the real error. */
3185 return anv_device_set_lost(device, "gem wait failed: %m");
3186 }
3187
3188 /* Query for device status after the busy call. If the BO we're checking
3189 * got caught in a GPU hang we don't want to return VK_SUCCESS to the
3190 * client because it clearly doesn't have valid data. Yes, this most
3191 * likely means an ioctl, but we just did an ioctl to query the busy status
3192 * so it's no great loss.
3193 */
3194 return anv_device_query_status(device);
3195 }
3196
3197 VkResult
3198 anv_device_wait(struct anv_device *device, struct anv_bo *bo,
3199 int64_t timeout)
3200 {
3201 int ret = anv_gem_wait(device, bo->gem_handle, &timeout);
3202 if (ret == -1 && errno == ETIME) {
3203 return VK_TIMEOUT;
3204 } else if (ret == -1) {
3205 /* We don't know the real error. */
3206 return anv_device_set_lost(device, "gem wait failed: %m");
3207 }
3208
3209 /* Query for device status after the wait. If the BO we're waiting on got
3210 * caught in a GPU hang we don't want to return VK_SUCCESS to the client
3211 * because it clearly doesn't have valid data. Yes, this most likely means
3212 * an ioctl, but we just did an ioctl to wait so it's no great loss.
3213 */
3214 return anv_device_query_status(device);
3215 }
3216
3217 VkResult anv_DeviceWaitIdle(
3218 VkDevice _device)
3219 {
3220 ANV_FROM_HANDLE(anv_device, device, _device);
3221
3222 if (anv_device_is_lost(device))
3223 return VK_ERROR_DEVICE_LOST;
3224
3225 return anv_queue_submit_simple_batch(&device->queue, NULL);
3226 }
3227
3228 uint64_t
3229 anv_vma_alloc(struct anv_device *device,
3230 uint64_t size, uint64_t align,
3231 enum anv_bo_alloc_flags alloc_flags,
3232 uint64_t client_address)
3233 {
3234 pthread_mutex_lock(&device->vma_mutex);
3235
3236 uint64_t addr = 0;
3237
3238 if (alloc_flags & ANV_BO_ALLOC_CLIENT_VISIBLE_ADDRESS) {
3239 if (client_address) {
3240 if (util_vma_heap_alloc_addr(&device->vma_cva,
3241 client_address, size)) {
3242 addr = client_address;
3243 }
3244 } else {
3245 addr = util_vma_heap_alloc(&device->vma_cva, size, align);
3246 }
3247 /* We don't want to fall back to other heaps */
3248 goto done;
3249 }
3250
3251 assert(client_address == 0);
3252
3253 if (!(alloc_flags & ANV_BO_ALLOC_32BIT_ADDRESS))
3254 addr = util_vma_heap_alloc(&device->vma_hi, size, align);
3255
3256 if (addr == 0)
3257 addr = util_vma_heap_alloc(&device->vma_lo, size, align);
3258
3259 done:
3260 pthread_mutex_unlock(&device->vma_mutex);
3261
3262 assert(addr == gen_48b_address(addr));
3263 return gen_canonical_address(addr);
3264 }
3265
3266 void
3267 anv_vma_free(struct anv_device *device,
3268 uint64_t address, uint64_t size)
3269 {
3270 const uint64_t addr_48b = gen_48b_address(address);
3271
3272 pthread_mutex_lock(&device->vma_mutex);
3273
3274 if (addr_48b >= LOW_HEAP_MIN_ADDRESS &&
3275 addr_48b <= LOW_HEAP_MAX_ADDRESS) {
3276 util_vma_heap_free(&device->vma_lo, addr_48b, size);
3277 } else if (addr_48b >= CLIENT_VISIBLE_HEAP_MIN_ADDRESS &&
3278 addr_48b <= CLIENT_VISIBLE_HEAP_MAX_ADDRESS) {
3279 util_vma_heap_free(&device->vma_cva, addr_48b, size);
3280 } else {
3281 assert(addr_48b >= HIGH_HEAP_MIN_ADDRESS);
3282 util_vma_heap_free(&device->vma_hi, addr_48b, size);
3283 }
3284
3285 pthread_mutex_unlock(&device->vma_mutex);
3286 }
3287
3288 VkResult anv_AllocateMemory(
3289 VkDevice _device,
3290 const VkMemoryAllocateInfo* pAllocateInfo,
3291 const VkAllocationCallbacks* pAllocator,
3292 VkDeviceMemory* pMem)
3293 {
3294 ANV_FROM_HANDLE(anv_device, device, _device);
3295 struct anv_physical_device *pdevice = device->physical;
3296 struct anv_device_memory *mem;
3297 VkResult result = VK_SUCCESS;
3298
3299 assert(pAllocateInfo->sType == VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO);
3300
3301 /* The Vulkan 1.0.33 spec says "allocationSize must be greater than 0". */
3302 assert(pAllocateInfo->allocationSize > 0);
3303
3304 VkDeviceSize aligned_alloc_size =
3305 align_u64(pAllocateInfo->allocationSize, 4096);
3306
3307 if (aligned_alloc_size > MAX_MEMORY_ALLOCATION_SIZE)
3308 return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
3309
3310 assert(pAllocateInfo->memoryTypeIndex < pdevice->memory.type_count);
3311 struct anv_memory_type *mem_type =
3312 &pdevice->memory.types[pAllocateInfo->memoryTypeIndex];
3313 assert(mem_type->heapIndex < pdevice->memory.heap_count);
3314 struct anv_memory_heap *mem_heap =
3315 &pdevice->memory.heaps[mem_type->heapIndex];
3316
3317 uint64_t mem_heap_used = p_atomic_read(&mem_heap->used);
3318 if (mem_heap_used + aligned_alloc_size > mem_heap->size)
3319 return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
3320
3321 mem = vk_alloc2(&device->alloc, pAllocator, sizeof(*mem), 8,
3322 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
3323 if (mem == NULL)
3324 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
3325
3326 assert(pAllocateInfo->memoryTypeIndex < pdevice->memory.type_count);
3327 mem->type = mem_type;
3328 mem->map = NULL;
3329 mem->map_size = 0;
3330 mem->ahw = NULL;
3331 mem->host_ptr = NULL;
3332
3333 enum anv_bo_alloc_flags alloc_flags = 0;
3334
3335 const VkExportMemoryAllocateInfo *export_info = NULL;
3336 const VkImportAndroidHardwareBufferInfoANDROID *ahw_import_info = NULL;
3337 const VkImportMemoryFdInfoKHR *fd_info = NULL;
3338 const VkImportMemoryHostPointerInfoEXT *host_ptr_info = NULL;
3339 const VkMemoryDedicatedAllocateInfo *dedicated_info = NULL;
3340 VkMemoryAllocateFlags vk_flags = 0;
3341 uint64_t client_address = 0;
3342
3343 vk_foreach_struct_const(ext, pAllocateInfo->pNext) {
3344 switch (ext->sType) {
3345 case VK_STRUCTURE_TYPE_EXPORT_MEMORY_ALLOCATE_INFO:
3346 export_info = (void *)ext;
3347 break;
3348
3349 case VK_STRUCTURE_TYPE_IMPORT_ANDROID_HARDWARE_BUFFER_INFO_ANDROID:
3350 ahw_import_info = (void *)ext;
3351 break;
3352
3353 case VK_STRUCTURE_TYPE_IMPORT_MEMORY_FD_INFO_KHR:
3354 fd_info = (void *)ext;
3355 break;
3356
3357 case VK_STRUCTURE_TYPE_IMPORT_MEMORY_HOST_POINTER_INFO_EXT:
3358 host_ptr_info = (void *)ext;
3359 break;
3360
3361 case VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_FLAGS_INFO: {
3362 const VkMemoryAllocateFlagsInfo *flags_info = (void *)ext;
3363 vk_flags = flags_info->flags;
3364 break;
3365 }
3366
3367 case VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO:
3368 dedicated_info = (void *)ext;
3369 break;
3370
3371 case VK_STRUCTURE_TYPE_MEMORY_OPAQUE_CAPTURE_ADDRESS_ALLOCATE_INFO_KHR: {
3372 const VkMemoryOpaqueCaptureAddressAllocateInfoKHR *addr_info =
3373 (const VkMemoryOpaqueCaptureAddressAllocateInfoKHR *)ext;
3374 client_address = addr_info->opaqueCaptureAddress;
3375 break;
3376 }
3377
3378 default:
3379 anv_debug_ignored_stype(ext->sType);
3380 break;
3381 }
3382 }
3383
3384 /* By default, we want all VkDeviceMemory objects to support CCS */
3385 if (device->physical->has_implicit_ccs)
3386 alloc_flags |= ANV_BO_ALLOC_IMPLICIT_CCS;
3387
3388 if (vk_flags & VK_MEMORY_ALLOCATE_DEVICE_ADDRESS_BIT_KHR)
3389 alloc_flags |= ANV_BO_ALLOC_CLIENT_VISIBLE_ADDRESS;
3390
3391 if ((export_info && export_info->handleTypes) ||
3392 (fd_info && fd_info->handleType) ||
3393 (host_ptr_info && host_ptr_info->handleType)) {
3394 /* Anything imported or exported is EXTERNAL */
3395 alloc_flags |= ANV_BO_ALLOC_EXTERNAL;
3396
3397 /* We can't have implicit CCS on external memory with an AUX-table.
3398 * Doing so would require us to sync the aux tables across processes
3399 * which is impractical.
3400 */
3401 if (device->info.has_aux_map)
3402 alloc_flags &= ~ANV_BO_ALLOC_IMPLICIT_CCS;
3403 }
3404
3405 /* Check if we need to support Android HW buffer export. If so,
3406 * create AHardwareBuffer and import memory from it.
3407 */
3408 bool android_export = false;
3409 if (export_info && export_info->handleTypes &
3410 VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID)
3411 android_export = true;
3412
3413 if (ahw_import_info) {
3414 result = anv_import_ahw_memory(_device, mem, ahw_import_info);
3415 if (result != VK_SUCCESS)
3416 goto fail;
3417
3418 goto success;
3419 } else if (android_export) {
3420 result = anv_create_ahw_memory(_device, mem, pAllocateInfo);
3421 if (result != VK_SUCCESS)
3422 goto fail;
3423
3424 const VkImportAndroidHardwareBufferInfoANDROID import_info = {
3425 .buffer = mem->ahw,
3426 };
3427 result = anv_import_ahw_memory(_device, mem, &import_info);
3428 if (result != VK_SUCCESS)
3429 goto fail;
3430
3431 goto success;
3432 }
3433
3434 /* The Vulkan spec permits handleType to be 0, in which case the struct is
3435 * ignored.
3436 */
3437 if (fd_info && fd_info->handleType) {
3438 /* At the moment, we support only the below handle types. */
3439 assert(fd_info->handleType ==
3440 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT ||
3441 fd_info->handleType ==
3442 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT);
3443
3444 result = anv_device_import_bo(device, fd_info->fd, alloc_flags,
3445 client_address, &mem->bo);
3446 if (result != VK_SUCCESS)
3447 goto fail;
3448
3449 /* For security purposes, we reject importing the bo if it's smaller
3450 * than the requested allocation size. This prevents a malicious client
3451 * from passing a buffer to a trusted client, lying about the size, and
3452 * telling the trusted client to try and texture from an image that goes
3453 * out-of-bounds. This sort of thing could lead to GPU hangs or worse
3454 * in the trusted client. The trusted client can protect itself against
3455 * this sort of attack but only if it can trust the buffer size.
3456 */
3457 if (mem->bo->size < aligned_alloc_size) {
3458 result = vk_errorf(device, device, VK_ERROR_INVALID_EXTERNAL_HANDLE,
3459 "aligned allocationSize too large for "
3460 "VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT: "
3461 "%"PRIu64"B > %"PRIu64"B",
3462 aligned_alloc_size, mem->bo->size);
3463 anv_device_release_bo(device, mem->bo);
3464 goto fail;
3465 }
3466
3467 /* From the Vulkan spec:
3468 *
3469 * "Importing memory from a file descriptor transfers ownership of
3470 * the file descriptor from the application to the Vulkan
3471 * implementation. The application must not perform any operations on
3472 * the file descriptor after a successful import."
3473 *
3474 * If the import fails, we leave the file descriptor open.
3475 */
3476 close(fd_info->fd);
3477 goto success;
3478 }
3479
3480 if (host_ptr_info && host_ptr_info->handleType) {
3481 if (host_ptr_info->handleType ==
3482 VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_MAPPED_FOREIGN_MEMORY_BIT_EXT) {
3483 result = vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE);
3484 goto fail;
3485 }
3486
3487 assert(host_ptr_info->handleType ==
3488 VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT);
3489
3490 result = anv_device_import_bo_from_host_ptr(device,
3491 host_ptr_info->pHostPointer,
3492 pAllocateInfo->allocationSize,
3493 alloc_flags,
3494 client_address,
3495 &mem->bo);
3496 if (result != VK_SUCCESS)
3497 goto fail;
3498
3499 mem->host_ptr = host_ptr_info->pHostPointer;
3500 goto success;
3501 }
3502
3503 /* Regular allocate (not importing memory). */
3504
3505 result = anv_device_alloc_bo(device, pAllocateInfo->allocationSize,
3506 alloc_flags, client_address, &mem->bo);
3507 if (result != VK_SUCCESS)
3508 goto fail;
3509
3510 if (dedicated_info && dedicated_info->image != VK_NULL_HANDLE) {
3511 ANV_FROM_HANDLE(anv_image, image, dedicated_info->image);
3512
3513 /* Some legacy (non-modifiers) consumers need the tiling to be set on
3514 * the BO. In this case, we have a dedicated allocation.
3515 */
3516 if (image->needs_set_tiling) {
3517 const uint32_t i915_tiling =
3518 isl_tiling_to_i915_tiling(image->planes[0].surface.isl.tiling);
3519 int ret = anv_gem_set_tiling(device, mem->bo->gem_handle,
3520 image->planes[0].surface.isl.row_pitch_B,
3521 i915_tiling);
3522 if (ret) {
3523 anv_device_release_bo(device, mem->bo);
3524 result = vk_errorf(device, device, VK_ERROR_OUT_OF_DEVICE_MEMORY,
3525 "failed to set BO tiling: %m");
3526 goto fail;
3527 }
3528 }
3529 }
3530
3531 success:
3532 mem_heap_used = p_atomic_add_return(&mem_heap->used, mem->bo->size);
3533 if (mem_heap_used > mem_heap->size) {
3534 p_atomic_add(&mem_heap->used, -mem->bo->size);
3535 anv_device_release_bo(device, mem->bo);
3536 result = vk_errorf(device, device, VK_ERROR_OUT_OF_DEVICE_MEMORY,
3537 "Out of heap memory");
3538 goto fail;
3539 }
3540
3541 pthread_mutex_lock(&device->mutex);
3542 list_addtail(&mem->link, &device->memory_objects);
3543 pthread_mutex_unlock(&device->mutex);
3544
3545 *pMem = anv_device_memory_to_handle(mem);
3546
3547 return VK_SUCCESS;
3548
3549 fail:
3550 vk_free2(&device->alloc, pAllocator, mem);
3551
3552 return result;
3553 }
3554
3555 VkResult anv_GetMemoryFdKHR(
3556 VkDevice device_h,
3557 const VkMemoryGetFdInfoKHR* pGetFdInfo,
3558 int* pFd)
3559 {
3560 ANV_FROM_HANDLE(anv_device, dev, device_h);
3561 ANV_FROM_HANDLE(anv_device_memory, mem, pGetFdInfo->memory);
3562
3563 assert(pGetFdInfo->sType == VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR);
3564
3565 assert(pGetFdInfo->handleType == VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT ||
3566 pGetFdInfo->handleType == VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT);
3567
3568 return anv_device_export_bo(dev, mem->bo, pFd);
3569 }
3570
3571 VkResult anv_GetMemoryFdPropertiesKHR(
3572 VkDevice _device,
3573 VkExternalMemoryHandleTypeFlagBits handleType,
3574 int fd,
3575 VkMemoryFdPropertiesKHR* pMemoryFdProperties)
3576 {
3577 ANV_FROM_HANDLE(anv_device, device, _device);
3578
3579 switch (handleType) {
3580 case VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT:
3581 /* dma-buf can be imported as any memory type */
3582 pMemoryFdProperties->memoryTypeBits =
3583 (1 << device->physical->memory.type_count) - 1;
3584 return VK_SUCCESS;
3585
3586 default:
3587 /* The valid usage section for this function says:
3588 *
3589 * "handleType must not be one of the handle types defined as
3590 * opaque."
3591 *
3592 * So opaque handle types fall into the default "unsupported" case.
3593 */
3594 return vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE);
3595 }
3596 }
3597
3598 VkResult anv_GetMemoryHostPointerPropertiesEXT(
3599 VkDevice _device,
3600 VkExternalMemoryHandleTypeFlagBits handleType,
3601 const void* pHostPointer,
3602 VkMemoryHostPointerPropertiesEXT* pMemoryHostPointerProperties)
3603 {
3604 ANV_FROM_HANDLE(anv_device, device, _device);
3605
3606 assert(pMemoryHostPointerProperties->sType ==
3607 VK_STRUCTURE_TYPE_MEMORY_HOST_POINTER_PROPERTIES_EXT);
3608
3609 switch (handleType) {
3610 case VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT:
3611 /* Host memory can be imported as any memory type. */
3612 pMemoryHostPointerProperties->memoryTypeBits =
3613 (1ull << device->physical->memory.type_count) - 1;
3614
3615 return VK_SUCCESS;
3616
3617 default:
3618 return VK_ERROR_INVALID_EXTERNAL_HANDLE;
3619 }
3620 }
3621
3622 void anv_FreeMemory(
3623 VkDevice _device,
3624 VkDeviceMemory _mem,
3625 const VkAllocationCallbacks* pAllocator)
3626 {
3627 ANV_FROM_HANDLE(anv_device, device, _device);
3628 ANV_FROM_HANDLE(anv_device_memory, mem, _mem);
3629
3630 if (mem == NULL)
3631 return;
3632
3633 pthread_mutex_lock(&device->mutex);
3634 list_del(&mem->link);
3635 pthread_mutex_unlock(&device->mutex);
3636
3637 if (mem->map)
3638 anv_UnmapMemory(_device, _mem);
3639
3640 p_atomic_add(&device->physical->memory.heaps[mem->type->heapIndex].used,
3641 -mem->bo->size);
3642
3643 anv_device_release_bo(device, mem->bo);
3644
3645 #if defined(ANDROID) && ANDROID_API_LEVEL >= 26
3646 if (mem->ahw)
3647 AHardwareBuffer_release(mem->ahw);
3648 #endif
3649
3650 vk_free2(&device->alloc, pAllocator, mem);
3651 }
3652
3653 VkResult anv_MapMemory(
3654 VkDevice _device,
3655 VkDeviceMemory _memory,
3656 VkDeviceSize offset,
3657 VkDeviceSize size,
3658 VkMemoryMapFlags flags,
3659 void** ppData)
3660 {
3661 ANV_FROM_HANDLE(anv_device, device, _device);
3662 ANV_FROM_HANDLE(anv_device_memory, mem, _memory);
3663
3664 if (mem == NULL) {
3665 *ppData = NULL;
3666 return VK_SUCCESS;
3667 }
3668
3669 if (mem->host_ptr) {
3670 *ppData = mem->host_ptr + offset;
3671 return VK_SUCCESS;
3672 }
3673
3674 if (size == VK_WHOLE_SIZE)
3675 size = mem->bo->size - offset;
3676
3677 /* From the Vulkan spec version 1.0.32 docs for MapMemory:
3678 *
3679 * * If size is not equal to VK_WHOLE_SIZE, size must be greater than 0
3680 * assert(size != 0);
3681 * * If size is not equal to VK_WHOLE_SIZE, size must be less than or
3682 * equal to the size of the memory minus offset
3683 */
3684 assert(size > 0);
3685 assert(offset + size <= mem->bo->size);
3686
3687 /* FIXME: Is this supposed to be thread safe? Since vkUnmapMemory() only
3688 * takes a VkDeviceMemory pointer, it seems like only one map of the memory
3689 * at a time is valid. We could just mmap up front and return an offset
3690 * pointer here, but that may exhaust virtual memory on 32 bit
3691 * userspace. */
3692
3693 uint32_t gem_flags = 0;
3694
3695 if (!device->info.has_llc &&
3696 (mem->type->propertyFlags & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT))
3697 gem_flags |= I915_MMAP_WC;
3698
3699 /* GEM will fail to map if the offset isn't 4k-aligned. Round down. */
3700 uint64_t map_offset = offset & ~4095ull;
3701 assert(offset >= map_offset);
3702 uint64_t map_size = (offset + size) - map_offset;
3703
3704 /* Let's map whole pages */
3705 map_size = align_u64(map_size, 4096);
3706
3707 void *map = anv_gem_mmap(device, mem->bo->gem_handle,
3708 map_offset, map_size, gem_flags);
3709 if (map == MAP_FAILED)
3710 return vk_error(VK_ERROR_MEMORY_MAP_FAILED);
3711
3712 mem->map = map;
3713 mem->map_size = map_size;
3714
3715 *ppData = mem->map + (offset - map_offset);
3716
3717 return VK_SUCCESS;
3718 }
3719
3720 void anv_UnmapMemory(
3721 VkDevice _device,
3722 VkDeviceMemory _memory)
3723 {
3724 ANV_FROM_HANDLE(anv_device_memory, mem, _memory);
3725
3726 if (mem == NULL || mem->host_ptr)
3727 return;
3728
3729 anv_gem_munmap(mem->map, mem->map_size);
3730
3731 mem->map = NULL;
3732 mem->map_size = 0;
3733 }
3734
3735 static void
3736 clflush_mapped_ranges(struct anv_device *device,
3737 uint32_t count,
3738 const VkMappedMemoryRange *ranges)
3739 {
3740 for (uint32_t i = 0; i < count; i++) {
3741 ANV_FROM_HANDLE(anv_device_memory, mem, ranges[i].memory);
3742 if (ranges[i].offset >= mem->map_size)
3743 continue;
3744
3745 gen_clflush_range(mem->map + ranges[i].offset,
3746 MIN2(ranges[i].size, mem->map_size - ranges[i].offset));
3747 }
3748 }
3749
3750 VkResult anv_FlushMappedMemoryRanges(
3751 VkDevice _device,
3752 uint32_t memoryRangeCount,
3753 const VkMappedMemoryRange* pMemoryRanges)
3754 {
3755 ANV_FROM_HANDLE(anv_device, device, _device);
3756
3757 if (device->info.has_llc)
3758 return VK_SUCCESS;
3759
3760 /* Make sure the writes we're flushing have landed. */
3761 __builtin_ia32_mfence();
3762
3763 clflush_mapped_ranges(device, memoryRangeCount, pMemoryRanges);
3764
3765 return VK_SUCCESS;
3766 }
3767
3768 VkResult anv_InvalidateMappedMemoryRanges(
3769 VkDevice _device,
3770 uint32_t memoryRangeCount,
3771 const VkMappedMemoryRange* pMemoryRanges)
3772 {
3773 ANV_FROM_HANDLE(anv_device, device, _device);
3774
3775 if (device->info.has_llc)
3776 return VK_SUCCESS;
3777
3778 clflush_mapped_ranges(device, memoryRangeCount, pMemoryRanges);
3779
3780 /* Make sure no reads get moved up above the invalidate. */
3781 __builtin_ia32_mfence();
3782
3783 return VK_SUCCESS;
3784 }
3785
3786 void anv_GetBufferMemoryRequirements(
3787 VkDevice _device,
3788 VkBuffer _buffer,
3789 VkMemoryRequirements* pMemoryRequirements)
3790 {
3791 ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
3792 ANV_FROM_HANDLE(anv_device, device, _device);
3793
3794 /* The Vulkan spec (git aaed022) says:
3795 *
3796 * memoryTypeBits is a bitfield and contains one bit set for every
3797 * supported memory type for the resource. The bit `1<<i` is set if and
3798 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
3799 * structure for the physical device is supported.
3800 */
3801 uint32_t memory_types = (1ull << device->physical->memory.type_count) - 1;
3802
3803 /* Base alignment requirement of a cache line */
3804 uint32_t alignment = 16;
3805
3806 /* We need an alignment of 32 for pushing UBOs */
3807 if (buffer->usage & VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT)
3808 alignment = MAX2(alignment, 32);
3809
3810 pMemoryRequirements->size = buffer->size;
3811 pMemoryRequirements->alignment = alignment;
3812
3813 /* Storage and Uniform buffers should have their size aligned to
3814 * 32-bits to avoid boundary checks when last DWord is not complete.
3815 * This would ensure that not internal padding would be needed for
3816 * 16-bit types.
3817 */
3818 if (device->robust_buffer_access &&
3819 (buffer->usage & VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT ||
3820 buffer->usage & VK_BUFFER_USAGE_STORAGE_BUFFER_BIT))
3821 pMemoryRequirements->size = align_u64(buffer->size, 4);
3822
3823 pMemoryRequirements->memoryTypeBits = memory_types;
3824 }
3825
3826 void anv_GetBufferMemoryRequirements2(
3827 VkDevice _device,
3828 const VkBufferMemoryRequirementsInfo2* pInfo,
3829 VkMemoryRequirements2* pMemoryRequirements)
3830 {
3831 anv_GetBufferMemoryRequirements(_device, pInfo->buffer,
3832 &pMemoryRequirements->memoryRequirements);
3833
3834 vk_foreach_struct(ext, pMemoryRequirements->pNext) {
3835 switch (ext->sType) {
3836 case VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS: {
3837 VkMemoryDedicatedRequirements *requirements = (void *)ext;
3838 requirements->prefersDedicatedAllocation = false;
3839 requirements->requiresDedicatedAllocation = false;
3840 break;
3841 }
3842
3843 default:
3844 anv_debug_ignored_stype(ext->sType);
3845 break;
3846 }
3847 }
3848 }
3849
3850 void anv_GetImageMemoryRequirements(
3851 VkDevice _device,
3852 VkImage _image,
3853 VkMemoryRequirements* pMemoryRequirements)
3854 {
3855 ANV_FROM_HANDLE(anv_image, image, _image);
3856 ANV_FROM_HANDLE(anv_device, device, _device);
3857
3858 /* The Vulkan spec (git aaed022) says:
3859 *
3860 * memoryTypeBits is a bitfield and contains one bit set for every
3861 * supported memory type for the resource. The bit `1<<i` is set if and
3862 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
3863 * structure for the physical device is supported.
3864 *
3865 * All types are currently supported for images.
3866 */
3867 uint32_t memory_types = (1ull << device->physical->memory.type_count) - 1;
3868
3869 /* We must have image allocated or imported at this point. According to the
3870 * specification, external images must have been bound to memory before
3871 * calling GetImageMemoryRequirements.
3872 */
3873 assert(image->size > 0);
3874
3875 pMemoryRequirements->size = image->size;
3876 pMemoryRequirements->alignment = image->alignment;
3877 pMemoryRequirements->memoryTypeBits = memory_types;
3878 }
3879
3880 void anv_GetImageMemoryRequirements2(
3881 VkDevice _device,
3882 const VkImageMemoryRequirementsInfo2* pInfo,
3883 VkMemoryRequirements2* pMemoryRequirements)
3884 {
3885 ANV_FROM_HANDLE(anv_device, device, _device);
3886 ANV_FROM_HANDLE(anv_image, image, pInfo->image);
3887
3888 anv_GetImageMemoryRequirements(_device, pInfo->image,
3889 &pMemoryRequirements->memoryRequirements);
3890
3891 vk_foreach_struct_const(ext, pInfo->pNext) {
3892 switch (ext->sType) {
3893 case VK_STRUCTURE_TYPE_IMAGE_PLANE_MEMORY_REQUIREMENTS_INFO: {
3894 const VkImagePlaneMemoryRequirementsInfo *plane_reqs =
3895 (const VkImagePlaneMemoryRequirementsInfo *) ext;
3896 uint32_t plane = anv_image_aspect_to_plane(image->aspects,
3897 plane_reqs->planeAspect);
3898
3899 assert(image->planes[plane].offset == 0);
3900
3901 /* The Vulkan spec (git aaed022) says:
3902 *
3903 * memoryTypeBits is a bitfield and contains one bit set for every
3904 * supported memory type for the resource. The bit `1<<i` is set
3905 * if and only if the memory type `i` in the
3906 * VkPhysicalDeviceMemoryProperties structure for the physical
3907 * device is supported.
3908 *
3909 * All types are currently supported for images.
3910 */
3911 pMemoryRequirements->memoryRequirements.memoryTypeBits =
3912 (1ull << device->physical->memory.type_count) - 1;
3913
3914 /* We must have image allocated or imported at this point. According to the
3915 * specification, external images must have been bound to memory before
3916 * calling GetImageMemoryRequirements.
3917 */
3918 assert(image->planes[plane].size > 0);
3919
3920 pMemoryRequirements->memoryRequirements.size = image->planes[plane].size;
3921 pMemoryRequirements->memoryRequirements.alignment =
3922 image->planes[plane].alignment;
3923 break;
3924 }
3925
3926 default:
3927 anv_debug_ignored_stype(ext->sType);
3928 break;
3929 }
3930 }
3931
3932 vk_foreach_struct(ext, pMemoryRequirements->pNext) {
3933 switch (ext->sType) {
3934 case VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS: {
3935 VkMemoryDedicatedRequirements *requirements = (void *)ext;
3936 if (image->needs_set_tiling || image->external_format) {
3937 /* If we need to set the tiling for external consumers, we need a
3938 * dedicated allocation.
3939 *
3940 * See also anv_AllocateMemory.
3941 */
3942 requirements->prefersDedicatedAllocation = true;
3943 requirements->requiresDedicatedAllocation = true;
3944 } else {
3945 requirements->prefersDedicatedAllocation = false;
3946 requirements->requiresDedicatedAllocation = false;
3947 }
3948 break;
3949 }
3950
3951 default:
3952 anv_debug_ignored_stype(ext->sType);
3953 break;
3954 }
3955 }
3956 }
3957
3958 void anv_GetImageSparseMemoryRequirements(
3959 VkDevice device,
3960 VkImage image,
3961 uint32_t* pSparseMemoryRequirementCount,
3962 VkSparseImageMemoryRequirements* pSparseMemoryRequirements)
3963 {
3964 *pSparseMemoryRequirementCount = 0;
3965 }
3966
3967 void anv_GetImageSparseMemoryRequirements2(
3968 VkDevice device,
3969 const VkImageSparseMemoryRequirementsInfo2* pInfo,
3970 uint32_t* pSparseMemoryRequirementCount,
3971 VkSparseImageMemoryRequirements2* pSparseMemoryRequirements)
3972 {
3973 *pSparseMemoryRequirementCount = 0;
3974 }
3975
3976 void anv_GetDeviceMemoryCommitment(
3977 VkDevice device,
3978 VkDeviceMemory memory,
3979 VkDeviceSize* pCommittedMemoryInBytes)
3980 {
3981 *pCommittedMemoryInBytes = 0;
3982 }
3983
3984 static void
3985 anv_bind_buffer_memory(const VkBindBufferMemoryInfo *pBindInfo)
3986 {
3987 ANV_FROM_HANDLE(anv_device_memory, mem, pBindInfo->memory);
3988 ANV_FROM_HANDLE(anv_buffer, buffer, pBindInfo->buffer);
3989
3990 assert(pBindInfo->sType == VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO);
3991
3992 if (mem) {
3993 buffer->address = (struct anv_address) {
3994 .bo = mem->bo,
3995 .offset = pBindInfo->memoryOffset,
3996 };
3997 } else {
3998 buffer->address = ANV_NULL_ADDRESS;
3999 }
4000 }
4001
4002 VkResult anv_BindBufferMemory(
4003 VkDevice device,
4004 VkBuffer buffer,
4005 VkDeviceMemory memory,
4006 VkDeviceSize memoryOffset)
4007 {
4008 anv_bind_buffer_memory(
4009 &(VkBindBufferMemoryInfo) {
4010 .sType = VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO,
4011 .buffer = buffer,
4012 .memory = memory,
4013 .memoryOffset = memoryOffset,
4014 });
4015
4016 return VK_SUCCESS;
4017 }
4018
4019 VkResult anv_BindBufferMemory2(
4020 VkDevice device,
4021 uint32_t bindInfoCount,
4022 const VkBindBufferMemoryInfo* pBindInfos)
4023 {
4024 for (uint32_t i = 0; i < bindInfoCount; i++)
4025 anv_bind_buffer_memory(&pBindInfos[i]);
4026
4027 return VK_SUCCESS;
4028 }
4029
4030 VkResult anv_QueueBindSparse(
4031 VkQueue _queue,
4032 uint32_t bindInfoCount,
4033 const VkBindSparseInfo* pBindInfo,
4034 VkFence fence)
4035 {
4036 ANV_FROM_HANDLE(anv_queue, queue, _queue);
4037 if (anv_device_is_lost(queue->device))
4038 return VK_ERROR_DEVICE_LOST;
4039
4040 return vk_error(VK_ERROR_FEATURE_NOT_PRESENT);
4041 }
4042
4043 // Event functions
4044
4045 VkResult anv_CreateEvent(
4046 VkDevice _device,
4047 const VkEventCreateInfo* pCreateInfo,
4048 const VkAllocationCallbacks* pAllocator,
4049 VkEvent* pEvent)
4050 {
4051 ANV_FROM_HANDLE(anv_device, device, _device);
4052 struct anv_state state;
4053 struct anv_event *event;
4054
4055 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_EVENT_CREATE_INFO);
4056
4057 state = anv_state_pool_alloc(&device->dynamic_state_pool,
4058 sizeof(*event), 8);
4059 event = state.map;
4060 event->state = state;
4061 event->semaphore = VK_EVENT_RESET;
4062
4063 if (!device->info.has_llc) {
4064 /* Make sure the writes we're flushing have landed. */
4065 __builtin_ia32_mfence();
4066 __builtin_ia32_clflush(event);
4067 }
4068
4069 *pEvent = anv_event_to_handle(event);
4070
4071 return VK_SUCCESS;
4072 }
4073
4074 void anv_DestroyEvent(
4075 VkDevice _device,
4076 VkEvent _event,
4077 const VkAllocationCallbacks* pAllocator)
4078 {
4079 ANV_FROM_HANDLE(anv_device, device, _device);
4080 ANV_FROM_HANDLE(anv_event, event, _event);
4081
4082 if (!event)
4083 return;
4084
4085 anv_state_pool_free(&device->dynamic_state_pool, event->state);
4086 }
4087
4088 VkResult anv_GetEventStatus(
4089 VkDevice _device,
4090 VkEvent _event)
4091 {
4092 ANV_FROM_HANDLE(anv_device, device, _device);
4093 ANV_FROM_HANDLE(anv_event, event, _event);
4094
4095 if (anv_device_is_lost(device))
4096 return VK_ERROR_DEVICE_LOST;
4097
4098 if (!device->info.has_llc) {
4099 /* Invalidate read cache before reading event written by GPU. */
4100 __builtin_ia32_clflush(event);
4101 __builtin_ia32_mfence();
4102
4103 }
4104
4105 return event->semaphore;
4106 }
4107
4108 VkResult anv_SetEvent(
4109 VkDevice _device,
4110 VkEvent _event)
4111 {
4112 ANV_FROM_HANDLE(anv_device, device, _device);
4113 ANV_FROM_HANDLE(anv_event, event, _event);
4114
4115 event->semaphore = VK_EVENT_SET;
4116
4117 if (!device->info.has_llc) {
4118 /* Make sure the writes we're flushing have landed. */
4119 __builtin_ia32_mfence();
4120 __builtin_ia32_clflush(event);
4121 }
4122
4123 return VK_SUCCESS;
4124 }
4125
4126 VkResult anv_ResetEvent(
4127 VkDevice _device,
4128 VkEvent _event)
4129 {
4130 ANV_FROM_HANDLE(anv_device, device, _device);
4131 ANV_FROM_HANDLE(anv_event, event, _event);
4132
4133 event->semaphore = VK_EVENT_RESET;
4134
4135 if (!device->info.has_llc) {
4136 /* Make sure the writes we're flushing have landed. */
4137 __builtin_ia32_mfence();
4138 __builtin_ia32_clflush(event);
4139 }
4140
4141 return VK_SUCCESS;
4142 }
4143
4144 // Buffer functions
4145
4146 VkResult anv_CreateBuffer(
4147 VkDevice _device,
4148 const VkBufferCreateInfo* pCreateInfo,
4149 const VkAllocationCallbacks* pAllocator,
4150 VkBuffer* pBuffer)
4151 {
4152 ANV_FROM_HANDLE(anv_device, device, _device);
4153 struct anv_buffer *buffer;
4154
4155 /* Don't allow creating buffers bigger than our address space. The real
4156 * issue here is that we may align up the buffer size and we don't want
4157 * doing so to cause roll-over. However, no one has any business
4158 * allocating a buffer larger than our GTT size.
4159 */
4160 if (pCreateInfo->size > device->physical->gtt_size)
4161 return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
4162
4163 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO);
4164
4165 buffer = vk_alloc2(&device->alloc, pAllocator, sizeof(*buffer), 8,
4166 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
4167 if (buffer == NULL)
4168 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
4169
4170 buffer->size = pCreateInfo->size;
4171 buffer->usage = pCreateInfo->usage;
4172 buffer->address = ANV_NULL_ADDRESS;
4173
4174 *pBuffer = anv_buffer_to_handle(buffer);
4175
4176 return VK_SUCCESS;
4177 }
4178
4179 void anv_DestroyBuffer(
4180 VkDevice _device,
4181 VkBuffer _buffer,
4182 const VkAllocationCallbacks* pAllocator)
4183 {
4184 ANV_FROM_HANDLE(anv_device, device, _device);
4185 ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
4186
4187 if (!buffer)
4188 return;
4189
4190 vk_free2(&device->alloc, pAllocator, buffer);
4191 }
4192
4193 VkDeviceAddress anv_GetBufferDeviceAddress(
4194 VkDevice device,
4195 const VkBufferDeviceAddressInfoKHR* pInfo)
4196 {
4197 ANV_FROM_HANDLE(anv_buffer, buffer, pInfo->buffer);
4198
4199 assert(!anv_address_is_null(buffer->address));
4200 assert(buffer->address.bo->flags & EXEC_OBJECT_PINNED);
4201
4202 return anv_address_physical(buffer->address);
4203 }
4204
4205 uint64_t anv_GetBufferOpaqueCaptureAddress(
4206 VkDevice device,
4207 const VkBufferDeviceAddressInfoKHR* pInfo)
4208 {
4209 return 0;
4210 }
4211
4212 uint64_t anv_GetDeviceMemoryOpaqueCaptureAddress(
4213 VkDevice device,
4214 const VkDeviceMemoryOpaqueCaptureAddressInfoKHR* pInfo)
4215 {
4216 ANV_FROM_HANDLE(anv_device_memory, memory, pInfo->memory);
4217
4218 assert(memory->bo->flags & EXEC_OBJECT_PINNED);
4219 assert(memory->bo->has_client_visible_address);
4220
4221 return gen_48b_address(memory->bo->offset);
4222 }
4223
4224 void
4225 anv_fill_buffer_surface_state(struct anv_device *device, struct anv_state state,
4226 enum isl_format format,
4227 struct anv_address address,
4228 uint32_t range, uint32_t stride)
4229 {
4230 isl_buffer_fill_state(&device->isl_dev, state.map,
4231 .address = anv_address_physical(address),
4232 .mocs = device->isl_dev.mocs.internal,
4233 .size_B = range,
4234 .format = format,
4235 .swizzle = ISL_SWIZZLE_IDENTITY,
4236 .stride_B = stride);
4237 }
4238
4239 void anv_DestroySampler(
4240 VkDevice _device,
4241 VkSampler _sampler,
4242 const VkAllocationCallbacks* pAllocator)
4243 {
4244 ANV_FROM_HANDLE(anv_device, device, _device);
4245 ANV_FROM_HANDLE(anv_sampler, sampler, _sampler);
4246
4247 if (!sampler)
4248 return;
4249
4250 if (sampler->bindless_state.map) {
4251 anv_state_pool_free(&device->dynamic_state_pool,
4252 sampler->bindless_state);
4253 }
4254
4255 vk_free2(&device->alloc, pAllocator, sampler);
4256 }
4257
4258 VkResult anv_CreateFramebuffer(
4259 VkDevice _device,
4260 const VkFramebufferCreateInfo* pCreateInfo,
4261 const VkAllocationCallbacks* pAllocator,
4262 VkFramebuffer* pFramebuffer)
4263 {
4264 ANV_FROM_HANDLE(anv_device, device, _device);
4265 struct anv_framebuffer *framebuffer;
4266
4267 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO);
4268
4269 size_t size = sizeof(*framebuffer);
4270
4271 /* VK_KHR_imageless_framebuffer extension says:
4272 *
4273 * If flags includes VK_FRAMEBUFFER_CREATE_IMAGELESS_BIT_KHR,
4274 * parameter pAttachments is ignored.
4275 */
4276 if (!(pCreateInfo->flags & VK_FRAMEBUFFER_CREATE_IMAGELESS_BIT_KHR)) {
4277 size += sizeof(struct anv_image_view *) * pCreateInfo->attachmentCount;
4278 framebuffer = vk_alloc2(&device->alloc, pAllocator, size, 8,
4279 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
4280 if (framebuffer == NULL)
4281 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
4282
4283 for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
4284 ANV_FROM_HANDLE(anv_image_view, iview, pCreateInfo->pAttachments[i]);
4285 framebuffer->attachments[i] = iview;
4286 }
4287 framebuffer->attachment_count = pCreateInfo->attachmentCount;
4288 } else {
4289 assert(device->enabled_extensions.KHR_imageless_framebuffer);
4290 framebuffer = vk_alloc2(&device->alloc, pAllocator, size, 8,
4291 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
4292 if (framebuffer == NULL)
4293 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
4294
4295 framebuffer->attachment_count = 0;
4296 }
4297
4298 framebuffer->width = pCreateInfo->width;
4299 framebuffer->height = pCreateInfo->height;
4300 framebuffer->layers = pCreateInfo->layers;
4301
4302 *pFramebuffer = anv_framebuffer_to_handle(framebuffer);
4303
4304 return VK_SUCCESS;
4305 }
4306
4307 void anv_DestroyFramebuffer(
4308 VkDevice _device,
4309 VkFramebuffer _fb,
4310 const VkAllocationCallbacks* pAllocator)
4311 {
4312 ANV_FROM_HANDLE(anv_device, device, _device);
4313 ANV_FROM_HANDLE(anv_framebuffer, fb, _fb);
4314
4315 if (!fb)
4316 return;
4317
4318 vk_free2(&device->alloc, pAllocator, fb);
4319 }
4320
4321 static const VkTimeDomainEXT anv_time_domains[] = {
4322 VK_TIME_DOMAIN_DEVICE_EXT,
4323 VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT,
4324 VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT,
4325 };
4326
4327 VkResult anv_GetPhysicalDeviceCalibrateableTimeDomainsEXT(
4328 VkPhysicalDevice physicalDevice,
4329 uint32_t *pTimeDomainCount,
4330 VkTimeDomainEXT *pTimeDomains)
4331 {
4332 int d;
4333 VK_OUTARRAY_MAKE(out, pTimeDomains, pTimeDomainCount);
4334
4335 for (d = 0; d < ARRAY_SIZE(anv_time_domains); d++) {
4336 vk_outarray_append(&out, i) {
4337 *i = anv_time_domains[d];
4338 }
4339 }
4340
4341 return vk_outarray_status(&out);
4342 }
4343
4344 static uint64_t
4345 anv_clock_gettime(clockid_t clock_id)
4346 {
4347 struct timespec current;
4348 int ret;
4349
4350 ret = clock_gettime(clock_id, &current);
4351 if (ret < 0 && clock_id == CLOCK_MONOTONIC_RAW)
4352 ret = clock_gettime(CLOCK_MONOTONIC, &current);
4353 if (ret < 0)
4354 return 0;
4355
4356 return (uint64_t) current.tv_sec * 1000000000ULL + current.tv_nsec;
4357 }
4358
4359 #define TIMESTAMP 0x2358
4360
4361 VkResult anv_GetCalibratedTimestampsEXT(
4362 VkDevice _device,
4363 uint32_t timestampCount,
4364 const VkCalibratedTimestampInfoEXT *pTimestampInfos,
4365 uint64_t *pTimestamps,
4366 uint64_t *pMaxDeviation)
4367 {
4368 ANV_FROM_HANDLE(anv_device, device, _device);
4369 uint64_t timestamp_frequency = device->info.timestamp_frequency;
4370 int ret;
4371 int d;
4372 uint64_t begin, end;
4373 uint64_t max_clock_period = 0;
4374
4375 begin = anv_clock_gettime(CLOCK_MONOTONIC_RAW);
4376
4377 for (d = 0; d < timestampCount; d++) {
4378 switch (pTimestampInfos[d].timeDomain) {
4379 case VK_TIME_DOMAIN_DEVICE_EXT:
4380 ret = anv_gem_reg_read(device, TIMESTAMP | 1,
4381 &pTimestamps[d]);
4382
4383 if (ret != 0) {
4384 return anv_device_set_lost(device, "Failed to read the TIMESTAMP "
4385 "register: %m");
4386 }
4387 uint64_t device_period = DIV_ROUND_UP(1000000000, timestamp_frequency);
4388 max_clock_period = MAX2(max_clock_period, device_period);
4389 break;
4390 case VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT:
4391 pTimestamps[d] = anv_clock_gettime(CLOCK_MONOTONIC);
4392 max_clock_period = MAX2(max_clock_period, 1);
4393 break;
4394
4395 case VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT:
4396 pTimestamps[d] = begin;
4397 break;
4398 default:
4399 pTimestamps[d] = 0;
4400 break;
4401 }
4402 }
4403
4404 end = anv_clock_gettime(CLOCK_MONOTONIC_RAW);
4405
4406 /*
4407 * The maximum deviation is the sum of the interval over which we
4408 * perform the sampling and the maximum period of any sampled
4409 * clock. That's because the maximum skew between any two sampled
4410 * clock edges is when the sampled clock with the largest period is
4411 * sampled at the end of that period but right at the beginning of the
4412 * sampling interval and some other clock is sampled right at the
4413 * begining of its sampling period and right at the end of the
4414 * sampling interval. Let's assume the GPU has the longest clock
4415 * period and that the application is sampling GPU and monotonic:
4416 *
4417 * s e
4418 * w x y z 0 1 2 3 4 5 6 7 8 9 a b c d e f
4419 * Raw -_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-
4420 *
4421 * g
4422 * 0 1 2 3
4423 * GPU -----_____-----_____-----_____-----_____
4424 *
4425 * m
4426 * x y z 0 1 2 3 4 5 6 7 8 9 a b c
4427 * Monotonic -_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-
4428 *
4429 * Interval <----------------->
4430 * Deviation <-------------------------->
4431 *
4432 * s = read(raw) 2
4433 * g = read(GPU) 1
4434 * m = read(monotonic) 2
4435 * e = read(raw) b
4436 *
4437 * We round the sample interval up by one tick to cover sampling error
4438 * in the interval clock
4439 */
4440
4441 uint64_t sample_interval = end - begin + 1;
4442
4443 *pMaxDeviation = sample_interval + max_clock_period;
4444
4445 return VK_SUCCESS;
4446 }
4447
4448 /* vk_icd.h does not declare this function, so we declare it here to
4449 * suppress Wmissing-prototypes.
4450 */
4451 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
4452 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t* pSupportedVersion);
4453
4454 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
4455 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t* pSupportedVersion)
4456 {
4457 /* For the full details on loader interface versioning, see
4458 * <https://github.com/KhronosGroup/Vulkan-LoaderAndValidationLayers/blob/master/loader/LoaderAndLayerInterface.md>.
4459 * What follows is a condensed summary, to help you navigate the large and
4460 * confusing official doc.
4461 *
4462 * - Loader interface v0 is incompatible with later versions. We don't
4463 * support it.
4464 *
4465 * - In loader interface v1:
4466 * - The first ICD entrypoint called by the loader is
4467 * vk_icdGetInstanceProcAddr(). The ICD must statically expose this
4468 * entrypoint.
4469 * - The ICD must statically expose no other Vulkan symbol unless it is
4470 * linked with -Bsymbolic.
4471 * - Each dispatchable Vulkan handle created by the ICD must be
4472 * a pointer to a struct whose first member is VK_LOADER_DATA. The
4473 * ICD must initialize VK_LOADER_DATA.loadMagic to ICD_LOADER_MAGIC.
4474 * - The loader implements vkCreate{PLATFORM}SurfaceKHR() and
4475 * vkDestroySurfaceKHR(). The ICD must be capable of working with
4476 * such loader-managed surfaces.
4477 *
4478 * - Loader interface v2 differs from v1 in:
4479 * - The first ICD entrypoint called by the loader is
4480 * vk_icdNegotiateLoaderICDInterfaceVersion(). The ICD must
4481 * statically expose this entrypoint.
4482 *
4483 * - Loader interface v3 differs from v2 in:
4484 * - The ICD must implement vkCreate{PLATFORM}SurfaceKHR(),
4485 * vkDestroySurfaceKHR(), and other API which uses VKSurfaceKHR,
4486 * because the loader no longer does so.
4487 *
4488 * - Loader interface v4 differs from v3 in:
4489 * - The ICD must implement vk_icdGetPhysicalDeviceProcAddr().
4490 */
4491 *pSupportedVersion = MIN2(*pSupportedVersion, 4u);
4492 return VK_SUCCESS;
4493 }