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
->apiVersion
;
218 if (VK_MAKE_VERSION(1, 0, 0) > client_version
||
219 client_version
> VK_MAKE_VERSION(1, 0, 2)) {
220 return vk_error(VK_ERROR_INCOMPATIBLE_DRIVER
);
223 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
225 for (uint32_t j
= 0; j
< ARRAY_SIZE(global_extensions
); j
++) {
226 if (strcmp(pCreateInfo
->ppEnabledExtensionNames
[i
],
227 global_extensions
[j
].extensionName
) == 0) {
233 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
236 instance
= anv_alloc2(&default_alloc
, pAllocator
, sizeof(*instance
), 8,
237 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE
);
239 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
241 instance
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
244 instance
->alloc
= *pAllocator
;
246 instance
->alloc
= default_alloc
;
248 instance
->apiVersion
= pCreateInfo
->pApplicationInfo
->apiVersion
;
249 instance
->physicalDeviceCount
= -1;
253 VG(VALGRIND_CREATE_MEMPOOL(instance
, 0, false));
255 anv_init_wsi(instance
);
257 *pInstance
= anv_instance_to_handle(instance
);
262 void anv_DestroyInstance(
263 VkInstance _instance
,
264 const VkAllocationCallbacks
* pAllocator
)
266 ANV_FROM_HANDLE(anv_instance
, instance
, _instance
);
268 if (instance
->physicalDeviceCount
> 0) {
269 /* We support at most one physical device. */
270 assert(instance
->physicalDeviceCount
== 1);
271 anv_physical_device_finish(&instance
->physicalDevice
);
274 anv_finish_wsi(instance
);
276 VG(VALGRIND_DESTROY_MEMPOOL(instance
));
280 anv_free(&instance
->alloc
, instance
);
283 VkResult
anv_EnumeratePhysicalDevices(
284 VkInstance _instance
,
285 uint32_t* pPhysicalDeviceCount
,
286 VkPhysicalDevice
* pPhysicalDevices
)
288 ANV_FROM_HANDLE(anv_instance
, instance
, _instance
);
291 if (instance
->physicalDeviceCount
< 0) {
292 result
= anv_physical_device_init(&instance
->physicalDevice
,
293 instance
, "/dev/dri/renderD128");
294 if (result
== VK_ERROR_INCOMPATIBLE_DRIVER
) {
295 instance
->physicalDeviceCount
= 0;
296 } else if (result
== VK_SUCCESS
) {
297 instance
->physicalDeviceCount
= 1;
303 /* pPhysicalDeviceCount is an out parameter if pPhysicalDevices is NULL;
304 * otherwise it's an inout parameter.
306 * The Vulkan spec (git aaed022) says:
308 * pPhysicalDeviceCount is a pointer to an unsigned integer variable
309 * that is initialized with the number of devices the application is
310 * prepared to receive handles to. pname:pPhysicalDevices is pointer to
311 * an array of at least this many VkPhysicalDevice handles [...].
313 * Upon success, if pPhysicalDevices is NULL, vkEnumeratePhysicalDevices
314 * overwrites the contents of the variable pointed to by
315 * pPhysicalDeviceCount with the number of physical devices in in the
316 * instance; otherwise, vkEnumeratePhysicalDevices overwrites
317 * pPhysicalDeviceCount with the number of physical handles written to
320 if (!pPhysicalDevices
) {
321 *pPhysicalDeviceCount
= instance
->physicalDeviceCount
;
322 } else if (*pPhysicalDeviceCount
>= 1) {
323 pPhysicalDevices
[0] = anv_physical_device_to_handle(&instance
->physicalDevice
);
324 *pPhysicalDeviceCount
= 1;
326 *pPhysicalDeviceCount
= 0;
332 void anv_GetPhysicalDeviceFeatures(
333 VkPhysicalDevice physicalDevice
,
334 VkPhysicalDeviceFeatures
* pFeatures
)
336 anv_finishme("Get correct values for PhysicalDeviceFeatures");
338 *pFeatures
= (VkPhysicalDeviceFeatures
) {
339 .robustBufferAccess
= false,
340 .fullDrawIndexUint32
= false,
341 .imageCubeArray
= false,
342 .independentBlend
= false,
343 .geometryShader
= true,
344 .tessellationShader
= false,
345 .sampleRateShading
= false,
346 .dualSrcBlend
= true,
348 .multiDrawIndirect
= false,
349 .drawIndirectFirstInstance
= false,
351 .depthBiasClamp
= false,
352 .fillModeNonSolid
= true,
353 .depthBounds
= false,
357 .multiViewport
= true,
358 .samplerAnisotropy
= false, /* FINISHME */
359 .textureCompressionETC2
= true,
360 .textureCompressionASTC_LDR
= true,
361 .textureCompressionBC
= true,
362 .occlusionQueryPrecise
= false, /* FINISHME */
363 .pipelineStatisticsQuery
= true,
364 .vertexPipelineStoresAndAtomics
= false,
365 .fragmentStoresAndAtomics
= true,
366 .shaderTessellationAndGeometryPointSize
= true,
367 .shaderImageGatherExtended
= true,
368 .shaderStorageImageExtendedFormats
= false,
369 .shaderStorageImageMultisample
= false,
370 .shaderUniformBufferArrayDynamicIndexing
= true,
371 .shaderSampledImageArrayDynamicIndexing
= false,
372 .shaderStorageBufferArrayDynamicIndexing
= false,
373 .shaderStorageImageArrayDynamicIndexing
= false,
374 .shaderStorageImageReadWithoutFormat
= false,
375 .shaderStorageImageWriteWithoutFormat
= true,
376 .shaderClipDistance
= false,
377 .shaderCullDistance
= false,
378 .shaderFloat64
= false,
379 .shaderInt64
= false,
380 .shaderInt16
= false,
382 .variableMultisampleRate
= false,
383 .inheritedQueries
= false,
387 void anv_GetPhysicalDeviceProperties(
388 VkPhysicalDevice physicalDevice
,
389 VkPhysicalDeviceProperties
* pProperties
)
391 ANV_FROM_HANDLE(anv_physical_device
, pdevice
, physicalDevice
);
392 const struct brw_device_info
*devinfo
= pdevice
->info
;
394 anv_finishme("Get correct values for VkPhysicalDeviceLimits");
396 const float time_stamp_base
= devinfo
->gen
>= 9 ? 83.333 : 80.0;
398 VkSampleCountFlags sample_counts
=
399 isl_device_get_sample_counts(&pdevice
->isl_dev
);
401 VkPhysicalDeviceLimits limits
= {
402 .maxImageDimension1D
= (1 << 14),
403 .maxImageDimension2D
= (1 << 14),
404 .maxImageDimension3D
= (1 << 10),
405 .maxImageDimensionCube
= (1 << 14),
406 .maxImageArrayLayers
= (1 << 10),
407 .maxTexelBufferElements
= 128 * 1024 * 1024,
408 .maxUniformBufferRange
= UINT32_MAX
,
409 .maxStorageBufferRange
= UINT32_MAX
,
410 .maxPushConstantsSize
= MAX_PUSH_CONSTANTS_SIZE
,
411 .maxMemoryAllocationCount
= UINT32_MAX
,
412 .maxSamplerAllocationCount
= 64 * 1024,
413 .bufferImageGranularity
= 64, /* A cache line */
414 .sparseAddressSpaceSize
= 0,
415 .maxBoundDescriptorSets
= MAX_SETS
,
416 .maxPerStageDescriptorSamplers
= 64,
417 .maxPerStageDescriptorUniformBuffers
= 64,
418 .maxPerStageDescriptorStorageBuffers
= 64,
419 .maxPerStageDescriptorSampledImages
= 64,
420 .maxPerStageDescriptorStorageImages
= 64,
421 .maxPerStageDescriptorInputAttachments
= 64,
422 .maxPerStageResources
= 128,
423 .maxDescriptorSetSamplers
= 256,
424 .maxDescriptorSetUniformBuffers
= 256,
425 .maxDescriptorSetUniformBuffersDynamic
= 256,
426 .maxDescriptorSetStorageBuffers
= 256,
427 .maxDescriptorSetStorageBuffersDynamic
= 256,
428 .maxDescriptorSetSampledImages
= 256,
429 .maxDescriptorSetStorageImages
= 256,
430 .maxDescriptorSetInputAttachments
= 256,
431 .maxVertexInputAttributes
= 32,
432 .maxVertexInputBindings
= 32,
433 .maxVertexInputAttributeOffset
= 2047,
434 .maxVertexInputBindingStride
= 2048,
435 .maxVertexOutputComponents
= 128,
436 .maxTessellationGenerationLevel
= 0,
437 .maxTessellationPatchSize
= 0,
438 .maxTessellationControlPerVertexInputComponents
= 0,
439 .maxTessellationControlPerVertexOutputComponents
= 0,
440 .maxTessellationControlPerPatchOutputComponents
= 0,
441 .maxTessellationControlTotalOutputComponents
= 0,
442 .maxTessellationEvaluationInputComponents
= 0,
443 .maxTessellationEvaluationOutputComponents
= 0,
444 .maxGeometryShaderInvocations
= 32,
445 .maxGeometryInputComponents
= 64,
446 .maxGeometryOutputComponents
= 128,
447 .maxGeometryOutputVertices
= 256,
448 .maxGeometryTotalOutputComponents
= 1024,
449 .maxFragmentInputComponents
= 128,
450 .maxFragmentOutputAttachments
= 8,
451 .maxFragmentDualSrcAttachments
= 2,
452 .maxFragmentCombinedOutputResources
= 8,
453 .maxComputeSharedMemorySize
= 32768,
454 .maxComputeWorkGroupCount
= { 65535, 65535, 65535 },
455 .maxComputeWorkGroupInvocations
= 16 * devinfo
->max_cs_threads
,
456 .maxComputeWorkGroupSize
= {
457 16 * devinfo
->max_cs_threads
,
458 16 * devinfo
->max_cs_threads
,
459 16 * devinfo
->max_cs_threads
,
461 .subPixelPrecisionBits
= 4 /* FIXME */,
462 .subTexelPrecisionBits
= 4 /* FIXME */,
463 .mipmapPrecisionBits
= 4 /* FIXME */,
464 .maxDrawIndexedIndexValue
= UINT32_MAX
,
465 .maxDrawIndirectCount
= UINT32_MAX
,
466 .maxSamplerLodBias
= 16,
467 .maxSamplerAnisotropy
= 16,
468 .maxViewports
= MAX_VIEWPORTS
,
469 .maxViewportDimensions
= { (1 << 14), (1 << 14) },
470 .viewportBoundsRange
= { -16384.0, 16384.0 },
471 .viewportSubPixelBits
= 13, /* We take a float? */
472 .minMemoryMapAlignment
= 4096, /* A page */
473 .minTexelBufferOffsetAlignment
= 1,
474 .minUniformBufferOffsetAlignment
= 1,
475 .minStorageBufferOffsetAlignment
= 1,
476 .minTexelOffset
= -8,
478 .minTexelGatherOffset
= -8,
479 .maxTexelGatherOffset
= 7,
480 .minInterpolationOffset
= 0, /* FIXME */
481 .maxInterpolationOffset
= 0, /* FIXME */
482 .subPixelInterpolationOffsetBits
= 0, /* FIXME */
483 .maxFramebufferWidth
= (1 << 14),
484 .maxFramebufferHeight
= (1 << 14),
485 .maxFramebufferLayers
= (1 << 10),
486 .framebufferColorSampleCounts
= sample_counts
,
487 .framebufferDepthSampleCounts
= sample_counts
,
488 .framebufferStencilSampleCounts
= sample_counts
,
489 .framebufferNoAttachmentsSampleCounts
= sample_counts
,
490 .maxColorAttachments
= MAX_RTS
,
491 .sampledImageColorSampleCounts
= sample_counts
,
492 .sampledImageIntegerSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
493 .sampledImageDepthSampleCounts
= sample_counts
,
494 .sampledImageStencilSampleCounts
= sample_counts
,
495 .storageImageSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
496 .maxSampleMaskWords
= 1,
497 .timestampComputeAndGraphics
= false,
498 .timestampPeriod
= time_stamp_base
/ (1000 * 1000 * 1000),
499 .maxClipDistances
= 0 /* FIXME */,
500 .maxCullDistances
= 0 /* FIXME */,
501 .maxCombinedClipAndCullDistances
= 0 /* FIXME */,
502 .discreteQueuePriorities
= 1,
503 .pointSizeRange
= { 0.125, 255.875 },
504 .lineWidthRange
= { 0.0, 7.9921875 },
505 .pointSizeGranularity
= (1.0 / 8.0),
506 .lineWidthGranularity
= (1.0 / 128.0),
507 .strictLines
= false, /* FINISHME */
508 .standardSampleLocations
= true,
509 .optimalBufferCopyOffsetAlignment
= 128,
510 .optimalBufferCopyRowPitchAlignment
= 128,
511 .nonCoherentAtomSize
= 64,
514 *pProperties
= (VkPhysicalDeviceProperties
) {
515 .apiVersion
= VK_MAKE_VERSION(1, 0, 2),
518 .deviceID
= pdevice
->chipset_id
,
519 .deviceType
= VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
,
521 .sparseProperties
= {0}, /* Broadwell doesn't do sparse. */
524 strcpy(pProperties
->deviceName
, pdevice
->name
);
525 snprintf((char *)pProperties
->pipelineCacheUUID
, VK_UUID_SIZE
,
526 "anv-%s", MESA_GIT_SHA1
+ 4);
529 void anv_GetPhysicalDeviceQueueFamilyProperties(
530 VkPhysicalDevice physicalDevice
,
532 VkQueueFamilyProperties
* pQueueFamilyProperties
)
534 if (pQueueFamilyProperties
== NULL
) {
539 assert(*pCount
>= 1);
541 *pQueueFamilyProperties
= (VkQueueFamilyProperties
) {
542 .queueFlags
= VK_QUEUE_GRAPHICS_BIT
|
543 VK_QUEUE_COMPUTE_BIT
|
544 VK_QUEUE_TRANSFER_BIT
,
546 .timestampValidBits
= 36, /* XXX: Real value here */
547 .minImageTransferGranularity
= (VkExtent3D
) { 1, 1, 1 },
551 void anv_GetPhysicalDeviceMemoryProperties(
552 VkPhysicalDevice physicalDevice
,
553 VkPhysicalDeviceMemoryProperties
* pMemoryProperties
)
555 ANV_FROM_HANDLE(anv_physical_device
, physical_device
, physicalDevice
);
556 VkDeviceSize heap_size
;
558 /* Reserve some wiggle room for the driver by exposing only 75% of the
559 * aperture to the heap.
561 heap_size
= 3 * physical_device
->aperture_size
/ 4;
563 if (physical_device
->info
->has_llc
) {
564 /* Big core GPUs share LLC with the CPU and thus one memory type can be
565 * both cached and coherent at the same time.
567 pMemoryProperties
->memoryTypeCount
= 1;
568 pMemoryProperties
->memoryTypes
[0] = (VkMemoryType
) {
569 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
570 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
571 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
|
572 VK_MEMORY_PROPERTY_HOST_CACHED_BIT
,
576 /* The spec requires that we expose a host-visible, coherent memory
577 * type, but Atom GPUs don't share LLC. Thus we offer two memory types
578 * to give the application a choice between cached, but not coherent and
579 * coherent but uncached (WC though).
581 pMemoryProperties
->memoryTypeCount
= 2;
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
,
588 pMemoryProperties
->memoryTypes
[1] = (VkMemoryType
) {
589 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
590 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
591 VK_MEMORY_PROPERTY_HOST_CACHED_BIT
,
596 pMemoryProperties
->memoryHeapCount
= 1;
597 pMemoryProperties
->memoryHeaps
[0] = (VkMemoryHeap
) {
599 .flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
,
603 PFN_vkVoidFunction
anv_GetInstanceProcAddr(
607 return anv_lookup_entrypoint(pName
);
610 PFN_vkVoidFunction
anv_GetDeviceProcAddr(
614 return anv_lookup_entrypoint(pName
);
618 anv_queue_init(struct anv_device
*device
, struct anv_queue
*queue
)
620 queue
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
621 queue
->device
= device
;
622 queue
->pool
= &device
->surface_state_pool
;
628 anv_queue_finish(struct anv_queue
*queue
)
632 static struct anv_state
633 anv_state_pool_emit_data(struct anv_state_pool
*pool
, size_t size
, size_t align
, const void *p
)
635 struct anv_state state
;
637 state
= anv_state_pool_alloc(pool
, size
, align
);
638 memcpy(state
.map
, p
, size
);
640 if (!pool
->block_pool
->device
->info
.has_llc
)
641 anv_state_clflush(state
);
646 struct gen8_border_color
{
651 /* Pad out to 64 bytes */
656 anv_device_init_border_colors(struct anv_device
*device
)
658 static const struct gen8_border_color border_colors
[] = {
659 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
] = { .float32
= { 0.0, 0.0, 0.0, 0.0 } },
660 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
] = { .float32
= { 0.0, 0.0, 0.0, 1.0 } },
661 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
] = { .float32
= { 1.0, 1.0, 1.0, 1.0 } },
662 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
] = { .uint32
= { 0, 0, 0, 0 } },
663 [VK_BORDER_COLOR_INT_OPAQUE_BLACK
] = { .uint32
= { 0, 0, 0, 1 } },
664 [VK_BORDER_COLOR_INT_OPAQUE_WHITE
] = { .uint32
= { 1, 1, 1, 1 } },
667 device
->border_colors
= anv_state_pool_emit_data(&device
->dynamic_state_pool
,
668 sizeof(border_colors
), 64,
672 VkResult
anv_CreateDevice(
673 VkPhysicalDevice physicalDevice
,
674 const VkDeviceCreateInfo
* pCreateInfo
,
675 const VkAllocationCallbacks
* pAllocator
,
678 ANV_FROM_HANDLE(anv_physical_device
, physical_device
, physicalDevice
);
680 struct anv_device
*device
;
682 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO
);
684 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
686 for (uint32_t j
= 0; j
< ARRAY_SIZE(device_extensions
); j
++) {
687 if (strcmp(pCreateInfo
->ppEnabledExtensionNames
[i
],
688 device_extensions
[j
].extensionName
) == 0) {
694 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
697 anv_set_dispatch_devinfo(physical_device
->info
);
699 device
= anv_alloc2(&physical_device
->instance
->alloc
, pAllocator
,
701 VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
703 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
705 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
706 device
->instance
= physical_device
->instance
;
709 device
->alloc
= *pAllocator
;
711 device
->alloc
= physical_device
->instance
->alloc
;
713 /* XXX(chadv): Can we dup() physicalDevice->fd here? */
714 device
->fd
= open(physical_device
->path
, O_RDWR
| O_CLOEXEC
);
715 if (device
->fd
== -1) {
716 result
= vk_error(VK_ERROR_INITIALIZATION_FAILED
);
720 device
->context_id
= anv_gem_create_context(device
);
721 if (device
->context_id
== -1) {
722 result
= vk_error(VK_ERROR_INITIALIZATION_FAILED
);
726 device
->info
= *physical_device
->info
;
727 device
->isl_dev
= physical_device
->isl_dev
;
729 pthread_mutex_init(&device
->mutex
, NULL
);
731 anv_bo_pool_init(&device
->batch_bo_pool
, device
, ANV_CMD_BUFFER_BATCH_SIZE
);
733 anv_block_pool_init(&device
->dynamic_state_block_pool
, device
, 16384);
735 anv_state_pool_init(&device
->dynamic_state_pool
,
736 &device
->dynamic_state_block_pool
);
738 anv_block_pool_init(&device
->instruction_block_pool
, device
, 128 * 1024);
739 anv_pipeline_cache_init(&device
->default_pipeline_cache
, device
);
741 anv_block_pool_init(&device
->surface_state_block_pool
, device
, 4096);
743 anv_state_pool_init(&device
->surface_state_pool
,
744 &device
->surface_state_block_pool
);
746 anv_bo_init_new(&device
->workaround_bo
, device
, 1024);
748 anv_block_pool_init(&device
->scratch_block_pool
, device
, 0x10000);
750 anv_queue_init(device
, &device
->queue
);
752 result
= anv_device_init_meta(device
);
753 if (result
!= VK_SUCCESS
)
756 anv_device_init_border_colors(device
);
758 *pDevice
= anv_device_to_handle(device
);
765 anv_free(&device
->alloc
, device
);
770 void anv_DestroyDevice(
772 const VkAllocationCallbacks
* pAllocator
)
774 ANV_FROM_HANDLE(anv_device
, device
, _device
);
776 anv_queue_finish(&device
->queue
);
778 anv_device_finish_meta(device
);
781 /* We only need to free these to prevent valgrind errors. The backing
782 * BO will go away in a couple of lines so we don't actually leak.
784 anv_state_pool_free(&device
->dynamic_state_pool
, device
->border_colors
);
787 anv_gem_munmap(device
->workaround_bo
.map
, device
->workaround_bo
.size
);
788 anv_gem_close(device
, device
->workaround_bo
.gem_handle
);
790 anv_bo_pool_finish(&device
->batch_bo_pool
);
791 anv_state_pool_finish(&device
->dynamic_state_pool
);
792 anv_block_pool_finish(&device
->dynamic_state_block_pool
);
793 anv_block_pool_finish(&device
->instruction_block_pool
);
794 anv_state_pool_finish(&device
->surface_state_pool
);
795 anv_block_pool_finish(&device
->surface_state_block_pool
);
796 anv_block_pool_finish(&device
->scratch_block_pool
);
800 pthread_mutex_destroy(&device
->mutex
);
802 anv_free(&device
->alloc
, device
);
805 VkResult
anv_EnumerateInstanceExtensionProperties(
806 const char* pLayerName
,
807 uint32_t* pPropertyCount
,
808 VkExtensionProperties
* pProperties
)
810 if (pProperties
== NULL
) {
811 *pPropertyCount
= ARRAY_SIZE(global_extensions
);
815 assert(*pPropertyCount
>= ARRAY_SIZE(global_extensions
));
817 *pPropertyCount
= ARRAY_SIZE(global_extensions
);
818 memcpy(pProperties
, global_extensions
, sizeof(global_extensions
));
823 VkResult
anv_EnumerateDeviceExtensionProperties(
824 VkPhysicalDevice physicalDevice
,
825 const char* pLayerName
,
826 uint32_t* pPropertyCount
,
827 VkExtensionProperties
* pProperties
)
829 if (pProperties
== NULL
) {
830 *pPropertyCount
= ARRAY_SIZE(device_extensions
);
834 assert(*pPropertyCount
>= ARRAY_SIZE(device_extensions
));
836 *pPropertyCount
= ARRAY_SIZE(device_extensions
);
837 memcpy(pProperties
, device_extensions
, sizeof(device_extensions
));
842 VkResult
anv_EnumerateInstanceLayerProperties(
843 uint32_t* pPropertyCount
,
844 VkLayerProperties
* pProperties
)
846 if (pProperties
== NULL
) {
851 /* None supported at this time */
852 return vk_error(VK_ERROR_LAYER_NOT_PRESENT
);
855 VkResult
anv_EnumerateDeviceLayerProperties(
856 VkPhysicalDevice physicalDevice
,
857 uint32_t* pPropertyCount
,
858 VkLayerProperties
* pProperties
)
860 if (pProperties
== NULL
) {
865 /* None supported at this time */
866 return vk_error(VK_ERROR_LAYER_NOT_PRESENT
);
869 void anv_GetDeviceQueue(
871 uint32_t queueNodeIndex
,
875 ANV_FROM_HANDLE(anv_device
, device
, _device
);
877 assert(queueIndex
== 0);
879 *pQueue
= anv_queue_to_handle(&device
->queue
);
882 VkResult
anv_QueueSubmit(
884 uint32_t submitCount
,
885 const VkSubmitInfo
* pSubmits
,
888 ANV_FROM_HANDLE(anv_queue
, queue
, _queue
);
889 ANV_FROM_HANDLE(anv_fence
, fence
, _fence
);
890 struct anv_device
*device
= queue
->device
;
893 for (uint32_t i
= 0; i
< submitCount
; i
++) {
894 for (uint32_t j
= 0; j
< pSubmits
[i
].commandBufferCount
; j
++) {
895 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
,
896 pSubmits
[i
].pCommandBuffers
[j
]);
897 assert(cmd_buffer
->level
== VK_COMMAND_BUFFER_LEVEL_PRIMARY
);
899 ret
= anv_gem_execbuffer(device
, &cmd_buffer
->execbuf2
.execbuf
);
901 /* We don't know the real error. */
902 return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
,
903 "execbuf2 failed: %m");
907 ret
= anv_gem_execbuffer(device
, &fence
->execbuf
);
909 /* We don't know the real error. */
910 return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
,
911 "execbuf2 failed: %m");
915 for (uint32_t k
= 0; k
< cmd_buffer
->execbuf2
.bo_count
; k
++)
916 cmd_buffer
->execbuf2
.bos
[k
]->offset
= cmd_buffer
->execbuf2
.objects
[k
].offset
;
923 VkResult
anv_QueueWaitIdle(
926 ANV_FROM_HANDLE(anv_queue
, queue
, _queue
);
928 return ANV_CALL(DeviceWaitIdle
)(anv_device_to_handle(queue
->device
));
931 VkResult
anv_DeviceWaitIdle(
934 ANV_FROM_HANDLE(anv_device
, device
, _device
);
935 struct anv_state state
;
936 struct anv_batch batch
;
937 struct drm_i915_gem_execbuffer2 execbuf
;
938 struct drm_i915_gem_exec_object2 exec2_objects
[1];
939 struct anv_bo
*bo
= NULL
;
944 state
= anv_state_pool_alloc(&device
->dynamic_state_pool
, 32, 32);
945 bo
= &device
->dynamic_state_pool
.block_pool
->bo
;
946 batch
.start
= batch
.next
= state
.map
;
947 batch
.end
= state
.map
+ 32;
948 anv_batch_emit(&batch
, GEN7_MI_BATCH_BUFFER_END
);
949 anv_batch_emit(&batch
, GEN7_MI_NOOP
);
951 if (!device
->info
.has_llc
)
952 anv_state_clflush(state
);
954 exec2_objects
[0].handle
= bo
->gem_handle
;
955 exec2_objects
[0].relocation_count
= 0;
956 exec2_objects
[0].relocs_ptr
= 0;
957 exec2_objects
[0].alignment
= 0;
958 exec2_objects
[0].offset
= bo
->offset
;
959 exec2_objects
[0].flags
= 0;
960 exec2_objects
[0].rsvd1
= 0;
961 exec2_objects
[0].rsvd2
= 0;
963 execbuf
.buffers_ptr
= (uintptr_t) exec2_objects
;
964 execbuf
.buffer_count
= 1;
965 execbuf
.batch_start_offset
= state
.offset
;
966 execbuf
.batch_len
= batch
.next
- state
.map
;
967 execbuf
.cliprects_ptr
= 0;
968 execbuf
.num_cliprects
= 0;
973 I915_EXEC_HANDLE_LUT
| I915_EXEC_NO_RELOC
| I915_EXEC_RENDER
;
974 execbuf
.rsvd1
= device
->context_id
;
977 ret
= anv_gem_execbuffer(device
, &execbuf
);
979 /* We don't know the real error. */
980 result
= vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
, "execbuf2 failed: %m");
985 ret
= anv_gem_wait(device
, bo
->gem_handle
, &timeout
);
987 /* We don't know the real error. */
988 result
= vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
, "execbuf2 failed: %m");
992 anv_state_pool_free(&device
->dynamic_state_pool
, state
);
997 anv_state_pool_free(&device
->dynamic_state_pool
, state
);
1003 anv_bo_init_new(struct anv_bo
*bo
, struct anv_device
*device
, uint64_t size
)
1005 bo
->gem_handle
= anv_gem_create(device
, size
);
1006 if (!bo
->gem_handle
)
1007 return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY
);
1017 VkResult
anv_AllocateMemory(
1019 const VkMemoryAllocateInfo
* pAllocateInfo
,
1020 const VkAllocationCallbacks
* pAllocator
,
1021 VkDeviceMemory
* pMem
)
1023 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1024 struct anv_device_memory
*mem
;
1027 assert(pAllocateInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO
);
1029 if (pAllocateInfo
->allocationSize
== 0) {
1030 /* Apparently, this is allowed */
1031 *pMem
= VK_NULL_HANDLE
;
1035 /* We support exactly one memory heap. */
1036 assert(pAllocateInfo
->memoryTypeIndex
== 0 ||
1037 (!device
->info
.has_llc
&& pAllocateInfo
->memoryTypeIndex
< 2));
1039 /* FINISHME: Fail if allocation request exceeds heap size. */
1041 mem
= anv_alloc2(&device
->alloc
, pAllocator
, sizeof(*mem
), 8,
1042 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1044 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1046 /* The kernel is going to give us whole pages anyway */
1047 uint64_t alloc_size
= align_u64(pAllocateInfo
->allocationSize
, 4096);
1049 result
= anv_bo_init_new(&mem
->bo
, device
, alloc_size
);
1050 if (result
!= VK_SUCCESS
)
1053 mem
->type_index
= pAllocateInfo
->memoryTypeIndex
;
1055 *pMem
= anv_device_memory_to_handle(mem
);
1060 anv_free2(&device
->alloc
, pAllocator
, mem
);
1065 void anv_FreeMemory(
1067 VkDeviceMemory _mem
,
1068 const VkAllocationCallbacks
* pAllocator
)
1070 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1071 ANV_FROM_HANDLE(anv_device_memory
, mem
, _mem
);
1077 anv_gem_munmap(mem
->bo
.map
, mem
->bo
.size
);
1079 if (mem
->bo
.gem_handle
!= 0)
1080 anv_gem_close(device
, mem
->bo
.gem_handle
);
1082 anv_free2(&device
->alloc
, pAllocator
, mem
);
1085 VkResult
anv_MapMemory(
1087 VkDeviceMemory _memory
,
1088 VkDeviceSize offset
,
1090 VkMemoryMapFlags flags
,
1093 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1094 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
1101 if (size
== VK_WHOLE_SIZE
)
1102 size
= mem
->bo
.size
- offset
;
1104 /* FIXME: Is this supposed to be thread safe? Since vkUnmapMemory() only
1105 * takes a VkDeviceMemory pointer, it seems like only one map of the memory
1106 * at a time is valid. We could just mmap up front and return an offset
1107 * pointer here, but that may exhaust virtual memory on 32 bit
1110 uint32_t gem_flags
= 0;
1111 if (!device
->info
.has_llc
&& mem
->type_index
== 0)
1112 gem_flags
|= I915_MMAP_WC
;
1114 /* GEM will fail to map if the offset isn't 4k-aligned. Round down. */
1115 uint64_t map_offset
= offset
& ~4095ull;
1116 assert(offset
>= map_offset
);
1117 uint64_t map_size
= (offset
+ size
) - map_offset
;
1119 /* Let's map whole pages */
1120 map_size
= align_u64(map_size
, 4096);
1122 mem
->map
= anv_gem_mmap(device
, mem
->bo
.gem_handle
,
1123 map_offset
, map_size
, gem_flags
);
1124 mem
->map_size
= map_size
;
1126 *ppData
= mem
->map
+ (offset
- map_offset
);
1131 void anv_UnmapMemory(
1133 VkDeviceMemory _memory
)
1135 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
1140 anv_gem_munmap(mem
->map
, mem
->map_size
);
1144 clflush_mapped_ranges(struct anv_device
*device
,
1146 const VkMappedMemoryRange
*ranges
)
1148 for (uint32_t i
= 0; i
< count
; i
++) {
1149 ANV_FROM_HANDLE(anv_device_memory
, mem
, ranges
[i
].memory
);
1150 void *p
= mem
->map
+ (ranges
[i
].offset
& ~CACHELINE_MASK
);
1153 if (ranges
[i
].offset
+ ranges
[i
].size
> mem
->map_size
)
1154 end
= mem
->map
+ mem
->map_size
;
1156 end
= mem
->map
+ ranges
[i
].offset
+ ranges
[i
].size
;
1159 __builtin_ia32_clflush(p
);
1160 p
+= CACHELINE_SIZE
;
1165 VkResult
anv_FlushMappedMemoryRanges(
1167 uint32_t memoryRangeCount
,
1168 const VkMappedMemoryRange
* pMemoryRanges
)
1170 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1172 if (device
->info
.has_llc
)
1175 /* Make sure the writes we're flushing have landed. */
1176 __builtin_ia32_mfence();
1178 clflush_mapped_ranges(device
, memoryRangeCount
, pMemoryRanges
);
1183 VkResult
anv_InvalidateMappedMemoryRanges(
1185 uint32_t memoryRangeCount
,
1186 const VkMappedMemoryRange
* pMemoryRanges
)
1188 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1190 if (device
->info
.has_llc
)
1193 clflush_mapped_ranges(device
, memoryRangeCount
, pMemoryRanges
);
1195 /* Make sure no reads get moved up above the invalidate. */
1196 __builtin_ia32_mfence();
1201 void anv_GetBufferMemoryRequirements(
1204 VkMemoryRequirements
* pMemoryRequirements
)
1206 ANV_FROM_HANDLE(anv_buffer
, buffer
, _buffer
);
1208 /* The Vulkan spec (git aaed022) says:
1210 * memoryTypeBits is a bitfield and contains one bit set for every
1211 * supported memory type for the resource. The bit `1<<i` is set if and
1212 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
1213 * structure for the physical device is supported.
1215 * We support exactly one memory type.
1217 pMemoryRequirements
->memoryTypeBits
= 1;
1219 pMemoryRequirements
->size
= buffer
->size
;
1220 pMemoryRequirements
->alignment
= 16;
1223 void anv_GetImageMemoryRequirements(
1226 VkMemoryRequirements
* pMemoryRequirements
)
1228 ANV_FROM_HANDLE(anv_image
, image
, _image
);
1230 /* The Vulkan spec (git aaed022) says:
1232 * memoryTypeBits is a bitfield and contains one bit set for every
1233 * supported memory type for the resource. The bit `1<<i` is set if and
1234 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
1235 * structure for the physical device is supported.
1237 * We support exactly one memory type.
1239 pMemoryRequirements
->memoryTypeBits
= 1;
1241 pMemoryRequirements
->size
= image
->size
;
1242 pMemoryRequirements
->alignment
= image
->alignment
;
1245 void anv_GetImageSparseMemoryRequirements(
1248 uint32_t* pSparseMemoryRequirementCount
,
1249 VkSparseImageMemoryRequirements
* pSparseMemoryRequirements
)
1254 void anv_GetDeviceMemoryCommitment(
1256 VkDeviceMemory memory
,
1257 VkDeviceSize
* pCommittedMemoryInBytes
)
1259 *pCommittedMemoryInBytes
= 0;
1262 VkResult
anv_BindBufferMemory(
1265 VkDeviceMemory _memory
,
1266 VkDeviceSize memoryOffset
)
1268 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
1269 ANV_FROM_HANDLE(anv_buffer
, buffer
, _buffer
);
1272 buffer
->bo
= &mem
->bo
;
1273 buffer
->offset
= memoryOffset
;
1282 VkResult
anv_BindImageMemory(
1285 VkDeviceMemory _memory
,
1286 VkDeviceSize memoryOffset
)
1288 ANV_FROM_HANDLE(anv_device_memory
, mem
, _memory
);
1289 ANV_FROM_HANDLE(anv_image
, image
, _image
);
1292 image
->bo
= &mem
->bo
;
1293 image
->offset
= memoryOffset
;
1302 VkResult
anv_QueueBindSparse(
1304 uint32_t bindInfoCount
,
1305 const VkBindSparseInfo
* pBindInfo
,
1308 stub_return(VK_ERROR_INCOMPATIBLE_DRIVER
);
1311 VkResult
anv_CreateFence(
1313 const VkFenceCreateInfo
* pCreateInfo
,
1314 const VkAllocationCallbacks
* pAllocator
,
1317 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1318 struct anv_fence
*fence
;
1319 struct anv_batch batch
;
1322 const uint32_t fence_size
= 128;
1324 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FENCE_CREATE_INFO
);
1326 fence
= anv_alloc2(&device
->alloc
, pAllocator
, sizeof(*fence
), 8,
1327 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1329 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1331 result
= anv_bo_init_new(&fence
->bo
, device
, fence_size
);
1332 if (result
!= VK_SUCCESS
)
1336 anv_gem_mmap(device
, fence
->bo
.gem_handle
, 0, fence
->bo
.size
, 0);
1337 batch
.next
= batch
.start
= fence
->bo
.map
;
1338 batch
.end
= fence
->bo
.map
+ fence
->bo
.size
;
1339 anv_batch_emit(&batch
, GEN7_MI_BATCH_BUFFER_END
);
1340 anv_batch_emit(&batch
, GEN7_MI_NOOP
);
1342 if (!device
->info
.has_llc
) {
1343 assert(((uintptr_t) fence
->bo
.map
& CACHELINE_MASK
) == 0);
1344 assert(batch
.next
- fence
->bo
.map
<= CACHELINE_SIZE
);
1345 __builtin_ia32_mfence();
1346 __builtin_ia32_clflush(fence
->bo
.map
);
1349 fence
->exec2_objects
[0].handle
= fence
->bo
.gem_handle
;
1350 fence
->exec2_objects
[0].relocation_count
= 0;
1351 fence
->exec2_objects
[0].relocs_ptr
= 0;
1352 fence
->exec2_objects
[0].alignment
= 0;
1353 fence
->exec2_objects
[0].offset
= fence
->bo
.offset
;
1354 fence
->exec2_objects
[0].flags
= 0;
1355 fence
->exec2_objects
[0].rsvd1
= 0;
1356 fence
->exec2_objects
[0].rsvd2
= 0;
1358 fence
->execbuf
.buffers_ptr
= (uintptr_t) fence
->exec2_objects
;
1359 fence
->execbuf
.buffer_count
= 1;
1360 fence
->execbuf
.batch_start_offset
= 0;
1361 fence
->execbuf
.batch_len
= batch
.next
- fence
->bo
.map
;
1362 fence
->execbuf
.cliprects_ptr
= 0;
1363 fence
->execbuf
.num_cliprects
= 0;
1364 fence
->execbuf
.DR1
= 0;
1365 fence
->execbuf
.DR4
= 0;
1367 fence
->execbuf
.flags
=
1368 I915_EXEC_HANDLE_LUT
| I915_EXEC_NO_RELOC
| I915_EXEC_RENDER
;
1369 fence
->execbuf
.rsvd1
= device
->context_id
;
1370 fence
->execbuf
.rsvd2
= 0;
1372 *pFence
= anv_fence_to_handle(fence
);
1377 anv_free2(&device
->alloc
, pAllocator
, fence
);
1382 void anv_DestroyFence(
1385 const VkAllocationCallbacks
* pAllocator
)
1387 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1388 ANV_FROM_HANDLE(anv_fence
, fence
, _fence
);
1390 anv_gem_munmap(fence
->bo
.map
, fence
->bo
.size
);
1391 anv_gem_close(device
, fence
->bo
.gem_handle
);
1392 anv_free2(&device
->alloc
, pAllocator
, fence
);
1395 VkResult
anv_ResetFences(
1397 uint32_t fenceCount
,
1398 const VkFence
* pFences
)
1400 for (uint32_t i
= 0; i
< fenceCount
; i
++) {
1401 ANV_FROM_HANDLE(anv_fence
, fence
, pFences
[i
]);
1402 fence
->ready
= false;
1408 VkResult
anv_GetFenceStatus(
1412 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1413 ANV_FROM_HANDLE(anv_fence
, fence
, _fence
);
1420 ret
= anv_gem_wait(device
, fence
->bo
.gem_handle
, &t
);
1422 fence
->ready
= true;
1426 return VK_NOT_READY
;
1429 VkResult
anv_WaitForFences(
1431 uint32_t fenceCount
,
1432 const VkFence
* pFences
,
1436 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1438 /* DRM_IOCTL_I915_GEM_WAIT uses a signed 64 bit timeout and is supposed
1439 * to block indefinitely timeouts <= 0. Unfortunately, this was broken
1440 * for a couple of kernel releases. Since there's no way to know
1441 * whether or not the kernel we're using is one of the broken ones, the
1442 * best we can do is to clamp the timeout to INT64_MAX. This limits the
1443 * maximum timeout from 584 years to 292 years - likely not a big deal.
1445 if (timeout
> INT64_MAX
)
1446 timeout
= INT64_MAX
;
1448 int64_t t
= timeout
;
1450 /* FIXME: handle !waitAll */
1452 for (uint32_t i
= 0; i
< fenceCount
; i
++) {
1453 ANV_FROM_HANDLE(anv_fence
, fence
, pFences
[i
]);
1454 int ret
= anv_gem_wait(device
, fence
->bo
.gem_handle
, &t
);
1455 if (ret
== -1 && errno
== ETIME
) {
1457 } else if (ret
== -1) {
1458 /* We don't know the real error. */
1459 return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY
,
1460 "gem wait failed: %m");
1467 // Queue semaphore functions
1469 VkResult
anv_CreateSemaphore(
1471 const VkSemaphoreCreateInfo
* pCreateInfo
,
1472 const VkAllocationCallbacks
* pAllocator
,
1473 VkSemaphore
* pSemaphore
)
1475 /* The DRM execbuffer ioctl always execute in-oder, even between different
1476 * rings. As such, there's nothing to do for the user space semaphore.
1479 *pSemaphore
= (VkSemaphore
)1;
1484 void anv_DestroySemaphore(
1486 VkSemaphore semaphore
,
1487 const VkAllocationCallbacks
* pAllocator
)
1493 VkResult
anv_CreateEvent(
1495 const VkEventCreateInfo
* pCreateInfo
,
1496 const VkAllocationCallbacks
* pAllocator
,
1499 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1500 struct anv_state state
;
1501 struct anv_event
*event
;
1503 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_EVENT_CREATE_INFO
);
1505 state
= anv_state_pool_alloc(&device
->dynamic_state_pool
,
1508 event
->state
= state
;
1509 event
->semaphore
= VK_EVENT_RESET
;
1511 if (!device
->info
.has_llc
) {
1512 /* Make sure the writes we're flushing have landed. */
1513 __builtin_ia32_mfence();
1514 __builtin_ia32_clflush(event
);
1517 *pEvent
= anv_event_to_handle(event
);
1522 void anv_DestroyEvent(
1525 const VkAllocationCallbacks
* pAllocator
)
1527 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1528 ANV_FROM_HANDLE(anv_event
, event
, _event
);
1530 anv_state_pool_free(&device
->dynamic_state_pool
, event
->state
);
1533 VkResult
anv_GetEventStatus(
1537 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1538 ANV_FROM_HANDLE(anv_event
, event
, _event
);
1540 if (!device
->info
.has_llc
) {
1541 /* Invalidate read cache before reading event written by GPU. */
1542 __builtin_ia32_clflush(event
);
1543 __builtin_ia32_mfence();
1547 return event
->semaphore
;
1550 VkResult
anv_SetEvent(
1554 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1555 ANV_FROM_HANDLE(anv_event
, event
, _event
);
1557 event
->semaphore
= VK_EVENT_SET
;
1559 if (!device
->info
.has_llc
) {
1560 /* Make sure the writes we're flushing have landed. */
1561 __builtin_ia32_mfence();
1562 __builtin_ia32_clflush(event
);
1568 VkResult
anv_ResetEvent(
1572 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1573 ANV_FROM_HANDLE(anv_event
, event
, _event
);
1575 event
->semaphore
= VK_EVENT_RESET
;
1577 if (!device
->info
.has_llc
) {
1578 /* Make sure the writes we're flushing have landed. */
1579 __builtin_ia32_mfence();
1580 __builtin_ia32_clflush(event
);
1588 VkResult
anv_CreateBuffer(
1590 const VkBufferCreateInfo
* pCreateInfo
,
1591 const VkAllocationCallbacks
* pAllocator
,
1594 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1595 struct anv_buffer
*buffer
;
1597 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
1599 buffer
= anv_alloc2(&device
->alloc
, pAllocator
, sizeof(*buffer
), 8,
1600 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1602 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1604 buffer
->size
= pCreateInfo
->size
;
1605 buffer
->usage
= pCreateInfo
->usage
;
1609 *pBuffer
= anv_buffer_to_handle(buffer
);
1614 void anv_DestroyBuffer(
1617 const VkAllocationCallbacks
* pAllocator
)
1619 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1620 ANV_FROM_HANDLE(anv_buffer
, buffer
, _buffer
);
1622 anv_free2(&device
->alloc
, pAllocator
, buffer
);
1626 anv_fill_buffer_surface_state(struct anv_device
*device
, struct anv_state state
,
1627 enum isl_format format
,
1628 uint32_t offset
, uint32_t range
, uint32_t stride
)
1630 switch (device
->info
.gen
) {
1632 if (device
->info
.is_haswell
)
1633 gen75_fill_buffer_surface_state(state
.map
, format
, offset
, range
,
1636 gen7_fill_buffer_surface_state(state
.map
, format
, offset
, range
,
1640 gen8_fill_buffer_surface_state(state
.map
, format
, offset
, range
, stride
);
1643 gen9_fill_buffer_surface_state(state
.map
, format
, offset
, range
, stride
);
1646 unreachable("unsupported gen\n");
1649 if (!device
->info
.has_llc
)
1650 anv_state_clflush(state
);
1653 void anv_DestroySampler(
1656 const VkAllocationCallbacks
* pAllocator
)
1658 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1659 ANV_FROM_HANDLE(anv_sampler
, sampler
, _sampler
);
1661 anv_free2(&device
->alloc
, pAllocator
, sampler
);
1664 VkResult
anv_CreateFramebuffer(
1666 const VkFramebufferCreateInfo
* pCreateInfo
,
1667 const VkAllocationCallbacks
* pAllocator
,
1668 VkFramebuffer
* pFramebuffer
)
1670 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1671 struct anv_framebuffer
*framebuffer
;
1673 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
1675 size_t size
= sizeof(*framebuffer
) +
1676 sizeof(struct anv_image_view
*) * pCreateInfo
->attachmentCount
;
1677 framebuffer
= anv_alloc2(&device
->alloc
, pAllocator
, size
, 8,
1678 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1679 if (framebuffer
== NULL
)
1680 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1682 framebuffer
->attachment_count
= pCreateInfo
->attachmentCount
;
1683 for (uint32_t i
= 0; i
< pCreateInfo
->attachmentCount
; i
++) {
1684 VkImageView _iview
= pCreateInfo
->pAttachments
[i
];
1685 framebuffer
->attachments
[i
] = anv_image_view_from_handle(_iview
);
1688 framebuffer
->width
= pCreateInfo
->width
;
1689 framebuffer
->height
= pCreateInfo
->height
;
1690 framebuffer
->layers
= pCreateInfo
->layers
;
1692 *pFramebuffer
= anv_framebuffer_to_handle(framebuffer
);
1697 void anv_DestroyFramebuffer(
1700 const VkAllocationCallbacks
* pAllocator
)
1702 ANV_FROM_HANDLE(anv_device
, device
, _device
);
1703 ANV_FROM_HANDLE(anv_framebuffer
, fb
, _fb
);
1705 anv_free2(&device
->alloc
, pAllocator
, fb
);
1708 void vkCmdDbgMarkerBegin(
1709 VkCommandBuffer commandBuffer
,
1710 const char* pMarker
)
1711 __attribute__ ((visibility ("default")));
1713 void vkCmdDbgMarkerEnd(
1714 VkCommandBuffer commandBuffer
)
1715 __attribute__ ((visibility ("default")));
1717 void vkCmdDbgMarkerBegin(
1718 VkCommandBuffer commandBuffer
,
1719 const char* pMarker
)
1723 void vkCmdDbgMarkerEnd(
1724 VkCommandBuffer commandBuffer
)