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
30 #include "anv_private.h"
31 #include "mesa/main/git_sha1.h"
32 #include "util/strtod.h"
33 #include "util/debug.h"
35 #include "genxml/gen7_pack.h"
37 struct anv_dispatch_table dtable
;
40 compiler_debug_log(void *data
, const char *fmt
, ...)
44 compiler_perf_log(void *data
, const char *fmt
, ...)
49 if (unlikely(INTEL_DEBUG
& DEBUG_PERF
))
50 vfprintf(stderr
, fmt
, args
);
56 anv_physical_device_init(struct anv_physical_device
*device
,
57 struct anv_instance
*instance
,
63 fd
= open(path
, O_RDWR
| O_CLOEXEC
);
65 return vk_errorf(VK_ERROR_INITIALIZATION_FAILED
,
66 "failed to open %s: %m", path
);
68 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
69 device
->instance
= instance
;
72 device
->chipset_id
= anv_gem_get_param(fd
, I915_PARAM_CHIPSET_ID
);
73 if (!device
->chipset_id
) {
74 result
= vk_errorf(VK_ERROR_INITIALIZATION_FAILED
,
75 "failed to get chipset id: %m");
79 device
->name
= brw_get_device_name(device
->chipset_id
);
80 device
->info
= brw_get_device_info(device
->chipset_id
);
82 result
= vk_errorf(VK_ERROR_INITIALIZATION_FAILED
,
83 "failed to get device info");
87 if (device
->info
->is_haswell
) {
88 fprintf(stderr
, "WARNING: Haswell Vulkan support is incomplete\n");
89 } else if (device
->info
->gen
== 7 && !device
->info
->is_baytrail
) {
90 fprintf(stderr
, "WARNING: Ivy Bridge Vulkan support is incomplete\n");
91 } else if (device
->info
->gen
== 7 && device
->info
->is_baytrail
) {
92 fprintf(stderr
, "WARNING: Bay Trail Vulkan support is incomplete\n");
93 } else if (device
->info
->gen
>= 8) {
94 /* Broadwell, Cherryview, Skylake, Broxton, Kabylake is as fully
95 * supported as anything */
97 result
= vk_errorf(VK_ERROR_INCOMPATIBLE_DRIVER
,
98 "Vulkan not yet supported on %s", device
->name
);
102 if (anv_gem_get_aperture(fd
, &device
->aperture_size
) == -1) {
103 result
= vk_errorf(VK_ERROR_INITIALIZATION_FAILED
,
104 "failed to get aperture size: %m");
108 if (!anv_gem_get_param(fd
, I915_PARAM_HAS_WAIT_TIMEOUT
)) {
109 result
= vk_errorf(VK_ERROR_INITIALIZATION_FAILED
,
110 "kernel missing gem wait");
114 if (!anv_gem_get_param(fd
, I915_PARAM_HAS_EXECBUF2
)) {
115 result
= vk_errorf(VK_ERROR_INITIALIZATION_FAILED
,
116 "kernel missing execbuf2");
120 if (!device
->info
->has_llc
&&
121 anv_gem_get_param(fd
, I915_PARAM_MMAP_VERSION
) < 1) {
122 result
= vk_errorf(VK_ERROR_INITIALIZATION_FAILED
,
123 "kernel missing wc mmap");
127 bool swizzled
= anv_gem_get_bit6_swizzle(fd
, I915_TILING_X
);
131 brw_process_intel_debug_variable();
133 device
->compiler
= brw_compiler_create(NULL
, device
->info
);
134 if (device
->compiler
== NULL
) {
135 result
= vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
138 device
->compiler
->shader_debug_log
= compiler_debug_log
;
139 device
->compiler
->shader_perf_log
= compiler_perf_log
;
141 /* XXX: Actually detect bit6 swizzling */
142 isl_device_init(&device
->isl_dev
, device
->info
, swizzled
);
152 anv_physical_device_finish(struct anv_physical_device
*device
)
154 ralloc_free(device
->compiler
);
157 static const VkExtensionProperties global_extensions
[] = {
159 .extensionName
= VK_KHR_SURFACE_EXTENSION_NAME
,
163 .extensionName
= VK_KHR_XCB_SURFACE_EXTENSION_NAME
,
166 #ifdef HAVE_WAYLAND_PLATFORM
168 .extensionName
= VK_KHR_WAYLAND_SURFACE_EXTENSION_NAME
,
174 static const VkExtensionProperties device_extensions
[] = {
176 .extensionName
= VK_KHR_SWAPCHAIN_EXTENSION_NAME
,
182 default_alloc_func(void *pUserData
, size_t size
, size_t align
,
183 VkSystemAllocationScope allocationScope
)
189 default_realloc_func(void *pUserData
, void *pOriginal
, size_t size
,
190 size_t align
, VkSystemAllocationScope allocationScope
)
192 return realloc(pOriginal
, size
);
196 default_free_func(void *pUserData
, void *pMemory
)
201 static const VkAllocationCallbacks default_alloc
= {
203 .pfnAllocation
= default_alloc_func
,
204 .pfnReallocation
= default_realloc_func
,
205 .pfnFree
= default_free_func
,
208 VkResult
anv_CreateInstance(
209 const VkInstanceCreateInfo
* pCreateInfo
,
210 const VkAllocationCallbacks
* pAllocator
,
211 VkInstance
* pInstance
)
213 struct anv_instance
*instance
;
215 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO
);
217 uint32_t client_version
= pCreateInfo
->pApplicationInfo
?
218 pCreateInfo
->pApplicationInfo
->apiVersion
:
219 VK_MAKE_VERSION(1, 0, 0);
220 if (VK_MAKE_VERSION(1, 0, 0) > client_version
||
221 client_version
> VK_MAKE_VERSION(1, 0, 3)) {
222 return vk_errorf(VK_ERROR_INCOMPATIBLE_DRIVER
,
223 "Client requested version %d.%d.%d",
224 VK_VERSION_MAJOR(client_version
),
225 VK_VERSION_MINOR(client_version
),
226 VK_VERSION_PATCH(client_version
));
229 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
231 for (uint32_t j
= 0; j
< ARRAY_SIZE(global_extensions
); j
++) {
232 if (strcmp(pCreateInfo
->ppEnabledExtensionNames
[i
],
233 global_extensions
[j
].extensionName
) == 0) {
239 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
242 instance
= anv_alloc2(&default_alloc
, pAllocator
, sizeof(*instance
), 8,
243 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE
);
245 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
247 instance
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
250 instance
->alloc
= *pAllocator
;
252 instance
->alloc
= default_alloc
;
254 instance
->apiVersion
= client_version
;
255 instance
->physicalDeviceCount
= -1;
257 memset(instance
->wsi
, 0, sizeof(instance
->wsi
));
261 VG(VALGRIND_CREATE_MEMPOOL(instance
, 0, false));
263 anv_init_wsi(instance
);
265 *pInstance
= anv_instance_to_handle(instance
);
270 void anv_DestroyInstance(
271 VkInstance _instance
,
272 const VkAllocationCallbacks
* pAllocator
)
274 ANV_FROM_HANDLE(anv_instance
, instance
, _instance
);
276 if (instance
->physicalDeviceCount
> 0) {
277 /* We support at most one physical device. */
278 assert(instance
->physicalDeviceCount
== 1);
279 anv_physical_device_finish(&instance
->physicalDevice
);
282 anv_finish_wsi(instance
);
284 VG(VALGRIND_DESTROY_MEMPOOL(instance
));
288 anv_free(&instance
->alloc
, instance
);
291 VkResult
anv_EnumeratePhysicalDevices(
292 VkInstance _instance
,
293 uint32_t* pPhysicalDeviceCount
,
294 VkPhysicalDevice
* pPhysicalDevices
)
296 ANV_FROM_HANDLE(anv_instance
, instance
, _instance
);
299 if (instance
->physicalDeviceCount
< 0) {
300 result
= anv_physical_device_init(&instance
->physicalDevice
,
301 instance
, "/dev/dri/renderD128");
302 if (result
== VK_ERROR_INCOMPATIBLE_DRIVER
) {
303 instance
->physicalDeviceCount
= 0;
304 } else if (result
== VK_SUCCESS
) {
305 instance
->physicalDeviceCount
= 1;
311 /* pPhysicalDeviceCount is an out parameter if pPhysicalDevices is NULL;
312 * otherwise it's an inout parameter.
314 * The Vulkan spec (git aaed022) says:
316 * pPhysicalDeviceCount is a pointer to an unsigned integer variable
317 * that is initialized with the number of devices the application is
318 * prepared to receive handles to. pname:pPhysicalDevices is pointer to
319 * an array of at least this many VkPhysicalDevice handles [...].
321 * Upon success, if pPhysicalDevices is NULL, vkEnumeratePhysicalDevices
322 * overwrites the contents of the variable pointed to by
323 * pPhysicalDeviceCount with the number of physical devices in in the
324 * instance; otherwise, vkEnumeratePhysicalDevices overwrites
325 * pPhysicalDeviceCount with the number of physical handles written to
328 if (!pPhysicalDevices
) {
329 *pPhysicalDeviceCount
= instance
->physicalDeviceCount
;
330 } else if (*pPhysicalDeviceCount
>= 1) {
331 pPhysicalDevices
[0] = anv_physical_device_to_handle(&instance
->physicalDevice
);
332 *pPhysicalDeviceCount
= 1;
334 *pPhysicalDeviceCount
= 0;
340 void anv_GetPhysicalDeviceFeatures(
341 VkPhysicalDevice physicalDevice
,
342 VkPhysicalDeviceFeatures
* pFeatures
)
344 ANV_FROM_HANDLE(anv_physical_device
, pdevice
, physicalDevice
);
346 *pFeatures
= (VkPhysicalDeviceFeatures
) {
347 .robustBufferAccess
= true,
348 .fullDrawIndexUint32
= true,
349 .imageCubeArray
= false,
350 .independentBlend
= pdevice
->info
->gen
>= 8,
351 .geometryShader
= true,
352 .tessellationShader
= false,
353 .sampleRateShading
= false,
354 .dualSrcBlend
= true,
356 .multiDrawIndirect
= false,
357 .drawIndirectFirstInstance
= false,
359 .depthBiasClamp
= false,
360 .fillModeNonSolid
= true,
361 .depthBounds
= false,
365 .multiViewport
= true,
366 .samplerAnisotropy
= false, /* FINISHME */
367 .textureCompressionETC2
= true,
368 .textureCompressionASTC_LDR
= true,
369 .textureCompressionBC
= true,
370 .occlusionQueryPrecise
= true,
371 .pipelineStatisticsQuery
= true,
372 .vertexPipelineStoresAndAtomics
= pdevice
->info
->gen
>= 8,
373 .fragmentStoresAndAtomics
= true,
374 .shaderTessellationAndGeometryPointSize
= true,
375 .shaderImageGatherExtended
= true,
376 .shaderStorageImageExtendedFormats
= false,
377 .shaderStorageImageMultisample
= false,
378 .shaderUniformBufferArrayDynamicIndexing
= true,
379 .shaderSampledImageArrayDynamicIndexing
= true,
380 .shaderStorageBufferArrayDynamicIndexing
= true,
381 .shaderStorageImageArrayDynamicIndexing
= true,
382 .shaderStorageImageReadWithoutFormat
= false,
383 .shaderStorageImageWriteWithoutFormat
= true,
384 .shaderClipDistance
= false,
385 .shaderCullDistance
= false,
386 .shaderFloat64
= false,
387 .shaderInt64
= false,
388 .shaderInt16
= false,
390 .variableMultisampleRate
= false,
391 .inheritedQueries
= false,
396 anv_device_get_cache_uuid(void *uuid
)
398 memset(uuid
, 0, VK_UUID_SIZE
);
399 snprintf(uuid
, VK_UUID_SIZE
, "anv-%s", MESA_GIT_SHA1
+ 4);
402 void anv_GetPhysicalDeviceProperties(
403 VkPhysicalDevice physicalDevice
,
404 VkPhysicalDeviceProperties
* pProperties
)
406 ANV_FROM_HANDLE(anv_physical_device
, pdevice
, physicalDevice
);
407 const struct brw_device_info
*devinfo
= pdevice
->info
;
409 anv_finishme("Get correct values for VkPhysicalDeviceLimits");
411 const float time_stamp_base
= devinfo
->gen
>= 9 ? 83.333 : 80.0;
413 VkSampleCountFlags sample_counts
=
414 isl_device_get_sample_counts(&pdevice
->isl_dev
);
416 VkPhysicalDeviceLimits limits
= {
417 .maxImageDimension1D
= (1 << 14),
418 .maxImageDimension2D
= (1 << 14),
419 .maxImageDimension3D
= (1 << 10),
420 .maxImageDimensionCube
= (1 << 14),
421 .maxImageArrayLayers
= (1 << 10),
422 .maxTexelBufferElements
= 128 * 1024 * 1024,
423 .maxUniformBufferRange
= UINT32_MAX
,
424 .maxStorageBufferRange
= UINT32_MAX
,
425 .maxPushConstantsSize
= MAX_PUSH_CONSTANTS_SIZE
,
426 .maxMemoryAllocationCount
= UINT32_MAX
,
427 .maxSamplerAllocationCount
= 64 * 1024,
428 .bufferImageGranularity
= 64, /* A cache line */
429 .sparseAddressSpaceSize
= 0,
430 .maxBoundDescriptorSets
= MAX_SETS
,
431 .maxPerStageDescriptorSamplers
= 64,
432 .maxPerStageDescriptorUniformBuffers
= 64,
433 .maxPerStageDescriptorStorageBuffers
= 64,
434 .maxPerStageDescriptorSampledImages
= 64,
435 .maxPerStageDescriptorStorageImages
= 64,
436 .maxPerStageDescriptorInputAttachments
= 64,
437 .maxPerStageResources
= 128,
438 .maxDescriptorSetSamplers
= 256,
439 .maxDescriptorSetUniformBuffers
= 256,
440 .maxDescriptorSetUniformBuffersDynamic
= 256,
441 .maxDescriptorSetStorageBuffers
= 256,
442 .maxDescriptorSetStorageBuffersDynamic
= 256,
443 .maxDescriptorSetSampledImages
= 256,
444 .maxDescriptorSetStorageImages
= 256,
445 .maxDescriptorSetInputAttachments
= 256,
446 .maxVertexInputAttributes
= 32,
447 .maxVertexInputBindings
= 32,
448 .maxVertexInputAttributeOffset
= 2047,
449 .maxVertexInputBindingStride
= 2048,
450 .maxVertexOutputComponents
= 128,
451 .maxTessellationGenerationLevel
= 0,
452 .maxTessellationPatchSize
= 0,
453 .maxTessellationControlPerVertexInputComponents
= 0,
454 .maxTessellationControlPerVertexOutputComponents
= 0,
455 .maxTessellationControlPerPatchOutputComponents
= 0,
456 .maxTessellationControlTotalOutputComponents
= 0,
457 .maxTessellationEvaluationInputComponents
= 0,
458 .maxTessellationEvaluationOutputComponents
= 0,
459 .maxGeometryShaderInvocations
= 32,
460 .maxGeometryInputComponents
= 64,
461 .maxGeometryOutputComponents
= 128,
462 .maxGeometryOutputVertices
= 256,
463 .maxGeometryTotalOutputComponents
= 1024,
464 .maxFragmentInputComponents
= 128,
465 .maxFragmentOutputAttachments
= 8,
466 .maxFragmentDualSrcAttachments
= 2,
467 .maxFragmentCombinedOutputResources
= 8,
468 .maxComputeSharedMemorySize
= 32768,
469 .maxComputeWorkGroupCount
= { 65535, 65535, 65535 },
470 .maxComputeWorkGroupInvocations
= 16 * devinfo
->max_cs_threads
,
471 .maxComputeWorkGroupSize
= {
472 16 * devinfo
->max_cs_threads
,
473 16 * devinfo
->max_cs_threads
,
474 16 * devinfo
->max_cs_threads
,
476 .subPixelPrecisionBits
= 4 /* FIXME */,
477 .subTexelPrecisionBits
= 4 /* FIXME */,
478 .mipmapPrecisionBits
= 4 /* FIXME */,
479 .maxDrawIndexedIndexValue
= UINT32_MAX
,
480 .maxDrawIndirectCount
= UINT32_MAX
,
481 .maxSamplerLodBias
= 16,
482 .maxSamplerAnisotropy
= 16,
483 .maxViewports
= MAX_VIEWPORTS
,
484 .maxViewportDimensions
= { (1 << 14), (1 << 14) },
485 .viewportBoundsRange
= { -16384.0, 16384.0 },
486 .viewportSubPixelBits
= 13, /* We take a float? */
487 .minMemoryMapAlignment
= 4096, /* A page */
488 .minTexelBufferOffsetAlignment
= 1,
489 .minUniformBufferOffsetAlignment
= 1,
490 .minStorageBufferOffsetAlignment
= 1,
491 .minTexelOffset
= -8,
493 .minTexelGatherOffset
= -8,
494 .maxTexelGatherOffset
= 7,
495 .minInterpolationOffset
= 0, /* FIXME */
496 .maxInterpolationOffset
= 0, /* FIXME */
497 .subPixelInterpolationOffsetBits
= 0, /* FIXME */
498 .maxFramebufferWidth
= (1 << 14),
499 .maxFramebufferHeight
= (1 << 14),
500 .maxFramebufferLayers
= (1 << 10),
501 .framebufferColorSampleCounts
= sample_counts
,
502 .framebufferDepthSampleCounts
= sample_counts
,
503 .framebufferStencilSampleCounts
= sample_counts
,
504 .framebufferNoAttachmentsSampleCounts
= sample_counts
,
505 .maxColorAttachments
= MAX_RTS
,
506 .sampledImageColorSampleCounts
= sample_counts
,
507 .sampledImageIntegerSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
508 .sampledImageDepthSampleCounts
= sample_counts
,
509 .sampledImageStencilSampleCounts
= sample_counts
,
510 .storageImageSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
511 .maxSampleMaskWords
= 1,
512 .timestampComputeAndGraphics
= false,
513 .timestampPeriod
= time_stamp_base
/ (1000 * 1000 * 1000),
514 .maxClipDistances
= 0 /* FIXME */,
515 .maxCullDistances
= 0 /* FIXME */,
516 .maxCombinedClipAndCullDistances
= 0 /* FIXME */,
517 .discreteQueuePriorities
= 1,
518 .pointSizeRange
= { 0.125, 255.875 },
519 .lineWidthRange
= { 0.0, 7.9921875 },
520 .pointSizeGranularity
= (1.0 / 8.0),
521 .lineWidthGranularity
= (1.0 / 128.0),
522 .strictLines
= false, /* FINISHME */
523 .standardSampleLocations
= true,
524 .optimalBufferCopyOffsetAlignment
= 128,
525 .optimalBufferCopyRowPitchAlignment
= 128,
526 .nonCoherentAtomSize
= 64,
529 *pProperties
= (VkPhysicalDeviceProperties
) {
530 .apiVersion
= VK_MAKE_VERSION(1, 0, 2),
533 .deviceID
= pdevice
->chipset_id
,
534 .deviceType
= VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
,
536 .sparseProperties
= {0}, /* Broadwell doesn't do sparse. */
539 strcpy(pProperties
->deviceName
, pdevice
->name
);
540 anv_device_get_cache_uuid(pProperties
->pipelineCacheUUID
);
543 void anv_GetPhysicalDeviceQueueFamilyProperties(
544 VkPhysicalDevice physicalDevice
,
546 VkQueueFamilyProperties
* pQueueFamilyProperties
)
548 if (pQueueFamilyProperties
== NULL
) {
553 assert(*pCount
>= 1);
555 *pQueueFamilyProperties
= (VkQueueFamilyProperties
) {
556 .queueFlags
= VK_QUEUE_GRAPHICS_BIT
|
557 VK_QUEUE_COMPUTE_BIT
|
558 VK_QUEUE_TRANSFER_BIT
,
560 .timestampValidBits
= 36, /* XXX: Real value here */
561 .minImageTransferGranularity
= (VkExtent3D
) { 1, 1, 1 },
565 void anv_GetPhysicalDeviceMemoryProperties(
566 VkPhysicalDevice physicalDevice
,
567 VkPhysicalDeviceMemoryProperties
* pMemoryProperties
)
569 ANV_FROM_HANDLE(anv_physical_device
, physical_device
, physicalDevice
);
570 VkDeviceSize heap_size
;
572 /* Reserve some wiggle room for the driver by exposing only 75% of the
573 * aperture to the heap.
575 heap_size
= 3 * physical_device
->aperture_size
/ 4;
577 if (physical_device
->info
->has_llc
) {
578 /* Big core GPUs share LLC with the CPU and thus one memory type can be
579 * both cached and coherent at the same time.
581 pMemoryProperties
->memoryTypeCount
= 1;
582 pMemoryProperties
->memoryTypes
[0] = (VkMemoryType
) {
583 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
584 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
585 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
|
586 VK_MEMORY_PROPERTY_HOST_CACHED_BIT
,
590 /* The spec requires that we expose a host-visible, coherent memory
591 * type, but Atom GPUs don't share LLC. Thus we offer two memory types
592 * to give the application a choice between cached, but not coherent and
593 * coherent but uncached (WC though).
595 pMemoryProperties
->memoryTypeCount
= 2;
596 pMemoryProperties
->memoryTypes
[0] = (VkMemoryType
) {
597 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
598 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
599 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
,
602 pMemoryProperties
->memoryTypes
[1] = (VkMemoryType
) {
603 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
604 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
605 VK_MEMORY_PROPERTY_HOST_CACHED_BIT
,
610 pMemoryProperties
->memoryHeapCount
= 1;
611 pMemoryProperties
->memoryHeaps
[0] = (VkMemoryHeap
) {
613 .flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
,
617 PFN_vkVoidFunction
anv_GetInstanceProcAddr(
621 return anv_lookup_entrypoint(pName
);
624 /* The loader wants us to expose a second GetInstanceProcAddr function
625 * to work around certain LD_PRELOAD issues seen in apps.
627 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
631 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
635 return anv_GetInstanceProcAddr(instance
, pName
);
638 PFN_vkVoidFunction
anv_GetDeviceProcAddr(
642 return anv_lookup_entrypoint(pName
);
646 anv_queue_init(struct anv_device
*device
, struct anv_queue
*queue
)
648 queue
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
649 queue
->device
= device
;
650 queue
->pool
= &device
->surface_state_pool
;
656 anv_queue_finish(struct anv_queue
*queue
)
660 static struct anv_state
661 anv_state_pool_emit_data(struct anv_state_pool
*pool
, size_t size
, size_t align
, const void *p
)
663 struct anv_state state
;
665 state
= anv_state_pool_alloc(pool
, size
, align
);
666 memcpy(state
.map
, p
, size
);
668 if (!pool
->block_pool
->device
->info
.has_llc
)
669 anv_state_clflush(state
);
674 struct gen8_border_color
{
679 /* Pad out to 64 bytes */
684 anv_device_init_border_colors(struct anv_device
*device
)
686 static const struct gen8_border_color border_colors
[] = {
687 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
] = { .float32
= { 0.0, 0.0, 0.0, 0.0 } },
688 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
] = { .float32
= { 0.0, 0.0, 0.0, 1.0 } },
689 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
] = { .float32
= { 1.0, 1.0, 1.0, 1.0 } },
690 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
] = { .uint32
= { 0, 0, 0, 0 } },
691 [VK_BORDER_COLOR_INT_OPAQUE_BLACK
] = { .uint32
= { 0, 0, 0, 1 } },
692 [VK_BORDER_COLOR_INT_OPAQUE_WHITE
] = { .uint32
= { 1, 1, 1, 1 } },
695 device
->border_colors
= anv_state_pool_emit_data(&device
->dynamic_state_pool
,
696 sizeof(border_colors
), 64,
701 anv_device_submit_simple_batch(struct anv_device
*device
,
702 struct anv_batch
*batch
)
704 struct drm_i915_gem_execbuffer2 execbuf
;
705 struct drm_i915_gem_exec_object2 exec2_objects
[1];
707 VkResult result
= VK_SUCCESS
;
712 /* Kernel driver requires 8 byte aligned batch length */
713 size
= align_u32(batch
->next
- batch
->start
, 8);
714 assert(size
< device
->batch_bo_pool
.bo_size
);
715 result
= anv_bo_pool_alloc(&device
->batch_bo_pool
, &bo
);
716 if (result
!= VK_SUCCESS
)
719 memcpy(bo
.map
, batch
->start
, size
);
720 if (!device
->info
.has_llc
)
721 anv_clflush_range(bo
.map
, size
);
723 exec2_objects
[0].handle
= bo
.gem_handle
;
724 exec2_objects
[0].relocation_count
= 0;
725 exec2_objects
[0].relocs_ptr
= 0;
726 exec2_objects
[0].alignment
= 0;
727 exec2_objects
[0].offset
= bo
.offset
;
728 exec2_objects
[0].flags
= 0;
729 exec2_objects
[0].rsvd1
= 0;
730 exec2_objects
[0].rsvd2
= 0;
732 execbuf
.buffers_ptr
= (uintptr_t) exec2_objects
;
733 execbuf
.buffer_count
= 1;
734 execbuf
.batch_start_offset
= 0;
735 execbuf
.batch_len
= size
;
736 execbuf
.cliprects_ptr
= 0;
737 execbuf
.num_cliprects
= 0;
742 I915_EXEC_HANDLE_LUT
| I915_EXEC_NO_RELOC
| I915_EXEC_RENDER
;
743 execbuf
.rsvd1
= device
->context_id
;
746 ret
= anv_gem_execbuffer(device
, &execbuf
);
748 /* We don't know the real error. */
749 result
= vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
, "execbuf2 failed: %m");
754 ret
= anv_gem_wait(device
, bo
.gem_handle
, &timeout
);
756 /* We don't know the real error. */
757 result
= vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
, "execbuf2 failed: %m");
762 anv_bo_pool_free(&device
->batch_bo_pool
, &bo
);
767 VkResult
anv_CreateDevice(
768 VkPhysicalDevice physicalDevice
,
769 const VkDeviceCreateInfo
* pCreateInfo
,
770 const VkAllocationCallbacks
* pAllocator
,
773 ANV_FROM_HANDLE(anv_physical_device
, physical_device
, physicalDevice
);
775 struct anv_device
*device
;
777 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO
);
779 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
781 for (uint32_t j
= 0; j
< ARRAY_SIZE(device_extensions
); j
++) {
782 if (strcmp(pCreateInfo
->ppEnabledExtensionNames
[i
],
783 device_extensions
[j
].extensionName
) == 0) {
789 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
792 anv_set_dispatch_devinfo(physical_device
->info
);
794 device
= anv_alloc2(&physical_device
->instance
->alloc
, pAllocator
,
796 VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
798 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
800 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
801 device
->instance
= physical_device
->instance
;
802 device
->chipset_id
= physical_device
->chipset_id
;
805 device
->alloc
= *pAllocator
;
807 device
->alloc
= physical_device
->instance
->alloc
;
809 /* XXX(chadv): Can we dup() physicalDevice->fd here? */
810 device
->fd
= open(physical_device
->path
, O_RDWR
| O_CLOEXEC
);
811 if (device
->fd
== -1) {
812 result
= vk_error(VK_ERROR_INITIALIZATION_FAILED
);
816 device
->context_id
= anv_gem_create_context(device
);
817 if (device
->context_id
== -1) {
818 result
= vk_error(VK_ERROR_INITIALIZATION_FAILED
);
822 device
->info
= *physical_device
->info
;
823 device
->isl_dev
= physical_device
->isl_dev
;
825 pthread_mutex_init(&device
->mutex
, NULL
);
827 anv_bo_pool_init(&device
->batch_bo_pool
, device
, ANV_CMD_BUFFER_BATCH_SIZE
);
829 anv_block_pool_init(&device
->dynamic_state_block_pool
, device
, 16384);
831 anv_state_pool_init(&device
->dynamic_state_pool
,
832 &device
->dynamic_state_block_pool
);
834 anv_block_pool_init(&device
->instruction_block_pool
, device
, 128 * 1024);
835 anv_pipeline_cache_init(&device
->default_pipeline_cache
, device
);
837 anv_block_pool_init(&device
->surface_state_block_pool
, device
, 4096);
839 anv_state_pool_init(&device
->surface_state_pool
,
840 &device
->surface_state_block_pool
);
842 anv_bo_init_new(&device
->workaround_bo
, device
, 1024);
844 anv_block_pool_init(&device
->scratch_block_pool
, device
, 0x10000);
846 anv_queue_init(device
, &device
->queue
);
848 switch (device
->info
.gen
) {
850 if (!device
->info
.is_haswell
)
851 result
= gen7_init_device_state(device
);
853 result
= gen75_init_device_state(device
);
856 result
= gen8_init_device_state(device
);
859 result
= gen9_init_device_state(device
);
862 /* Shouldn't get here as we don't create physical devices for any other
864 unreachable("unhandled gen");
866 if (result
!= VK_SUCCESS
)
869 result
= anv_device_init_meta(device
);
870 if (result
!= VK_SUCCESS
)
873 anv_device_init_border_colors(device
);
875 *pDevice
= anv_device_to_handle(device
);
882 anv_free(&device
->alloc
, device
);
887 void anv_DestroyDevice(
889 const VkAllocationCallbacks
* pAllocator
)
891 ANV_FROM_HANDLE(anv_device
, device
, _device
);
893 anv_queue_finish(&device
->queue
);
895 anv_device_finish_meta(device
);
898 /* We only need to free these to prevent valgrind errors. The backing
899 * BO will go away in a couple of lines so we don't actually leak.
901 anv_state_pool_free(&device
->dynamic_state_pool
, device
->border_colors
);
904 anv_gem_munmap(device
->workaround_bo
.map
, device
->workaround_bo
.size
);
905 anv_gem_close(device
, device
->workaround_bo
.gem_handle
);
907 anv_bo_pool_finish(&device
->batch_bo_pool
);
908 anv_state_pool_finish(&device
->dynamic_state_pool
);
909 anv_block_pool_finish(&device
->dynamic_state_block_pool
);
910 anv_block_pool_finish(&device
->instruction_block_pool
);
911 anv_state_pool_finish(&device
->surface_state_pool
);
912 anv_block_pool_finish(&device
->surface_state_block_pool
);
913 anv_block_pool_finish(&device
->scratch_block_pool
);
917 pthread_mutex_destroy(&device
->mutex
);
919 anv_free(&device
->alloc
, device
);
922 VkResult
anv_EnumerateInstanceExtensionProperties(
923 const char* pLayerName
,
924 uint32_t* pPropertyCount
,
925 VkExtensionProperties
* pProperties
)
927 if (pProperties
== NULL
) {
928 *pPropertyCount
= ARRAY_SIZE(global_extensions
);
932 assert(*pPropertyCount
>= ARRAY_SIZE(global_extensions
));
934 *pPropertyCount
= ARRAY_SIZE(global_extensions
);
935 memcpy(pProperties
, global_extensions
, sizeof(global_extensions
));
940 VkResult
anv_EnumerateDeviceExtensionProperties(
941 VkPhysicalDevice physicalDevice
,
942 const char* pLayerName
,
943 uint32_t* pPropertyCount
,
944 VkExtensionProperties
* pProperties
)
946 if (pProperties
== NULL
) {
947 *pPropertyCount
= ARRAY_SIZE(device_extensions
);
951 assert(*pPropertyCount
>= ARRAY_SIZE(device_extensions
));
953 *pPropertyCount
= ARRAY_SIZE(device_extensions
);
954 memcpy(pProperties
, device_extensions
, sizeof(device_extensions
));
959 VkResult
anv_EnumerateInstanceLayerProperties(
960 uint32_t* pPropertyCount
,
961 VkLayerProperties
* pProperties
)
963 if (pProperties
== NULL
) {
968 /* None supported at this time */
969 return vk_error(VK_ERROR_LAYER_NOT_PRESENT
);
972 VkResult
anv_EnumerateDeviceLayerProperties(
973 VkPhysicalDevice physicalDevice
,
974 uint32_t* pPropertyCount
,
975 VkLayerProperties
* pProperties
)
977 if (pProperties
== NULL
) {
982 /* None supported at this time */
983 return vk_error(VK_ERROR_LAYER_NOT_PRESENT
);
986 void anv_GetDeviceQueue(
988 uint32_t queueNodeIndex
,
992 ANV_FROM_HANDLE(anv_device
, device
, _device
);
994 assert(queueIndex
== 0);
996 *pQueue
= anv_queue_to_handle(&device
->queue
);
999 VkResult
anv_QueueSubmit(
1001 uint32_t submitCount
,
1002 const VkSubmitInfo
* pSubmits
,
1005 ANV_FROM_HANDLE(anv_queue
, queue
, _queue
);
1006 ANV_FROM_HANDLE(anv_fence
, fence
, _fence
);
1007 struct anv_device
*device
= queue
->device
;
1010 for (uint32_t i
= 0; i
< submitCount
; i
++) {
1011 for (uint32_t j
= 0; j
< pSubmits
[i
].commandBufferCount
; j
++) {
1012 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
,
1013 pSubmits
[i
].pCommandBuffers
[j
]);
1014 assert(cmd_buffer
->level
== VK_COMMAND_BUFFER_LEVEL_PRIMARY
);
1016 ret
= anv_gem_execbuffer(device
, &cmd_buffer
->execbuf2
.execbuf
);
1018 /* We don't know the real error. */
1019 return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
,
1020 "execbuf2 failed: %m");
1023 for (uint32_t k
= 0; k
< cmd_buffer
->execbuf2
.bo_count
; k
++)
1024 cmd_buffer
->execbuf2
.bos
[k
]->offset
= cmd_buffer
->execbuf2
.objects
[k
].offset
;
1029 ret
= anv_gem_execbuffer(device
, &fence
->execbuf
);
1031 /* We don't know the real error. */
1032 return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
,
1033 "execbuf2 failed: %m");
1040 VkResult
anv_QueueWaitIdle(
1043 ANV_FROM_HANDLE(anv_queue
, queue
, _queue
);
1045 return ANV_CALL(DeviceWaitIdle
)(anv_device_to_handle(queue
->device
));
1048 VkResult
anv_DeviceWaitIdle(
1051 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1052 struct anv_batch batch
;
1055 batch
.start
= batch
.next
= cmds
;
1056 batch
.end
= (void *) cmds
+ sizeof(cmds
);
1058 anv_batch_emit(&batch
, GEN7_MI_BATCH_BUFFER_END
);
1059 anv_batch_emit(&batch
, GEN7_MI_NOOP
);
1061 return anv_device_submit_simple_batch(device
, &batch
);
1065 anv_bo_init_new(struct anv_bo
*bo
, struct anv_device
*device
, uint64_t size
)
1067 bo
->gem_handle
= anv_gem_create(device
, size
);
1068 if (!bo
->gem_handle
)
1069 return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY
);
1075 bo
->is_winsys_bo
= false;
1080 VkResult
anv_AllocateMemory(
1082 const VkMemoryAllocateInfo
* pAllocateInfo
,
1083 const VkAllocationCallbacks
* pAllocator
,
1084 VkDeviceMemory
* pMem
)
1086 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1087 struct anv_device_memory
*mem
;
1090 assert(pAllocateInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO
);
1092 if (pAllocateInfo
->allocationSize
== 0) {
1093 /* Apparently, this is allowed */
1094 *pMem
= VK_NULL_HANDLE
;
1098 /* We support exactly one memory heap. */
1099 assert(pAllocateInfo
->memoryTypeIndex
== 0 ||
1100 (!device
->info
.has_llc
&& pAllocateInfo
->memoryTypeIndex
< 2));
1102 /* FINISHME: Fail if allocation request exceeds heap size. */
1104 mem
= anv_alloc2(&device
->alloc
, pAllocator
, sizeof(*mem
), 8,
1105 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1107 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1109 /* The kernel is going to give us whole pages anyway */
1110 uint64_t alloc_size
= align_u64(pAllocateInfo
->allocationSize
, 4096);
1112 result
= anv_bo_init_new(&mem
->bo
, device
, alloc_size
);
1113 if (result
!= VK_SUCCESS
)
1116 mem
->type_index
= pAllocateInfo
->memoryTypeIndex
;
1118 *pMem
= anv_device_memory_to_handle(mem
);
1123 anv_free2(&device
->alloc
, pAllocator
, mem
);
1128 void anv_FreeMemory(
1130 VkDeviceMemory _mem
,
1131 const VkAllocationCallbacks
* pAllocator
)
1133 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1134 ANV_FROM_HANDLE(anv_device_memory
, mem
, _mem
);
1140 anv_gem_munmap(mem
->bo
.map
, mem
->bo
.size
);
1142 if (mem
->bo
.gem_handle
!= 0)
1143 anv_gem_close(device
, mem
->bo
.gem_handle
);
1145 anv_free2(&device
->alloc
, pAllocator
, mem
);
1148 VkResult
anv_MapMemory(
1150 VkDeviceMemory _memory
,
1151 VkDeviceSize offset
,
1153 VkMemoryMapFlags flags
,
1156 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1157 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
1164 if (size
== VK_WHOLE_SIZE
)
1165 size
= mem
->bo
.size
- offset
;
1167 /* FIXME: Is this supposed to be thread safe? Since vkUnmapMemory() only
1168 * takes a VkDeviceMemory pointer, it seems like only one map of the memory
1169 * at a time is valid. We could just mmap up front and return an offset
1170 * pointer here, but that may exhaust virtual memory on 32 bit
1173 uint32_t gem_flags
= 0;
1174 if (!device
->info
.has_llc
&& mem
->type_index
== 0)
1175 gem_flags
|= I915_MMAP_WC
;
1177 /* GEM will fail to map if the offset isn't 4k-aligned. Round down. */
1178 uint64_t map_offset
= offset
& ~4095ull;
1179 assert(offset
>= map_offset
);
1180 uint64_t map_size
= (offset
+ size
) - map_offset
;
1182 /* Let's map whole pages */
1183 map_size
= align_u64(map_size
, 4096);
1185 mem
->map
= anv_gem_mmap(device
, mem
->bo
.gem_handle
,
1186 map_offset
, map_size
, gem_flags
);
1187 mem
->map_size
= map_size
;
1189 *ppData
= mem
->map
+ (offset
- map_offset
);
1194 void anv_UnmapMemory(
1196 VkDeviceMemory _memory
)
1198 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
1203 anv_gem_munmap(mem
->map
, mem
->map_size
);
1207 clflush_mapped_ranges(struct anv_device
*device
,
1209 const VkMappedMemoryRange
*ranges
)
1211 for (uint32_t i
= 0; i
< count
; i
++) {
1212 ANV_FROM_HANDLE(anv_device_memory
, mem
, ranges
[i
].memory
);
1213 void *p
= mem
->map
+ (ranges
[i
].offset
& ~CACHELINE_MASK
);
1216 if (ranges
[i
].offset
+ ranges
[i
].size
> mem
->map_size
)
1217 end
= mem
->map
+ mem
->map_size
;
1219 end
= mem
->map
+ ranges
[i
].offset
+ ranges
[i
].size
;
1222 __builtin_ia32_clflush(p
);
1223 p
+= CACHELINE_SIZE
;
1228 VkResult
anv_FlushMappedMemoryRanges(
1230 uint32_t memoryRangeCount
,
1231 const VkMappedMemoryRange
* pMemoryRanges
)
1233 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1235 if (device
->info
.has_llc
)
1238 /* Make sure the writes we're flushing have landed. */
1239 __builtin_ia32_mfence();
1241 clflush_mapped_ranges(device
, memoryRangeCount
, pMemoryRanges
);
1246 VkResult
anv_InvalidateMappedMemoryRanges(
1248 uint32_t memoryRangeCount
,
1249 const VkMappedMemoryRange
* pMemoryRanges
)
1251 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1253 if (device
->info
.has_llc
)
1256 clflush_mapped_ranges(device
, memoryRangeCount
, pMemoryRanges
);
1258 /* Make sure no reads get moved up above the invalidate. */
1259 __builtin_ia32_mfence();
1264 void anv_GetBufferMemoryRequirements(
1267 VkMemoryRequirements
* pMemoryRequirements
)
1269 ANV_FROM_HANDLE(anv_buffer
, buffer
, _buffer
);
1271 /* The Vulkan spec (git aaed022) says:
1273 * memoryTypeBits is a bitfield and contains one bit set for every
1274 * supported memory type for the resource. The bit `1<<i` is set if and
1275 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
1276 * structure for the physical device is supported.
1278 * We support exactly one memory type.
1280 pMemoryRequirements
->memoryTypeBits
= 1;
1282 pMemoryRequirements
->size
= buffer
->size
;
1283 pMemoryRequirements
->alignment
= 16;
1286 void anv_GetImageMemoryRequirements(
1289 VkMemoryRequirements
* pMemoryRequirements
)
1291 ANV_FROM_HANDLE(anv_image
, image
, _image
);
1293 /* The Vulkan spec (git aaed022) says:
1295 * memoryTypeBits is a bitfield and contains one bit set for every
1296 * supported memory type for the resource. The bit `1<<i` is set if and
1297 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
1298 * structure for the physical device is supported.
1300 * We support exactly one memory type.
1302 pMemoryRequirements
->memoryTypeBits
= 1;
1304 pMemoryRequirements
->size
= image
->size
;
1305 pMemoryRequirements
->alignment
= image
->alignment
;
1308 void anv_GetImageSparseMemoryRequirements(
1311 uint32_t* pSparseMemoryRequirementCount
,
1312 VkSparseImageMemoryRequirements
* pSparseMemoryRequirements
)
1317 void anv_GetDeviceMemoryCommitment(
1319 VkDeviceMemory memory
,
1320 VkDeviceSize
* pCommittedMemoryInBytes
)
1322 *pCommittedMemoryInBytes
= 0;
1325 VkResult
anv_BindBufferMemory(
1328 VkDeviceMemory _memory
,
1329 VkDeviceSize memoryOffset
)
1331 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
1332 ANV_FROM_HANDLE(anv_buffer
, buffer
, _buffer
);
1335 buffer
->bo
= &mem
->bo
;
1336 buffer
->offset
= memoryOffset
;
1345 VkResult
anv_BindImageMemory(
1348 VkDeviceMemory _memory
,
1349 VkDeviceSize memoryOffset
)
1351 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
1352 ANV_FROM_HANDLE(anv_image
, image
, _image
);
1355 image
->bo
= &mem
->bo
;
1356 image
->offset
= memoryOffset
;
1365 VkResult
anv_QueueBindSparse(
1367 uint32_t bindInfoCount
,
1368 const VkBindSparseInfo
* pBindInfo
,
1371 stub_return(VK_ERROR_INCOMPATIBLE_DRIVER
);
1374 VkResult
anv_CreateFence(
1376 const VkFenceCreateInfo
* pCreateInfo
,
1377 const VkAllocationCallbacks
* pAllocator
,
1380 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1381 struct anv_fence
*fence
;
1382 struct anv_batch batch
;
1385 const uint32_t fence_size
= 128;
1387 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FENCE_CREATE_INFO
);
1389 fence
= anv_alloc2(&device
->alloc
, pAllocator
, sizeof(*fence
), 8,
1390 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1392 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1394 result
= anv_bo_init_new(&fence
->bo
, device
, fence_size
);
1395 if (result
!= VK_SUCCESS
)
1399 anv_gem_mmap(device
, fence
->bo
.gem_handle
, 0, fence
->bo
.size
, 0);
1400 batch
.next
= batch
.start
= fence
->bo
.map
;
1401 batch
.end
= fence
->bo
.map
+ fence
->bo
.size
;
1402 anv_batch_emit(&batch
, GEN7_MI_BATCH_BUFFER_END
);
1403 anv_batch_emit(&batch
, GEN7_MI_NOOP
);
1405 if (!device
->info
.has_llc
) {
1406 assert(((uintptr_t) fence
->bo
.map
& CACHELINE_MASK
) == 0);
1407 assert(batch
.next
- fence
->bo
.map
<= CACHELINE_SIZE
);
1408 __builtin_ia32_mfence();
1409 __builtin_ia32_clflush(fence
->bo
.map
);
1412 fence
->exec2_objects
[0].handle
= fence
->bo
.gem_handle
;
1413 fence
->exec2_objects
[0].relocation_count
= 0;
1414 fence
->exec2_objects
[0].relocs_ptr
= 0;
1415 fence
->exec2_objects
[0].alignment
= 0;
1416 fence
->exec2_objects
[0].offset
= fence
->bo
.offset
;
1417 fence
->exec2_objects
[0].flags
= 0;
1418 fence
->exec2_objects
[0].rsvd1
= 0;
1419 fence
->exec2_objects
[0].rsvd2
= 0;
1421 fence
->execbuf
.buffers_ptr
= (uintptr_t) fence
->exec2_objects
;
1422 fence
->execbuf
.buffer_count
= 1;
1423 fence
->execbuf
.batch_start_offset
= 0;
1424 fence
->execbuf
.batch_len
= batch
.next
- fence
->bo
.map
;
1425 fence
->execbuf
.cliprects_ptr
= 0;
1426 fence
->execbuf
.num_cliprects
= 0;
1427 fence
->execbuf
.DR1
= 0;
1428 fence
->execbuf
.DR4
= 0;
1430 fence
->execbuf
.flags
=
1431 I915_EXEC_HANDLE_LUT
| I915_EXEC_NO_RELOC
| I915_EXEC_RENDER
;
1432 fence
->execbuf
.rsvd1
= device
->context_id
;
1433 fence
->execbuf
.rsvd2
= 0;
1435 fence
->ready
= false;
1437 *pFence
= anv_fence_to_handle(fence
);
1442 anv_free2(&device
->alloc
, pAllocator
, fence
);
1447 void anv_DestroyFence(
1450 const VkAllocationCallbacks
* pAllocator
)
1452 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1453 ANV_FROM_HANDLE(anv_fence
, fence
, _fence
);
1455 anv_gem_munmap(fence
->bo
.map
, fence
->bo
.size
);
1456 anv_gem_close(device
, fence
->bo
.gem_handle
);
1457 anv_free2(&device
->alloc
, pAllocator
, fence
);
1460 VkResult
anv_ResetFences(
1462 uint32_t fenceCount
,
1463 const VkFence
* pFences
)
1465 for (uint32_t i
= 0; i
< fenceCount
; i
++) {
1466 ANV_FROM_HANDLE(anv_fence
, fence
, pFences
[i
]);
1467 fence
->ready
= false;
1473 VkResult
anv_GetFenceStatus(
1477 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1478 ANV_FROM_HANDLE(anv_fence
, fence
, _fence
);
1485 ret
= anv_gem_wait(device
, fence
->bo
.gem_handle
, &t
);
1487 fence
->ready
= true;
1491 return VK_NOT_READY
;
1494 VkResult
anv_WaitForFences(
1496 uint32_t fenceCount
,
1497 const VkFence
* pFences
,
1501 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1503 /* DRM_IOCTL_I915_GEM_WAIT uses a signed 64 bit timeout and is supposed
1504 * to block indefinitely timeouts <= 0. Unfortunately, this was broken
1505 * for a couple of kernel releases. Since there's no way to know
1506 * whether or not the kernel we're using is one of the broken ones, the
1507 * best we can do is to clamp the timeout to INT64_MAX. This limits the
1508 * maximum timeout from 584 years to 292 years - likely not a big deal.
1510 if (timeout
> INT64_MAX
)
1511 timeout
= INT64_MAX
;
1513 int64_t t
= timeout
;
1515 /* FIXME: handle !waitAll */
1517 for (uint32_t i
= 0; i
< fenceCount
; i
++) {
1518 ANV_FROM_HANDLE(anv_fence
, fence
, pFences
[i
]);
1519 int ret
= anv_gem_wait(device
, fence
->bo
.gem_handle
, &t
);
1520 if (ret
== -1 && errno
== ETIME
) {
1522 } else if (ret
== -1) {
1523 /* We don't know the real error. */
1524 return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
,
1525 "gem wait failed: %m");
1532 // Queue semaphore functions
1534 VkResult
anv_CreateSemaphore(
1536 const VkSemaphoreCreateInfo
* pCreateInfo
,
1537 const VkAllocationCallbacks
* pAllocator
,
1538 VkSemaphore
* pSemaphore
)
1540 /* The DRM execbuffer ioctl always execute in-oder, even between different
1541 * rings. As such, there's nothing to do for the user space semaphore.
1544 *pSemaphore
= (VkSemaphore
)1;
1549 void anv_DestroySemaphore(
1551 VkSemaphore semaphore
,
1552 const VkAllocationCallbacks
* pAllocator
)
1558 VkResult
anv_CreateEvent(
1560 const VkEventCreateInfo
* pCreateInfo
,
1561 const VkAllocationCallbacks
* pAllocator
,
1564 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1565 struct anv_state state
;
1566 struct anv_event
*event
;
1568 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_EVENT_CREATE_INFO
);
1570 state
= anv_state_pool_alloc(&device
->dynamic_state_pool
,
1573 event
->state
= state
;
1574 event
->semaphore
= VK_EVENT_RESET
;
1576 if (!device
->info
.has_llc
) {
1577 /* Make sure the writes we're flushing have landed. */
1578 __builtin_ia32_mfence();
1579 __builtin_ia32_clflush(event
);
1582 *pEvent
= anv_event_to_handle(event
);
1587 void anv_DestroyEvent(
1590 const VkAllocationCallbacks
* pAllocator
)
1592 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1593 ANV_FROM_HANDLE(anv_event
, event
, _event
);
1595 anv_state_pool_free(&device
->dynamic_state_pool
, event
->state
);
1598 VkResult
anv_GetEventStatus(
1602 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1603 ANV_FROM_HANDLE(anv_event
, event
, _event
);
1605 if (!device
->info
.has_llc
) {
1606 /* Invalidate read cache before reading event written by GPU. */
1607 __builtin_ia32_clflush(event
);
1608 __builtin_ia32_mfence();
1612 return event
->semaphore
;
1615 VkResult
anv_SetEvent(
1619 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1620 ANV_FROM_HANDLE(anv_event
, event
, _event
);
1622 event
->semaphore
= VK_EVENT_SET
;
1624 if (!device
->info
.has_llc
) {
1625 /* Make sure the writes we're flushing have landed. */
1626 __builtin_ia32_mfence();
1627 __builtin_ia32_clflush(event
);
1633 VkResult
anv_ResetEvent(
1637 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1638 ANV_FROM_HANDLE(anv_event
, event
, _event
);
1640 event
->semaphore
= VK_EVENT_RESET
;
1642 if (!device
->info
.has_llc
) {
1643 /* Make sure the writes we're flushing have landed. */
1644 __builtin_ia32_mfence();
1645 __builtin_ia32_clflush(event
);
1653 VkResult
anv_CreateBuffer(
1655 const VkBufferCreateInfo
* pCreateInfo
,
1656 const VkAllocationCallbacks
* pAllocator
,
1659 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1660 struct anv_buffer
*buffer
;
1662 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
1664 buffer
= anv_alloc2(&device
->alloc
, pAllocator
, sizeof(*buffer
), 8,
1665 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1667 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1669 buffer
->size
= pCreateInfo
->size
;
1670 buffer
->usage
= pCreateInfo
->usage
;
1674 *pBuffer
= anv_buffer_to_handle(buffer
);
1679 void anv_DestroyBuffer(
1682 const VkAllocationCallbacks
* pAllocator
)
1684 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1685 ANV_FROM_HANDLE(anv_buffer
, buffer
, _buffer
);
1687 anv_free2(&device
->alloc
, pAllocator
, buffer
);
1691 anv_fill_buffer_surface_state(struct anv_device
*device
, struct anv_state state
,
1692 enum isl_format format
,
1693 uint32_t offset
, uint32_t range
, uint32_t stride
)
1695 switch (device
->info
.gen
) {
1697 if (device
->info
.is_haswell
)
1698 gen75_fill_buffer_surface_state(state
.map
, format
, offset
, range
,
1701 gen7_fill_buffer_surface_state(state
.map
, format
, offset
, range
,
1705 gen8_fill_buffer_surface_state(state
.map
, format
, offset
, range
, stride
);
1708 gen9_fill_buffer_surface_state(state
.map
, format
, offset
, range
, stride
);
1711 unreachable("unsupported gen\n");
1714 if (!device
->info
.has_llc
)
1715 anv_state_clflush(state
);
1718 void anv_DestroySampler(
1721 const VkAllocationCallbacks
* pAllocator
)
1723 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1724 ANV_FROM_HANDLE(anv_sampler
, sampler
, _sampler
);
1726 anv_free2(&device
->alloc
, pAllocator
, sampler
);
1729 VkResult
anv_CreateFramebuffer(
1731 const VkFramebufferCreateInfo
* pCreateInfo
,
1732 const VkAllocationCallbacks
* pAllocator
,
1733 VkFramebuffer
* pFramebuffer
)
1735 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1736 struct anv_framebuffer
*framebuffer
;
1738 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
1740 size_t size
= sizeof(*framebuffer
) +
1741 sizeof(struct anv_image_view
*) * pCreateInfo
->attachmentCount
;
1742 framebuffer
= anv_alloc2(&device
->alloc
, pAllocator
, size
, 8,
1743 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1744 if (framebuffer
== NULL
)
1745 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1747 framebuffer
->attachment_count
= pCreateInfo
->attachmentCount
;
1748 for (uint32_t i
= 0; i
< pCreateInfo
->attachmentCount
; i
++) {
1749 VkImageView _iview
= pCreateInfo
->pAttachments
[i
];
1750 framebuffer
->attachments
[i
] = anv_image_view_from_handle(_iview
);
1753 framebuffer
->width
= pCreateInfo
->width
;
1754 framebuffer
->height
= pCreateInfo
->height
;
1755 framebuffer
->layers
= pCreateInfo
->layers
;
1757 *pFramebuffer
= anv_framebuffer_to_handle(framebuffer
);
1762 void anv_DestroyFramebuffer(
1765 const VkAllocationCallbacks
* pAllocator
)
1767 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1768 ANV_FROM_HANDLE(anv_framebuffer
, fb
, _fb
);
1770 anv_free2(&device
->alloc
, pAllocator
, fb
);
1773 void vkCmdDbgMarkerBegin(
1774 VkCommandBuffer commandBuffer
,
1775 const char* pMarker
)
1776 __attribute__ ((visibility ("default")));
1778 void vkCmdDbgMarkerEnd(
1779 VkCommandBuffer commandBuffer
)
1780 __attribute__ ((visibility ("default")));
1782 void vkCmdDbgMarkerBegin(
1783 VkCommandBuffer commandBuffer
,
1784 const char* pMarker
)
1788 void vkCmdDbgMarkerEnd(
1789 VkCommandBuffer commandBuffer
)