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 "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;
259 VG(VALGRIND_CREATE_MEMPOOL(instance
, 0, false));
261 anv_init_wsi(instance
);
263 *pInstance
= anv_instance_to_handle(instance
);
268 void anv_DestroyInstance(
269 VkInstance _instance
,
270 const VkAllocationCallbacks
* pAllocator
)
272 ANV_FROM_HANDLE(anv_instance
, instance
, _instance
);
274 if (instance
->physicalDeviceCount
> 0) {
275 /* We support at most one physical device. */
276 assert(instance
->physicalDeviceCount
== 1);
277 anv_physical_device_finish(&instance
->physicalDevice
);
280 anv_finish_wsi(instance
);
282 VG(VALGRIND_DESTROY_MEMPOOL(instance
));
286 anv_free(&instance
->alloc
, instance
);
289 VkResult
anv_EnumeratePhysicalDevices(
290 VkInstance _instance
,
291 uint32_t* pPhysicalDeviceCount
,
292 VkPhysicalDevice
* pPhysicalDevices
)
294 ANV_FROM_HANDLE(anv_instance
, instance
, _instance
);
297 if (instance
->physicalDeviceCount
< 0) {
298 result
= anv_physical_device_init(&instance
->physicalDevice
,
299 instance
, "/dev/dri/renderD128");
300 if (result
== VK_ERROR_INCOMPATIBLE_DRIVER
) {
301 instance
->physicalDeviceCount
= 0;
302 } else if (result
== VK_SUCCESS
) {
303 instance
->physicalDeviceCount
= 1;
309 /* pPhysicalDeviceCount is an out parameter if pPhysicalDevices is NULL;
310 * otherwise it's an inout parameter.
312 * The Vulkan spec (git aaed022) says:
314 * pPhysicalDeviceCount is a pointer to an unsigned integer variable
315 * that is initialized with the number of devices the application is
316 * prepared to receive handles to. pname:pPhysicalDevices is pointer to
317 * an array of at least this many VkPhysicalDevice handles [...].
319 * Upon success, if pPhysicalDevices is NULL, vkEnumeratePhysicalDevices
320 * overwrites the contents of the variable pointed to by
321 * pPhysicalDeviceCount with the number of physical devices in in the
322 * instance; otherwise, vkEnumeratePhysicalDevices overwrites
323 * pPhysicalDeviceCount with the number of physical handles written to
326 if (!pPhysicalDevices
) {
327 *pPhysicalDeviceCount
= instance
->physicalDeviceCount
;
328 } else if (*pPhysicalDeviceCount
>= 1) {
329 pPhysicalDevices
[0] = anv_physical_device_to_handle(&instance
->physicalDevice
);
330 *pPhysicalDeviceCount
= 1;
332 *pPhysicalDeviceCount
= 0;
338 void anv_GetPhysicalDeviceFeatures(
339 VkPhysicalDevice physicalDevice
,
340 VkPhysicalDeviceFeatures
* pFeatures
)
342 anv_finishme("Get correct values for PhysicalDeviceFeatures");
344 *pFeatures
= (VkPhysicalDeviceFeatures
) {
345 .robustBufferAccess
= true,
346 .fullDrawIndexUint32
= false,
347 .imageCubeArray
= false,
348 .independentBlend
= false,
349 .geometryShader
= true,
350 .tessellationShader
= false,
351 .sampleRateShading
= false,
352 .dualSrcBlend
= true,
354 .multiDrawIndirect
= false,
355 .drawIndirectFirstInstance
= false,
357 .depthBiasClamp
= false,
358 .fillModeNonSolid
= true,
359 .depthBounds
= false,
363 .multiViewport
= true,
364 .samplerAnisotropy
= false, /* FINISHME */
365 .textureCompressionETC2
= true,
366 .textureCompressionASTC_LDR
= true,
367 .textureCompressionBC
= true,
368 .occlusionQueryPrecise
= false, /* FINISHME */
369 .pipelineStatisticsQuery
= true,
370 .vertexPipelineStoresAndAtomics
= false,
371 .fragmentStoresAndAtomics
= true,
372 .shaderTessellationAndGeometryPointSize
= true,
373 .shaderImageGatherExtended
= true,
374 .shaderStorageImageExtendedFormats
= false,
375 .shaderStorageImageMultisample
= false,
376 .shaderUniformBufferArrayDynamicIndexing
= true,
377 .shaderSampledImageArrayDynamicIndexing
= false,
378 .shaderStorageBufferArrayDynamicIndexing
= false,
379 .shaderStorageImageArrayDynamicIndexing
= false,
380 .shaderStorageImageReadWithoutFormat
= false,
381 .shaderStorageImageWriteWithoutFormat
= true,
382 .shaderClipDistance
= false,
383 .shaderCullDistance
= false,
384 .shaderFloat64
= false,
385 .shaderInt64
= false,
386 .shaderInt16
= false,
388 .variableMultisampleRate
= false,
389 .inheritedQueries
= false,
394 anv_device_get_cache_uuid(void *uuid
)
396 memset(uuid
, 0, VK_UUID_SIZE
);
397 snprintf(uuid
, VK_UUID_SIZE
, "anv-%s", MESA_GIT_SHA1
+ 4);
400 void anv_GetPhysicalDeviceProperties(
401 VkPhysicalDevice physicalDevice
,
402 VkPhysicalDeviceProperties
* pProperties
)
404 ANV_FROM_HANDLE(anv_physical_device
, pdevice
, physicalDevice
);
405 const struct brw_device_info
*devinfo
= pdevice
->info
;
407 anv_finishme("Get correct values for VkPhysicalDeviceLimits");
409 const float time_stamp_base
= devinfo
->gen
>= 9 ? 83.333 : 80.0;
411 VkSampleCountFlags sample_counts
=
412 isl_device_get_sample_counts(&pdevice
->isl_dev
);
414 VkPhysicalDeviceLimits limits
= {
415 .maxImageDimension1D
= (1 << 14),
416 .maxImageDimension2D
= (1 << 14),
417 .maxImageDimension3D
= (1 << 10),
418 .maxImageDimensionCube
= (1 << 14),
419 .maxImageArrayLayers
= (1 << 10),
420 .maxTexelBufferElements
= 128 * 1024 * 1024,
421 .maxUniformBufferRange
= UINT32_MAX
,
422 .maxStorageBufferRange
= UINT32_MAX
,
423 .maxPushConstantsSize
= MAX_PUSH_CONSTANTS_SIZE
,
424 .maxMemoryAllocationCount
= UINT32_MAX
,
425 .maxSamplerAllocationCount
= 64 * 1024,
426 .bufferImageGranularity
= 64, /* A cache line */
427 .sparseAddressSpaceSize
= 0,
428 .maxBoundDescriptorSets
= MAX_SETS
,
429 .maxPerStageDescriptorSamplers
= 64,
430 .maxPerStageDescriptorUniformBuffers
= 64,
431 .maxPerStageDescriptorStorageBuffers
= 64,
432 .maxPerStageDescriptorSampledImages
= 64,
433 .maxPerStageDescriptorStorageImages
= 64,
434 .maxPerStageDescriptorInputAttachments
= 64,
435 .maxPerStageResources
= 128,
436 .maxDescriptorSetSamplers
= 256,
437 .maxDescriptorSetUniformBuffers
= 256,
438 .maxDescriptorSetUniformBuffersDynamic
= 256,
439 .maxDescriptorSetStorageBuffers
= 256,
440 .maxDescriptorSetStorageBuffersDynamic
= 256,
441 .maxDescriptorSetSampledImages
= 256,
442 .maxDescriptorSetStorageImages
= 256,
443 .maxDescriptorSetInputAttachments
= 256,
444 .maxVertexInputAttributes
= 32,
445 .maxVertexInputBindings
= 32,
446 .maxVertexInputAttributeOffset
= 2047,
447 .maxVertexInputBindingStride
= 2048,
448 .maxVertexOutputComponents
= 128,
449 .maxTessellationGenerationLevel
= 0,
450 .maxTessellationPatchSize
= 0,
451 .maxTessellationControlPerVertexInputComponents
= 0,
452 .maxTessellationControlPerVertexOutputComponents
= 0,
453 .maxTessellationControlPerPatchOutputComponents
= 0,
454 .maxTessellationControlTotalOutputComponents
= 0,
455 .maxTessellationEvaluationInputComponents
= 0,
456 .maxTessellationEvaluationOutputComponents
= 0,
457 .maxGeometryShaderInvocations
= 32,
458 .maxGeometryInputComponents
= 64,
459 .maxGeometryOutputComponents
= 128,
460 .maxGeometryOutputVertices
= 256,
461 .maxGeometryTotalOutputComponents
= 1024,
462 .maxFragmentInputComponents
= 128,
463 .maxFragmentOutputAttachments
= 8,
464 .maxFragmentDualSrcAttachments
= 2,
465 .maxFragmentCombinedOutputResources
= 8,
466 .maxComputeSharedMemorySize
= 32768,
467 .maxComputeWorkGroupCount
= { 65535, 65535, 65535 },
468 .maxComputeWorkGroupInvocations
= 16 * devinfo
->max_cs_threads
,
469 .maxComputeWorkGroupSize
= {
470 16 * devinfo
->max_cs_threads
,
471 16 * devinfo
->max_cs_threads
,
472 16 * devinfo
->max_cs_threads
,
474 .subPixelPrecisionBits
= 4 /* FIXME */,
475 .subTexelPrecisionBits
= 4 /* FIXME */,
476 .mipmapPrecisionBits
= 4 /* FIXME */,
477 .maxDrawIndexedIndexValue
= UINT32_MAX
,
478 .maxDrawIndirectCount
= UINT32_MAX
,
479 .maxSamplerLodBias
= 16,
480 .maxSamplerAnisotropy
= 16,
481 .maxViewports
= MAX_VIEWPORTS
,
482 .maxViewportDimensions
= { (1 << 14), (1 << 14) },
483 .viewportBoundsRange
= { -16384.0, 16384.0 },
484 .viewportSubPixelBits
= 13, /* We take a float? */
485 .minMemoryMapAlignment
= 4096, /* A page */
486 .minTexelBufferOffsetAlignment
= 1,
487 .minUniformBufferOffsetAlignment
= 1,
488 .minStorageBufferOffsetAlignment
= 1,
489 .minTexelOffset
= -8,
491 .minTexelGatherOffset
= -8,
492 .maxTexelGatherOffset
= 7,
493 .minInterpolationOffset
= 0, /* FIXME */
494 .maxInterpolationOffset
= 0, /* FIXME */
495 .subPixelInterpolationOffsetBits
= 0, /* FIXME */
496 .maxFramebufferWidth
= (1 << 14),
497 .maxFramebufferHeight
= (1 << 14),
498 .maxFramebufferLayers
= (1 << 10),
499 .framebufferColorSampleCounts
= sample_counts
,
500 .framebufferDepthSampleCounts
= sample_counts
,
501 .framebufferStencilSampleCounts
= sample_counts
,
502 .framebufferNoAttachmentsSampleCounts
= sample_counts
,
503 .maxColorAttachments
= MAX_RTS
,
504 .sampledImageColorSampleCounts
= sample_counts
,
505 .sampledImageIntegerSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
506 .sampledImageDepthSampleCounts
= sample_counts
,
507 .sampledImageStencilSampleCounts
= sample_counts
,
508 .storageImageSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
509 .maxSampleMaskWords
= 1,
510 .timestampComputeAndGraphics
= false,
511 .timestampPeriod
= time_stamp_base
/ (1000 * 1000 * 1000),
512 .maxClipDistances
= 0 /* FIXME */,
513 .maxCullDistances
= 0 /* FIXME */,
514 .maxCombinedClipAndCullDistances
= 0 /* FIXME */,
515 .discreteQueuePriorities
= 1,
516 .pointSizeRange
= { 0.125, 255.875 },
517 .lineWidthRange
= { 0.0, 7.9921875 },
518 .pointSizeGranularity
= (1.0 / 8.0),
519 .lineWidthGranularity
= (1.0 / 128.0),
520 .strictLines
= false, /* FINISHME */
521 .standardSampleLocations
= true,
522 .optimalBufferCopyOffsetAlignment
= 128,
523 .optimalBufferCopyRowPitchAlignment
= 128,
524 .nonCoherentAtomSize
= 64,
527 *pProperties
= (VkPhysicalDeviceProperties
) {
528 .apiVersion
= VK_MAKE_VERSION(1, 0, 2),
531 .deviceID
= pdevice
->chipset_id
,
532 .deviceType
= VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
,
534 .sparseProperties
= {0}, /* Broadwell doesn't do sparse. */
537 strcpy(pProperties
->deviceName
, pdevice
->name
);
538 anv_device_get_cache_uuid(pProperties
->pipelineCacheUUID
);
541 void anv_GetPhysicalDeviceQueueFamilyProperties(
542 VkPhysicalDevice physicalDevice
,
544 VkQueueFamilyProperties
* pQueueFamilyProperties
)
546 if (pQueueFamilyProperties
== NULL
) {
551 assert(*pCount
>= 1);
553 *pQueueFamilyProperties
= (VkQueueFamilyProperties
) {
554 .queueFlags
= VK_QUEUE_GRAPHICS_BIT
|
555 VK_QUEUE_COMPUTE_BIT
|
556 VK_QUEUE_TRANSFER_BIT
,
558 .timestampValidBits
= 36, /* XXX: Real value here */
559 .minImageTransferGranularity
= (VkExtent3D
) { 1, 1, 1 },
563 void anv_GetPhysicalDeviceMemoryProperties(
564 VkPhysicalDevice physicalDevice
,
565 VkPhysicalDeviceMemoryProperties
* pMemoryProperties
)
567 ANV_FROM_HANDLE(anv_physical_device
, physical_device
, physicalDevice
);
568 VkDeviceSize heap_size
;
570 /* Reserve some wiggle room for the driver by exposing only 75% of the
571 * aperture to the heap.
573 heap_size
= 3 * physical_device
->aperture_size
/ 4;
575 if (physical_device
->info
->has_llc
) {
576 /* Big core GPUs share LLC with the CPU and thus one memory type can be
577 * both cached and coherent at the same time.
579 pMemoryProperties
->memoryTypeCount
= 1;
580 pMemoryProperties
->memoryTypes
[0] = (VkMemoryType
) {
581 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
582 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
583 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
|
584 VK_MEMORY_PROPERTY_HOST_CACHED_BIT
,
588 /* The spec requires that we expose a host-visible, coherent memory
589 * type, but Atom GPUs don't share LLC. Thus we offer two memory types
590 * to give the application a choice between cached, but not coherent and
591 * coherent but uncached (WC though).
593 pMemoryProperties
->memoryTypeCount
= 2;
594 pMemoryProperties
->memoryTypes
[0] = (VkMemoryType
) {
595 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
596 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
597 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
,
600 pMemoryProperties
->memoryTypes
[1] = (VkMemoryType
) {
601 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
602 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
603 VK_MEMORY_PROPERTY_HOST_CACHED_BIT
,
608 pMemoryProperties
->memoryHeapCount
= 1;
609 pMemoryProperties
->memoryHeaps
[0] = (VkMemoryHeap
) {
611 .flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
,
615 PFN_vkVoidFunction
anv_GetInstanceProcAddr(
619 return anv_lookup_entrypoint(pName
);
622 /* The loader wants us to expose a second GetInstanceProcAddr function
623 * to work around certain LD_PRELOAD issues seen in apps.
625 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
629 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
633 return anv_GetInstanceProcAddr(instance
, pName
);
636 PFN_vkVoidFunction
anv_GetDeviceProcAddr(
640 return anv_lookup_entrypoint(pName
);
644 anv_queue_init(struct anv_device
*device
, struct anv_queue
*queue
)
646 queue
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
647 queue
->device
= device
;
648 queue
->pool
= &device
->surface_state_pool
;
654 anv_queue_finish(struct anv_queue
*queue
)
658 static struct anv_state
659 anv_state_pool_emit_data(struct anv_state_pool
*pool
, size_t size
, size_t align
, const void *p
)
661 struct anv_state state
;
663 state
= anv_state_pool_alloc(pool
, size
, align
);
664 memcpy(state
.map
, p
, size
);
666 if (!pool
->block_pool
->device
->info
.has_llc
)
667 anv_state_clflush(state
);
672 struct gen8_border_color
{
677 /* Pad out to 64 bytes */
682 anv_device_init_border_colors(struct anv_device
*device
)
684 static const struct gen8_border_color border_colors
[] = {
685 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
] = { .float32
= { 0.0, 0.0, 0.0, 0.0 } },
686 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
] = { .float32
= { 0.0, 0.0, 0.0, 1.0 } },
687 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
] = { .float32
= { 1.0, 1.0, 1.0, 1.0 } },
688 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
] = { .uint32
= { 0, 0, 0, 0 } },
689 [VK_BORDER_COLOR_INT_OPAQUE_BLACK
] = { .uint32
= { 0, 0, 0, 1 } },
690 [VK_BORDER_COLOR_INT_OPAQUE_WHITE
] = { .uint32
= { 1, 1, 1, 1 } },
693 device
->border_colors
= anv_state_pool_emit_data(&device
->dynamic_state_pool
,
694 sizeof(border_colors
), 64,
699 anv_device_submit_simple_batch(struct anv_device
*device
,
700 struct anv_batch
*batch
)
702 struct drm_i915_gem_execbuffer2 execbuf
;
703 struct drm_i915_gem_exec_object2 exec2_objects
[1];
705 VkResult result
= VK_SUCCESS
;
710 /* Kernel driver requires 8 byte aligned batch length */
711 size
= align_u32(batch
->next
- batch
->start
, 8);
712 assert(size
< device
->batch_bo_pool
.bo_size
);
713 result
= anv_bo_pool_alloc(&device
->batch_bo_pool
, &bo
);
714 if (result
!= VK_SUCCESS
)
717 memcpy(bo
.map
, batch
->start
, size
);
718 if (!device
->info
.has_llc
)
719 anv_clflush_range(bo
.map
, size
);
721 exec2_objects
[0].handle
= bo
.gem_handle
;
722 exec2_objects
[0].relocation_count
= 0;
723 exec2_objects
[0].relocs_ptr
= 0;
724 exec2_objects
[0].alignment
= 0;
725 exec2_objects
[0].offset
= bo
.offset
;
726 exec2_objects
[0].flags
= 0;
727 exec2_objects
[0].rsvd1
= 0;
728 exec2_objects
[0].rsvd2
= 0;
730 execbuf
.buffers_ptr
= (uintptr_t) exec2_objects
;
731 execbuf
.buffer_count
= 1;
732 execbuf
.batch_start_offset
= 0;
733 execbuf
.batch_len
= size
;
734 execbuf
.cliprects_ptr
= 0;
735 execbuf
.num_cliprects
= 0;
740 I915_EXEC_HANDLE_LUT
| I915_EXEC_NO_RELOC
| I915_EXEC_RENDER
;
741 execbuf
.rsvd1
= device
->context_id
;
744 ret
= anv_gem_execbuffer(device
, &execbuf
);
746 /* We don't know the real error. */
747 result
= vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
, "execbuf2 failed: %m");
752 ret
= anv_gem_wait(device
, bo
.gem_handle
, &timeout
);
754 /* We don't know the real error. */
755 result
= vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
, "execbuf2 failed: %m");
760 anv_bo_pool_free(&device
->batch_bo_pool
, &bo
);
765 VkResult
anv_CreateDevice(
766 VkPhysicalDevice physicalDevice
,
767 const VkDeviceCreateInfo
* pCreateInfo
,
768 const VkAllocationCallbacks
* pAllocator
,
771 ANV_FROM_HANDLE(anv_physical_device
, physical_device
, physicalDevice
);
773 struct anv_device
*device
;
775 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO
);
777 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
779 for (uint32_t j
= 0; j
< ARRAY_SIZE(device_extensions
); j
++) {
780 if (strcmp(pCreateInfo
->ppEnabledExtensionNames
[i
],
781 device_extensions
[j
].extensionName
) == 0) {
787 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
790 anv_set_dispatch_devinfo(physical_device
->info
);
792 device
= anv_alloc2(&physical_device
->instance
->alloc
, pAllocator
,
794 VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
796 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
798 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
799 device
->instance
= physical_device
->instance
;
800 device
->chipset_id
= physical_device
->chipset_id
;
803 device
->alloc
= *pAllocator
;
805 device
->alloc
= physical_device
->instance
->alloc
;
807 /* XXX(chadv): Can we dup() physicalDevice->fd here? */
808 device
->fd
= open(physical_device
->path
, O_RDWR
| O_CLOEXEC
);
809 if (device
->fd
== -1) {
810 result
= vk_error(VK_ERROR_INITIALIZATION_FAILED
);
814 device
->context_id
= anv_gem_create_context(device
);
815 if (device
->context_id
== -1) {
816 result
= vk_error(VK_ERROR_INITIALIZATION_FAILED
);
820 device
->info
= *physical_device
->info
;
821 device
->isl_dev
= physical_device
->isl_dev
;
823 pthread_mutex_init(&device
->mutex
, NULL
);
825 anv_bo_pool_init(&device
->batch_bo_pool
, device
, ANV_CMD_BUFFER_BATCH_SIZE
);
827 anv_block_pool_init(&device
->dynamic_state_block_pool
, device
, 16384);
829 anv_state_pool_init(&device
->dynamic_state_pool
,
830 &device
->dynamic_state_block_pool
);
832 anv_block_pool_init(&device
->instruction_block_pool
, device
, 128 * 1024);
833 anv_pipeline_cache_init(&device
->default_pipeline_cache
, device
);
835 anv_block_pool_init(&device
->surface_state_block_pool
, device
, 4096);
837 anv_state_pool_init(&device
->surface_state_pool
,
838 &device
->surface_state_block_pool
);
840 anv_bo_init_new(&device
->workaround_bo
, device
, 1024);
842 anv_block_pool_init(&device
->scratch_block_pool
, device
, 0x10000);
844 anv_queue_init(device
, &device
->queue
);
846 switch (device
->info
.gen
) {
848 if (!device
->info
.is_haswell
)
849 result
= gen7_init_device_state(device
);
851 result
= gen75_init_device_state(device
);
854 result
= gen8_init_device_state(device
);
857 result
= gen9_init_device_state(device
);
860 /* Shouldn't get here as we don't create physical devices for any other
862 unreachable("unhandled gen");
864 if (result
!= VK_SUCCESS
)
867 result
= anv_device_init_meta(device
);
868 if (result
!= VK_SUCCESS
)
871 anv_device_init_border_colors(device
);
873 *pDevice
= anv_device_to_handle(device
);
880 anv_free(&device
->alloc
, device
);
885 void anv_DestroyDevice(
887 const VkAllocationCallbacks
* pAllocator
)
889 ANV_FROM_HANDLE(anv_device
, device
, _device
);
891 anv_queue_finish(&device
->queue
);
893 anv_device_finish_meta(device
);
896 /* We only need to free these to prevent valgrind errors. The backing
897 * BO will go away in a couple of lines so we don't actually leak.
899 anv_state_pool_free(&device
->dynamic_state_pool
, device
->border_colors
);
902 anv_gem_munmap(device
->workaround_bo
.map
, device
->workaround_bo
.size
);
903 anv_gem_close(device
, device
->workaround_bo
.gem_handle
);
905 anv_bo_pool_finish(&device
->batch_bo_pool
);
906 anv_state_pool_finish(&device
->dynamic_state_pool
);
907 anv_block_pool_finish(&device
->dynamic_state_block_pool
);
908 anv_block_pool_finish(&device
->instruction_block_pool
);
909 anv_state_pool_finish(&device
->surface_state_pool
);
910 anv_block_pool_finish(&device
->surface_state_block_pool
);
911 anv_block_pool_finish(&device
->scratch_block_pool
);
915 pthread_mutex_destroy(&device
->mutex
);
917 anv_free(&device
->alloc
, device
);
920 VkResult
anv_EnumerateInstanceExtensionProperties(
921 const char* pLayerName
,
922 uint32_t* pPropertyCount
,
923 VkExtensionProperties
* pProperties
)
925 if (pProperties
== NULL
) {
926 *pPropertyCount
= ARRAY_SIZE(global_extensions
);
930 assert(*pPropertyCount
>= ARRAY_SIZE(global_extensions
));
932 *pPropertyCount
= ARRAY_SIZE(global_extensions
);
933 memcpy(pProperties
, global_extensions
, sizeof(global_extensions
));
938 VkResult
anv_EnumerateDeviceExtensionProperties(
939 VkPhysicalDevice physicalDevice
,
940 const char* pLayerName
,
941 uint32_t* pPropertyCount
,
942 VkExtensionProperties
* pProperties
)
944 if (pProperties
== NULL
) {
945 *pPropertyCount
= ARRAY_SIZE(device_extensions
);
949 assert(*pPropertyCount
>= ARRAY_SIZE(device_extensions
));
951 *pPropertyCount
= ARRAY_SIZE(device_extensions
);
952 memcpy(pProperties
, device_extensions
, sizeof(device_extensions
));
957 VkResult
anv_EnumerateInstanceLayerProperties(
958 uint32_t* pPropertyCount
,
959 VkLayerProperties
* pProperties
)
961 if (pProperties
== NULL
) {
966 /* None supported at this time */
967 return vk_error(VK_ERROR_LAYER_NOT_PRESENT
);
970 VkResult
anv_EnumerateDeviceLayerProperties(
971 VkPhysicalDevice physicalDevice
,
972 uint32_t* pPropertyCount
,
973 VkLayerProperties
* pProperties
)
975 if (pProperties
== NULL
) {
980 /* None supported at this time */
981 return vk_error(VK_ERROR_LAYER_NOT_PRESENT
);
984 void anv_GetDeviceQueue(
986 uint32_t queueNodeIndex
,
990 ANV_FROM_HANDLE(anv_device
, device
, _device
);
992 assert(queueIndex
== 0);
994 *pQueue
= anv_queue_to_handle(&device
->queue
);
997 VkResult
anv_QueueSubmit(
999 uint32_t submitCount
,
1000 const VkSubmitInfo
* pSubmits
,
1003 ANV_FROM_HANDLE(anv_queue
, queue
, _queue
);
1004 ANV_FROM_HANDLE(anv_fence
, fence
, _fence
);
1005 struct anv_device
*device
= queue
->device
;
1008 for (uint32_t i
= 0; i
< submitCount
; i
++) {
1009 for (uint32_t j
= 0; j
< pSubmits
[i
].commandBufferCount
; j
++) {
1010 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
,
1011 pSubmits
[i
].pCommandBuffers
[j
]);
1012 assert(cmd_buffer
->level
== VK_COMMAND_BUFFER_LEVEL_PRIMARY
);
1014 ret
= anv_gem_execbuffer(device
, &cmd_buffer
->execbuf2
.execbuf
);
1016 /* We don't know the real error. */
1017 return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
,
1018 "execbuf2 failed: %m");
1021 for (uint32_t k
= 0; k
< cmd_buffer
->execbuf2
.bo_count
; k
++)
1022 cmd_buffer
->execbuf2
.bos
[k
]->offset
= cmd_buffer
->execbuf2
.objects
[k
].offset
;
1027 ret
= anv_gem_execbuffer(device
, &fence
->execbuf
);
1029 /* We don't know the real error. */
1030 return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
,
1031 "execbuf2 failed: %m");
1038 VkResult
anv_QueueWaitIdle(
1041 ANV_FROM_HANDLE(anv_queue
, queue
, _queue
);
1043 return ANV_CALL(DeviceWaitIdle
)(anv_device_to_handle(queue
->device
));
1046 VkResult
anv_DeviceWaitIdle(
1049 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1050 struct anv_batch batch
;
1053 batch
.start
= batch
.next
= cmds
;
1054 batch
.end
= (void *) cmds
+ sizeof(cmds
);
1056 anv_batch_emit(&batch
, GEN7_MI_BATCH_BUFFER_END
);
1057 anv_batch_emit(&batch
, GEN7_MI_NOOP
);
1059 return anv_device_submit_simple_batch(device
, &batch
);
1063 anv_bo_init_new(struct anv_bo
*bo
, struct anv_device
*device
, uint64_t size
)
1065 bo
->gem_handle
= anv_gem_create(device
, size
);
1066 if (!bo
->gem_handle
)
1067 return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY
);
1077 VkResult
anv_AllocateMemory(
1079 const VkMemoryAllocateInfo
* pAllocateInfo
,
1080 const VkAllocationCallbacks
* pAllocator
,
1081 VkDeviceMemory
* pMem
)
1083 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1084 struct anv_device_memory
*mem
;
1087 assert(pAllocateInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO
);
1089 if (pAllocateInfo
->allocationSize
== 0) {
1090 /* Apparently, this is allowed */
1091 *pMem
= VK_NULL_HANDLE
;
1095 /* We support exactly one memory heap. */
1096 assert(pAllocateInfo
->memoryTypeIndex
== 0 ||
1097 (!device
->info
.has_llc
&& pAllocateInfo
->memoryTypeIndex
< 2));
1099 /* FINISHME: Fail if allocation request exceeds heap size. */
1101 mem
= anv_alloc2(&device
->alloc
, pAllocator
, sizeof(*mem
), 8,
1102 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1104 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1106 /* The kernel is going to give us whole pages anyway */
1107 uint64_t alloc_size
= align_u64(pAllocateInfo
->allocationSize
, 4096);
1109 result
= anv_bo_init_new(&mem
->bo
, device
, alloc_size
);
1110 if (result
!= VK_SUCCESS
)
1113 mem
->type_index
= pAllocateInfo
->memoryTypeIndex
;
1115 *pMem
= anv_device_memory_to_handle(mem
);
1120 anv_free2(&device
->alloc
, pAllocator
, mem
);
1125 void anv_FreeMemory(
1127 VkDeviceMemory _mem
,
1128 const VkAllocationCallbacks
* pAllocator
)
1130 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1131 ANV_FROM_HANDLE(anv_device_memory
, mem
, _mem
);
1137 anv_gem_munmap(mem
->bo
.map
, mem
->bo
.size
);
1139 if (mem
->bo
.gem_handle
!= 0)
1140 anv_gem_close(device
, mem
->bo
.gem_handle
);
1142 anv_free2(&device
->alloc
, pAllocator
, mem
);
1145 VkResult
anv_MapMemory(
1147 VkDeviceMemory _memory
,
1148 VkDeviceSize offset
,
1150 VkMemoryMapFlags flags
,
1153 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1154 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
1161 if (size
== VK_WHOLE_SIZE
)
1162 size
= mem
->bo
.size
- offset
;
1164 /* FIXME: Is this supposed to be thread safe? Since vkUnmapMemory() only
1165 * takes a VkDeviceMemory pointer, it seems like only one map of the memory
1166 * at a time is valid. We could just mmap up front and return an offset
1167 * pointer here, but that may exhaust virtual memory on 32 bit
1170 uint32_t gem_flags
= 0;
1171 if (!device
->info
.has_llc
&& mem
->type_index
== 0)
1172 gem_flags
|= I915_MMAP_WC
;
1174 /* GEM will fail to map if the offset isn't 4k-aligned. Round down. */
1175 uint64_t map_offset
= offset
& ~4095ull;
1176 assert(offset
>= map_offset
);
1177 uint64_t map_size
= (offset
+ size
) - map_offset
;
1179 /* Let's map whole pages */
1180 map_size
= align_u64(map_size
, 4096);
1182 mem
->map
= anv_gem_mmap(device
, mem
->bo
.gem_handle
,
1183 map_offset
, map_size
, gem_flags
);
1184 mem
->map_size
= map_size
;
1186 *ppData
= mem
->map
+ (offset
- map_offset
);
1191 void anv_UnmapMemory(
1193 VkDeviceMemory _memory
)
1195 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
1200 anv_gem_munmap(mem
->map
, mem
->map_size
);
1204 clflush_mapped_ranges(struct anv_device
*device
,
1206 const VkMappedMemoryRange
*ranges
)
1208 for (uint32_t i
= 0; i
< count
; i
++) {
1209 ANV_FROM_HANDLE(anv_device_memory
, mem
, ranges
[i
].memory
);
1210 void *p
= mem
->map
+ (ranges
[i
].offset
& ~CACHELINE_MASK
);
1213 if (ranges
[i
].offset
+ ranges
[i
].size
> mem
->map_size
)
1214 end
= mem
->map
+ mem
->map_size
;
1216 end
= mem
->map
+ ranges
[i
].offset
+ ranges
[i
].size
;
1219 __builtin_ia32_clflush(p
);
1220 p
+= CACHELINE_SIZE
;
1225 VkResult
anv_FlushMappedMemoryRanges(
1227 uint32_t memoryRangeCount
,
1228 const VkMappedMemoryRange
* pMemoryRanges
)
1230 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1232 if (device
->info
.has_llc
)
1235 /* Make sure the writes we're flushing have landed. */
1236 __builtin_ia32_mfence();
1238 clflush_mapped_ranges(device
, memoryRangeCount
, pMemoryRanges
);
1243 VkResult
anv_InvalidateMappedMemoryRanges(
1245 uint32_t memoryRangeCount
,
1246 const VkMappedMemoryRange
* pMemoryRanges
)
1248 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1250 if (device
->info
.has_llc
)
1253 clflush_mapped_ranges(device
, memoryRangeCount
, pMemoryRanges
);
1255 /* Make sure no reads get moved up above the invalidate. */
1256 __builtin_ia32_mfence();
1261 void anv_GetBufferMemoryRequirements(
1264 VkMemoryRequirements
* pMemoryRequirements
)
1266 ANV_FROM_HANDLE(anv_buffer
, buffer
, _buffer
);
1268 /* The Vulkan spec (git aaed022) says:
1270 * memoryTypeBits is a bitfield and contains one bit set for every
1271 * supported memory type for the resource. The bit `1<<i` is set if and
1272 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
1273 * structure for the physical device is supported.
1275 * We support exactly one memory type.
1277 pMemoryRequirements
->memoryTypeBits
= 1;
1279 pMemoryRequirements
->size
= buffer
->size
;
1280 pMemoryRequirements
->alignment
= 16;
1283 void anv_GetImageMemoryRequirements(
1286 VkMemoryRequirements
* pMemoryRequirements
)
1288 ANV_FROM_HANDLE(anv_image
, image
, _image
);
1290 /* The Vulkan spec (git aaed022) says:
1292 * memoryTypeBits is a bitfield and contains one bit set for every
1293 * supported memory type for the resource. The bit `1<<i` is set if and
1294 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
1295 * structure for the physical device is supported.
1297 * We support exactly one memory type.
1299 pMemoryRequirements
->memoryTypeBits
= 1;
1301 pMemoryRequirements
->size
= image
->size
;
1302 pMemoryRequirements
->alignment
= image
->alignment
;
1305 void anv_GetImageSparseMemoryRequirements(
1308 uint32_t* pSparseMemoryRequirementCount
,
1309 VkSparseImageMemoryRequirements
* pSparseMemoryRequirements
)
1314 void anv_GetDeviceMemoryCommitment(
1316 VkDeviceMemory memory
,
1317 VkDeviceSize
* pCommittedMemoryInBytes
)
1319 *pCommittedMemoryInBytes
= 0;
1322 VkResult
anv_BindBufferMemory(
1325 VkDeviceMemory _memory
,
1326 VkDeviceSize memoryOffset
)
1328 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
1329 ANV_FROM_HANDLE(anv_buffer
, buffer
, _buffer
);
1332 buffer
->bo
= &mem
->bo
;
1333 buffer
->offset
= memoryOffset
;
1342 VkResult
anv_BindImageMemory(
1345 VkDeviceMemory _memory
,
1346 VkDeviceSize memoryOffset
)
1348 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
1349 ANV_FROM_HANDLE(anv_image
, image
, _image
);
1352 image
->bo
= &mem
->bo
;
1353 image
->offset
= memoryOffset
;
1362 VkResult
anv_QueueBindSparse(
1364 uint32_t bindInfoCount
,
1365 const VkBindSparseInfo
* pBindInfo
,
1368 stub_return(VK_ERROR_INCOMPATIBLE_DRIVER
);
1371 VkResult
anv_CreateFence(
1373 const VkFenceCreateInfo
* pCreateInfo
,
1374 const VkAllocationCallbacks
* pAllocator
,
1377 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1378 struct anv_fence
*fence
;
1379 struct anv_batch batch
;
1382 const uint32_t fence_size
= 128;
1384 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FENCE_CREATE_INFO
);
1386 fence
= anv_alloc2(&device
->alloc
, pAllocator
, sizeof(*fence
), 8,
1387 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1389 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1391 result
= anv_bo_init_new(&fence
->bo
, device
, fence_size
);
1392 if (result
!= VK_SUCCESS
)
1396 anv_gem_mmap(device
, fence
->bo
.gem_handle
, 0, fence
->bo
.size
, 0);
1397 batch
.next
= batch
.start
= fence
->bo
.map
;
1398 batch
.end
= fence
->bo
.map
+ fence
->bo
.size
;
1399 anv_batch_emit(&batch
, GEN7_MI_BATCH_BUFFER_END
);
1400 anv_batch_emit(&batch
, GEN7_MI_NOOP
);
1402 if (!device
->info
.has_llc
) {
1403 assert(((uintptr_t) fence
->bo
.map
& CACHELINE_MASK
) == 0);
1404 assert(batch
.next
- fence
->bo
.map
<= CACHELINE_SIZE
);
1405 __builtin_ia32_mfence();
1406 __builtin_ia32_clflush(fence
->bo
.map
);
1409 fence
->exec2_objects
[0].handle
= fence
->bo
.gem_handle
;
1410 fence
->exec2_objects
[0].relocation_count
= 0;
1411 fence
->exec2_objects
[0].relocs_ptr
= 0;
1412 fence
->exec2_objects
[0].alignment
= 0;
1413 fence
->exec2_objects
[0].offset
= fence
->bo
.offset
;
1414 fence
->exec2_objects
[0].flags
= 0;
1415 fence
->exec2_objects
[0].rsvd1
= 0;
1416 fence
->exec2_objects
[0].rsvd2
= 0;
1418 fence
->execbuf
.buffers_ptr
= (uintptr_t) fence
->exec2_objects
;
1419 fence
->execbuf
.buffer_count
= 1;
1420 fence
->execbuf
.batch_start_offset
= 0;
1421 fence
->execbuf
.batch_len
= batch
.next
- fence
->bo
.map
;
1422 fence
->execbuf
.cliprects_ptr
= 0;
1423 fence
->execbuf
.num_cliprects
= 0;
1424 fence
->execbuf
.DR1
= 0;
1425 fence
->execbuf
.DR4
= 0;
1427 fence
->execbuf
.flags
=
1428 I915_EXEC_HANDLE_LUT
| I915_EXEC_NO_RELOC
| I915_EXEC_RENDER
;
1429 fence
->execbuf
.rsvd1
= device
->context_id
;
1430 fence
->execbuf
.rsvd2
= 0;
1432 fence
->ready
= false;
1434 *pFence
= anv_fence_to_handle(fence
);
1439 anv_free2(&device
->alloc
, pAllocator
, fence
);
1444 void anv_DestroyFence(
1447 const VkAllocationCallbacks
* pAllocator
)
1449 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1450 ANV_FROM_HANDLE(anv_fence
, fence
, _fence
);
1452 anv_gem_munmap(fence
->bo
.map
, fence
->bo
.size
);
1453 anv_gem_close(device
, fence
->bo
.gem_handle
);
1454 anv_free2(&device
->alloc
, pAllocator
, fence
);
1457 VkResult
anv_ResetFences(
1459 uint32_t fenceCount
,
1460 const VkFence
* pFences
)
1462 for (uint32_t i
= 0; i
< fenceCount
; i
++) {
1463 ANV_FROM_HANDLE(anv_fence
, fence
, pFences
[i
]);
1464 fence
->ready
= false;
1470 VkResult
anv_GetFenceStatus(
1474 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1475 ANV_FROM_HANDLE(anv_fence
, fence
, _fence
);
1482 ret
= anv_gem_wait(device
, fence
->bo
.gem_handle
, &t
);
1484 fence
->ready
= true;
1488 return VK_NOT_READY
;
1491 VkResult
anv_WaitForFences(
1493 uint32_t fenceCount
,
1494 const VkFence
* pFences
,
1498 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1500 /* DRM_IOCTL_I915_GEM_WAIT uses a signed 64 bit timeout and is supposed
1501 * to block indefinitely timeouts <= 0. Unfortunately, this was broken
1502 * for a couple of kernel releases. Since there's no way to know
1503 * whether or not the kernel we're using is one of the broken ones, the
1504 * best we can do is to clamp the timeout to INT64_MAX. This limits the
1505 * maximum timeout from 584 years to 292 years - likely not a big deal.
1507 if (timeout
> INT64_MAX
)
1508 timeout
= INT64_MAX
;
1510 int64_t t
= timeout
;
1512 /* FIXME: handle !waitAll */
1514 for (uint32_t i
= 0; i
< fenceCount
; i
++) {
1515 ANV_FROM_HANDLE(anv_fence
, fence
, pFences
[i
]);
1516 int ret
= anv_gem_wait(device
, fence
->bo
.gem_handle
, &t
);
1517 if (ret
== -1 && errno
== ETIME
) {
1519 } else if (ret
== -1) {
1520 /* We don't know the real error. */
1521 return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
,
1522 "gem wait failed: %m");
1529 // Queue semaphore functions
1531 VkResult
anv_CreateSemaphore(
1533 const VkSemaphoreCreateInfo
* pCreateInfo
,
1534 const VkAllocationCallbacks
* pAllocator
,
1535 VkSemaphore
* pSemaphore
)
1537 /* The DRM execbuffer ioctl always execute in-oder, even between different
1538 * rings. As such, there's nothing to do for the user space semaphore.
1541 *pSemaphore
= (VkSemaphore
)1;
1546 void anv_DestroySemaphore(
1548 VkSemaphore semaphore
,
1549 const VkAllocationCallbacks
* pAllocator
)
1555 VkResult
anv_CreateEvent(
1557 const VkEventCreateInfo
* pCreateInfo
,
1558 const VkAllocationCallbacks
* pAllocator
,
1561 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1562 struct anv_state state
;
1563 struct anv_event
*event
;
1565 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_EVENT_CREATE_INFO
);
1567 state
= anv_state_pool_alloc(&device
->dynamic_state_pool
,
1570 event
->state
= state
;
1571 event
->semaphore
= VK_EVENT_RESET
;
1573 if (!device
->info
.has_llc
) {
1574 /* Make sure the writes we're flushing have landed. */
1575 __builtin_ia32_mfence();
1576 __builtin_ia32_clflush(event
);
1579 *pEvent
= anv_event_to_handle(event
);
1584 void anv_DestroyEvent(
1587 const VkAllocationCallbacks
* pAllocator
)
1589 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1590 ANV_FROM_HANDLE(anv_event
, event
, _event
);
1592 anv_state_pool_free(&device
->dynamic_state_pool
, event
->state
);
1595 VkResult
anv_GetEventStatus(
1599 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1600 ANV_FROM_HANDLE(anv_event
, event
, _event
);
1602 if (!device
->info
.has_llc
) {
1603 /* Invalidate read cache before reading event written by GPU. */
1604 __builtin_ia32_clflush(event
);
1605 __builtin_ia32_mfence();
1609 return event
->semaphore
;
1612 VkResult
anv_SetEvent(
1616 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1617 ANV_FROM_HANDLE(anv_event
, event
, _event
);
1619 event
->semaphore
= VK_EVENT_SET
;
1621 if (!device
->info
.has_llc
) {
1622 /* Make sure the writes we're flushing have landed. */
1623 __builtin_ia32_mfence();
1624 __builtin_ia32_clflush(event
);
1630 VkResult
anv_ResetEvent(
1634 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1635 ANV_FROM_HANDLE(anv_event
, event
, _event
);
1637 event
->semaphore
= VK_EVENT_RESET
;
1639 if (!device
->info
.has_llc
) {
1640 /* Make sure the writes we're flushing have landed. */
1641 __builtin_ia32_mfence();
1642 __builtin_ia32_clflush(event
);
1650 VkResult
anv_CreateBuffer(
1652 const VkBufferCreateInfo
* pCreateInfo
,
1653 const VkAllocationCallbacks
* pAllocator
,
1656 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1657 struct anv_buffer
*buffer
;
1659 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
1661 buffer
= anv_alloc2(&device
->alloc
, pAllocator
, sizeof(*buffer
), 8,
1662 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1664 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1666 buffer
->size
= pCreateInfo
->size
;
1667 buffer
->usage
= pCreateInfo
->usage
;
1671 *pBuffer
= anv_buffer_to_handle(buffer
);
1676 void anv_DestroyBuffer(
1679 const VkAllocationCallbacks
* pAllocator
)
1681 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1682 ANV_FROM_HANDLE(anv_buffer
, buffer
, _buffer
);
1684 anv_free2(&device
->alloc
, pAllocator
, buffer
);
1688 anv_fill_buffer_surface_state(struct anv_device
*device
, struct anv_state state
,
1689 enum isl_format format
,
1690 uint32_t offset
, uint32_t range
, uint32_t stride
)
1692 switch (device
->info
.gen
) {
1694 if (device
->info
.is_haswell
)
1695 gen75_fill_buffer_surface_state(state
.map
, format
, offset
, range
,
1698 gen7_fill_buffer_surface_state(state
.map
, format
, offset
, range
,
1702 gen8_fill_buffer_surface_state(state
.map
, format
, offset
, range
, stride
);
1705 gen9_fill_buffer_surface_state(state
.map
, format
, offset
, range
, stride
);
1708 unreachable("unsupported gen\n");
1711 if (!device
->info
.has_llc
)
1712 anv_state_clflush(state
);
1715 void anv_DestroySampler(
1718 const VkAllocationCallbacks
* pAllocator
)
1720 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1721 ANV_FROM_HANDLE(anv_sampler
, sampler
, _sampler
);
1723 anv_free2(&device
->alloc
, pAllocator
, sampler
);
1726 VkResult
anv_CreateFramebuffer(
1728 const VkFramebufferCreateInfo
* pCreateInfo
,
1729 const VkAllocationCallbacks
* pAllocator
,
1730 VkFramebuffer
* pFramebuffer
)
1732 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1733 struct anv_framebuffer
*framebuffer
;
1735 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
1737 size_t size
= sizeof(*framebuffer
) +
1738 sizeof(struct anv_image_view
*) * pCreateInfo
->attachmentCount
;
1739 framebuffer
= anv_alloc2(&device
->alloc
, pAllocator
, size
, 8,
1740 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1741 if (framebuffer
== NULL
)
1742 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1744 framebuffer
->attachment_count
= pCreateInfo
->attachmentCount
;
1745 for (uint32_t i
= 0; i
< pCreateInfo
->attachmentCount
; i
++) {
1746 VkImageView _iview
= pCreateInfo
->pAttachments
[i
];
1747 framebuffer
->attachments
[i
] = anv_image_view_from_handle(_iview
);
1750 framebuffer
->width
= pCreateInfo
->width
;
1751 framebuffer
->height
= pCreateInfo
->height
;
1752 framebuffer
->layers
= pCreateInfo
->layers
;
1754 *pFramebuffer
= anv_framebuffer_to_handle(framebuffer
);
1759 void anv_DestroyFramebuffer(
1762 const VkAllocationCallbacks
* pAllocator
)
1764 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1765 ANV_FROM_HANDLE(anv_framebuffer
, fb
, _fb
);
1767 anv_free2(&device
->alloc
, pAllocator
, fb
);
1770 void vkCmdDbgMarkerBegin(
1771 VkCommandBuffer commandBuffer
,
1772 const char* pMarker
)
1773 __attribute__ ((visibility ("default")));
1775 void vkCmdDbgMarkerEnd(
1776 VkCommandBuffer commandBuffer
)
1777 __attribute__ ((visibility ("default")));
1779 void vkCmdDbgMarkerBegin(
1780 VkCommandBuffer commandBuffer
,
1781 const char* pMarker
)
1785 void vkCmdDbgMarkerEnd(
1786 VkCommandBuffer commandBuffer
)