2 * Copyright © 2015 Intel Corporation
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:
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
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
28 #include <sys/sysinfo.h>
32 #include "drm-uapi/drm_fourcc.h"
34 #include "anv_private.h"
35 #include "util/strtod.h"
36 #include "util/debug.h"
37 #include "util/build_id.h"
38 #include "util/disk_cache.h"
39 #include "util/mesa-sha1.h"
40 #include "util/os_file.h"
41 #include "util/u_atomic.h"
42 #include "util/u_string.h"
45 #include "common/gen_defines.h"
46 #include "compiler/glsl_types.h"
48 #include "genxml/gen7_pack.h"
50 /* This is probably far to big but it reflects the max size used for messages
51 * in OpenGLs KHR_debug.
53 #define MAX_DEBUG_MESSAGE_LENGTH 4096
56 compiler_debug_log(void *data
, const char *fmt
, ...)
58 char str
[MAX_DEBUG_MESSAGE_LENGTH
];
59 struct anv_device
*device
= (struct anv_device
*)data
;
61 if (list_empty(&device
->instance
->debug_report_callbacks
.callbacks
))
66 (void) vsnprintf(str
, MAX_DEBUG_MESSAGE_LENGTH
, fmt
, args
);
69 vk_debug_report(&device
->instance
->debug_report_callbacks
,
70 VK_DEBUG_REPORT_DEBUG_BIT_EXT
,
71 VK_DEBUG_REPORT_OBJECT_TYPE_UNKNOWN_EXT
,
76 compiler_perf_log(void *data
, const char *fmt
, ...)
81 if (unlikely(INTEL_DEBUG
& DEBUG_PERF
))
82 intel_logd_v(fmt
, args
);
88 anv_compute_heap_size(int fd
, uint64_t gtt_size
)
90 /* Query the total ram from the system */
94 uint64_t total_ram
= (uint64_t)info
.totalram
* (uint64_t)info
.mem_unit
;
96 /* We don't want to burn too much ram with the GPU. If the user has 4GiB
97 * or less, we use at most half. If they have more than 4GiB, we use 3/4.
99 uint64_t available_ram
;
100 if (total_ram
<= 4ull * 1024ull * 1024ull * 1024ull)
101 available_ram
= total_ram
/ 2;
103 available_ram
= total_ram
* 3 / 4;
105 /* We also want to leave some padding for things we allocate in the driver,
106 * so don't go over 3/4 of the GTT either.
108 uint64_t available_gtt
= gtt_size
* 3 / 4;
110 return MIN2(available_ram
, available_gtt
);
114 anv_physical_device_init_heaps(struct anv_physical_device
*device
, int fd
)
117 if (anv_gem_get_context_param(fd
, 0, I915_CONTEXT_PARAM_GTT_SIZE
,
119 /* If, for whatever reason, we can't actually get the GTT size from the
120 * kernel (too old?) fall back to the aperture size.
122 anv_perf_warn(NULL
, NULL
,
123 "Failed to get I915_CONTEXT_PARAM_GTT_SIZE: %m");
125 if (anv_gem_get_aperture(fd
, >t_size
) == -1) {
126 return vk_errorf(NULL
, NULL
, VK_ERROR_INITIALIZATION_FAILED
,
127 "failed to get aperture size: %m");
131 device
->supports_48bit_addresses
= (device
->info
.gen
>= 8) &&
132 gtt_size
> (4ULL << 30 /* GiB */);
134 uint64_t heap_size
= anv_compute_heap_size(fd
, gtt_size
);
136 if (heap_size
> (2ull << 30) && !device
->supports_48bit_addresses
) {
137 /* When running with an overridden PCI ID, we may get a GTT size from
138 * the kernel that is greater than 2 GiB but the execbuf check for 48bit
139 * address support can still fail. Just clamp the address space size to
140 * 2 GiB if we don't have 48-bit support.
142 intel_logw("%s:%d: The kernel reported a GTT size larger than 2 GiB but "
143 "not support for 48-bit addresses",
145 heap_size
= 2ull << 30;
148 if (heap_size
<= 3ull * (1ull << 30)) {
149 /* In this case, everything fits nicely into the 32-bit address space,
150 * so there's no need for supporting 48bit addresses on client-allocated
153 device
->memory
.heap_count
= 1;
154 device
->memory
.heaps
[0] = (struct anv_memory_heap
) {
155 .vma_start
= LOW_HEAP_MIN_ADDRESS
,
156 .vma_size
= LOW_HEAP_SIZE
,
158 .flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
,
159 .supports_48bit_addresses
= false,
162 /* Not everything will fit nicely into a 32-bit address space. In this
163 * case we need a 64-bit heap. Advertise a small 32-bit heap and a
164 * larger 48-bit heap. If we're in this case, then we have a total heap
165 * size larger than 3GiB which most likely means they have 8 GiB of
166 * video memory and so carving off 1 GiB for the 32-bit heap should be
169 const uint64_t heap_size_32bit
= 1ull << 30;
170 const uint64_t heap_size_48bit
= heap_size
- heap_size_32bit
;
172 assert(device
->supports_48bit_addresses
);
174 device
->memory
.heap_count
= 2;
175 device
->memory
.heaps
[0] = (struct anv_memory_heap
) {
176 .vma_start
= HIGH_HEAP_MIN_ADDRESS
,
177 /* Leave the last 4GiB out of the high vma range, so that no state
178 * base address + size can overflow 48 bits. For more information see
179 * the comment about Wa32bitGeneralStateOffset in anv_allocator.c
181 .vma_size
= gtt_size
- (1ull << 32) - HIGH_HEAP_MIN_ADDRESS
,
182 .size
= heap_size_48bit
,
183 .flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
,
184 .supports_48bit_addresses
= true,
186 device
->memory
.heaps
[1] = (struct anv_memory_heap
) {
187 .vma_start
= LOW_HEAP_MIN_ADDRESS
,
188 .vma_size
= LOW_HEAP_SIZE
,
189 .size
= heap_size_32bit
,
190 .flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
,
191 .supports_48bit_addresses
= false,
195 uint32_t type_count
= 0;
196 for (uint32_t heap
= 0; heap
< device
->memory
.heap_count
; heap
++) {
197 uint32_t valid_buffer_usage
= ~0;
199 /* There appears to be a hardware issue in the VF cache where it only
200 * considers the bottom 32 bits of memory addresses. If you happen to
201 * have two vertex buffers which get placed exactly 4 GiB apart and use
202 * them in back-to-back draw calls, you can get collisions. In order to
203 * solve this problem, we require vertex and index buffers be bound to
204 * memory allocated out of the 32-bit heap.
206 if (device
->memory
.heaps
[heap
].supports_48bit_addresses
) {
207 valid_buffer_usage
&= ~(VK_BUFFER_USAGE_INDEX_BUFFER_BIT
|
208 VK_BUFFER_USAGE_VERTEX_BUFFER_BIT
);
211 if (device
->info
.has_llc
) {
212 /* Big core GPUs share LLC with the CPU and thus one memory type can be
213 * both cached and coherent at the same time.
215 device
->memory
.types
[type_count
++] = (struct anv_memory_type
) {
216 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
217 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
218 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
|
219 VK_MEMORY_PROPERTY_HOST_CACHED_BIT
,
221 .valid_buffer_usage
= valid_buffer_usage
,
224 /* The spec requires that we expose a host-visible, coherent memory
225 * type, but Atom GPUs don't share LLC. Thus we offer two memory types
226 * to give the application a choice between cached, but not coherent and
227 * coherent but uncached (WC though).
229 device
->memory
.types
[type_count
++] = (struct anv_memory_type
) {
230 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
231 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
232 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
,
234 .valid_buffer_usage
= valid_buffer_usage
,
236 device
->memory
.types
[type_count
++] = (struct anv_memory_type
) {
237 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
238 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
239 VK_MEMORY_PROPERTY_HOST_CACHED_BIT
,
241 .valid_buffer_usage
= valid_buffer_usage
,
245 device
->memory
.type_count
= type_count
;
251 anv_physical_device_init_uuids(struct anv_physical_device
*device
)
253 const struct build_id_note
*note
=
254 build_id_find_nhdr_for_addr(anv_physical_device_init_uuids
);
256 return vk_errorf(device
->instance
, device
,
257 VK_ERROR_INITIALIZATION_FAILED
,
258 "Failed to find build-id");
261 unsigned build_id_len
= build_id_length(note
);
262 if (build_id_len
< 20) {
263 return vk_errorf(device
->instance
, device
,
264 VK_ERROR_INITIALIZATION_FAILED
,
265 "build-id too short. It needs to be a SHA");
268 memcpy(device
->driver_build_sha1
, build_id_data(note
), 20);
270 struct mesa_sha1 sha1_ctx
;
272 STATIC_ASSERT(VK_UUID_SIZE
<= sizeof(sha1
));
274 /* The pipeline cache UUID is used for determining when a pipeline cache is
275 * invalid. It needs both a driver build and the PCI ID of the device.
277 _mesa_sha1_init(&sha1_ctx
);
278 _mesa_sha1_update(&sha1_ctx
, build_id_data(note
), build_id_len
);
279 _mesa_sha1_update(&sha1_ctx
, &device
->chipset_id
,
280 sizeof(device
->chipset_id
));
281 _mesa_sha1_update(&sha1_ctx
, &device
->always_use_bindless
,
282 sizeof(device
->always_use_bindless
));
283 _mesa_sha1_update(&sha1_ctx
, &device
->has_a64_buffer_access
,
284 sizeof(device
->has_a64_buffer_access
));
285 _mesa_sha1_update(&sha1_ctx
, &device
->has_bindless_images
,
286 sizeof(device
->has_bindless_images
));
287 _mesa_sha1_update(&sha1_ctx
, &device
->has_bindless_samplers
,
288 sizeof(device
->has_bindless_samplers
));
289 _mesa_sha1_final(&sha1_ctx
, sha1
);
290 memcpy(device
->pipeline_cache_uuid
, sha1
, VK_UUID_SIZE
);
292 /* The driver UUID is used for determining sharability of images and memory
293 * between two Vulkan instances in separate processes. People who want to
294 * share memory need to also check the device UUID (below) so all this
295 * needs to be is the build-id.
297 memcpy(device
->driver_uuid
, build_id_data(note
), VK_UUID_SIZE
);
299 /* The device UUID uniquely identifies the given device within the machine.
300 * Since we never have more than one device, this doesn't need to be a real
301 * UUID. However, on the off-chance that someone tries to use this to
302 * cache pre-tiled images or something of the like, we use the PCI ID and
303 * some bits of ISL info to ensure that this is safe.
305 _mesa_sha1_init(&sha1_ctx
);
306 _mesa_sha1_update(&sha1_ctx
, &device
->chipset_id
,
307 sizeof(device
->chipset_id
));
308 _mesa_sha1_update(&sha1_ctx
, &device
->isl_dev
.has_bit6_swizzling
,
309 sizeof(device
->isl_dev
.has_bit6_swizzling
));
310 _mesa_sha1_final(&sha1_ctx
, sha1
);
311 memcpy(device
->device_uuid
, sha1
, VK_UUID_SIZE
);
317 anv_physical_device_init_disk_cache(struct anv_physical_device
*device
)
319 #ifdef ENABLE_SHADER_CACHE
321 ASSERTED
int len
= snprintf(renderer
, sizeof(renderer
), "anv_%04x",
323 assert(len
== sizeof(renderer
) - 2);
326 _mesa_sha1_format(timestamp
, device
->driver_build_sha1
);
328 const uint64_t driver_flags
=
329 brw_get_compiler_config_value(device
->compiler
);
330 device
->disk_cache
= disk_cache_create(renderer
, timestamp
, driver_flags
);
332 device
->disk_cache
= NULL
;
337 anv_physical_device_free_disk_cache(struct anv_physical_device
*device
)
339 #ifdef ENABLE_SHADER_CACHE
340 if (device
->disk_cache
)
341 disk_cache_destroy(device
->disk_cache
);
343 assert(device
->disk_cache
== NULL
);
348 get_available_system_memory()
350 char *meminfo
= os_read_file("/proc/meminfo");
354 char *str
= strstr(meminfo
, "MemAvailable:");
360 uint64_t kb_mem_available
;
361 if (sscanf(str
, "MemAvailable: %" PRIx64
, &kb_mem_available
) == 1) {
363 return kb_mem_available
<< 10;
371 anv_physical_device_init(struct anv_physical_device
*device
,
372 struct anv_instance
*instance
,
373 drmDevicePtr drm_device
)
375 const char *primary_path
= drm_device
->nodes
[DRM_NODE_PRIMARY
];
376 const char *path
= drm_device
->nodes
[DRM_NODE_RENDER
];
381 brw_process_intel_debug_variable();
383 fd
= open(path
, O_RDWR
| O_CLOEXEC
);
385 return vk_error(VK_ERROR_INCOMPATIBLE_DRIVER
);
387 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
388 device
->instance
= instance
;
390 assert(strlen(path
) < ARRAY_SIZE(device
->path
));
391 snprintf(device
->path
, ARRAY_SIZE(device
->path
), "%s", path
);
393 if (!gen_get_device_info_from_fd(fd
, &device
->info
)) {
394 result
= vk_error(VK_ERROR_INCOMPATIBLE_DRIVER
);
397 device
->chipset_id
= device
->info
.chipset_id
;
398 device
->no_hw
= device
->info
.no_hw
;
400 if (getenv("INTEL_NO_HW") != NULL
)
401 device
->no_hw
= true;
403 device
->pci_info
.domain
= drm_device
->businfo
.pci
->domain
;
404 device
->pci_info
.bus
= drm_device
->businfo
.pci
->bus
;
405 device
->pci_info
.device
= drm_device
->businfo
.pci
->dev
;
406 device
->pci_info
.function
= drm_device
->businfo
.pci
->func
;
408 device
->name
= gen_get_device_name(device
->chipset_id
);
410 if (device
->info
.is_haswell
) {
411 intel_logw("Haswell Vulkan support is incomplete");
412 } else if (device
->info
.gen
== 7 && !device
->info
.is_baytrail
) {
413 intel_logw("Ivy Bridge Vulkan support is incomplete");
414 } else if (device
->info
.gen
== 7 && device
->info
.is_baytrail
) {
415 intel_logw("Bay Trail Vulkan support is incomplete");
416 } else if (device
->info
.gen
>= 8 && device
->info
.gen
<= 11) {
417 /* Gen8-11 fully supported */
418 } else if (device
->info
.gen
== 12) {
419 intel_logw("Vulkan is not yet fully supported on gen12");
421 result
= vk_errorf(device
->instance
, device
,
422 VK_ERROR_INCOMPATIBLE_DRIVER
,
423 "Vulkan not yet supported on %s", device
->name
);
427 device
->cmd_parser_version
= -1;
428 if (device
->info
.gen
== 7) {
429 device
->cmd_parser_version
=
430 anv_gem_get_param(fd
, I915_PARAM_CMD_PARSER_VERSION
);
431 if (device
->cmd_parser_version
== -1) {
432 result
= vk_errorf(device
->instance
, device
,
433 VK_ERROR_INITIALIZATION_FAILED
,
434 "failed to get command parser version");
439 if (!anv_gem_get_param(fd
, I915_PARAM_HAS_WAIT_TIMEOUT
)) {
440 result
= vk_errorf(device
->instance
, device
,
441 VK_ERROR_INITIALIZATION_FAILED
,
442 "kernel missing gem wait");
446 if (!anv_gem_get_param(fd
, I915_PARAM_HAS_EXECBUF2
)) {
447 result
= vk_errorf(device
->instance
, device
,
448 VK_ERROR_INITIALIZATION_FAILED
,
449 "kernel missing execbuf2");
453 if (!device
->info
.has_llc
&&
454 anv_gem_get_param(fd
, I915_PARAM_MMAP_VERSION
) < 1) {
455 result
= vk_errorf(device
->instance
, device
,
456 VK_ERROR_INITIALIZATION_FAILED
,
457 "kernel missing wc mmap");
461 result
= anv_physical_device_init_heaps(device
, fd
);
462 if (result
!= VK_SUCCESS
)
465 device
->has_exec_async
= anv_gem_get_param(fd
, I915_PARAM_HAS_EXEC_ASYNC
);
466 device
->has_exec_capture
= anv_gem_get_param(fd
, I915_PARAM_HAS_EXEC_CAPTURE
);
467 device
->has_exec_fence
= anv_gem_get_param(fd
, I915_PARAM_HAS_EXEC_FENCE
);
468 device
->has_syncobj
= anv_gem_get_param(fd
, I915_PARAM_HAS_EXEC_FENCE_ARRAY
);
469 device
->has_syncobj_wait
= device
->has_syncobj
&&
470 anv_gem_supports_syncobj_wait(fd
);
471 device
->has_context_priority
= anv_gem_has_context_priority(fd
);
473 device
->use_softpin
= anv_gem_get_param(fd
, I915_PARAM_HAS_EXEC_SOFTPIN
)
474 && device
->supports_48bit_addresses
;
476 device
->has_context_isolation
=
477 anv_gem_get_param(fd
, I915_PARAM_HAS_CONTEXT_ISOLATION
);
479 device
->always_use_bindless
=
480 env_var_as_boolean("ANV_ALWAYS_BINDLESS", false);
482 /* We first got the A64 messages on broadwell and we can only use them if
483 * we can pass addresses directly into the shader which requires softpin.
485 device
->has_a64_buffer_access
= device
->info
.gen
>= 8 &&
488 /* We first get bindless image access on Skylake and we can only really do
489 * it if we don't have any relocations so we need softpin.
491 device
->has_bindless_images
= device
->info
.gen
>= 9 &&
494 /* We've had bindless samplers since Ivy Bridge (forever in Vulkan terms)
495 * because it's just a matter of setting the sampler address in the sample
496 * message header. However, we've not bothered to wire it up for vec4 so
497 * we leave it disabled on gen7.
499 device
->has_bindless_samplers
= device
->info
.gen
>= 8;
501 device
->has_mem_available
= get_available_system_memory() != 0;
503 /* Starting with Gen10, the timestamp frequency of the command streamer may
504 * vary from one part to another. We can query the value from the kernel.
506 if (device
->info
.gen
>= 10) {
507 int timestamp_frequency
=
508 anv_gem_get_param(fd
, I915_PARAM_CS_TIMESTAMP_FREQUENCY
);
510 if (timestamp_frequency
< 0)
511 intel_logw("Kernel 4.16-rc1+ required to properly query CS timestamp frequency");
513 device
->info
.timestamp_frequency
= timestamp_frequency
;
516 /* GENs prior to 8 do not support EU/Subslice info */
517 if (device
->info
.gen
>= 8) {
518 device
->subslice_total
= anv_gem_get_param(fd
, I915_PARAM_SUBSLICE_TOTAL
);
519 device
->eu_total
= anv_gem_get_param(fd
, I915_PARAM_EU_TOTAL
);
521 /* Without this information, we cannot get the right Braswell
522 * brandstrings, and we have to use conservative numbers for GPGPU on
523 * many platforms, but otherwise, things will just work.
525 if (device
->subslice_total
< 1 || device
->eu_total
< 1) {
526 intel_logw("Kernel 4.1 required to properly query GPU properties");
528 } else if (device
->info
.gen
== 7) {
529 device
->subslice_total
= 1 << (device
->info
.gt
- 1);
532 if (device
->info
.is_cherryview
&&
533 device
->subslice_total
> 0 && device
->eu_total
> 0) {
534 /* Logical CS threads = EUs per subslice * num threads per EU */
535 uint32_t max_cs_threads
=
536 device
->eu_total
/ device
->subslice_total
* device
->info
.num_thread_per_eu
;
538 /* Fuse configurations may give more threads than expected, never less. */
539 if (max_cs_threads
> device
->info
.max_cs_threads
)
540 device
->info
.max_cs_threads
= max_cs_threads
;
543 device
->compiler
= brw_compiler_create(NULL
, &device
->info
);
544 if (device
->compiler
== NULL
) {
545 result
= vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
548 device
->compiler
->shader_debug_log
= compiler_debug_log
;
549 device
->compiler
->shader_perf_log
= compiler_perf_log
;
550 device
->compiler
->supports_pull_constants
= false;
551 device
->compiler
->constant_buffer_0_is_relative
=
552 device
->info
.gen
< 8 || !device
->has_context_isolation
;
553 device
->compiler
->supports_shader_constants
= true;
555 /* Broadwell PRM says:
557 * "Before Gen8, there was a historical configuration control field to
558 * swizzle address bit[6] for in X/Y tiling modes. This was set in three
559 * different places: TILECTL[1:0], ARB_MODE[5:4], and
560 * DISP_ARB_CTL[14:13].
562 * For Gen8 and subsequent generations, the swizzle fields are all
563 * reserved, and the CPU's memory controller performs all address
564 * swizzling modifications."
567 device
->info
.gen
< 8 && anv_gem_get_bit6_swizzle(fd
, I915_TILING_X
);
569 isl_device_init(&device
->isl_dev
, &device
->info
, swizzled
);
571 result
= anv_physical_device_init_uuids(device
);
572 if (result
!= VK_SUCCESS
)
575 anv_physical_device_init_disk_cache(device
);
577 if (instance
->enabled_extensions
.KHR_display
) {
578 master_fd
= open(primary_path
, O_RDWR
| O_CLOEXEC
);
579 if (master_fd
>= 0) {
580 /* prod the device with a GETPARAM call which will fail if
581 * we don't have permission to even render on this device
583 if (anv_gem_get_param(master_fd
, I915_PARAM_CHIPSET_ID
) == 0) {
589 device
->master_fd
= master_fd
;
591 result
= anv_init_wsi(device
);
592 if (result
!= VK_SUCCESS
) {
593 ralloc_free(device
->compiler
);
594 anv_physical_device_free_disk_cache(device
);
598 anv_physical_device_get_supported_extensions(device
,
599 &device
->supported_extensions
);
602 device
->local_fd
= fd
;
614 anv_physical_device_finish(struct anv_physical_device
*device
)
616 anv_finish_wsi(device
);
617 anv_physical_device_free_disk_cache(device
);
618 ralloc_free(device
->compiler
);
619 close(device
->local_fd
);
620 if (device
->master_fd
>= 0)
621 close(device
->master_fd
);
625 default_alloc_func(void *pUserData
, size_t size
, size_t align
,
626 VkSystemAllocationScope allocationScope
)
632 default_realloc_func(void *pUserData
, void *pOriginal
, size_t size
,
633 size_t align
, VkSystemAllocationScope allocationScope
)
635 return realloc(pOriginal
, size
);
639 default_free_func(void *pUserData
, void *pMemory
)
644 static const VkAllocationCallbacks default_alloc
= {
646 .pfnAllocation
= default_alloc_func
,
647 .pfnReallocation
= default_realloc_func
,
648 .pfnFree
= default_free_func
,
651 VkResult
anv_EnumerateInstanceExtensionProperties(
652 const char* pLayerName
,
653 uint32_t* pPropertyCount
,
654 VkExtensionProperties
* pProperties
)
656 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
658 for (int i
= 0; i
< ANV_INSTANCE_EXTENSION_COUNT
; i
++) {
659 if (anv_instance_extensions_supported
.extensions
[i
]) {
660 vk_outarray_append(&out
, prop
) {
661 *prop
= anv_instance_extensions
[i
];
666 return vk_outarray_status(&out
);
669 VkResult
anv_CreateInstance(
670 const VkInstanceCreateInfo
* pCreateInfo
,
671 const VkAllocationCallbacks
* pAllocator
,
672 VkInstance
* pInstance
)
674 struct anv_instance
*instance
;
677 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO
);
679 struct anv_instance_extension_table enabled_extensions
= {};
680 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
682 for (idx
= 0; idx
< ANV_INSTANCE_EXTENSION_COUNT
; idx
++) {
683 if (strcmp(pCreateInfo
->ppEnabledExtensionNames
[i
],
684 anv_instance_extensions
[idx
].extensionName
) == 0)
688 if (idx
>= ANV_INSTANCE_EXTENSION_COUNT
)
689 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
691 if (!anv_instance_extensions_supported
.extensions
[idx
])
692 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
694 enabled_extensions
.extensions
[idx
] = true;
697 instance
= vk_alloc2(&default_alloc
, pAllocator
, sizeof(*instance
), 8,
698 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE
);
700 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
702 instance
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
705 instance
->alloc
= *pAllocator
;
707 instance
->alloc
= default_alloc
;
709 instance
->app_info
= (struct anv_app_info
) { .api_version
= 0 };
710 if (pCreateInfo
->pApplicationInfo
) {
711 const VkApplicationInfo
*app
= pCreateInfo
->pApplicationInfo
;
713 instance
->app_info
.app_name
=
714 vk_strdup(&instance
->alloc
, app
->pApplicationName
,
715 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE
);
716 instance
->app_info
.app_version
= app
->applicationVersion
;
718 instance
->app_info
.engine_name
=
719 vk_strdup(&instance
->alloc
, app
->pEngineName
,
720 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE
);
721 instance
->app_info
.engine_version
= app
->engineVersion
;
723 instance
->app_info
.api_version
= app
->apiVersion
;
726 if (instance
->app_info
.api_version
== 0)
727 instance
->app_info
.api_version
= VK_API_VERSION_1_0
;
729 instance
->enabled_extensions
= enabled_extensions
;
731 for (unsigned i
= 0; i
< ARRAY_SIZE(instance
->dispatch
.entrypoints
); i
++) {
732 /* Vulkan requires that entrypoints for extensions which have not been
733 * enabled must not be advertised.
735 if (!anv_instance_entrypoint_is_enabled(i
, instance
->app_info
.api_version
,
736 &instance
->enabled_extensions
)) {
737 instance
->dispatch
.entrypoints
[i
] = NULL
;
739 instance
->dispatch
.entrypoints
[i
] =
740 anv_instance_dispatch_table
.entrypoints
[i
];
744 for (unsigned i
= 0; i
< ARRAY_SIZE(instance
->device_dispatch
.entrypoints
); i
++) {
745 /* Vulkan requires that entrypoints for extensions which have not been
746 * enabled must not be advertised.
748 if (!anv_device_entrypoint_is_enabled(i
, instance
->app_info
.api_version
,
749 &instance
->enabled_extensions
, NULL
)) {
750 instance
->device_dispatch
.entrypoints
[i
] = NULL
;
752 instance
->device_dispatch
.entrypoints
[i
] =
753 anv_device_dispatch_table
.entrypoints
[i
];
757 instance
->physicalDeviceCount
= -1;
759 result
= vk_debug_report_instance_init(&instance
->debug_report_callbacks
);
760 if (result
!= VK_SUCCESS
) {
761 vk_free2(&default_alloc
, pAllocator
, instance
);
762 return vk_error(result
);
765 instance
->pipeline_cache_enabled
=
766 env_var_as_boolean("ANV_ENABLE_PIPELINE_CACHE", true);
769 glsl_type_singleton_init_or_ref();
771 VG(VALGRIND_CREATE_MEMPOOL(instance
, 0, false));
773 *pInstance
= anv_instance_to_handle(instance
);
778 void anv_DestroyInstance(
779 VkInstance _instance
,
780 const VkAllocationCallbacks
* pAllocator
)
782 ANV_FROM_HANDLE(anv_instance
, instance
, _instance
);
787 if (instance
->physicalDeviceCount
> 0) {
788 /* We support at most one physical device. */
789 assert(instance
->physicalDeviceCount
== 1);
790 anv_physical_device_finish(&instance
->physicalDevice
);
793 vk_free(&instance
->alloc
, (char *)instance
->app_info
.app_name
);
794 vk_free(&instance
->alloc
, (char *)instance
->app_info
.engine_name
);
796 VG(VALGRIND_DESTROY_MEMPOOL(instance
));
798 vk_debug_report_instance_destroy(&instance
->debug_report_callbacks
);
800 glsl_type_singleton_decref();
803 vk_free(&instance
->alloc
, instance
);
807 anv_enumerate_devices(struct anv_instance
*instance
)
809 /* TODO: Check for more devices ? */
810 drmDevicePtr devices
[8];
811 VkResult result
= VK_ERROR_INCOMPATIBLE_DRIVER
;
814 instance
->physicalDeviceCount
= 0;
816 max_devices
= drmGetDevices2(0, devices
, ARRAY_SIZE(devices
));
818 return VK_ERROR_INCOMPATIBLE_DRIVER
;
820 for (unsigned i
= 0; i
< (unsigned)max_devices
; i
++) {
821 if (devices
[i
]->available_nodes
& 1 << DRM_NODE_RENDER
&&
822 devices
[i
]->bustype
== DRM_BUS_PCI
&&
823 devices
[i
]->deviceinfo
.pci
->vendor_id
== 0x8086) {
825 result
= anv_physical_device_init(&instance
->physicalDevice
,
826 instance
, devices
[i
]);
827 if (result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
831 drmFreeDevices(devices
, max_devices
);
833 if (result
== VK_SUCCESS
)
834 instance
->physicalDeviceCount
= 1;
840 anv_instance_ensure_physical_device(struct anv_instance
*instance
)
842 if (instance
->physicalDeviceCount
< 0) {
843 VkResult result
= anv_enumerate_devices(instance
);
844 if (result
!= VK_SUCCESS
&&
845 result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
852 VkResult
anv_EnumeratePhysicalDevices(
853 VkInstance _instance
,
854 uint32_t* pPhysicalDeviceCount
,
855 VkPhysicalDevice
* pPhysicalDevices
)
857 ANV_FROM_HANDLE(anv_instance
, instance
, _instance
);
858 VK_OUTARRAY_MAKE(out
, pPhysicalDevices
, pPhysicalDeviceCount
);
860 VkResult result
= anv_instance_ensure_physical_device(instance
);
861 if (result
!= VK_SUCCESS
)
864 if (instance
->physicalDeviceCount
== 0)
867 assert(instance
->physicalDeviceCount
== 1);
868 vk_outarray_append(&out
, i
) {
869 *i
= anv_physical_device_to_handle(&instance
->physicalDevice
);
872 return vk_outarray_status(&out
);
875 VkResult
anv_EnumeratePhysicalDeviceGroups(
876 VkInstance _instance
,
877 uint32_t* pPhysicalDeviceGroupCount
,
878 VkPhysicalDeviceGroupProperties
* pPhysicalDeviceGroupProperties
)
880 ANV_FROM_HANDLE(anv_instance
, instance
, _instance
);
881 VK_OUTARRAY_MAKE(out
, pPhysicalDeviceGroupProperties
,
882 pPhysicalDeviceGroupCount
);
884 VkResult result
= anv_instance_ensure_physical_device(instance
);
885 if (result
!= VK_SUCCESS
)
888 if (instance
->physicalDeviceCount
== 0)
891 assert(instance
->physicalDeviceCount
== 1);
893 vk_outarray_append(&out
, p
) {
894 p
->physicalDeviceCount
= 1;
895 memset(p
->physicalDevices
, 0, sizeof(p
->physicalDevices
));
896 p
->physicalDevices
[0] =
897 anv_physical_device_to_handle(&instance
->physicalDevice
);
898 p
->subsetAllocation
= false;
900 vk_foreach_struct(ext
, p
->pNext
)
901 anv_debug_ignored_stype(ext
->sType
);
904 return vk_outarray_status(&out
);
907 void anv_GetPhysicalDeviceFeatures(
908 VkPhysicalDevice physicalDevice
,
909 VkPhysicalDeviceFeatures
* pFeatures
)
911 ANV_FROM_HANDLE(anv_physical_device
, pdevice
, physicalDevice
);
913 *pFeatures
= (VkPhysicalDeviceFeatures
) {
914 .robustBufferAccess
= true,
915 .fullDrawIndexUint32
= true,
916 .imageCubeArray
= true,
917 .independentBlend
= true,
918 .geometryShader
= true,
919 .tessellationShader
= true,
920 .sampleRateShading
= true,
921 .dualSrcBlend
= true,
923 .multiDrawIndirect
= true,
924 .drawIndirectFirstInstance
= true,
926 .depthBiasClamp
= true,
927 .fillModeNonSolid
= true,
928 .depthBounds
= false,
932 .multiViewport
= true,
933 .samplerAnisotropy
= true,
934 .textureCompressionETC2
= pdevice
->info
.gen
>= 8 ||
935 pdevice
->info
.is_baytrail
,
936 .textureCompressionASTC_LDR
= pdevice
->info
.gen
>= 9, /* FINISHME CHV */
937 .textureCompressionBC
= true,
938 .occlusionQueryPrecise
= true,
939 .pipelineStatisticsQuery
= true,
940 .fragmentStoresAndAtomics
= true,
941 .shaderTessellationAndGeometryPointSize
= true,
942 .shaderImageGatherExtended
= true,
943 .shaderStorageImageExtendedFormats
= true,
944 .shaderStorageImageMultisample
= false,
945 .shaderStorageImageReadWithoutFormat
= false,
946 .shaderStorageImageWriteWithoutFormat
= true,
947 .shaderUniformBufferArrayDynamicIndexing
= true,
948 .shaderSampledImageArrayDynamicIndexing
= true,
949 .shaderStorageBufferArrayDynamicIndexing
= true,
950 .shaderStorageImageArrayDynamicIndexing
= true,
951 .shaderClipDistance
= true,
952 .shaderCullDistance
= true,
953 .shaderFloat64
= pdevice
->info
.gen
>= 8 &&
954 pdevice
->info
.has_64bit_types
,
955 .shaderInt64
= pdevice
->info
.gen
>= 8 &&
956 pdevice
->info
.has_64bit_types
,
957 .shaderInt16
= pdevice
->info
.gen
>= 8,
958 .shaderResourceMinLod
= pdevice
->info
.gen
>= 9,
959 .variableMultisampleRate
= true,
960 .inheritedQueries
= true,
963 /* We can't do image stores in vec4 shaders */
964 pFeatures
->vertexPipelineStoresAndAtomics
=
965 pdevice
->compiler
->scalar_stage
[MESA_SHADER_VERTEX
] &&
966 pdevice
->compiler
->scalar_stage
[MESA_SHADER_GEOMETRY
];
968 struct anv_app_info
*app_info
= &pdevice
->instance
->app_info
;
970 /* The new DOOM and Wolfenstein games require depthBounds without
971 * checking for it. They seem to run fine without it so just claim it's
972 * there and accept the consequences.
974 if (app_info
->engine_name
&& strcmp(app_info
->engine_name
, "idTech") == 0)
975 pFeatures
->depthBounds
= true;
978 void anv_GetPhysicalDeviceFeatures2(
979 VkPhysicalDevice physicalDevice
,
980 VkPhysicalDeviceFeatures2
* pFeatures
)
982 ANV_FROM_HANDLE(anv_physical_device
, pdevice
, physicalDevice
);
983 anv_GetPhysicalDeviceFeatures(physicalDevice
, &pFeatures
->features
);
985 vk_foreach_struct(ext
, pFeatures
->pNext
) {
986 switch (ext
->sType
) {
987 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_8BIT_STORAGE_FEATURES_KHR
: {
988 VkPhysicalDevice8BitStorageFeaturesKHR
*features
=
989 (VkPhysicalDevice8BitStorageFeaturesKHR
*)ext
;
990 features
->storageBuffer8BitAccess
= pdevice
->info
.gen
>= 8;
991 features
->uniformAndStorageBuffer8BitAccess
= pdevice
->info
.gen
>= 8;
992 features
->storagePushConstant8
= pdevice
->info
.gen
>= 8;
996 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES
: {
997 VkPhysicalDevice16BitStorageFeatures
*features
=
998 (VkPhysicalDevice16BitStorageFeatures
*)ext
;
999 features
->storageBuffer16BitAccess
= pdevice
->info
.gen
>= 8;
1000 features
->uniformAndStorageBuffer16BitAccess
= pdevice
->info
.gen
>= 8;
1001 features
->storagePushConstant16
= pdevice
->info
.gen
>= 8;
1002 features
->storageInputOutput16
= false;
1006 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES_EXT
: {
1007 VkPhysicalDeviceBufferDeviceAddressFeaturesEXT
*features
= (void *)ext
;
1008 features
->bufferDeviceAddress
= pdevice
->has_a64_buffer_access
;
1009 features
->bufferDeviceAddressCaptureReplay
= false;
1010 features
->bufferDeviceAddressMultiDevice
= false;
1014 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_COMPUTE_SHADER_DERIVATIVES_FEATURES_NV
: {
1015 VkPhysicalDeviceComputeShaderDerivativesFeaturesNV
*features
=
1016 (VkPhysicalDeviceComputeShaderDerivativesFeaturesNV
*)ext
;
1017 features
->computeDerivativeGroupQuads
= true;
1018 features
->computeDerivativeGroupLinear
= true;
1022 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONDITIONAL_RENDERING_FEATURES_EXT
: {
1023 VkPhysicalDeviceConditionalRenderingFeaturesEXT
*features
=
1024 (VkPhysicalDeviceConditionalRenderingFeaturesEXT
*)ext
;
1025 features
->conditionalRendering
= pdevice
->info
.gen
>= 8 ||
1026 pdevice
->info
.is_haswell
;
1027 features
->inheritedConditionalRendering
= pdevice
->info
.gen
>= 8 ||
1028 pdevice
->info
.is_haswell
;
1032 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_CLIP_ENABLE_FEATURES_EXT
: {
1033 VkPhysicalDeviceDepthClipEnableFeaturesEXT
*features
=
1034 (VkPhysicalDeviceDepthClipEnableFeaturesEXT
*)ext
;
1035 features
->depthClipEnable
= true;
1039 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR
: {
1040 VkPhysicalDeviceFloat16Int8FeaturesKHR
*features
= (void *)ext
;
1041 features
->shaderFloat16
= pdevice
->info
.gen
>= 8;
1042 features
->shaderInt8
= pdevice
->info
.gen
>= 8;
1046 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FRAGMENT_SHADER_INTERLOCK_FEATURES_EXT
: {
1047 VkPhysicalDeviceFragmentShaderInterlockFeaturesEXT
*features
=
1048 (VkPhysicalDeviceFragmentShaderInterlockFeaturesEXT
*)ext
;
1049 features
->fragmentShaderSampleInterlock
= pdevice
->info
.gen
>= 9;
1050 features
->fragmentShaderPixelInterlock
= pdevice
->info
.gen
>= 9;
1051 features
->fragmentShaderShadingRateInterlock
= false;
1055 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_HOST_QUERY_RESET_FEATURES_EXT
: {
1056 VkPhysicalDeviceHostQueryResetFeaturesEXT
*features
=
1057 (VkPhysicalDeviceHostQueryResetFeaturesEXT
*)ext
;
1058 features
->hostQueryReset
= true;
1062 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT
: {
1063 VkPhysicalDeviceDescriptorIndexingFeaturesEXT
*features
=
1064 (VkPhysicalDeviceDescriptorIndexingFeaturesEXT
*)ext
;
1065 features
->shaderInputAttachmentArrayDynamicIndexing
= false;
1066 features
->shaderUniformTexelBufferArrayDynamicIndexing
= true;
1067 features
->shaderStorageTexelBufferArrayDynamicIndexing
= true;
1068 features
->shaderUniformBufferArrayNonUniformIndexing
= false;
1069 features
->shaderSampledImageArrayNonUniformIndexing
= true;
1070 features
->shaderStorageBufferArrayNonUniformIndexing
= true;
1071 features
->shaderStorageImageArrayNonUniformIndexing
= true;
1072 features
->shaderInputAttachmentArrayNonUniformIndexing
= false;
1073 features
->shaderUniformTexelBufferArrayNonUniformIndexing
= true;
1074 features
->shaderStorageTexelBufferArrayNonUniformIndexing
= true;
1075 features
->descriptorBindingUniformBufferUpdateAfterBind
= false;
1076 features
->descriptorBindingSampledImageUpdateAfterBind
= true;
1077 features
->descriptorBindingStorageImageUpdateAfterBind
= true;
1078 features
->descriptorBindingStorageBufferUpdateAfterBind
= true;
1079 features
->descriptorBindingUniformTexelBufferUpdateAfterBind
= true;
1080 features
->descriptorBindingStorageTexelBufferUpdateAfterBind
= true;
1081 features
->descriptorBindingUpdateUnusedWhilePending
= true;
1082 features
->descriptorBindingPartiallyBound
= true;
1083 features
->descriptorBindingVariableDescriptorCount
= false;
1084 features
->runtimeDescriptorArray
= true;
1088 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INDEX_TYPE_UINT8_FEATURES_EXT
: {
1089 VkPhysicalDeviceIndexTypeUint8FeaturesEXT
*features
=
1090 (VkPhysicalDeviceIndexTypeUint8FeaturesEXT
*)ext
;
1091 features
->indexTypeUint8
= true;
1095 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INLINE_UNIFORM_BLOCK_FEATURES_EXT
: {
1096 VkPhysicalDeviceInlineUniformBlockFeaturesEXT
*features
=
1097 (VkPhysicalDeviceInlineUniformBlockFeaturesEXT
*)ext
;
1098 features
->inlineUniformBlock
= true;
1099 features
->descriptorBindingInlineUniformBlockUpdateAfterBind
= true;
1103 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_LINE_RASTERIZATION_FEATURES_EXT
: {
1104 VkPhysicalDeviceLineRasterizationFeaturesEXT
*features
=
1105 (VkPhysicalDeviceLineRasterizationFeaturesEXT
*)ext
;
1106 features
->rectangularLines
= true;
1107 features
->bresenhamLines
= true;
1108 features
->smoothLines
= true;
1109 features
->stippledRectangularLines
= false;
1110 features
->stippledBresenhamLines
= true;
1111 features
->stippledSmoothLines
= false;
1115 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES
: {
1116 VkPhysicalDeviceMultiviewFeatures
*features
=
1117 (VkPhysicalDeviceMultiviewFeatures
*)ext
;
1118 features
->multiview
= true;
1119 features
->multiviewGeometryShader
= true;
1120 features
->multiviewTessellationShader
= true;
1124 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_IMAGELESS_FRAMEBUFFER_FEATURES_KHR
: {
1125 VkPhysicalDeviceImagelessFramebufferFeaturesKHR
*features
=
1126 (VkPhysicalDeviceImagelessFramebufferFeaturesKHR
*)ext
;
1127 features
->imagelessFramebuffer
= true;
1131 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PIPELINE_EXECUTABLE_PROPERTIES_FEATURES_KHR
: {
1132 VkPhysicalDevicePipelineExecutablePropertiesFeaturesKHR
*features
=
1133 (VkPhysicalDevicePipelineExecutablePropertiesFeaturesKHR
*)ext
;
1134 features
->pipelineExecutableInfo
= true;
1138 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES
: {
1139 VkPhysicalDeviceProtectedMemoryFeatures
*features
= (void *)ext
;
1140 features
->protectedMemory
= false;
1144 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES
: {
1145 VkPhysicalDeviceSamplerYcbcrConversionFeatures
*features
=
1146 (VkPhysicalDeviceSamplerYcbcrConversionFeatures
*) ext
;
1147 features
->samplerYcbcrConversion
= true;
1151 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SCALAR_BLOCK_LAYOUT_FEATURES_EXT
: {
1152 VkPhysicalDeviceScalarBlockLayoutFeaturesEXT
*features
=
1153 (VkPhysicalDeviceScalarBlockLayoutFeaturesEXT
*)ext
;
1154 features
->scalarBlockLayout
= true;
1158 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_INT64_FEATURES_KHR
: {
1159 VkPhysicalDeviceShaderAtomicInt64FeaturesKHR
*features
= (void *)ext
;
1160 features
->shaderBufferInt64Atomics
=
1161 pdevice
->info
.gen
>= 9 && pdevice
->use_softpin
;
1162 features
->shaderSharedInt64Atomics
= VK_FALSE
;
1166 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DEMOTE_TO_HELPER_INVOCATION_FEATURES_EXT
: {
1167 VkPhysicalDeviceShaderDemoteToHelperInvocationFeaturesEXT
*features
= (void *)ext
;
1168 features
->shaderDemoteToHelperInvocation
= true;
1172 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETERS_FEATURES
: {
1173 VkPhysicalDeviceShaderDrawParametersFeatures
*features
= (void *)ext
;
1174 features
->shaderDrawParameters
= true;
1178 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_FEATURES_EXT
: {
1179 VkPhysicalDeviceSubgroupSizeControlFeaturesEXT
*features
=
1180 (VkPhysicalDeviceSubgroupSizeControlFeaturesEXT
*)ext
;
1181 features
->subgroupSizeControl
= true;
1182 features
->computeFullSubgroups
= true;
1186 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TEXEL_BUFFER_ALIGNMENT_FEATURES_EXT
: {
1187 VkPhysicalDeviceTexelBufferAlignmentFeaturesEXT
*features
=
1188 (VkPhysicalDeviceTexelBufferAlignmentFeaturesEXT
*)ext
;
1189 features
->texelBufferAlignment
= true;
1193 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTERS_FEATURES
: {
1194 VkPhysicalDeviceVariablePointersFeatures
*features
= (void *)ext
;
1195 features
->variablePointersStorageBuffer
= true;
1196 features
->variablePointers
= true;
1200 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT
: {
1201 VkPhysicalDeviceTransformFeedbackFeaturesEXT
*features
=
1202 (VkPhysicalDeviceTransformFeedbackFeaturesEXT
*)ext
;
1203 features
->transformFeedback
= true;
1204 features
->geometryStreams
= true;
1208 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_UNIFORM_BUFFER_STANDARD_LAYOUT_FEATURES_KHR
: {
1209 VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR
*features
=
1210 (VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR
*)ext
;
1211 features
->uniformBufferStandardLayout
= true;
1215 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_FEATURES_EXT
: {
1216 VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT
*features
=
1217 (VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT
*)ext
;
1218 features
->vertexAttributeInstanceRateDivisor
= true;
1219 features
->vertexAttributeInstanceRateZeroDivisor
= true;
1223 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_YCBCR_IMAGE_ARRAYS_FEATURES_EXT
: {
1224 VkPhysicalDeviceYcbcrImageArraysFeaturesEXT
*features
=
1225 (VkPhysicalDeviceYcbcrImageArraysFeaturesEXT
*)ext
;
1226 features
->ycbcrImageArrays
= true;
1231 anv_debug_ignored_stype(ext
->sType
);
1237 #define MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS 64
1239 #define MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS 64
1240 #define MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS 256
1242 void anv_GetPhysicalDeviceProperties(
1243 VkPhysicalDevice physicalDevice
,
1244 VkPhysicalDeviceProperties
* pProperties
)
1246 ANV_FROM_HANDLE(anv_physical_device
, pdevice
, physicalDevice
);
1247 const struct gen_device_info
*devinfo
= &pdevice
->info
;
1249 /* See assertions made when programming the buffer surface state. */
1250 const uint32_t max_raw_buffer_sz
= devinfo
->gen
>= 7 ?
1251 (1ul << 30) : (1ul << 27);
1253 const uint32_t max_ssbos
= pdevice
->has_a64_buffer_access
? UINT16_MAX
: 64;
1254 const uint32_t max_textures
=
1255 pdevice
->has_bindless_images
? UINT16_MAX
: 128;
1256 const uint32_t max_samplers
=
1257 pdevice
->has_bindless_samplers
? UINT16_MAX
:
1258 (devinfo
->gen
>= 8 || devinfo
->is_haswell
) ? 128 : 16;
1259 const uint32_t max_images
=
1260 pdevice
->has_bindless_images
? UINT16_MAX
: MAX_IMAGES
;
1262 /* The moment we have anything bindless, claim a high per-stage limit */
1263 const uint32_t max_per_stage
=
1264 pdevice
->has_a64_buffer_access
? UINT32_MAX
:
1265 MAX_BINDING_TABLE_SIZE
- MAX_RTS
;
1267 VkSampleCountFlags sample_counts
=
1268 isl_device_get_sample_counts(&pdevice
->isl_dev
);
1271 VkPhysicalDeviceLimits limits
= {
1272 .maxImageDimension1D
= (1 << 14),
1273 .maxImageDimension2D
= (1 << 14),
1274 .maxImageDimension3D
= (1 << 11),
1275 .maxImageDimensionCube
= (1 << 14),
1276 .maxImageArrayLayers
= (1 << 11),
1277 .maxTexelBufferElements
= 128 * 1024 * 1024,
1278 .maxUniformBufferRange
= (1ul << 27),
1279 .maxStorageBufferRange
= max_raw_buffer_sz
,
1280 .maxPushConstantsSize
= MAX_PUSH_CONSTANTS_SIZE
,
1281 .maxMemoryAllocationCount
= UINT32_MAX
,
1282 .maxSamplerAllocationCount
= 64 * 1024,
1283 .bufferImageGranularity
= 64, /* A cache line */
1284 .sparseAddressSpaceSize
= 0,
1285 .maxBoundDescriptorSets
= MAX_SETS
,
1286 .maxPerStageDescriptorSamplers
= max_samplers
,
1287 .maxPerStageDescriptorUniformBuffers
= MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS
,
1288 .maxPerStageDescriptorStorageBuffers
= max_ssbos
,
1289 .maxPerStageDescriptorSampledImages
= max_textures
,
1290 .maxPerStageDescriptorStorageImages
= max_images
,
1291 .maxPerStageDescriptorInputAttachments
= MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS
,
1292 .maxPerStageResources
= max_per_stage
,
1293 .maxDescriptorSetSamplers
= 6 * max_samplers
, /* number of stages * maxPerStageDescriptorSamplers */
1294 .maxDescriptorSetUniformBuffers
= 6 * MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS
, /* number of stages * maxPerStageDescriptorUniformBuffers */
1295 .maxDescriptorSetUniformBuffersDynamic
= MAX_DYNAMIC_BUFFERS
/ 2,
1296 .maxDescriptorSetStorageBuffers
= 6 * max_ssbos
, /* number of stages * maxPerStageDescriptorStorageBuffers */
1297 .maxDescriptorSetStorageBuffersDynamic
= MAX_DYNAMIC_BUFFERS
/ 2,
1298 .maxDescriptorSetSampledImages
= 6 * max_textures
, /* number of stages * maxPerStageDescriptorSampledImages */
1299 .maxDescriptorSetStorageImages
= 6 * max_images
, /* number of stages * maxPerStageDescriptorStorageImages */
1300 .maxDescriptorSetInputAttachments
= MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS
,
1301 .maxVertexInputAttributes
= MAX_VBS
,
1302 .maxVertexInputBindings
= MAX_VBS
,
1303 .maxVertexInputAttributeOffset
= 2047,
1304 .maxVertexInputBindingStride
= 2048,
1305 .maxVertexOutputComponents
= 128,
1306 .maxTessellationGenerationLevel
= 64,
1307 .maxTessellationPatchSize
= 32,
1308 .maxTessellationControlPerVertexInputComponents
= 128,
1309 .maxTessellationControlPerVertexOutputComponents
= 128,
1310 .maxTessellationControlPerPatchOutputComponents
= 128,
1311 .maxTessellationControlTotalOutputComponents
= 2048,
1312 .maxTessellationEvaluationInputComponents
= 128,
1313 .maxTessellationEvaluationOutputComponents
= 128,
1314 .maxGeometryShaderInvocations
= 32,
1315 .maxGeometryInputComponents
= 64,
1316 .maxGeometryOutputComponents
= 128,
1317 .maxGeometryOutputVertices
= 256,
1318 .maxGeometryTotalOutputComponents
= 1024,
1319 .maxFragmentInputComponents
= 116, /* 128 components - (PSIZ, CLIP_DIST0, CLIP_DIST1) */
1320 .maxFragmentOutputAttachments
= 8,
1321 .maxFragmentDualSrcAttachments
= 1,
1322 .maxFragmentCombinedOutputResources
= 8,
1323 .maxComputeSharedMemorySize
= 64 * 1024,
1324 .maxComputeWorkGroupCount
= { 65535, 65535, 65535 },
1325 .maxComputeWorkGroupInvocations
= 32 * devinfo
->max_cs_threads
,
1326 .maxComputeWorkGroupSize
= {
1327 16 * devinfo
->max_cs_threads
,
1328 16 * devinfo
->max_cs_threads
,
1329 16 * devinfo
->max_cs_threads
,
1331 .subPixelPrecisionBits
= 8,
1332 .subTexelPrecisionBits
= 8,
1333 .mipmapPrecisionBits
= 8,
1334 .maxDrawIndexedIndexValue
= UINT32_MAX
,
1335 .maxDrawIndirectCount
= UINT32_MAX
,
1336 .maxSamplerLodBias
= 16,
1337 .maxSamplerAnisotropy
= 16,
1338 .maxViewports
= MAX_VIEWPORTS
,
1339 .maxViewportDimensions
= { (1 << 14), (1 << 14) },
1340 .viewportBoundsRange
= { INT16_MIN
, INT16_MAX
},
1341 .viewportSubPixelBits
= 13, /* We take a float? */
1342 .minMemoryMapAlignment
= 4096, /* A page */
1343 /* The dataport requires texel alignment so we need to assume a worst
1344 * case of R32G32B32A32 which is 16 bytes.
1346 .minTexelBufferOffsetAlignment
= 16,
1347 /* We need 16 for UBO block reads to work and 32 for push UBOs */
1348 .minUniformBufferOffsetAlignment
= 32,
1349 .minStorageBufferOffsetAlignment
= 4,
1350 .minTexelOffset
= -8,
1351 .maxTexelOffset
= 7,
1352 .minTexelGatherOffset
= -32,
1353 .maxTexelGatherOffset
= 31,
1354 .minInterpolationOffset
= -0.5,
1355 .maxInterpolationOffset
= 0.4375,
1356 .subPixelInterpolationOffsetBits
= 4,
1357 .maxFramebufferWidth
= (1 << 14),
1358 .maxFramebufferHeight
= (1 << 14),
1359 .maxFramebufferLayers
= (1 << 11),
1360 .framebufferColorSampleCounts
= sample_counts
,
1361 .framebufferDepthSampleCounts
= sample_counts
,
1362 .framebufferStencilSampleCounts
= sample_counts
,
1363 .framebufferNoAttachmentsSampleCounts
= sample_counts
,
1364 .maxColorAttachments
= MAX_RTS
,
1365 .sampledImageColorSampleCounts
= sample_counts
,
1366 .sampledImageIntegerSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
1367 .sampledImageDepthSampleCounts
= sample_counts
,
1368 .sampledImageStencilSampleCounts
= sample_counts
,
1369 .storageImageSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
1370 .maxSampleMaskWords
= 1,
1371 .timestampComputeAndGraphics
= true,
1372 .timestampPeriod
= 1000000000.0 / devinfo
->timestamp_frequency
,
1373 .maxClipDistances
= 8,
1374 .maxCullDistances
= 8,
1375 .maxCombinedClipAndCullDistances
= 8,
1376 .discreteQueuePriorities
= 2,
1377 .pointSizeRange
= { 0.125, 255.875 },
1380 (devinfo
->gen
>= 9 || devinfo
->is_cherryview
) ?
1381 2047.9921875 : 7.9921875,
1383 .pointSizeGranularity
= (1.0 / 8.0),
1384 .lineWidthGranularity
= (1.0 / 128.0),
1385 .strictLines
= false,
1386 .standardSampleLocations
= true,
1387 .optimalBufferCopyOffsetAlignment
= 128,
1388 .optimalBufferCopyRowPitchAlignment
= 128,
1389 .nonCoherentAtomSize
= 64,
1392 *pProperties
= (VkPhysicalDeviceProperties
) {
1393 .apiVersion
= anv_physical_device_api_version(pdevice
),
1394 .driverVersion
= vk_get_driver_version(),
1396 .deviceID
= pdevice
->chipset_id
,
1397 .deviceType
= VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
,
1399 .sparseProperties
= {0}, /* Broadwell doesn't do sparse. */
1402 snprintf(pProperties
->deviceName
, sizeof(pProperties
->deviceName
),
1403 "%s", pdevice
->name
);
1404 memcpy(pProperties
->pipelineCacheUUID
,
1405 pdevice
->pipeline_cache_uuid
, VK_UUID_SIZE
);
1408 void anv_GetPhysicalDeviceProperties2(
1409 VkPhysicalDevice physicalDevice
,
1410 VkPhysicalDeviceProperties2
* pProperties
)
1412 ANV_FROM_HANDLE(anv_physical_device
, pdevice
, physicalDevice
);
1414 anv_GetPhysicalDeviceProperties(physicalDevice
, &pProperties
->properties
);
1416 vk_foreach_struct(ext
, pProperties
->pNext
) {
1417 switch (ext
->sType
) {
1418 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_STENCIL_RESOLVE_PROPERTIES_KHR
: {
1419 VkPhysicalDeviceDepthStencilResolvePropertiesKHR
*props
=
1420 (VkPhysicalDeviceDepthStencilResolvePropertiesKHR
*)ext
;
1422 /* We support all of the depth resolve modes */
1423 props
->supportedDepthResolveModes
=
1424 VK_RESOLVE_MODE_SAMPLE_ZERO_BIT_KHR
|
1425 VK_RESOLVE_MODE_AVERAGE_BIT_KHR
|
1426 VK_RESOLVE_MODE_MIN_BIT_KHR
|
1427 VK_RESOLVE_MODE_MAX_BIT_KHR
;
1429 /* Average doesn't make sense for stencil so we don't support that */
1430 props
->supportedStencilResolveModes
=
1431 VK_RESOLVE_MODE_SAMPLE_ZERO_BIT_KHR
;
1432 if (pdevice
->info
.gen
>= 8) {
1433 /* The advanced stencil resolve modes currently require stencil
1434 * sampling be supported by the hardware.
1436 props
->supportedStencilResolveModes
|=
1437 VK_RESOLVE_MODE_MIN_BIT_KHR
|
1438 VK_RESOLVE_MODE_MAX_BIT_KHR
;
1441 props
->independentResolveNone
= true;
1442 props
->independentResolve
= true;
1446 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_PROPERTIES_EXT
: {
1447 VkPhysicalDeviceDescriptorIndexingPropertiesEXT
*props
=
1448 (VkPhysicalDeviceDescriptorIndexingPropertiesEXT
*)ext
;
1450 /* It's a bit hard to exactly map our implementation to the limits
1451 * described here. The bindless surface handle in the extended
1452 * message descriptors is 20 bits and it's an index into the table of
1453 * RENDER_SURFACE_STATE structs that starts at bindless surface base
1454 * address. Given that most things consume two surface states per
1455 * view (general/sampled for textures and write-only/read-write for
1456 * images), we claim 2^19 things.
1458 * For SSBOs, we just use A64 messages so there is no real limit
1459 * there beyond the limit on the total size of a descriptor set.
1461 const unsigned max_bindless_views
= 1 << 19;
1463 props
->maxUpdateAfterBindDescriptorsInAllPools
= max_bindless_views
;
1464 props
->shaderUniformBufferArrayNonUniformIndexingNative
= false;
1465 props
->shaderSampledImageArrayNonUniformIndexingNative
= false;
1466 props
->shaderStorageBufferArrayNonUniformIndexingNative
= true;
1467 props
->shaderStorageImageArrayNonUniformIndexingNative
= false;
1468 props
->shaderInputAttachmentArrayNonUniformIndexingNative
= false;
1469 props
->robustBufferAccessUpdateAfterBind
= true;
1470 props
->quadDivergentImplicitLod
= false;
1471 props
->maxPerStageDescriptorUpdateAfterBindSamplers
= max_bindless_views
;
1472 props
->maxPerStageDescriptorUpdateAfterBindUniformBuffers
= MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS
;
1473 props
->maxPerStageDescriptorUpdateAfterBindStorageBuffers
= UINT32_MAX
;
1474 props
->maxPerStageDescriptorUpdateAfterBindSampledImages
= max_bindless_views
;
1475 props
->maxPerStageDescriptorUpdateAfterBindStorageImages
= max_bindless_views
;
1476 props
->maxPerStageDescriptorUpdateAfterBindInputAttachments
= MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS
;
1477 props
->maxPerStageUpdateAfterBindResources
= UINT32_MAX
;
1478 props
->maxDescriptorSetUpdateAfterBindSamplers
= max_bindless_views
;
1479 props
->maxDescriptorSetUpdateAfterBindUniformBuffers
= 6 * MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS
;
1480 props
->maxDescriptorSetUpdateAfterBindUniformBuffersDynamic
= MAX_DYNAMIC_BUFFERS
/ 2;
1481 props
->maxDescriptorSetUpdateAfterBindStorageBuffers
= UINT32_MAX
;
1482 props
->maxDescriptorSetUpdateAfterBindStorageBuffersDynamic
= MAX_DYNAMIC_BUFFERS
/ 2;
1483 props
->maxDescriptorSetUpdateAfterBindSampledImages
= max_bindless_views
;
1484 props
->maxDescriptorSetUpdateAfterBindStorageImages
= max_bindless_views
;
1485 props
->maxDescriptorSetUpdateAfterBindInputAttachments
= MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS
;
1489 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES_KHR
: {
1490 VkPhysicalDeviceDriverPropertiesKHR
*driver_props
=
1491 (VkPhysicalDeviceDriverPropertiesKHR
*) ext
;
1493 driver_props
->driverID
= VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA_KHR
;
1494 snprintf(driver_props
->driverName
, VK_MAX_DRIVER_NAME_SIZE_KHR
,
1495 "Intel open-source Mesa driver");
1497 snprintf(driver_props
->driverInfo
, VK_MAX_DRIVER_INFO_SIZE_KHR
,
1498 "Mesa " PACKAGE_VERSION MESA_GIT_SHA1
);
1500 driver_props
->conformanceVersion
= (VkConformanceVersionKHR
) {
1509 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_MEMORY_HOST_PROPERTIES_EXT
: {
1510 VkPhysicalDeviceExternalMemoryHostPropertiesEXT
*props
=
1511 (VkPhysicalDeviceExternalMemoryHostPropertiesEXT
*) ext
;
1512 /* Userptr needs page aligned memory. */
1513 props
->minImportedHostPointerAlignment
= 4096;
1517 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES
: {
1518 VkPhysicalDeviceIDProperties
*id_props
=
1519 (VkPhysicalDeviceIDProperties
*)ext
;
1520 memcpy(id_props
->deviceUUID
, pdevice
->device_uuid
, VK_UUID_SIZE
);
1521 memcpy(id_props
->driverUUID
, pdevice
->driver_uuid
, VK_UUID_SIZE
);
1522 /* The LUID is for Windows. */
1523 id_props
->deviceLUIDValid
= false;
1527 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INLINE_UNIFORM_BLOCK_PROPERTIES_EXT
: {
1528 VkPhysicalDeviceInlineUniformBlockPropertiesEXT
*props
=
1529 (VkPhysicalDeviceInlineUniformBlockPropertiesEXT
*)ext
;
1530 props
->maxInlineUniformBlockSize
= MAX_INLINE_UNIFORM_BLOCK_SIZE
;
1531 props
->maxPerStageDescriptorInlineUniformBlocks
=
1532 MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS
;
1533 props
->maxPerStageDescriptorUpdateAfterBindInlineUniformBlocks
=
1534 MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS
;
1535 props
->maxDescriptorSetInlineUniformBlocks
=
1536 MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS
;
1537 props
->maxDescriptorSetUpdateAfterBindInlineUniformBlocks
=
1538 MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS
;
1542 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_LINE_RASTERIZATION_PROPERTIES_EXT
: {
1543 VkPhysicalDeviceLineRasterizationPropertiesEXT
*props
=
1544 (VkPhysicalDeviceLineRasterizationPropertiesEXT
*)ext
;
1545 /* In the Skylake PRM Vol. 7, subsection titled "GIQ (Diamond)
1546 * Sampling Rules - Legacy Mode", it says the following:
1548 * "Note that the device divides a pixel into a 16x16 array of
1549 * subpixels, referenced by their upper left corners."
1551 * This is the only known reference in the PRMs to the subpixel
1552 * precision of line rasterization and a "16x16 array of subpixels"
1553 * implies 4 subpixel precision bits. Empirical testing has shown
1554 * that 4 subpixel precision bits applies to all line rasterization
1557 props
->lineSubPixelPrecisionBits
= 4;
1561 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES
: {
1562 VkPhysicalDeviceMaintenance3Properties
*props
=
1563 (VkPhysicalDeviceMaintenance3Properties
*)ext
;
1564 /* This value doesn't matter for us today as our per-stage
1565 * descriptors are the real limit.
1567 props
->maxPerSetDescriptors
= 1024;
1568 props
->maxMemoryAllocationSize
= MAX_MEMORY_ALLOCATION_SIZE
;
1572 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_PROPERTIES
: {
1573 VkPhysicalDeviceMultiviewProperties
*properties
=
1574 (VkPhysicalDeviceMultiviewProperties
*)ext
;
1575 properties
->maxMultiviewViewCount
= 16;
1576 properties
->maxMultiviewInstanceIndex
= UINT32_MAX
/ 16;
1580 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PCI_BUS_INFO_PROPERTIES_EXT
: {
1581 VkPhysicalDevicePCIBusInfoPropertiesEXT
*properties
=
1582 (VkPhysicalDevicePCIBusInfoPropertiesEXT
*)ext
;
1583 properties
->pciDomain
= pdevice
->pci_info
.domain
;
1584 properties
->pciBus
= pdevice
->pci_info
.bus
;
1585 properties
->pciDevice
= pdevice
->pci_info
.device
;
1586 properties
->pciFunction
= pdevice
->pci_info
.function
;
1590 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_POINT_CLIPPING_PROPERTIES
: {
1591 VkPhysicalDevicePointClippingProperties
*properties
=
1592 (VkPhysicalDevicePointClippingProperties
*) ext
;
1593 properties
->pointClippingBehavior
= VK_POINT_CLIPPING_BEHAVIOR_USER_CLIP_PLANES_ONLY
;
1597 #pragma GCC diagnostic push
1598 #pragma GCC diagnostic ignored "-Wswitch"
1599 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRESENTATION_PROPERTIES_ANDROID
: {
1600 VkPhysicalDevicePresentationPropertiesANDROID
*props
=
1601 (VkPhysicalDevicePresentationPropertiesANDROID
*)ext
;
1602 props
->sharedImage
= VK_FALSE
;
1605 #pragma GCC diagnostic pop
1607 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_PROPERTIES
: {
1608 VkPhysicalDeviceProtectedMemoryProperties
*props
=
1609 (VkPhysicalDeviceProtectedMemoryProperties
*)ext
;
1610 props
->protectedNoFault
= false;
1614 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR
: {
1615 VkPhysicalDevicePushDescriptorPropertiesKHR
*properties
=
1616 (VkPhysicalDevicePushDescriptorPropertiesKHR
*) ext
;
1618 properties
->maxPushDescriptors
= MAX_PUSH_DESCRIPTORS
;
1622 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_FILTER_MINMAX_PROPERTIES_EXT
: {
1623 VkPhysicalDeviceSamplerFilterMinmaxPropertiesEXT
*properties
=
1624 (VkPhysicalDeviceSamplerFilterMinmaxPropertiesEXT
*)ext
;
1625 properties
->filterMinmaxImageComponentMapping
= pdevice
->info
.gen
>= 9;
1626 properties
->filterMinmaxSingleComponentFormats
= true;
1630 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES
: {
1631 VkPhysicalDeviceSubgroupProperties
*properties
= (void *)ext
;
1633 properties
->subgroupSize
= BRW_SUBGROUP_SIZE
;
1635 VkShaderStageFlags scalar_stages
= 0;
1636 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
1637 if (pdevice
->compiler
->scalar_stage
[stage
])
1638 scalar_stages
|= mesa_to_vk_shader_stage(stage
);
1640 properties
->supportedStages
= scalar_stages
;
1642 properties
->supportedOperations
= VK_SUBGROUP_FEATURE_BASIC_BIT
|
1643 VK_SUBGROUP_FEATURE_VOTE_BIT
|
1644 VK_SUBGROUP_FEATURE_BALLOT_BIT
|
1645 VK_SUBGROUP_FEATURE_SHUFFLE_BIT
|
1646 VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT
|
1647 VK_SUBGROUP_FEATURE_QUAD_BIT
;
1648 if (pdevice
->info
.gen
>= 8) {
1649 /* TODO: There's no technical reason why these can't be made to
1650 * work on gen7 but they don't at the moment so it's best to leave
1651 * the feature disabled than enabled and broken.
1653 properties
->supportedOperations
|=
1654 VK_SUBGROUP_FEATURE_ARITHMETIC_BIT
|
1655 VK_SUBGROUP_FEATURE_CLUSTERED_BIT
;
1657 properties
->quadOperationsInAllStages
= pdevice
->info
.gen
>= 8;
1661 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_PROPERTIES_EXT
: {
1662 VkPhysicalDeviceSubgroupSizeControlPropertiesEXT
*props
=
1663 (VkPhysicalDeviceSubgroupSizeControlPropertiesEXT
*)ext
;
1664 STATIC_ASSERT(8 <= BRW_SUBGROUP_SIZE
&& BRW_SUBGROUP_SIZE
<= 32);
1665 props
->minSubgroupSize
= 8;
1666 props
->maxSubgroupSize
= 32;
1667 props
->maxComputeWorkgroupSubgroups
= pdevice
->info
.max_cs_threads
;
1668 props
->requiredSubgroupSizeStages
= VK_SHADER_STAGE_COMPUTE_BIT
;
1672 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TEXEL_BUFFER_ALIGNMENT_PROPERTIES_EXT
: {
1673 VkPhysicalDeviceTexelBufferAlignmentPropertiesEXT
*props
=
1674 (VkPhysicalDeviceTexelBufferAlignmentPropertiesEXT
*)ext
;
1676 /* From the SKL PRM Vol. 2d, docs for RENDER_SURFACE_STATE::Surface
1679 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field
1680 * specifies the base address of the first element of the surface,
1681 * computed in software by adding the surface base address to the
1682 * byte offset of the element in the buffer. The base address must
1683 * be aligned to element size."
1685 * The typed dataport messages require that things be texel aligned.
1686 * Otherwise, we may just load/store the wrong data or, in the worst
1687 * case, there may be hangs.
1689 props
->storageTexelBufferOffsetAlignmentBytes
= 16;
1690 props
->storageTexelBufferOffsetSingleTexelAlignment
= true;
1692 /* The sampler, however, is much more forgiving and it can handle
1693 * arbitrary byte alignment for linear and buffer surfaces. It's
1694 * hard to find a good PRM citation for this but years of empirical
1695 * experience demonstrate that this is true.
1697 props
->uniformTexelBufferOffsetAlignmentBytes
= 1;
1698 props
->uniformTexelBufferOffsetSingleTexelAlignment
= false;
1702 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT
: {
1703 VkPhysicalDeviceTransformFeedbackPropertiesEXT
*props
=
1704 (VkPhysicalDeviceTransformFeedbackPropertiesEXT
*)ext
;
1706 props
->maxTransformFeedbackStreams
= MAX_XFB_STREAMS
;
1707 props
->maxTransformFeedbackBuffers
= MAX_XFB_BUFFERS
;
1708 props
->maxTransformFeedbackBufferSize
= (1ull << 32);
1709 props
->maxTransformFeedbackStreamDataSize
= 128 * 4;
1710 props
->maxTransformFeedbackBufferDataSize
= 128 * 4;
1711 props
->maxTransformFeedbackBufferDataStride
= 2048;
1712 props
->transformFeedbackQueries
= true;
1713 props
->transformFeedbackStreamsLinesTriangles
= false;
1714 props
->transformFeedbackRasterizationStreamSelect
= false;
1715 props
->transformFeedbackDraw
= true;
1719 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT
: {
1720 VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT
*props
=
1721 (VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT
*)ext
;
1722 /* We have to restrict this a bit for multiview */
1723 props
->maxVertexAttribDivisor
= UINT32_MAX
/ 16;
1728 anv_debug_ignored_stype(ext
->sType
);
1734 /* We support exactly one queue family. */
1735 static const VkQueueFamilyProperties
1736 anv_queue_family_properties
= {
1737 .queueFlags
= VK_QUEUE_GRAPHICS_BIT
|
1738 VK_QUEUE_COMPUTE_BIT
|
1739 VK_QUEUE_TRANSFER_BIT
,
1741 .timestampValidBits
= 36, /* XXX: Real value here */
1742 .minImageTransferGranularity
= { 1, 1, 1 },
1745 void anv_GetPhysicalDeviceQueueFamilyProperties(
1746 VkPhysicalDevice physicalDevice
,
1748 VkQueueFamilyProperties
* pQueueFamilyProperties
)
1750 VK_OUTARRAY_MAKE(out
, pQueueFamilyProperties
, pCount
);
1752 vk_outarray_append(&out
, p
) {
1753 *p
= anv_queue_family_properties
;
1757 void anv_GetPhysicalDeviceQueueFamilyProperties2(
1758 VkPhysicalDevice physicalDevice
,
1759 uint32_t* pQueueFamilyPropertyCount
,
1760 VkQueueFamilyProperties2
* pQueueFamilyProperties
)
1763 VK_OUTARRAY_MAKE(out
, pQueueFamilyProperties
, pQueueFamilyPropertyCount
);
1765 vk_outarray_append(&out
, p
) {
1766 p
->queueFamilyProperties
= anv_queue_family_properties
;
1768 vk_foreach_struct(s
, p
->pNext
) {
1769 anv_debug_ignored_stype(s
->sType
);
1774 void anv_GetPhysicalDeviceMemoryProperties(
1775 VkPhysicalDevice physicalDevice
,
1776 VkPhysicalDeviceMemoryProperties
* pMemoryProperties
)
1778 ANV_FROM_HANDLE(anv_physical_device
, physical_device
, physicalDevice
);
1780 pMemoryProperties
->memoryTypeCount
= physical_device
->memory
.type_count
;
1781 for (uint32_t i
= 0; i
< physical_device
->memory
.type_count
; i
++) {
1782 pMemoryProperties
->memoryTypes
[i
] = (VkMemoryType
) {
1783 .propertyFlags
= physical_device
->memory
.types
[i
].propertyFlags
,
1784 .heapIndex
= physical_device
->memory
.types
[i
].heapIndex
,
1788 pMemoryProperties
->memoryHeapCount
= physical_device
->memory
.heap_count
;
1789 for (uint32_t i
= 0; i
< physical_device
->memory
.heap_count
; i
++) {
1790 pMemoryProperties
->memoryHeaps
[i
] = (VkMemoryHeap
) {
1791 .size
= physical_device
->memory
.heaps
[i
].size
,
1792 .flags
= physical_device
->memory
.heaps
[i
].flags
,
1798 anv_get_memory_budget(VkPhysicalDevice physicalDevice
,
1799 VkPhysicalDeviceMemoryBudgetPropertiesEXT
*memoryBudget
)
1801 ANV_FROM_HANDLE(anv_physical_device
, device
, physicalDevice
);
1802 uint64_t sys_available
= get_available_system_memory();
1803 assert(sys_available
> 0);
1805 VkDeviceSize total_heaps_size
= 0;
1806 for (size_t i
= 0; i
< device
->memory
.heap_count
; i
++)
1807 total_heaps_size
+= device
->memory
.heaps
[i
].size
;
1809 for (size_t i
= 0; i
< device
->memory
.heap_count
; i
++) {
1810 VkDeviceSize heap_size
= device
->memory
.heaps
[i
].size
;
1811 VkDeviceSize heap_used
= device
->memory
.heaps
[i
].used
;
1812 VkDeviceSize heap_budget
;
1814 double heap_proportion
= (double) heap_size
/ total_heaps_size
;
1815 VkDeviceSize sys_available_prop
= sys_available
* heap_proportion
;
1818 * Let's not incite the app to starve the system: report at most 90% of
1819 * available system memory.
1821 uint64_t heap_available
= sys_available_prop
* 9 / 10;
1822 heap_budget
= MIN2(heap_size
, heap_used
+ heap_available
);
1825 * Round down to the nearest MB
1827 heap_budget
&= ~((1ull << 20) - 1);
1830 * The heapBudget value must be non-zero for array elements less than
1831 * VkPhysicalDeviceMemoryProperties::memoryHeapCount. The heapBudget
1832 * value must be less than or equal to VkMemoryHeap::size for each heap.
1834 assert(0 < heap_budget
&& heap_budget
<= heap_size
);
1836 memoryBudget
->heapUsage
[i
] = heap_used
;
1837 memoryBudget
->heapBudget
[i
] = heap_budget
;
1840 /* The heapBudget and heapUsage values must be zero for array elements
1841 * greater than or equal to VkPhysicalDeviceMemoryProperties::memoryHeapCount
1843 for (uint32_t i
= device
->memory
.heap_count
; i
< VK_MAX_MEMORY_HEAPS
; i
++) {
1844 memoryBudget
->heapBudget
[i
] = 0;
1845 memoryBudget
->heapUsage
[i
] = 0;
1849 void anv_GetPhysicalDeviceMemoryProperties2(
1850 VkPhysicalDevice physicalDevice
,
1851 VkPhysicalDeviceMemoryProperties2
* pMemoryProperties
)
1853 anv_GetPhysicalDeviceMemoryProperties(physicalDevice
,
1854 &pMemoryProperties
->memoryProperties
);
1856 vk_foreach_struct(ext
, pMemoryProperties
->pNext
) {
1857 switch (ext
->sType
) {
1858 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_BUDGET_PROPERTIES_EXT
:
1859 anv_get_memory_budget(physicalDevice
, (void*)ext
);
1862 anv_debug_ignored_stype(ext
->sType
);
1869 anv_GetDeviceGroupPeerMemoryFeatures(
1872 uint32_t localDeviceIndex
,
1873 uint32_t remoteDeviceIndex
,
1874 VkPeerMemoryFeatureFlags
* pPeerMemoryFeatures
)
1876 assert(localDeviceIndex
== 0 && remoteDeviceIndex
== 0);
1877 *pPeerMemoryFeatures
= VK_PEER_MEMORY_FEATURE_COPY_SRC_BIT
|
1878 VK_PEER_MEMORY_FEATURE_COPY_DST_BIT
|
1879 VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT
|
1880 VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT
;
1883 PFN_vkVoidFunction
anv_GetInstanceProcAddr(
1884 VkInstance _instance
,
1887 ANV_FROM_HANDLE(anv_instance
, instance
, _instance
);
1889 /* The Vulkan 1.0 spec for vkGetInstanceProcAddr has a table of exactly
1890 * when we have to return valid function pointers, NULL, or it's left
1891 * undefined. See the table for exact details.
1896 #define LOOKUP_ANV_ENTRYPOINT(entrypoint) \
1897 if (strcmp(pName, "vk" #entrypoint) == 0) \
1898 return (PFN_vkVoidFunction)anv_##entrypoint
1900 LOOKUP_ANV_ENTRYPOINT(EnumerateInstanceExtensionProperties
);
1901 LOOKUP_ANV_ENTRYPOINT(EnumerateInstanceLayerProperties
);
1902 LOOKUP_ANV_ENTRYPOINT(EnumerateInstanceVersion
);
1903 LOOKUP_ANV_ENTRYPOINT(CreateInstance
);
1905 #undef LOOKUP_ANV_ENTRYPOINT
1907 if (instance
== NULL
)
1910 int idx
= anv_get_instance_entrypoint_index(pName
);
1912 return instance
->dispatch
.entrypoints
[idx
];
1914 idx
= anv_get_device_entrypoint_index(pName
);
1916 return instance
->device_dispatch
.entrypoints
[idx
];
1921 /* With version 1+ of the loader interface the ICD should expose
1922 * vk_icdGetInstanceProcAddr to work around certain LD_PRELOAD issues seen in apps.
1925 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
1926 VkInstance instance
,
1930 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
1931 VkInstance instance
,
1934 return anv_GetInstanceProcAddr(instance
, pName
);
1937 PFN_vkVoidFunction
anv_GetDeviceProcAddr(
1941 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1943 if (!device
|| !pName
)
1946 int idx
= anv_get_device_entrypoint_index(pName
);
1950 return device
->dispatch
.entrypoints
[idx
];
1954 anv_CreateDebugReportCallbackEXT(VkInstance _instance
,
1955 const VkDebugReportCallbackCreateInfoEXT
* pCreateInfo
,
1956 const VkAllocationCallbacks
* pAllocator
,
1957 VkDebugReportCallbackEXT
* pCallback
)
1959 ANV_FROM_HANDLE(anv_instance
, instance
, _instance
);
1960 return vk_create_debug_report_callback(&instance
->debug_report_callbacks
,
1961 pCreateInfo
, pAllocator
, &instance
->alloc
,
1966 anv_DestroyDebugReportCallbackEXT(VkInstance _instance
,
1967 VkDebugReportCallbackEXT _callback
,
1968 const VkAllocationCallbacks
* pAllocator
)
1970 ANV_FROM_HANDLE(anv_instance
, instance
, _instance
);
1971 vk_destroy_debug_report_callback(&instance
->debug_report_callbacks
,
1972 _callback
, pAllocator
, &instance
->alloc
);
1976 anv_DebugReportMessageEXT(VkInstance _instance
,
1977 VkDebugReportFlagsEXT flags
,
1978 VkDebugReportObjectTypeEXT objectType
,
1981 int32_t messageCode
,
1982 const char* pLayerPrefix
,
1983 const char* pMessage
)
1985 ANV_FROM_HANDLE(anv_instance
, instance
, _instance
);
1986 vk_debug_report(&instance
->debug_report_callbacks
, flags
, objectType
,
1987 object
, location
, messageCode
, pLayerPrefix
, pMessage
);
1991 anv_queue_init(struct anv_device
*device
, struct anv_queue
*queue
)
1993 queue
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
1994 queue
->device
= device
;
1999 anv_queue_finish(struct anv_queue
*queue
)
2003 static struct anv_state
2004 anv_state_pool_emit_data(struct anv_state_pool
*pool
, size_t size
, size_t align
, const void *p
)
2006 struct anv_state state
;
2008 state
= anv_state_pool_alloc(pool
, size
, align
);
2009 memcpy(state
.map
, p
, size
);
2014 /* Haswell border color is a bit of a disaster. Float and unorm formats use a
2015 * straightforward 32-bit float color in the first 64 bytes. Instead of using
2016 * a nice float/integer union like Gen8+, Haswell specifies the integer border
2017 * color as a separate entry /after/ the float color. The layout of this entry
2018 * also depends on the format's bpp (with extra hacks for RG32), and overlaps.
2020 * Since we don't know the format/bpp, we can't make any of the border colors
2021 * containing '1' work for all formats, as it would be in the wrong place for
2022 * some of them. We opt to make 32-bit integers work as this seems like the
2023 * most common option. Fortunately, transparent black works regardless, as
2024 * all zeroes is the same in every bit-size.
2026 struct hsw_border_color
{
2030 uint32_t _pad1
[108];
2033 struct gen8_border_color
{
2038 /* Pad out to 64 bytes */
2043 anv_device_init_border_colors(struct anv_device
*device
)
2045 if (device
->info
.is_haswell
) {
2046 static const struct hsw_border_color border_colors
[] = {
2047 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
] = { .float32
= { 0.0, 0.0, 0.0, 0.0 } },
2048 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
] = { .float32
= { 0.0, 0.0, 0.0, 1.0 } },
2049 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
] = { .float32
= { 1.0, 1.0, 1.0, 1.0 } },
2050 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
] = { .uint32
= { 0, 0, 0, 0 } },
2051 [VK_BORDER_COLOR_INT_OPAQUE_BLACK
] = { .uint32
= { 0, 0, 0, 1 } },
2052 [VK_BORDER_COLOR_INT_OPAQUE_WHITE
] = { .uint32
= { 1, 1, 1, 1 } },
2055 device
->border_colors
=
2056 anv_state_pool_emit_data(&device
->dynamic_state_pool
,
2057 sizeof(border_colors
), 512, border_colors
);
2059 static const struct gen8_border_color border_colors
[] = {
2060 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
] = { .float32
= { 0.0, 0.0, 0.0, 0.0 } },
2061 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
] = { .float32
= { 0.0, 0.0, 0.0, 1.0 } },
2062 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
] = { .float32
= { 1.0, 1.0, 1.0, 1.0 } },
2063 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
] = { .uint32
= { 0, 0, 0, 0 } },
2064 [VK_BORDER_COLOR_INT_OPAQUE_BLACK
] = { .uint32
= { 0, 0, 0, 1 } },
2065 [VK_BORDER_COLOR_INT_OPAQUE_WHITE
] = { .uint32
= { 1, 1, 1, 1 } },
2068 device
->border_colors
=
2069 anv_state_pool_emit_data(&device
->dynamic_state_pool
,
2070 sizeof(border_colors
), 64, border_colors
);
2075 anv_device_init_trivial_batch(struct anv_device
*device
)
2077 anv_bo_init_new(&device
->trivial_batch_bo
, device
, 4096);
2079 if (device
->instance
->physicalDevice
.has_exec_async
)
2080 device
->trivial_batch_bo
.flags
|= EXEC_OBJECT_ASYNC
;
2082 if (device
->instance
->physicalDevice
.use_softpin
)
2083 device
->trivial_batch_bo
.flags
|= EXEC_OBJECT_PINNED
;
2085 anv_vma_alloc(device
, &device
->trivial_batch_bo
);
2087 void *map
= anv_gem_mmap(device
, device
->trivial_batch_bo
.gem_handle
,
2090 struct anv_batch batch
= {
2096 anv_batch_emit(&batch
, GEN7_MI_BATCH_BUFFER_END
, bbe
);
2097 anv_batch_emit(&batch
, GEN7_MI_NOOP
, noop
);
2099 if (!device
->info
.has_llc
)
2100 gen_clflush_range(map
, batch
.next
- map
);
2102 anv_gem_munmap(map
, device
->trivial_batch_bo
.size
);
2105 VkResult
anv_EnumerateDeviceExtensionProperties(
2106 VkPhysicalDevice physicalDevice
,
2107 const char* pLayerName
,
2108 uint32_t* pPropertyCount
,
2109 VkExtensionProperties
* pProperties
)
2111 ANV_FROM_HANDLE(anv_physical_device
, device
, physicalDevice
);
2112 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
2114 for (int i
= 0; i
< ANV_DEVICE_EXTENSION_COUNT
; i
++) {
2115 if (device
->supported_extensions
.extensions
[i
]) {
2116 vk_outarray_append(&out
, prop
) {
2117 *prop
= anv_device_extensions
[i
];
2122 return vk_outarray_status(&out
);
2126 anv_device_init_dispatch(struct anv_device
*device
)
2128 const struct anv_device_dispatch_table
*genX_table
;
2129 switch (device
->info
.gen
) {
2131 genX_table
= &gen12_device_dispatch_table
;
2134 genX_table
= &gen11_device_dispatch_table
;
2137 genX_table
= &gen10_device_dispatch_table
;
2140 genX_table
= &gen9_device_dispatch_table
;
2143 genX_table
= &gen8_device_dispatch_table
;
2146 if (device
->info
.is_haswell
)
2147 genX_table
= &gen75_device_dispatch_table
;
2149 genX_table
= &gen7_device_dispatch_table
;
2152 unreachable("unsupported gen\n");
2155 for (unsigned i
= 0; i
< ARRAY_SIZE(device
->dispatch
.entrypoints
); i
++) {
2156 /* Vulkan requires that entrypoints for extensions which have not been
2157 * enabled must not be advertised.
2159 if (!anv_device_entrypoint_is_enabled(i
, device
->instance
->app_info
.api_version
,
2160 &device
->instance
->enabled_extensions
,
2161 &device
->enabled_extensions
)) {
2162 device
->dispatch
.entrypoints
[i
] = NULL
;
2163 } else if (genX_table
->entrypoints
[i
]) {
2164 device
->dispatch
.entrypoints
[i
] = genX_table
->entrypoints
[i
];
2166 device
->dispatch
.entrypoints
[i
] =
2167 anv_device_dispatch_table
.entrypoints
[i
];
2173 vk_priority_to_gen(int priority
)
2176 case VK_QUEUE_GLOBAL_PRIORITY_LOW_EXT
:
2177 return GEN_CONTEXT_LOW_PRIORITY
;
2178 case VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_EXT
:
2179 return GEN_CONTEXT_MEDIUM_PRIORITY
;
2180 case VK_QUEUE_GLOBAL_PRIORITY_HIGH_EXT
:
2181 return GEN_CONTEXT_HIGH_PRIORITY
;
2182 case VK_QUEUE_GLOBAL_PRIORITY_REALTIME_EXT
:
2183 return GEN_CONTEXT_REALTIME_PRIORITY
;
2185 unreachable("Invalid priority");
2190 anv_device_init_hiz_clear_value_bo(struct anv_device
*device
)
2192 anv_bo_init_new(&device
->hiz_clear_bo
, device
, 4096);
2194 if (device
->instance
->physicalDevice
.has_exec_async
)
2195 device
->hiz_clear_bo
.flags
|= EXEC_OBJECT_ASYNC
;
2197 if (device
->instance
->physicalDevice
.use_softpin
)
2198 device
->hiz_clear_bo
.flags
|= EXEC_OBJECT_PINNED
;
2200 anv_vma_alloc(device
, &device
->hiz_clear_bo
);
2202 uint32_t *map
= anv_gem_mmap(device
, device
->hiz_clear_bo
.gem_handle
,
2205 union isl_color_value hiz_clear
= { .u32
= { 0, } };
2206 hiz_clear
.f32
[0] = ANV_HZ_FC_VAL
;
2208 memcpy(map
, hiz_clear
.u32
, sizeof(hiz_clear
.u32
));
2209 anv_gem_munmap(map
, device
->hiz_clear_bo
.size
);
2213 get_bo_from_pool(struct gen_batch_decode_bo
*ret
,
2214 struct anv_block_pool
*pool
,
2217 for (uint32_t i
= 0; i
< pool
->nbos
; i
++) {
2218 uint64_t bo_address
= pool
->bos
[i
].offset
& (~0ull >> 16);
2219 uint32_t bo_size
= pool
->bos
[i
].size
;
2220 if (address
>= bo_address
&& address
< (bo_address
+ bo_size
)) {
2221 *ret
= (struct gen_batch_decode_bo
) {
2224 .map
= pool
->bos
[i
].map
,
2232 /* Finding a buffer for batch decoding */
2233 static struct gen_batch_decode_bo
2234 decode_get_bo(void *v_batch
, bool ppgtt
, uint64_t address
)
2236 struct anv_device
*device
= v_batch
;
2237 struct gen_batch_decode_bo ret_bo
= {};
2241 if (get_bo_from_pool(&ret_bo
, &device
->dynamic_state_pool
.block_pool
, address
))
2243 if (get_bo_from_pool(&ret_bo
, &device
->instruction_state_pool
.block_pool
, address
))
2245 if (get_bo_from_pool(&ret_bo
, &device
->binding_table_pool
.block_pool
, address
))
2247 if (get_bo_from_pool(&ret_bo
, &device
->surface_state_pool
.block_pool
, address
))
2250 if (!device
->cmd_buffer_being_decoded
)
2251 return (struct gen_batch_decode_bo
) { };
2253 struct anv_batch_bo
**bo
;
2255 u_vector_foreach(bo
, &device
->cmd_buffer_being_decoded
->seen_bbos
) {
2256 /* The decoder zeroes out the top 16 bits, so we need to as well */
2257 uint64_t bo_address
= (*bo
)->bo
.offset
& (~0ull >> 16);
2259 if (address
>= bo_address
&& address
< bo_address
+ (*bo
)->bo
.size
) {
2260 return (struct gen_batch_decode_bo
) {
2262 .size
= (*bo
)->bo
.size
,
2263 .map
= (*bo
)->bo
.map
,
2268 return (struct gen_batch_decode_bo
) { };
2271 VkResult
anv_CreateDevice(
2272 VkPhysicalDevice physicalDevice
,
2273 const VkDeviceCreateInfo
* pCreateInfo
,
2274 const VkAllocationCallbacks
* pAllocator
,
2277 ANV_FROM_HANDLE(anv_physical_device
, physical_device
, physicalDevice
);
2279 struct anv_device
*device
;
2281 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO
);
2283 struct anv_device_extension_table enabled_extensions
= { };
2284 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
2286 for (idx
= 0; idx
< ANV_DEVICE_EXTENSION_COUNT
; idx
++) {
2287 if (strcmp(pCreateInfo
->ppEnabledExtensionNames
[i
],
2288 anv_device_extensions
[idx
].extensionName
) == 0)
2292 if (idx
>= ANV_DEVICE_EXTENSION_COUNT
)
2293 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
2295 if (!physical_device
->supported_extensions
.extensions
[idx
])
2296 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
2298 enabled_extensions
.extensions
[idx
] = true;
2301 /* Check enabled features */
2302 if (pCreateInfo
->pEnabledFeatures
) {
2303 VkPhysicalDeviceFeatures supported_features
;
2304 anv_GetPhysicalDeviceFeatures(physicalDevice
, &supported_features
);
2305 VkBool32
*supported_feature
= (VkBool32
*)&supported_features
;
2306 VkBool32
*enabled_feature
= (VkBool32
*)pCreateInfo
->pEnabledFeatures
;
2307 unsigned num_features
= sizeof(VkPhysicalDeviceFeatures
) / sizeof(VkBool32
);
2308 for (uint32_t i
= 0; i
< num_features
; i
++) {
2309 if (enabled_feature
[i
] && !supported_feature
[i
])
2310 return vk_error(VK_ERROR_FEATURE_NOT_PRESENT
);
2314 /* Check requested queues and fail if we are requested to create any
2315 * queues with flags we don't support.
2317 assert(pCreateInfo
->queueCreateInfoCount
> 0);
2318 for (uint32_t i
= 0; i
< pCreateInfo
->queueCreateInfoCount
; i
++) {
2319 if (pCreateInfo
->pQueueCreateInfos
[i
].flags
!= 0)
2320 return vk_error(VK_ERROR_INITIALIZATION_FAILED
);
2323 /* Check if client specified queue priority. */
2324 const VkDeviceQueueGlobalPriorityCreateInfoEXT
*queue_priority
=
2325 vk_find_struct_const(pCreateInfo
->pQueueCreateInfos
[0].pNext
,
2326 DEVICE_QUEUE_GLOBAL_PRIORITY_CREATE_INFO_EXT
);
2328 VkQueueGlobalPriorityEXT priority
=
2329 queue_priority
? queue_priority
->globalPriority
:
2330 VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_EXT
;
2332 device
= vk_alloc2(&physical_device
->instance
->alloc
, pAllocator
,
2334 VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
2336 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
2338 if (INTEL_DEBUG
& DEBUG_BATCH
) {
2339 const unsigned decode_flags
=
2340 GEN_BATCH_DECODE_FULL
|
2341 ((INTEL_DEBUG
& DEBUG_COLOR
) ? GEN_BATCH_DECODE_IN_COLOR
: 0) |
2342 GEN_BATCH_DECODE_OFFSETS
|
2343 GEN_BATCH_DECODE_FLOATS
;
2345 gen_batch_decode_ctx_init(&device
->decoder_ctx
,
2346 &physical_device
->info
,
2347 stderr
, decode_flags
, NULL
,
2348 decode_get_bo
, NULL
, device
);
2351 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
2352 device
->instance
= physical_device
->instance
;
2353 device
->chipset_id
= physical_device
->chipset_id
;
2354 device
->no_hw
= physical_device
->no_hw
;
2355 device
->_lost
= false;
2358 device
->alloc
= *pAllocator
;
2360 device
->alloc
= physical_device
->instance
->alloc
;
2362 /* XXX(chadv): Can we dup() physicalDevice->fd here? */
2363 device
->fd
= open(physical_device
->path
, O_RDWR
| O_CLOEXEC
);
2364 if (device
->fd
== -1) {
2365 result
= vk_error(VK_ERROR_INITIALIZATION_FAILED
);
2369 device
->context_id
= anv_gem_create_context(device
);
2370 if (device
->context_id
== -1) {
2371 result
= vk_error(VK_ERROR_INITIALIZATION_FAILED
);
2375 if (physical_device
->use_softpin
) {
2376 if (pthread_mutex_init(&device
->vma_mutex
, NULL
) != 0) {
2377 result
= vk_error(VK_ERROR_INITIALIZATION_FAILED
);
2381 /* keep the page with address zero out of the allocator */
2382 struct anv_memory_heap
*low_heap
=
2383 &physical_device
->memory
.heaps
[physical_device
->memory
.heap_count
- 1];
2384 util_vma_heap_init(&device
->vma_lo
, low_heap
->vma_start
, low_heap
->vma_size
);
2385 device
->vma_lo_available
= low_heap
->size
;
2387 struct anv_memory_heap
*high_heap
=
2388 &physical_device
->memory
.heaps
[0];
2389 util_vma_heap_init(&device
->vma_hi
, high_heap
->vma_start
, high_heap
->vma_size
);
2390 device
->vma_hi_available
= physical_device
->memory
.heap_count
== 1 ? 0 :
2394 list_inithead(&device
->memory_objects
);
2396 /* As per spec, the driver implementation may deny requests to acquire
2397 * a priority above the default priority (MEDIUM) if the caller does not
2398 * have sufficient privileges. In this scenario VK_ERROR_NOT_PERMITTED_EXT
2401 if (physical_device
->has_context_priority
) {
2402 int err
= anv_gem_set_context_param(device
->fd
, device
->context_id
,
2403 I915_CONTEXT_PARAM_PRIORITY
,
2404 vk_priority_to_gen(priority
));
2405 if (err
!= 0 && priority
> VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_EXT
) {
2406 result
= vk_error(VK_ERROR_NOT_PERMITTED_EXT
);
2411 device
->info
= physical_device
->info
;
2412 device
->isl_dev
= physical_device
->isl_dev
;
2414 /* On Broadwell and later, we can use batch chaining to more efficiently
2415 * implement growing command buffers. Prior to Haswell, the kernel
2416 * command parser gets in the way and we have to fall back to growing
2419 device
->can_chain_batches
= device
->info
.gen
>= 8;
2421 device
->robust_buffer_access
= pCreateInfo
->pEnabledFeatures
&&
2422 pCreateInfo
->pEnabledFeatures
->robustBufferAccess
;
2423 device
->enabled_extensions
= enabled_extensions
;
2425 anv_device_init_dispatch(device
);
2427 if (pthread_mutex_init(&device
->mutex
, NULL
) != 0) {
2428 result
= vk_error(VK_ERROR_INITIALIZATION_FAILED
);
2429 goto fail_context_id
;
2432 pthread_condattr_t condattr
;
2433 if (pthread_condattr_init(&condattr
) != 0) {
2434 result
= vk_error(VK_ERROR_INITIALIZATION_FAILED
);
2437 if (pthread_condattr_setclock(&condattr
, CLOCK_MONOTONIC
) != 0) {
2438 pthread_condattr_destroy(&condattr
);
2439 result
= vk_error(VK_ERROR_INITIALIZATION_FAILED
);
2442 if (pthread_cond_init(&device
->queue_submit
, &condattr
) != 0) {
2443 pthread_condattr_destroy(&condattr
);
2444 result
= vk_error(VK_ERROR_INITIALIZATION_FAILED
);
2447 pthread_condattr_destroy(&condattr
);
2450 (physical_device
->supports_48bit_addresses
? EXEC_OBJECT_SUPPORTS_48B_ADDRESS
: 0) |
2451 (physical_device
->has_exec_async
? EXEC_OBJECT_ASYNC
: 0) |
2452 (physical_device
->has_exec_capture
? EXEC_OBJECT_CAPTURE
: 0) |
2453 (physical_device
->use_softpin
? EXEC_OBJECT_PINNED
: 0);
2455 anv_bo_pool_init(&device
->batch_bo_pool
, device
, bo_flags
);
2457 result
= anv_bo_cache_init(&device
->bo_cache
);
2458 if (result
!= VK_SUCCESS
)
2459 goto fail_batch_bo_pool
;
2461 if (!physical_device
->use_softpin
)
2462 bo_flags
&= ~EXEC_OBJECT_SUPPORTS_48B_ADDRESS
;
2464 result
= anv_state_pool_init(&device
->dynamic_state_pool
, device
,
2465 DYNAMIC_STATE_POOL_MIN_ADDRESS
,
2468 if (result
!= VK_SUCCESS
)
2471 result
= anv_state_pool_init(&device
->instruction_state_pool
, device
,
2472 INSTRUCTION_STATE_POOL_MIN_ADDRESS
,
2475 if (result
!= VK_SUCCESS
)
2476 goto fail_dynamic_state_pool
;
2478 result
= anv_state_pool_init(&device
->surface_state_pool
, device
,
2479 SURFACE_STATE_POOL_MIN_ADDRESS
,
2482 if (result
!= VK_SUCCESS
)
2483 goto fail_instruction_state_pool
;
2485 if (physical_device
->use_softpin
) {
2486 result
= anv_state_pool_init(&device
->binding_table_pool
, device
,
2487 BINDING_TABLE_POOL_MIN_ADDRESS
,
2490 if (result
!= VK_SUCCESS
)
2491 goto fail_surface_state_pool
;
2494 result
= anv_bo_init_new(&device
->workaround_bo
, device
, 4096);
2495 if (result
!= VK_SUCCESS
)
2496 goto fail_binding_table_pool
;
2498 if (physical_device
->use_softpin
)
2499 device
->workaround_bo
.flags
|= EXEC_OBJECT_PINNED
;
2501 if (!anv_vma_alloc(device
, &device
->workaround_bo
))
2502 goto fail_workaround_bo
;
2504 anv_device_init_trivial_batch(device
);
2506 if (device
->info
.gen
>= 10)
2507 anv_device_init_hiz_clear_value_bo(device
);
2509 anv_scratch_pool_init(device
, &device
->scratch_pool
);
2511 anv_queue_init(device
, &device
->queue
);
2513 switch (device
->info
.gen
) {
2515 if (!device
->info
.is_haswell
)
2516 result
= gen7_init_device_state(device
);
2518 result
= gen75_init_device_state(device
);
2521 result
= gen8_init_device_state(device
);
2524 result
= gen9_init_device_state(device
);
2527 result
= gen10_init_device_state(device
);
2530 result
= gen11_init_device_state(device
);
2533 result
= gen12_init_device_state(device
);
2536 /* Shouldn't get here as we don't create physical devices for any other
2538 unreachable("unhandled gen");
2540 if (result
!= VK_SUCCESS
)
2541 goto fail_workaround_bo
;
2543 anv_pipeline_cache_init(&device
->default_pipeline_cache
, device
, true);
2545 anv_device_init_blorp(device
);
2547 anv_device_init_border_colors(device
);
2549 *pDevice
= anv_device_to_handle(device
);
2554 anv_queue_finish(&device
->queue
);
2555 anv_scratch_pool_finish(device
, &device
->scratch_pool
);
2556 anv_gem_munmap(device
->workaround_bo
.map
, device
->workaround_bo
.size
);
2557 anv_gem_close(device
, device
->workaround_bo
.gem_handle
);
2558 fail_binding_table_pool
:
2559 if (physical_device
->use_softpin
)
2560 anv_state_pool_finish(&device
->binding_table_pool
);
2561 fail_surface_state_pool
:
2562 anv_state_pool_finish(&device
->surface_state_pool
);
2563 fail_instruction_state_pool
:
2564 anv_state_pool_finish(&device
->instruction_state_pool
);
2565 fail_dynamic_state_pool
:
2566 anv_state_pool_finish(&device
->dynamic_state_pool
);
2568 anv_bo_cache_finish(&device
->bo_cache
);
2570 anv_bo_pool_finish(&device
->batch_bo_pool
);
2571 pthread_cond_destroy(&device
->queue_submit
);
2573 pthread_mutex_destroy(&device
->mutex
);
2575 anv_gem_destroy_context(device
, device
->context_id
);
2579 vk_free(&device
->alloc
, device
);
2584 void anv_DestroyDevice(
2586 const VkAllocationCallbacks
* pAllocator
)
2588 ANV_FROM_HANDLE(anv_device
, device
, _device
);
2589 struct anv_physical_device
*physical_device
;
2594 physical_device
= &device
->instance
->physicalDevice
;
2596 anv_device_finish_blorp(device
);
2598 anv_pipeline_cache_finish(&device
->default_pipeline_cache
);
2600 anv_queue_finish(&device
->queue
);
2602 #ifdef HAVE_VALGRIND
2603 /* We only need to free these to prevent valgrind errors. The backing
2604 * BO will go away in a couple of lines so we don't actually leak.
2606 anv_state_pool_free(&device
->dynamic_state_pool
, device
->border_colors
);
2607 anv_state_pool_free(&device
->dynamic_state_pool
, device
->slice_hash
);
2610 anv_scratch_pool_finish(device
, &device
->scratch_pool
);
2612 anv_gem_munmap(device
->workaround_bo
.map
, device
->workaround_bo
.size
);
2613 anv_vma_free(device
, &device
->workaround_bo
);
2614 anv_gem_close(device
, device
->workaround_bo
.gem_handle
);
2616 anv_vma_free(device
, &device
->trivial_batch_bo
);
2617 anv_gem_close(device
, device
->trivial_batch_bo
.gem_handle
);
2618 if (device
->info
.gen
>= 10)
2619 anv_gem_close(device
, device
->hiz_clear_bo
.gem_handle
);
2621 if (physical_device
->use_softpin
)
2622 anv_state_pool_finish(&device
->binding_table_pool
);
2623 anv_state_pool_finish(&device
->surface_state_pool
);
2624 anv_state_pool_finish(&device
->instruction_state_pool
);
2625 anv_state_pool_finish(&device
->dynamic_state_pool
);
2627 anv_bo_cache_finish(&device
->bo_cache
);
2629 anv_bo_pool_finish(&device
->batch_bo_pool
);
2631 pthread_cond_destroy(&device
->queue_submit
);
2632 pthread_mutex_destroy(&device
->mutex
);
2634 anv_gem_destroy_context(device
, device
->context_id
);
2636 if (INTEL_DEBUG
& DEBUG_BATCH
)
2637 gen_batch_decode_ctx_finish(&device
->decoder_ctx
);
2641 vk_free(&device
->alloc
, device
);
2644 VkResult
anv_EnumerateInstanceLayerProperties(
2645 uint32_t* pPropertyCount
,
2646 VkLayerProperties
* pProperties
)
2648 if (pProperties
== NULL
) {
2649 *pPropertyCount
= 0;
2653 /* None supported at this time */
2654 return vk_error(VK_ERROR_LAYER_NOT_PRESENT
);
2657 VkResult
anv_EnumerateDeviceLayerProperties(
2658 VkPhysicalDevice physicalDevice
,
2659 uint32_t* pPropertyCount
,
2660 VkLayerProperties
* pProperties
)
2662 if (pProperties
== NULL
) {
2663 *pPropertyCount
= 0;
2667 /* None supported at this time */
2668 return vk_error(VK_ERROR_LAYER_NOT_PRESENT
);
2671 void anv_GetDeviceQueue(
2673 uint32_t queueNodeIndex
,
2674 uint32_t queueIndex
,
2677 ANV_FROM_HANDLE(anv_device
, device
, _device
);
2679 assert(queueIndex
== 0);
2681 *pQueue
= anv_queue_to_handle(&device
->queue
);
2684 void anv_GetDeviceQueue2(
2686 const VkDeviceQueueInfo2
* pQueueInfo
,
2689 ANV_FROM_HANDLE(anv_device
, device
, _device
);
2691 assert(pQueueInfo
->queueIndex
== 0);
2693 if (pQueueInfo
->flags
== device
->queue
.flags
)
2694 *pQueue
= anv_queue_to_handle(&device
->queue
);
2700 _anv_device_set_lost(struct anv_device
*device
,
2701 const char *file
, int line
,
2702 const char *msg
, ...)
2707 device
->_lost
= true;
2710 err
= __vk_errorv(device
->instance
, device
,
2711 VK_DEBUG_REPORT_OBJECT_TYPE_DEVICE_EXT
,
2712 VK_ERROR_DEVICE_LOST
, file
, line
, msg
, ap
);
2715 if (env_var_as_boolean("ANV_ABORT_ON_DEVICE_LOSS", false))
2722 anv_device_query_status(struct anv_device
*device
)
2724 /* This isn't likely as most of the callers of this function already check
2725 * for it. However, it doesn't hurt to check and it potentially lets us
2728 if (anv_device_is_lost(device
))
2729 return VK_ERROR_DEVICE_LOST
;
2731 uint32_t active
, pending
;
2732 int ret
= anv_gem_gpu_get_reset_stats(device
, &active
, &pending
);
2734 /* We don't know the real error. */
2735 return anv_device_set_lost(device
, "get_reset_stats failed: %m");
2739 return anv_device_set_lost(device
, "GPU hung on one of our command buffers");
2740 } else if (pending
) {
2741 return anv_device_set_lost(device
, "GPU hung with commands in-flight");
2748 anv_device_bo_busy(struct anv_device
*device
, struct anv_bo
*bo
)
2750 /* Note: This only returns whether or not the BO is in use by an i915 GPU.
2751 * Other usages of the BO (such as on different hardware) will not be
2752 * flagged as "busy" by this ioctl. Use with care.
2754 int ret
= anv_gem_busy(device
, bo
->gem_handle
);
2756 return VK_NOT_READY
;
2757 } else if (ret
== -1) {
2758 /* We don't know the real error. */
2759 return anv_device_set_lost(device
, "gem wait failed: %m");
2762 /* Query for device status after the busy call. If the BO we're checking
2763 * got caught in a GPU hang we don't want to return VK_SUCCESS to the
2764 * client because it clearly doesn't have valid data. Yes, this most
2765 * likely means an ioctl, but we just did an ioctl to query the busy status
2766 * so it's no great loss.
2768 return anv_device_query_status(device
);
2772 anv_device_wait(struct anv_device
*device
, struct anv_bo
*bo
,
2775 int ret
= anv_gem_wait(device
, bo
->gem_handle
, &timeout
);
2776 if (ret
== -1 && errno
== ETIME
) {
2778 } else if (ret
== -1) {
2779 /* We don't know the real error. */
2780 return anv_device_set_lost(device
, "gem wait failed: %m");
2783 /* Query for device status after the wait. If the BO we're waiting on got
2784 * caught in a GPU hang we don't want to return VK_SUCCESS to the client
2785 * because it clearly doesn't have valid data. Yes, this most likely means
2786 * an ioctl, but we just did an ioctl to wait so it's no great loss.
2788 return anv_device_query_status(device
);
2791 VkResult
anv_DeviceWaitIdle(
2794 ANV_FROM_HANDLE(anv_device
, device
, _device
);
2795 if (anv_device_is_lost(device
))
2796 return VK_ERROR_DEVICE_LOST
;
2798 struct anv_batch batch
;
2801 batch
.start
= batch
.next
= cmds
;
2802 batch
.end
= (void *) cmds
+ sizeof(cmds
);
2804 anv_batch_emit(&batch
, GEN7_MI_BATCH_BUFFER_END
, bbe
);
2805 anv_batch_emit(&batch
, GEN7_MI_NOOP
, noop
);
2807 return anv_device_submit_simple_batch(device
, &batch
);
2811 anv_vma_alloc(struct anv_device
*device
, struct anv_bo
*bo
)
2813 if (!(bo
->flags
& EXEC_OBJECT_PINNED
))
2816 pthread_mutex_lock(&device
->vma_mutex
);
2820 if (bo
->flags
& EXEC_OBJECT_SUPPORTS_48B_ADDRESS
&&
2821 device
->vma_hi_available
>= bo
->size
) {
2822 uint64_t addr
= util_vma_heap_alloc(&device
->vma_hi
, bo
->size
, 4096);
2824 bo
->offset
= gen_canonical_address(addr
);
2825 assert(addr
== gen_48b_address(bo
->offset
));
2826 device
->vma_hi_available
-= bo
->size
;
2830 if (bo
->offset
== 0 && device
->vma_lo_available
>= bo
->size
) {
2831 uint64_t addr
= util_vma_heap_alloc(&device
->vma_lo
, bo
->size
, 4096);
2833 bo
->offset
= gen_canonical_address(addr
);
2834 assert(addr
== gen_48b_address(bo
->offset
));
2835 device
->vma_lo_available
-= bo
->size
;
2839 pthread_mutex_unlock(&device
->vma_mutex
);
2841 return bo
->offset
!= 0;
2845 anv_vma_free(struct anv_device
*device
, struct anv_bo
*bo
)
2847 if (!(bo
->flags
& EXEC_OBJECT_PINNED
))
2850 const uint64_t addr_48b
= gen_48b_address(bo
->offset
);
2852 pthread_mutex_lock(&device
->vma_mutex
);
2854 if (addr_48b
>= LOW_HEAP_MIN_ADDRESS
&&
2855 addr_48b
<= LOW_HEAP_MAX_ADDRESS
) {
2856 util_vma_heap_free(&device
->vma_lo
, addr_48b
, bo
->size
);
2857 device
->vma_lo_available
+= bo
->size
;
2859 ASSERTED
const struct anv_physical_device
*physical_device
=
2860 &device
->instance
->physicalDevice
;
2861 assert(addr_48b
>= physical_device
->memory
.heaps
[0].vma_start
&&
2862 addr_48b
< (physical_device
->memory
.heaps
[0].vma_start
+
2863 physical_device
->memory
.heaps
[0].vma_size
));
2864 util_vma_heap_free(&device
->vma_hi
, addr_48b
, bo
->size
);
2865 device
->vma_hi_available
+= bo
->size
;
2868 pthread_mutex_unlock(&device
->vma_mutex
);
2874 anv_bo_init_new(struct anv_bo
*bo
, struct anv_device
*device
, uint64_t size
)
2876 uint32_t gem_handle
= anv_gem_create(device
, size
);
2878 return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY
);
2880 anv_bo_init(bo
, gem_handle
, size
);
2885 VkResult
anv_AllocateMemory(
2887 const VkMemoryAllocateInfo
* pAllocateInfo
,
2888 const VkAllocationCallbacks
* pAllocator
,
2889 VkDeviceMemory
* pMem
)
2891 ANV_FROM_HANDLE(anv_device
, device
, _device
);
2892 struct anv_physical_device
*pdevice
= &device
->instance
->physicalDevice
;
2893 struct anv_device_memory
*mem
;
2894 VkResult result
= VK_SUCCESS
;
2896 assert(pAllocateInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO
);
2898 /* The Vulkan 1.0.33 spec says "allocationSize must be greater than 0". */
2899 assert(pAllocateInfo
->allocationSize
> 0);
2901 if (pAllocateInfo
->allocationSize
> MAX_MEMORY_ALLOCATION_SIZE
)
2902 return VK_ERROR_OUT_OF_DEVICE_MEMORY
;
2904 /* FINISHME: Fail if allocation request exceeds heap size. */
2906 mem
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*mem
), 8,
2907 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2909 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
2911 assert(pAllocateInfo
->memoryTypeIndex
< pdevice
->memory
.type_count
);
2912 mem
->type
= &pdevice
->memory
.types
[pAllocateInfo
->memoryTypeIndex
];
2916 mem
->host_ptr
= NULL
;
2918 uint64_t bo_flags
= 0;
2920 assert(mem
->type
->heapIndex
< pdevice
->memory
.heap_count
);
2921 if (pdevice
->memory
.heaps
[mem
->type
->heapIndex
].supports_48bit_addresses
)
2922 bo_flags
|= EXEC_OBJECT_SUPPORTS_48B_ADDRESS
;
2924 const struct wsi_memory_allocate_info
*wsi_info
=
2925 vk_find_struct_const(pAllocateInfo
->pNext
, WSI_MEMORY_ALLOCATE_INFO_MESA
);
2926 if (wsi_info
&& wsi_info
->implicit_sync
) {
2927 /* We need to set the WRITE flag on window system buffers so that GEM
2928 * will know we're writing to them and synchronize uses on other rings
2929 * (eg if the display server uses the blitter ring).
2931 bo_flags
|= EXEC_OBJECT_WRITE
;
2932 } else if (pdevice
->has_exec_async
) {
2933 bo_flags
|= EXEC_OBJECT_ASYNC
;
2936 if (pdevice
->use_softpin
)
2937 bo_flags
|= EXEC_OBJECT_PINNED
;
2939 const VkExportMemoryAllocateInfo
*export_info
=
2940 vk_find_struct_const(pAllocateInfo
->pNext
, EXPORT_MEMORY_ALLOCATE_INFO
);
2942 /* Check if we need to support Android HW buffer export. If so,
2943 * create AHardwareBuffer and import memory from it.
2945 bool android_export
= false;
2946 if (export_info
&& export_info
->handleTypes
&
2947 VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID
)
2948 android_export
= true;
2950 /* Android memory import. */
2951 const struct VkImportAndroidHardwareBufferInfoANDROID
*ahw_import_info
=
2952 vk_find_struct_const(pAllocateInfo
->pNext
,
2953 IMPORT_ANDROID_HARDWARE_BUFFER_INFO_ANDROID
);
2955 if (ahw_import_info
) {
2956 result
= anv_import_ahw_memory(_device
, mem
, ahw_import_info
);
2957 if (result
!= VK_SUCCESS
)
2961 } else if (android_export
) {
2962 result
= anv_create_ahw_memory(_device
, mem
, pAllocateInfo
);
2963 if (result
!= VK_SUCCESS
)
2966 const struct VkImportAndroidHardwareBufferInfoANDROID import_info
= {
2969 result
= anv_import_ahw_memory(_device
, mem
, &import_info
);
2970 if (result
!= VK_SUCCESS
)
2976 const VkImportMemoryFdInfoKHR
*fd_info
=
2977 vk_find_struct_const(pAllocateInfo
->pNext
, IMPORT_MEMORY_FD_INFO_KHR
);
2979 /* The Vulkan spec permits handleType to be 0, in which case the struct is
2982 if (fd_info
&& fd_info
->handleType
) {
2983 /* At the moment, we support only the below handle types. */
2984 assert(fd_info
->handleType
==
2985 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT
||
2986 fd_info
->handleType
==
2987 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
2989 result
= anv_bo_cache_import(device
, &device
->bo_cache
, fd_info
->fd
,
2990 bo_flags
| ANV_BO_EXTERNAL
, &mem
->bo
);
2991 if (result
!= VK_SUCCESS
)
2994 VkDeviceSize aligned_alloc_size
=
2995 align_u64(pAllocateInfo
->allocationSize
, 4096);
2997 /* For security purposes, we reject importing the bo if it's smaller
2998 * than the requested allocation size. This prevents a malicious client
2999 * from passing a buffer to a trusted client, lying about the size, and
3000 * telling the trusted client to try and texture from an image that goes
3001 * out-of-bounds. This sort of thing could lead to GPU hangs or worse
3002 * in the trusted client. The trusted client can protect itself against
3003 * this sort of attack but only if it can trust the buffer size.
3005 if (mem
->bo
->size
< aligned_alloc_size
) {
3006 result
= vk_errorf(device
->instance
, device
,
3007 VK_ERROR_INVALID_EXTERNAL_HANDLE
,
3008 "aligned allocationSize too large for "
3009 "VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT: "
3010 "%"PRIu64
"B > %"PRIu64
"B",
3011 aligned_alloc_size
, mem
->bo
->size
);
3012 anv_bo_cache_release(device
, &device
->bo_cache
, mem
->bo
);
3016 /* From the Vulkan spec:
3018 * "Importing memory from a file descriptor transfers ownership of
3019 * the file descriptor from the application to the Vulkan
3020 * implementation. The application must not perform any operations on
3021 * the file descriptor after a successful import."
3023 * If the import fails, we leave the file descriptor open.
3029 const VkImportMemoryHostPointerInfoEXT
*host_ptr_info
=
3030 vk_find_struct_const(pAllocateInfo
->pNext
,
3031 IMPORT_MEMORY_HOST_POINTER_INFO_EXT
);
3032 if (host_ptr_info
&& host_ptr_info
->handleType
) {
3033 if (host_ptr_info
->handleType
==
3034 VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_MAPPED_FOREIGN_MEMORY_BIT_EXT
) {
3035 result
= vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE
);
3039 assert(host_ptr_info
->handleType
==
3040 VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT
);
3042 result
= anv_bo_cache_import_host_ptr(
3043 device
, &device
->bo_cache
, host_ptr_info
->pHostPointer
,
3044 pAllocateInfo
->allocationSize
, bo_flags
, &mem
->bo
);
3046 if (result
!= VK_SUCCESS
)
3049 mem
->host_ptr
= host_ptr_info
->pHostPointer
;
3053 /* Regular allocate (not importing memory). */
3055 if (export_info
&& export_info
->handleTypes
)
3056 bo_flags
|= ANV_BO_EXTERNAL
;
3058 result
= anv_bo_cache_alloc(device
, &device
->bo_cache
,
3059 pAllocateInfo
->allocationSize
, bo_flags
,
3061 if (result
!= VK_SUCCESS
)
3064 const VkMemoryDedicatedAllocateInfo
*dedicated_info
=
3065 vk_find_struct_const(pAllocateInfo
->pNext
, MEMORY_DEDICATED_ALLOCATE_INFO
);
3066 if (dedicated_info
&& dedicated_info
->image
!= VK_NULL_HANDLE
) {
3067 ANV_FROM_HANDLE(anv_image
, image
, dedicated_info
->image
);
3069 /* Some legacy (non-modifiers) consumers need the tiling to be set on
3070 * the BO. In this case, we have a dedicated allocation.
3072 if (image
->needs_set_tiling
) {
3073 const uint32_t i915_tiling
=
3074 isl_tiling_to_i915_tiling(image
->planes
[0].surface
.isl
.tiling
);
3075 int ret
= anv_gem_set_tiling(device
, mem
->bo
->gem_handle
,
3076 image
->planes
[0].surface
.isl
.row_pitch_B
,
3079 anv_bo_cache_release(device
, &device
->bo_cache
, mem
->bo
);
3080 return vk_errorf(device
->instance
, NULL
,
3081 VK_ERROR_OUT_OF_DEVICE_MEMORY
,
3082 "failed to set BO tiling: %m");
3088 pthread_mutex_lock(&device
->mutex
);
3089 list_addtail(&mem
->link
, &device
->memory_objects
);
3090 pthread_mutex_unlock(&device
->mutex
);
3092 *pMem
= anv_device_memory_to_handle(mem
);
3094 p_atomic_add(&pdevice
->memory
.heaps
[mem
->type
->heapIndex
].used
,
3100 vk_free2(&device
->alloc
, pAllocator
, mem
);
3105 VkResult
anv_GetMemoryFdKHR(
3107 const VkMemoryGetFdInfoKHR
* pGetFdInfo
,
3110 ANV_FROM_HANDLE(anv_device
, dev
, device_h
);
3111 ANV_FROM_HANDLE(anv_device_memory
, mem
, pGetFdInfo
->memory
);
3113 assert(pGetFdInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR
);
3115 assert(pGetFdInfo
->handleType
== VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT
||
3116 pGetFdInfo
->handleType
== VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
3118 return anv_bo_cache_export(dev
, &dev
->bo_cache
, mem
->bo
, pFd
);
3121 VkResult
anv_GetMemoryFdPropertiesKHR(
3123 VkExternalMemoryHandleTypeFlagBits handleType
,
3125 VkMemoryFdPropertiesKHR
* pMemoryFdProperties
)
3127 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3128 struct anv_physical_device
*pdevice
= &device
->instance
->physicalDevice
;
3130 switch (handleType
) {
3131 case VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
:
3132 /* dma-buf can be imported as any memory type */
3133 pMemoryFdProperties
->memoryTypeBits
=
3134 (1 << pdevice
->memory
.type_count
) - 1;
3138 /* The valid usage section for this function says:
3140 * "handleType must not be one of the handle types defined as
3143 * So opaque handle types fall into the default "unsupported" case.
3145 return vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE
);
3149 VkResult
anv_GetMemoryHostPointerPropertiesEXT(
3151 VkExternalMemoryHandleTypeFlagBits handleType
,
3152 const void* pHostPointer
,
3153 VkMemoryHostPointerPropertiesEXT
* pMemoryHostPointerProperties
)
3155 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3157 assert(pMemoryHostPointerProperties
->sType
==
3158 VK_STRUCTURE_TYPE_MEMORY_HOST_POINTER_PROPERTIES_EXT
);
3160 switch (handleType
) {
3161 case VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT
: {
3162 struct anv_physical_device
*pdevice
= &device
->instance
->physicalDevice
;
3164 /* Host memory can be imported as any memory type. */
3165 pMemoryHostPointerProperties
->memoryTypeBits
=
3166 (1ull << pdevice
->memory
.type_count
) - 1;
3171 return VK_ERROR_INVALID_EXTERNAL_HANDLE
;
3175 void anv_FreeMemory(
3177 VkDeviceMemory _mem
,
3178 const VkAllocationCallbacks
* pAllocator
)
3180 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3181 ANV_FROM_HANDLE(anv_device_memory
, mem
, _mem
);
3182 struct anv_physical_device
*pdevice
= &device
->instance
->physicalDevice
;
3187 pthread_mutex_lock(&device
->mutex
);
3188 list_del(&mem
->link
);
3189 pthread_mutex_unlock(&device
->mutex
);
3192 anv_UnmapMemory(_device
, _mem
);
3194 p_atomic_add(&pdevice
->memory
.heaps
[mem
->type
->heapIndex
].used
,
3197 anv_bo_cache_release(device
, &device
->bo_cache
, mem
->bo
);
3199 #if defined(ANDROID) && ANDROID_API_LEVEL >= 26
3201 AHardwareBuffer_release(mem
->ahw
);
3204 vk_free2(&device
->alloc
, pAllocator
, mem
);
3207 VkResult
anv_MapMemory(
3209 VkDeviceMemory _memory
,
3210 VkDeviceSize offset
,
3212 VkMemoryMapFlags flags
,
3215 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3216 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
3223 if (mem
->host_ptr
) {
3224 *ppData
= mem
->host_ptr
+ offset
;
3228 if (size
== VK_WHOLE_SIZE
)
3229 size
= mem
->bo
->size
- offset
;
3231 /* From the Vulkan spec version 1.0.32 docs for MapMemory:
3233 * * If size is not equal to VK_WHOLE_SIZE, size must be greater than 0
3234 * assert(size != 0);
3235 * * If size is not equal to VK_WHOLE_SIZE, size must be less than or
3236 * equal to the size of the memory minus offset
3239 assert(offset
+ size
<= mem
->bo
->size
);
3241 /* FIXME: Is this supposed to be thread safe? Since vkUnmapMemory() only
3242 * takes a VkDeviceMemory pointer, it seems like only one map of the memory
3243 * at a time is valid. We could just mmap up front and return an offset
3244 * pointer here, but that may exhaust virtual memory on 32 bit
3247 uint32_t gem_flags
= 0;
3249 if (!device
->info
.has_llc
&&
3250 (mem
->type
->propertyFlags
& VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
))
3251 gem_flags
|= I915_MMAP_WC
;
3253 /* GEM will fail to map if the offset isn't 4k-aligned. Round down. */
3254 uint64_t map_offset
= offset
& ~4095ull;
3255 assert(offset
>= map_offset
);
3256 uint64_t map_size
= (offset
+ size
) - map_offset
;
3258 /* Let's map whole pages */
3259 map_size
= align_u64(map_size
, 4096);
3261 void *map
= anv_gem_mmap(device
, mem
->bo
->gem_handle
,
3262 map_offset
, map_size
, gem_flags
);
3263 if (map
== MAP_FAILED
)
3264 return vk_error(VK_ERROR_MEMORY_MAP_FAILED
);
3267 mem
->map_size
= map_size
;
3269 *ppData
= mem
->map
+ (offset
- map_offset
);
3274 void anv_UnmapMemory(
3276 VkDeviceMemory _memory
)
3278 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
3280 if (mem
== NULL
|| mem
->host_ptr
)
3283 anv_gem_munmap(mem
->map
, mem
->map_size
);
3290 clflush_mapped_ranges(struct anv_device
*device
,
3292 const VkMappedMemoryRange
*ranges
)
3294 for (uint32_t i
= 0; i
< count
; i
++) {
3295 ANV_FROM_HANDLE(anv_device_memory
, mem
, ranges
[i
].memory
);
3296 if (ranges
[i
].offset
>= mem
->map_size
)
3299 gen_clflush_range(mem
->map
+ ranges
[i
].offset
,
3300 MIN2(ranges
[i
].size
, mem
->map_size
- ranges
[i
].offset
));
3304 VkResult
anv_FlushMappedMemoryRanges(
3306 uint32_t memoryRangeCount
,
3307 const VkMappedMemoryRange
* pMemoryRanges
)
3309 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3311 if (device
->info
.has_llc
)
3314 /* Make sure the writes we're flushing have landed. */
3315 __builtin_ia32_mfence();
3317 clflush_mapped_ranges(device
, memoryRangeCount
, pMemoryRanges
);
3322 VkResult
anv_InvalidateMappedMemoryRanges(
3324 uint32_t memoryRangeCount
,
3325 const VkMappedMemoryRange
* pMemoryRanges
)
3327 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3329 if (device
->info
.has_llc
)
3332 clflush_mapped_ranges(device
, memoryRangeCount
, pMemoryRanges
);
3334 /* Make sure no reads get moved up above the invalidate. */
3335 __builtin_ia32_mfence();
3340 void anv_GetBufferMemoryRequirements(
3343 VkMemoryRequirements
* pMemoryRequirements
)
3345 ANV_FROM_HANDLE(anv_buffer
, buffer
, _buffer
);
3346 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3347 struct anv_physical_device
*pdevice
= &device
->instance
->physicalDevice
;
3349 /* The Vulkan spec (git aaed022) says:
3351 * memoryTypeBits is a bitfield and contains one bit set for every
3352 * supported memory type for the resource. The bit `1<<i` is set if and
3353 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
3354 * structure for the physical device is supported.
3356 uint32_t memory_types
= 0;
3357 for (uint32_t i
= 0; i
< pdevice
->memory
.type_count
; i
++) {
3358 uint32_t valid_usage
= pdevice
->memory
.types
[i
].valid_buffer_usage
;
3359 if ((valid_usage
& buffer
->usage
) == buffer
->usage
)
3360 memory_types
|= (1u << i
);
3363 /* Base alignment requirement of a cache line */
3364 uint32_t alignment
= 16;
3366 /* We need an alignment of 32 for pushing UBOs */
3367 if (buffer
->usage
& VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT
)
3368 alignment
= MAX2(alignment
, 32);
3370 pMemoryRequirements
->size
= buffer
->size
;
3371 pMemoryRequirements
->alignment
= alignment
;
3373 /* Storage and Uniform buffers should have their size aligned to
3374 * 32-bits to avoid boundary checks when last DWord is not complete.
3375 * This would ensure that not internal padding would be needed for
3378 if (device
->robust_buffer_access
&&
3379 (buffer
->usage
& VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT
||
3380 buffer
->usage
& VK_BUFFER_USAGE_STORAGE_BUFFER_BIT
))
3381 pMemoryRequirements
->size
= align_u64(buffer
->size
, 4);
3383 pMemoryRequirements
->memoryTypeBits
= memory_types
;
3386 void anv_GetBufferMemoryRequirements2(
3388 const VkBufferMemoryRequirementsInfo2
* pInfo
,
3389 VkMemoryRequirements2
* pMemoryRequirements
)
3391 anv_GetBufferMemoryRequirements(_device
, pInfo
->buffer
,
3392 &pMemoryRequirements
->memoryRequirements
);
3394 vk_foreach_struct(ext
, pMemoryRequirements
->pNext
) {
3395 switch (ext
->sType
) {
3396 case VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS
: {
3397 VkMemoryDedicatedRequirements
*requirements
= (void *)ext
;
3398 requirements
->prefersDedicatedAllocation
= false;
3399 requirements
->requiresDedicatedAllocation
= false;
3404 anv_debug_ignored_stype(ext
->sType
);
3410 void anv_GetImageMemoryRequirements(
3413 VkMemoryRequirements
* pMemoryRequirements
)
3415 ANV_FROM_HANDLE(anv_image
, image
, _image
);
3416 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3417 struct anv_physical_device
*pdevice
= &device
->instance
->physicalDevice
;
3419 /* The Vulkan spec (git aaed022) says:
3421 * memoryTypeBits is a bitfield and contains one bit set for every
3422 * supported memory type for the resource. The bit `1<<i` is set if and
3423 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
3424 * structure for the physical device is supported.
3426 * All types are currently supported for images.
3428 uint32_t memory_types
= (1ull << pdevice
->memory
.type_count
) - 1;
3430 /* We must have image allocated or imported at this point. According to the
3431 * specification, external images must have been bound to memory before
3432 * calling GetImageMemoryRequirements.
3434 assert(image
->size
> 0);
3436 pMemoryRequirements
->size
= image
->size
;
3437 pMemoryRequirements
->alignment
= image
->alignment
;
3438 pMemoryRequirements
->memoryTypeBits
= memory_types
;
3441 void anv_GetImageMemoryRequirements2(
3443 const VkImageMemoryRequirementsInfo2
* pInfo
,
3444 VkMemoryRequirements2
* pMemoryRequirements
)
3446 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3447 ANV_FROM_HANDLE(anv_image
, image
, pInfo
->image
);
3449 anv_GetImageMemoryRequirements(_device
, pInfo
->image
,
3450 &pMemoryRequirements
->memoryRequirements
);
3452 vk_foreach_struct_const(ext
, pInfo
->pNext
) {
3453 switch (ext
->sType
) {
3454 case VK_STRUCTURE_TYPE_IMAGE_PLANE_MEMORY_REQUIREMENTS_INFO
: {
3455 struct anv_physical_device
*pdevice
= &device
->instance
->physicalDevice
;
3456 const VkImagePlaneMemoryRequirementsInfo
*plane_reqs
=
3457 (const VkImagePlaneMemoryRequirementsInfo
*) ext
;
3458 uint32_t plane
= anv_image_aspect_to_plane(image
->aspects
,
3459 plane_reqs
->planeAspect
);
3461 assert(image
->planes
[plane
].offset
== 0);
3463 /* The Vulkan spec (git aaed022) says:
3465 * memoryTypeBits is a bitfield and contains one bit set for every
3466 * supported memory type for the resource. The bit `1<<i` is set
3467 * if and only if the memory type `i` in the
3468 * VkPhysicalDeviceMemoryProperties structure for the physical
3469 * device is supported.
3471 * All types are currently supported for images.
3473 pMemoryRequirements
->memoryRequirements
.memoryTypeBits
=
3474 (1ull << pdevice
->memory
.type_count
) - 1;
3476 /* We must have image allocated or imported at this point. According to the
3477 * specification, external images must have been bound to memory before
3478 * calling GetImageMemoryRequirements.
3480 assert(image
->planes
[plane
].size
> 0);
3482 pMemoryRequirements
->memoryRequirements
.size
= image
->planes
[plane
].size
;
3483 pMemoryRequirements
->memoryRequirements
.alignment
=
3484 image
->planes
[plane
].alignment
;
3489 anv_debug_ignored_stype(ext
->sType
);
3494 vk_foreach_struct(ext
, pMemoryRequirements
->pNext
) {
3495 switch (ext
->sType
) {
3496 case VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS
: {
3497 VkMemoryDedicatedRequirements
*requirements
= (void *)ext
;
3498 if (image
->needs_set_tiling
|| image
->external_format
) {
3499 /* If we need to set the tiling for external consumers, we need a
3500 * dedicated allocation.
3502 * See also anv_AllocateMemory.
3504 requirements
->prefersDedicatedAllocation
= true;
3505 requirements
->requiresDedicatedAllocation
= true;
3507 requirements
->prefersDedicatedAllocation
= false;
3508 requirements
->requiresDedicatedAllocation
= false;
3514 anv_debug_ignored_stype(ext
->sType
);
3520 void anv_GetImageSparseMemoryRequirements(
3523 uint32_t* pSparseMemoryRequirementCount
,
3524 VkSparseImageMemoryRequirements
* pSparseMemoryRequirements
)
3526 *pSparseMemoryRequirementCount
= 0;
3529 void anv_GetImageSparseMemoryRequirements2(
3531 const VkImageSparseMemoryRequirementsInfo2
* pInfo
,
3532 uint32_t* pSparseMemoryRequirementCount
,
3533 VkSparseImageMemoryRequirements2
* pSparseMemoryRequirements
)
3535 *pSparseMemoryRequirementCount
= 0;
3538 void anv_GetDeviceMemoryCommitment(
3540 VkDeviceMemory memory
,
3541 VkDeviceSize
* pCommittedMemoryInBytes
)
3543 *pCommittedMemoryInBytes
= 0;
3547 anv_bind_buffer_memory(const VkBindBufferMemoryInfo
*pBindInfo
)
3549 ANV_FROM_HANDLE(anv_device_memory
, mem
, pBindInfo
->memory
);
3550 ANV_FROM_HANDLE(anv_buffer
, buffer
, pBindInfo
->buffer
);
3552 assert(pBindInfo
->sType
== VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO
);
3555 assert((buffer
->usage
& mem
->type
->valid_buffer_usage
) == buffer
->usage
);
3556 buffer
->address
= (struct anv_address
) {
3558 .offset
= pBindInfo
->memoryOffset
,
3561 buffer
->address
= ANV_NULL_ADDRESS
;
3565 VkResult
anv_BindBufferMemory(
3568 VkDeviceMemory memory
,
3569 VkDeviceSize memoryOffset
)
3571 anv_bind_buffer_memory(
3572 &(VkBindBufferMemoryInfo
) {
3573 .sType
= VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO
,
3576 .memoryOffset
= memoryOffset
,
3582 VkResult
anv_BindBufferMemory2(
3584 uint32_t bindInfoCount
,
3585 const VkBindBufferMemoryInfo
* pBindInfos
)
3587 for (uint32_t i
= 0; i
< bindInfoCount
; i
++)
3588 anv_bind_buffer_memory(&pBindInfos
[i
]);
3593 VkResult
anv_QueueBindSparse(
3595 uint32_t bindInfoCount
,
3596 const VkBindSparseInfo
* pBindInfo
,
3599 ANV_FROM_HANDLE(anv_queue
, queue
, _queue
);
3600 if (anv_device_is_lost(queue
->device
))
3601 return VK_ERROR_DEVICE_LOST
;
3603 return vk_error(VK_ERROR_FEATURE_NOT_PRESENT
);
3608 VkResult
anv_CreateEvent(
3610 const VkEventCreateInfo
* pCreateInfo
,
3611 const VkAllocationCallbacks
* pAllocator
,
3614 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3615 struct anv_state state
;
3616 struct anv_event
*event
;
3618 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_EVENT_CREATE_INFO
);
3620 state
= anv_state_pool_alloc(&device
->dynamic_state_pool
,
3623 event
->state
= state
;
3624 event
->semaphore
= VK_EVENT_RESET
;
3626 if (!device
->info
.has_llc
) {
3627 /* Make sure the writes we're flushing have landed. */
3628 __builtin_ia32_mfence();
3629 __builtin_ia32_clflush(event
);
3632 *pEvent
= anv_event_to_handle(event
);
3637 void anv_DestroyEvent(
3640 const VkAllocationCallbacks
* pAllocator
)
3642 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3643 ANV_FROM_HANDLE(anv_event
, event
, _event
);
3648 anv_state_pool_free(&device
->dynamic_state_pool
, event
->state
);
3651 VkResult
anv_GetEventStatus(
3655 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3656 ANV_FROM_HANDLE(anv_event
, event
, _event
);
3658 if (anv_device_is_lost(device
))
3659 return VK_ERROR_DEVICE_LOST
;
3661 if (!device
->info
.has_llc
) {
3662 /* Invalidate read cache before reading event written by GPU. */
3663 __builtin_ia32_clflush(event
);
3664 __builtin_ia32_mfence();
3668 return event
->semaphore
;
3671 VkResult
anv_SetEvent(
3675 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3676 ANV_FROM_HANDLE(anv_event
, event
, _event
);
3678 event
->semaphore
= VK_EVENT_SET
;
3680 if (!device
->info
.has_llc
) {
3681 /* Make sure the writes we're flushing have landed. */
3682 __builtin_ia32_mfence();
3683 __builtin_ia32_clflush(event
);
3689 VkResult
anv_ResetEvent(
3693 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3694 ANV_FROM_HANDLE(anv_event
, event
, _event
);
3696 event
->semaphore
= VK_EVENT_RESET
;
3698 if (!device
->info
.has_llc
) {
3699 /* Make sure the writes we're flushing have landed. */
3700 __builtin_ia32_mfence();
3701 __builtin_ia32_clflush(event
);
3709 VkResult
anv_CreateBuffer(
3711 const VkBufferCreateInfo
* pCreateInfo
,
3712 const VkAllocationCallbacks
* pAllocator
,
3715 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3716 struct anv_buffer
*buffer
;
3718 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
3720 buffer
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*buffer
), 8,
3721 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
3723 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
3725 buffer
->size
= pCreateInfo
->size
;
3726 buffer
->usage
= pCreateInfo
->usage
;
3727 buffer
->address
= ANV_NULL_ADDRESS
;
3729 *pBuffer
= anv_buffer_to_handle(buffer
);
3734 void anv_DestroyBuffer(
3737 const VkAllocationCallbacks
* pAllocator
)
3739 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3740 ANV_FROM_HANDLE(anv_buffer
, buffer
, _buffer
);
3745 vk_free2(&device
->alloc
, pAllocator
, buffer
);
3748 VkDeviceAddress
anv_GetBufferDeviceAddressEXT(
3750 const VkBufferDeviceAddressInfoEXT
* pInfo
)
3752 ANV_FROM_HANDLE(anv_buffer
, buffer
, pInfo
->buffer
);
3754 assert(buffer
->address
.bo
->flags
& EXEC_OBJECT_PINNED
);
3756 return anv_address_physical(buffer
->address
);
3760 anv_fill_buffer_surface_state(struct anv_device
*device
, struct anv_state state
,
3761 enum isl_format format
,
3762 struct anv_address address
,
3763 uint32_t range
, uint32_t stride
)
3765 isl_buffer_fill_state(&device
->isl_dev
, state
.map
,
3766 .address
= anv_address_physical(address
),
3767 .mocs
= device
->default_mocs
,
3770 .swizzle
= ISL_SWIZZLE_IDENTITY
,
3771 .stride_B
= stride
);
3774 void anv_DestroySampler(
3777 const VkAllocationCallbacks
* pAllocator
)
3779 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3780 ANV_FROM_HANDLE(anv_sampler
, sampler
, _sampler
);
3785 if (sampler
->bindless_state
.map
) {
3786 anv_state_pool_free(&device
->dynamic_state_pool
,
3787 sampler
->bindless_state
);
3790 vk_free2(&device
->alloc
, pAllocator
, sampler
);
3793 VkResult
anv_CreateFramebuffer(
3795 const VkFramebufferCreateInfo
* pCreateInfo
,
3796 const VkAllocationCallbacks
* pAllocator
,
3797 VkFramebuffer
* pFramebuffer
)
3799 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3800 struct anv_framebuffer
*framebuffer
;
3802 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
3804 size_t size
= sizeof(*framebuffer
);
3806 /* VK_KHR_imageless_framebuffer extension says:
3808 * If flags includes VK_FRAMEBUFFER_CREATE_IMAGELESS_BIT_KHR,
3809 * parameter pAttachments is ignored.
3811 if (!(pCreateInfo
->flags
& VK_FRAMEBUFFER_CREATE_IMAGELESS_BIT_KHR
)) {
3812 size
+= sizeof(struct anv_image_view
*) * pCreateInfo
->attachmentCount
;
3813 framebuffer
= vk_alloc2(&device
->alloc
, pAllocator
, size
, 8,
3814 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
3815 if (framebuffer
== NULL
)
3816 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
3818 for (uint32_t i
= 0; i
< pCreateInfo
->attachmentCount
; i
++) {
3819 ANV_FROM_HANDLE(anv_image_view
, iview
, pCreateInfo
->pAttachments
[i
]);
3820 framebuffer
->attachments
[i
] = iview
;
3822 framebuffer
->attachment_count
= pCreateInfo
->attachmentCount
;
3824 assert(device
->enabled_extensions
.KHR_imageless_framebuffer
);
3825 framebuffer
= vk_alloc2(&device
->alloc
, pAllocator
, size
, 8,
3826 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
3827 if (framebuffer
== NULL
)
3828 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
3830 framebuffer
->attachment_count
= 0;
3833 framebuffer
->width
= pCreateInfo
->width
;
3834 framebuffer
->height
= pCreateInfo
->height
;
3835 framebuffer
->layers
= pCreateInfo
->layers
;
3837 *pFramebuffer
= anv_framebuffer_to_handle(framebuffer
);
3842 void anv_DestroyFramebuffer(
3845 const VkAllocationCallbacks
* pAllocator
)
3847 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3848 ANV_FROM_HANDLE(anv_framebuffer
, fb
, _fb
);
3853 vk_free2(&device
->alloc
, pAllocator
, fb
);
3856 static const VkTimeDomainEXT anv_time_domains
[] = {
3857 VK_TIME_DOMAIN_DEVICE_EXT
,
3858 VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT
,
3859 VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT
,
3862 VkResult
anv_GetPhysicalDeviceCalibrateableTimeDomainsEXT(
3863 VkPhysicalDevice physicalDevice
,
3864 uint32_t *pTimeDomainCount
,
3865 VkTimeDomainEXT
*pTimeDomains
)
3868 VK_OUTARRAY_MAKE(out
, pTimeDomains
, pTimeDomainCount
);
3870 for (d
= 0; d
< ARRAY_SIZE(anv_time_domains
); d
++) {
3871 vk_outarray_append(&out
, i
) {
3872 *i
= anv_time_domains
[d
];
3876 return vk_outarray_status(&out
);
3880 anv_clock_gettime(clockid_t clock_id
)
3882 struct timespec current
;
3885 ret
= clock_gettime(clock_id
, ¤t
);
3886 if (ret
< 0 && clock_id
== CLOCK_MONOTONIC_RAW
)
3887 ret
= clock_gettime(CLOCK_MONOTONIC
, ¤t
);
3891 return (uint64_t) current
.tv_sec
* 1000000000ULL + current
.tv_nsec
;
3894 #define TIMESTAMP 0x2358
3896 VkResult
anv_GetCalibratedTimestampsEXT(
3898 uint32_t timestampCount
,
3899 const VkCalibratedTimestampInfoEXT
*pTimestampInfos
,
3900 uint64_t *pTimestamps
,
3901 uint64_t *pMaxDeviation
)
3903 ANV_FROM_HANDLE(anv_device
, device
, _device
);
3904 uint64_t timestamp_frequency
= device
->info
.timestamp_frequency
;
3907 uint64_t begin
, end
;
3908 uint64_t max_clock_period
= 0;
3910 begin
= anv_clock_gettime(CLOCK_MONOTONIC_RAW
);
3912 for (d
= 0; d
< timestampCount
; d
++) {
3913 switch (pTimestampInfos
[d
].timeDomain
) {
3914 case VK_TIME_DOMAIN_DEVICE_EXT
:
3915 ret
= anv_gem_reg_read(device
, TIMESTAMP
| 1,
3919 return anv_device_set_lost(device
, "Failed to read the TIMESTAMP "
3922 uint64_t device_period
= DIV_ROUND_UP(1000000000, timestamp_frequency
);
3923 max_clock_period
= MAX2(max_clock_period
, device_period
);
3925 case VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT
:
3926 pTimestamps
[d
] = anv_clock_gettime(CLOCK_MONOTONIC
);
3927 max_clock_period
= MAX2(max_clock_period
, 1);
3930 case VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT
:
3931 pTimestamps
[d
] = begin
;
3939 end
= anv_clock_gettime(CLOCK_MONOTONIC_RAW
);
3942 * The maximum deviation is the sum of the interval over which we
3943 * perform the sampling and the maximum period of any sampled
3944 * clock. That's because the maximum skew between any two sampled
3945 * clock edges is when the sampled clock with the largest period is
3946 * sampled at the end of that period but right at the beginning of the
3947 * sampling interval and some other clock is sampled right at the
3948 * begining of its sampling period and right at the end of the
3949 * sampling interval. Let's assume the GPU has the longest clock
3950 * period and that the application is sampling GPU and monotonic:
3953 * w x y z 0 1 2 3 4 5 6 7 8 9 a b c d e f
3954 * Raw -_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-
3958 * GPU -----_____-----_____-----_____-----_____
3961 * x y z 0 1 2 3 4 5 6 7 8 9 a b c
3962 * Monotonic -_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-
3964 * Interval <----------------->
3965 * Deviation <-------------------------->
3969 * m = read(monotonic) 2
3972 * We round the sample interval up by one tick to cover sampling error
3973 * in the interval clock
3976 uint64_t sample_interval
= end
- begin
+ 1;
3978 *pMaxDeviation
= sample_interval
+ max_clock_period
;
3983 /* vk_icd.h does not declare this function, so we declare it here to
3984 * suppress Wmissing-prototypes.
3986 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
3987 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t* pSupportedVersion
);
3989 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
3990 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t* pSupportedVersion
)
3992 /* For the full details on loader interface versioning, see
3993 * <https://github.com/KhronosGroup/Vulkan-LoaderAndValidationLayers/blob/master/loader/LoaderAndLayerInterface.md>.
3994 * What follows is a condensed summary, to help you navigate the large and
3995 * confusing official doc.
3997 * - Loader interface v0 is incompatible with later versions. We don't
4000 * - In loader interface v1:
4001 * - The first ICD entrypoint called by the loader is
4002 * vk_icdGetInstanceProcAddr(). The ICD must statically expose this
4004 * - The ICD must statically expose no other Vulkan symbol unless it is
4005 * linked with -Bsymbolic.
4006 * - Each dispatchable Vulkan handle created by the ICD must be
4007 * a pointer to a struct whose first member is VK_LOADER_DATA. The
4008 * ICD must initialize VK_LOADER_DATA.loadMagic to ICD_LOADER_MAGIC.
4009 * - The loader implements vkCreate{PLATFORM}SurfaceKHR() and
4010 * vkDestroySurfaceKHR(). The ICD must be capable of working with
4011 * such loader-managed surfaces.
4013 * - Loader interface v2 differs from v1 in:
4014 * - The first ICD entrypoint called by the loader is
4015 * vk_icdNegotiateLoaderICDInterfaceVersion(). The ICD must
4016 * statically expose this entrypoint.
4018 * - Loader interface v3 differs from v2 in:
4019 * - The ICD must implement vkCreate{PLATFORM}SurfaceKHR(),
4020 * vkDestroySurfaceKHR(), and other API which uses VKSurfaceKHR,
4021 * because the loader no longer does so.
4023 *pSupportedVersion
= MIN2(*pSupportedVersion
, 3u);