2 * Copyright © 2016 Red Hat.
3 * Copyright © 2016 Bas Nieuwenhuizen
5 * based in part on anv driver which is:
6 * Copyright © 2015 Intel Corporation
8 * Permission is hereby granted, free of charge, to any person obtaining a
9 * copy of this software and associated documentation files (the "Software"),
10 * to deal in the Software without restriction, including without limitation
11 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
12 * and/or sell copies of the Software, and to permit persons to whom the
13 * Software is furnished to do so, subject to the following conditions:
15 * The above copyright notice and this permission notice (including the next
16 * paragraph) shall be included in all copies or substantial portions of the
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
22 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
23 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
24 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
32 #include "radv_private.h"
33 #include "util/strtod.h"
37 #include <amdgpu_drm.h>
38 #include "amdgpu_id.h"
39 #include "winsys/amdgpu/radv_amdgpu_winsys_public.h"
40 #include "ac_llvm_util.h"
41 #include "vk_format.h"
43 #include "radv_timestamp.h"
44 #include "util/debug.h"
45 struct radv_dispatch_table dtable
;
48 radv_physical_device_init(struct radv_physical_device
*device
,
49 struct radv_instance
*instance
,
53 drmVersionPtr version
;
56 fd
= open(path
, O_RDWR
| O_CLOEXEC
);
58 return vk_errorf(VK_ERROR_INCOMPATIBLE_DRIVER
,
59 "failed to open %s: %m", path
);
61 version
= drmGetVersion(fd
);
64 return vk_errorf(VK_ERROR_INCOMPATIBLE_DRIVER
,
65 "failed to get version %s: %m", path
);
68 if (strcmp(version
->name
, "amdgpu")) {
69 drmFreeVersion(version
);
71 return VK_ERROR_INCOMPATIBLE_DRIVER
;
73 drmFreeVersion(version
);
75 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
76 device
->instance
= instance
;
77 assert(strlen(path
) < ARRAY_SIZE(device
->path
));
78 strncpy(device
->path
, path
, ARRAY_SIZE(device
->path
));
80 device
->ws
= radv_amdgpu_winsys_create(fd
);
82 result
= VK_ERROR_INCOMPATIBLE_DRIVER
;
85 device
->ws
->query_info(device
->ws
, &device
->rad_info
);
86 result
= radv_init_wsi(device
);
87 if (result
!= VK_SUCCESS
) {
88 device
->ws
->destroy(device
->ws
);
92 fprintf(stderr
, "WARNING: radv is not a conformant vulkan implementation, testing use only.\n");
93 device
->name
= device
->rad_info
.name
;
102 radv_physical_device_finish(struct radv_physical_device
*device
)
104 radv_finish_wsi(device
);
105 device
->ws
->destroy(device
->ws
);
108 static const VkExtensionProperties global_extensions
[] = {
110 .extensionName
= VK_KHR_SURFACE_EXTENSION_NAME
,
113 #ifdef VK_USE_PLATFORM_XCB_KHR
115 .extensionName
= VK_KHR_XCB_SURFACE_EXTENSION_NAME
,
119 #ifdef VK_USE_PLATFORM_WAYLAND_KHR
121 .extensionName
= VK_KHR_WAYLAND_SURFACE_EXTENSION_NAME
,
127 static const VkExtensionProperties device_extensions
[] = {
129 .extensionName
= VK_KHR_SWAPCHAIN_EXTENSION_NAME
,
135 default_alloc_func(void *pUserData
, size_t size
, size_t align
,
136 VkSystemAllocationScope allocationScope
)
142 default_realloc_func(void *pUserData
, void *pOriginal
, size_t size
,
143 size_t align
, VkSystemAllocationScope allocationScope
)
145 return realloc(pOriginal
, size
);
149 default_free_func(void *pUserData
, void *pMemory
)
154 static const VkAllocationCallbacks default_alloc
= {
156 .pfnAllocation
= default_alloc_func
,
157 .pfnReallocation
= default_realloc_func
,
158 .pfnFree
= default_free_func
,
161 VkResult
radv_CreateInstance(
162 const VkInstanceCreateInfo
* pCreateInfo
,
163 const VkAllocationCallbacks
* pAllocator
,
164 VkInstance
* pInstance
)
166 struct radv_instance
*instance
;
168 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO
);
170 uint32_t client_version
;
171 if (pCreateInfo
->pApplicationInfo
&&
172 pCreateInfo
->pApplicationInfo
->apiVersion
!= 0) {
173 client_version
= pCreateInfo
->pApplicationInfo
->apiVersion
;
175 client_version
= VK_MAKE_VERSION(1, 0, 0);
178 if (VK_MAKE_VERSION(1, 0, 0) > client_version
||
179 client_version
> VK_MAKE_VERSION(1, 0, 0xfff)) {
180 return vk_errorf(VK_ERROR_INCOMPATIBLE_DRIVER
,
181 "Client requested version %d.%d.%d",
182 VK_VERSION_MAJOR(client_version
),
183 VK_VERSION_MINOR(client_version
),
184 VK_VERSION_PATCH(client_version
));
187 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
189 for (uint32_t j
= 0; j
< ARRAY_SIZE(global_extensions
); j
++) {
190 if (strcmp(pCreateInfo
->ppEnabledExtensionNames
[i
],
191 global_extensions
[j
].extensionName
) == 0) {
197 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
200 instance
= vk_alloc2(&default_alloc
, pAllocator
, sizeof(*instance
), 8,
201 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE
);
203 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
205 instance
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
208 instance
->alloc
= *pAllocator
;
210 instance
->alloc
= default_alloc
;
212 instance
->apiVersion
= client_version
;
213 instance
->physicalDeviceCount
= -1;
217 VG(VALGRIND_CREATE_MEMPOOL(instance
, 0, false));
219 *pInstance
= radv_instance_to_handle(instance
);
224 void radv_DestroyInstance(
225 VkInstance _instance
,
226 const VkAllocationCallbacks
* pAllocator
)
228 RADV_FROM_HANDLE(radv_instance
, instance
, _instance
);
230 if (instance
->physicalDeviceCount
> 0) {
231 /* We support at most one physical device. */
232 assert(instance
->physicalDeviceCount
== 1);
233 radv_physical_device_finish(&instance
->physicalDevice
);
236 VG(VALGRIND_DESTROY_MEMPOOL(instance
));
240 vk_free(&instance
->alloc
, instance
);
243 VkResult
radv_EnumeratePhysicalDevices(
244 VkInstance _instance
,
245 uint32_t* pPhysicalDeviceCount
,
246 VkPhysicalDevice
* pPhysicalDevices
)
248 RADV_FROM_HANDLE(radv_instance
, instance
, _instance
);
251 if (instance
->physicalDeviceCount
< 0) {
253 for (unsigned i
= 0; i
< 8; i
++) {
254 snprintf(path
, sizeof(path
), "/dev/dri/renderD%d", 128 + i
);
255 result
= radv_physical_device_init(&instance
->physicalDevice
,
257 if (result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
261 if (result
== VK_ERROR_INCOMPATIBLE_DRIVER
) {
262 instance
->physicalDeviceCount
= 0;
263 } else if (result
== VK_SUCCESS
) {
264 instance
->physicalDeviceCount
= 1;
270 /* pPhysicalDeviceCount is an out parameter if pPhysicalDevices is NULL;
271 * otherwise it's an inout parameter.
273 * The Vulkan spec (git aaed022) says:
275 * pPhysicalDeviceCount is a pointer to an unsigned integer variable
276 * that is initialized with the number of devices the application is
277 * prepared to receive handles to. pname:pPhysicalDevices is pointer to
278 * an array of at least this many VkPhysicalDevice handles [...].
280 * Upon success, if pPhysicalDevices is NULL, vkEnumeratePhysicalDevices
281 * overwrites the contents of the variable pointed to by
282 * pPhysicalDeviceCount with the number of physical devices in in the
283 * instance; otherwise, vkEnumeratePhysicalDevices overwrites
284 * pPhysicalDeviceCount with the number of physical handles written to
287 if (!pPhysicalDevices
) {
288 *pPhysicalDeviceCount
= instance
->physicalDeviceCount
;
289 } else if (*pPhysicalDeviceCount
>= 1) {
290 pPhysicalDevices
[0] = radv_physical_device_to_handle(&instance
->physicalDevice
);
291 *pPhysicalDeviceCount
= 1;
292 } else if (*pPhysicalDeviceCount
< instance
->physicalDeviceCount
) {
293 return VK_INCOMPLETE
;
295 *pPhysicalDeviceCount
= 0;
301 void radv_GetPhysicalDeviceFeatures(
302 VkPhysicalDevice physicalDevice
,
303 VkPhysicalDeviceFeatures
* pFeatures
)
305 // RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
307 memset(pFeatures
, 0, sizeof(*pFeatures
));
309 *pFeatures
= (VkPhysicalDeviceFeatures
) {
310 .robustBufferAccess
= true,
311 .fullDrawIndexUint32
= true,
312 .imageCubeArray
= true,
313 .independentBlend
= true,
314 .geometryShader
= false,
315 .tessellationShader
= false,
316 .sampleRateShading
= false,
317 .dualSrcBlend
= true,
319 .multiDrawIndirect
= true,
320 .drawIndirectFirstInstance
= true,
322 .depthBiasClamp
= true,
323 .fillModeNonSolid
= true,
328 .multiViewport
= false,
329 .samplerAnisotropy
= false, /* FINISHME */
330 .textureCompressionETC2
= false,
331 .textureCompressionASTC_LDR
= false,
332 .textureCompressionBC
= true,
333 .occlusionQueryPrecise
= true,
334 .pipelineStatisticsQuery
= false,
335 .vertexPipelineStoresAndAtomics
= true,
336 .fragmentStoresAndAtomics
= true,
337 .shaderTessellationAndGeometryPointSize
= true,
338 .shaderImageGatherExtended
= false,
339 .shaderStorageImageExtendedFormats
= false,
340 .shaderStorageImageMultisample
= false,
341 .shaderUniformBufferArrayDynamicIndexing
= true,
342 .shaderSampledImageArrayDynamicIndexing
= true,
343 .shaderStorageBufferArrayDynamicIndexing
= true,
344 .shaderStorageImageArrayDynamicIndexing
= true,
345 .shaderStorageImageReadWithoutFormat
= false,
346 .shaderStorageImageWriteWithoutFormat
= true,
347 .shaderClipDistance
= true,
348 .shaderCullDistance
= true,
349 .shaderFloat64
= false,
350 .shaderInt64
= false,
351 .shaderInt16
= false,
353 .variableMultisampleRate
= false,
354 .inheritedQueries
= false,
359 radv_device_get_cache_uuid(void *uuid
)
361 memset(uuid
, 0, VK_UUID_SIZE
);
362 snprintf(uuid
, VK_UUID_SIZE
, "radv-%s", RADV_TIMESTAMP
);
365 void radv_GetPhysicalDeviceProperties(
366 VkPhysicalDevice physicalDevice
,
367 VkPhysicalDeviceProperties
* pProperties
)
369 RADV_FROM_HANDLE(radv_physical_device
, pdevice
, physicalDevice
);
370 VkSampleCountFlags sample_counts
= 0xf;
371 VkPhysicalDeviceLimits limits
= {
372 .maxImageDimension1D
= (1 << 14),
373 .maxImageDimension2D
= (1 << 14),
374 .maxImageDimension3D
= (1 << 11),
375 .maxImageDimensionCube
= (1 << 14),
376 .maxImageArrayLayers
= (1 << 11),
377 .maxTexelBufferElements
= 128 * 1024 * 1024,
378 .maxUniformBufferRange
= UINT32_MAX
,
379 .maxStorageBufferRange
= UINT32_MAX
,
380 .maxPushConstantsSize
= MAX_PUSH_CONSTANTS_SIZE
,
381 .maxMemoryAllocationCount
= UINT32_MAX
,
382 .maxSamplerAllocationCount
= 64 * 1024,
383 .bufferImageGranularity
= 64, /* A cache line */
384 .sparseAddressSpaceSize
= 0,
385 .maxBoundDescriptorSets
= MAX_SETS
,
386 .maxPerStageDescriptorSamplers
= 64,
387 .maxPerStageDescriptorUniformBuffers
= 64,
388 .maxPerStageDescriptorStorageBuffers
= 64,
389 .maxPerStageDescriptorSampledImages
= 64,
390 .maxPerStageDescriptorStorageImages
= 64,
391 .maxPerStageDescriptorInputAttachments
= 64,
392 .maxPerStageResources
= 128,
393 .maxDescriptorSetSamplers
= 256,
394 .maxDescriptorSetUniformBuffers
= 256,
395 .maxDescriptorSetUniformBuffersDynamic
= 256,
396 .maxDescriptorSetStorageBuffers
= 256,
397 .maxDescriptorSetStorageBuffersDynamic
= 256,
398 .maxDescriptorSetSampledImages
= 256,
399 .maxDescriptorSetStorageImages
= 256,
400 .maxDescriptorSetInputAttachments
= 256,
401 .maxVertexInputAttributes
= 32,
402 .maxVertexInputBindings
= 32,
403 .maxVertexInputAttributeOffset
= 2047,
404 .maxVertexInputBindingStride
= 2048,
405 .maxVertexOutputComponents
= 128,
406 .maxTessellationGenerationLevel
= 0,
407 .maxTessellationPatchSize
= 0,
408 .maxTessellationControlPerVertexInputComponents
= 0,
409 .maxTessellationControlPerVertexOutputComponents
= 0,
410 .maxTessellationControlPerPatchOutputComponents
= 0,
411 .maxTessellationControlTotalOutputComponents
= 0,
412 .maxTessellationEvaluationInputComponents
= 0,
413 .maxTessellationEvaluationOutputComponents
= 0,
414 .maxGeometryShaderInvocations
= 32,
415 .maxGeometryInputComponents
= 64,
416 .maxGeometryOutputComponents
= 128,
417 .maxGeometryOutputVertices
= 256,
418 .maxGeometryTotalOutputComponents
= 1024,
419 .maxFragmentInputComponents
= 128,
420 .maxFragmentOutputAttachments
= 8,
421 .maxFragmentDualSrcAttachments
= 2,
422 .maxFragmentCombinedOutputResources
= 8,
423 .maxComputeSharedMemorySize
= 32768,
424 .maxComputeWorkGroupCount
= { 65535, 65535, 65535 },
425 .maxComputeWorkGroupInvocations
= 16 * 1024,
426 .maxComputeWorkGroupSize
= {
427 16 * 1024/*devinfo->max_cs_threads*/,
431 .subPixelPrecisionBits
= 4 /* FIXME */,
432 .subTexelPrecisionBits
= 4 /* FIXME */,
433 .mipmapPrecisionBits
= 4 /* FIXME */,
434 .maxDrawIndexedIndexValue
= UINT32_MAX
,
435 .maxDrawIndirectCount
= UINT32_MAX
,
436 .maxSamplerLodBias
= 16,
437 .maxSamplerAnisotropy
= 16,
438 .maxViewports
= MAX_VIEWPORTS
,
439 .maxViewportDimensions
= { (1 << 14), (1 << 14) },
440 .viewportBoundsRange
= { INT16_MIN
, INT16_MAX
},
441 .viewportSubPixelBits
= 13, /* We take a float? */
442 .minMemoryMapAlignment
= 4096, /* A page */
443 .minTexelBufferOffsetAlignment
= 1,
444 .minUniformBufferOffsetAlignment
= 4,
445 .minStorageBufferOffsetAlignment
= 4,
446 .minTexelOffset
= -8,
448 .minTexelGatherOffset
= -8,
449 .maxTexelGatherOffset
= 7,
450 .minInterpolationOffset
= 0, /* FIXME */
451 .maxInterpolationOffset
= 0, /* FIXME */
452 .subPixelInterpolationOffsetBits
= 0, /* FIXME */
453 .maxFramebufferWidth
= (1 << 14),
454 .maxFramebufferHeight
= (1 << 14),
455 .maxFramebufferLayers
= (1 << 10),
456 .framebufferColorSampleCounts
= sample_counts
,
457 .framebufferDepthSampleCounts
= sample_counts
,
458 .framebufferStencilSampleCounts
= sample_counts
,
459 .framebufferNoAttachmentsSampleCounts
= sample_counts
,
460 .maxColorAttachments
= MAX_RTS
,
461 .sampledImageColorSampleCounts
= sample_counts
,
462 .sampledImageIntegerSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
463 .sampledImageDepthSampleCounts
= sample_counts
,
464 .sampledImageStencilSampleCounts
= sample_counts
,
465 .storageImageSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
466 .maxSampleMaskWords
= 1,
467 .timestampComputeAndGraphics
= false,
468 .timestampPeriod
= 100000.0 / pdevice
->rad_info
.clock_crystal_freq
,
469 .maxClipDistances
= 8,
470 .maxCullDistances
= 8,
471 .maxCombinedClipAndCullDistances
= 8,
472 .discreteQueuePriorities
= 1,
473 .pointSizeRange
= { 0.125, 255.875 },
474 .lineWidthRange
= { 0.0, 7.9921875 },
475 .pointSizeGranularity
= (1.0 / 8.0),
476 .lineWidthGranularity
= (1.0 / 128.0),
477 .strictLines
= false, /* FINISHME */
478 .standardSampleLocations
= true,
479 .optimalBufferCopyOffsetAlignment
= 128,
480 .optimalBufferCopyRowPitchAlignment
= 128,
481 .nonCoherentAtomSize
= 64,
484 *pProperties
= (VkPhysicalDeviceProperties
) {
485 .apiVersion
= VK_MAKE_VERSION(1, 0, 5),
488 .deviceID
= pdevice
->rad_info
.pci_id
,
489 .deviceType
= VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU
,
491 .sparseProperties
= {0}, /* Broadwell doesn't do sparse. */
494 strcpy(pProperties
->deviceName
, pdevice
->name
);
495 radv_device_get_cache_uuid(pProperties
->pipelineCacheUUID
);
498 void radv_GetPhysicalDeviceQueueFamilyProperties(
499 VkPhysicalDevice physicalDevice
,
501 VkQueueFamilyProperties
* pQueueFamilyProperties
)
503 if (pQueueFamilyProperties
== NULL
) {
507 assert(*pCount
>= 1);
509 *pQueueFamilyProperties
= (VkQueueFamilyProperties
) {
510 .queueFlags
= VK_QUEUE_GRAPHICS_BIT
|
511 VK_QUEUE_COMPUTE_BIT
|
512 VK_QUEUE_TRANSFER_BIT
,
514 .timestampValidBits
= 64,
515 .minImageTransferGranularity
= (VkExtent3D
) { 1, 1, 1 },
519 void radv_GetPhysicalDeviceMemoryProperties(
520 VkPhysicalDevice physicalDevice
,
521 VkPhysicalDeviceMemoryProperties
* pMemoryProperties
)
523 RADV_FROM_HANDLE(radv_physical_device
, physical_device
, physicalDevice
);
525 pMemoryProperties
->memoryTypeCount
= 3;
526 pMemoryProperties
->memoryTypes
[0] = (VkMemoryType
) {
527 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
,
530 pMemoryProperties
->memoryTypes
[1] = (VkMemoryType
) {
531 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
532 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
533 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
,
536 pMemoryProperties
->memoryTypes
[2] = (VkMemoryType
) {
537 .propertyFlags
= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
538 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
|
539 VK_MEMORY_PROPERTY_HOST_CACHED_BIT
,
543 pMemoryProperties
->memoryHeapCount
= 2;
544 pMemoryProperties
->memoryHeaps
[0] = (VkMemoryHeap
) {
545 .size
= physical_device
->rad_info
.vram_size
,
546 .flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
,
548 pMemoryProperties
->memoryHeaps
[1] = (VkMemoryHeap
) {
549 .size
= physical_device
->rad_info
.gart_size
,
555 radv_queue_init(struct radv_device
*device
, struct radv_queue
*queue
)
557 queue
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
558 queue
->device
= device
;
564 radv_queue_finish(struct radv_queue
*queue
)
568 VkResult
radv_CreateDevice(
569 VkPhysicalDevice physicalDevice
,
570 const VkDeviceCreateInfo
* pCreateInfo
,
571 const VkAllocationCallbacks
* pAllocator
,
574 RADV_FROM_HANDLE(radv_physical_device
, physical_device
, physicalDevice
);
576 struct radv_device
*device
;
578 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
580 for (uint32_t j
= 0; j
< ARRAY_SIZE(device_extensions
); j
++) {
581 if (strcmp(pCreateInfo
->ppEnabledExtensionNames
[i
],
582 device_extensions
[j
].extensionName
) == 0) {
588 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
591 device
= vk_alloc2(&physical_device
->instance
->alloc
, pAllocator
,
593 VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
595 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
597 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
598 device
->instance
= physical_device
->instance
;
600 device
->ws
= physical_device
->ws
;
602 device
->alloc
= *pAllocator
;
604 device
->alloc
= physical_device
->instance
->alloc
;
606 device
->hw_ctx
= device
->ws
->ctx_create(device
->ws
);
607 if (!device
->hw_ctx
) {
608 result
= VK_ERROR_OUT_OF_HOST_MEMORY
;
612 radv_queue_init(device
, &device
->queue
);
614 result
= radv_device_init_meta(device
);
615 if (result
!= VK_SUCCESS
) {
616 device
->ws
->ctx_destroy(device
->hw_ctx
);
619 device
->allow_fast_clears
= env_var_as_boolean("RADV_FAST_CLEARS", false);
620 device
->allow_dcc
= !env_var_as_boolean("RADV_DCC_DISABLE", false);
622 if (device
->allow_fast_clears
&& device
->allow_dcc
)
623 radv_finishme("DCC fast clears have not been tested\n");
625 radv_device_init_msaa(device
);
626 device
->empty_cs
= device
->ws
->cs_create(device
->ws
, RING_GFX
);
627 radeon_emit(device
->empty_cs
, PKT3(PKT3_CONTEXT_CONTROL
, 1, 0));
628 radeon_emit(device
->empty_cs
, CONTEXT_CONTROL_LOAD_ENABLE(1));
629 radeon_emit(device
->empty_cs
, CONTEXT_CONTROL_SHADOW_ENABLE(1));
630 device
->ws
->cs_finalize(device
->empty_cs
);
631 *pDevice
= radv_device_to_handle(device
);
634 vk_free(&device
->alloc
, device
);
638 void radv_DestroyDevice(
640 const VkAllocationCallbacks
* pAllocator
)
642 RADV_FROM_HANDLE(radv_device
, device
, _device
);
644 device
->ws
->ctx_destroy(device
->hw_ctx
);
645 radv_queue_finish(&device
->queue
);
646 radv_device_finish_meta(device
);
648 vk_free(&device
->alloc
, device
);
651 VkResult
radv_EnumerateInstanceExtensionProperties(
652 const char* pLayerName
,
653 uint32_t* pPropertyCount
,
654 VkExtensionProperties
* pProperties
)
657 if (pProperties
== NULL
) {
658 *pPropertyCount
= ARRAY_SIZE(global_extensions
);
662 for (i
= 0; i
< *pPropertyCount
; i
++)
663 memcpy(&pProperties
[i
], &global_extensions
[i
], sizeof(VkExtensionProperties
));
666 if (i
< ARRAY_SIZE(global_extensions
))
667 return VK_INCOMPLETE
;
672 VkResult
radv_EnumerateDeviceExtensionProperties(
673 VkPhysicalDevice physicalDevice
,
674 const char* pLayerName
,
675 uint32_t* pPropertyCount
,
676 VkExtensionProperties
* pProperties
)
680 if (pProperties
== NULL
) {
681 *pPropertyCount
= ARRAY_SIZE(device_extensions
);
685 for (i
= 0; i
< *pPropertyCount
; i
++)
686 memcpy(&pProperties
[i
], &device_extensions
[i
], sizeof(VkExtensionProperties
));
689 if (i
< ARRAY_SIZE(device_extensions
))
690 return VK_INCOMPLETE
;
694 VkResult
radv_EnumerateInstanceLayerProperties(
695 uint32_t* pPropertyCount
,
696 VkLayerProperties
* pProperties
)
698 if (pProperties
== NULL
) {
703 /* None supported at this time */
704 return vk_error(VK_ERROR_LAYER_NOT_PRESENT
);
707 VkResult
radv_EnumerateDeviceLayerProperties(
708 VkPhysicalDevice physicalDevice
,
709 uint32_t* pPropertyCount
,
710 VkLayerProperties
* pProperties
)
712 if (pProperties
== NULL
) {
717 /* None supported at this time */
718 return vk_error(VK_ERROR_LAYER_NOT_PRESENT
);
721 void radv_GetDeviceQueue(
723 uint32_t queueNodeIndex
,
727 RADV_FROM_HANDLE(radv_device
, device
, _device
);
729 assert(queueIndex
== 0);
731 *pQueue
= radv_queue_to_handle(&device
->queue
);
734 VkResult
radv_QueueSubmit(
736 uint32_t submitCount
,
737 const VkSubmitInfo
* pSubmits
,
740 RADV_FROM_HANDLE(radv_queue
, queue
, _queue
);
741 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
742 struct radeon_winsys_fence
*base_fence
= fence
? fence
->fence
: NULL
;
743 struct radeon_winsys_ctx
*ctx
= queue
->device
->hw_ctx
;
746 for (uint32_t i
= 0; i
< submitCount
; i
++) {
747 struct radeon_winsys_cs
**cs_array
;
748 bool can_patch
= true;
750 if (!pSubmits
[i
].commandBufferCount
)
753 cs_array
= malloc(sizeof(struct radeon_winsys_cs
*) *
754 pSubmits
[i
].commandBufferCount
);
756 for (uint32_t j
= 0; j
< pSubmits
[i
].commandBufferCount
; j
++) {
757 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
,
758 pSubmits
[i
].pCommandBuffers
[j
]);
759 assert(cmd_buffer
->level
== VK_COMMAND_BUFFER_LEVEL_PRIMARY
);
761 cs_array
[j
] = cmd_buffer
->cs
;
762 if ((cmd_buffer
->usage_flags
& VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT
))
765 ret
= queue
->device
->ws
->cs_submit(ctx
, cs_array
,
766 pSubmits
[i
].commandBufferCount
,
767 can_patch
, base_fence
);
769 radv_loge("failed to submit CS %d\n", i
);
775 ret
= queue
->device
->ws
->cs_submit(ctx
, &queue
->device
->empty_cs
,
776 1, false, base_fence
);
778 fence
->submitted
= true;
784 VkResult
radv_QueueWaitIdle(
787 RADV_FROM_HANDLE(radv_queue
, queue
, _queue
);
789 queue
->device
->ws
->ctx_wait_idle(queue
->device
->hw_ctx
);
793 VkResult
radv_DeviceWaitIdle(
796 RADV_FROM_HANDLE(radv_device
, device
, _device
);
798 device
->ws
->ctx_wait_idle(device
->hw_ctx
);
802 PFN_vkVoidFunction
radv_GetInstanceProcAddr(
806 return radv_lookup_entrypoint(pName
);
809 /* The loader wants us to expose a second GetInstanceProcAddr function
810 * to work around certain LD_PRELOAD issues seen in apps.
813 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
818 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
822 return radv_GetInstanceProcAddr(instance
, pName
);
825 PFN_vkVoidFunction
radv_GetDeviceProcAddr(
829 return radv_lookup_entrypoint(pName
);
832 VkResult
radv_AllocateMemory(
834 const VkMemoryAllocateInfo
* pAllocateInfo
,
835 const VkAllocationCallbacks
* pAllocator
,
836 VkDeviceMemory
* pMem
)
838 RADV_FROM_HANDLE(radv_device
, device
, _device
);
839 struct radv_device_memory
*mem
;
841 enum radeon_bo_domain domain
;
843 assert(pAllocateInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO
);
845 if (pAllocateInfo
->allocationSize
== 0) {
846 /* Apparently, this is allowed */
847 *pMem
= VK_NULL_HANDLE
;
851 mem
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*mem
), 8,
852 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
854 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
856 uint64_t alloc_size
= align_u64(pAllocateInfo
->allocationSize
, 4096);
857 if (pAllocateInfo
->memoryTypeIndex
== 2)
858 domain
= RADEON_DOMAIN_GTT
;
860 domain
= RADEON_DOMAIN_VRAM
;
862 if (pAllocateInfo
->memoryTypeIndex
== 0)
863 flags
|= RADEON_FLAG_NO_CPU_ACCESS
;
865 flags
|= RADEON_FLAG_CPU_ACCESS
;
866 mem
->bo
= device
->ws
->buffer_create(device
->ws
, alloc_size
, 32768,
870 result
= VK_ERROR_OUT_OF_DEVICE_MEMORY
;
873 mem
->type_index
= pAllocateInfo
->memoryTypeIndex
;
875 *pMem
= radv_device_memory_to_handle(mem
);
880 vk_free2(&device
->alloc
, pAllocator
, mem
);
885 void radv_FreeMemory(
888 const VkAllocationCallbacks
* pAllocator
)
890 RADV_FROM_HANDLE(radv_device
, device
, _device
);
891 RADV_FROM_HANDLE(radv_device_memory
, mem
, _mem
);
896 device
->ws
->buffer_destroy(mem
->bo
);
899 vk_free2(&device
->alloc
, pAllocator
, mem
);
902 VkResult
radv_MapMemory(
904 VkDeviceMemory _memory
,
907 VkMemoryMapFlags flags
,
910 RADV_FROM_HANDLE(radv_device
, device
, _device
);
911 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
918 *ppData
= device
->ws
->buffer_map(mem
->bo
);
924 return VK_ERROR_MEMORY_MAP_FAILED
;
927 void radv_UnmapMemory(
929 VkDeviceMemory _memory
)
931 RADV_FROM_HANDLE(radv_device
, device
, _device
);
932 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
937 device
->ws
->buffer_unmap(mem
->bo
);
940 VkResult
radv_FlushMappedMemoryRanges(
942 uint32_t memoryRangeCount
,
943 const VkMappedMemoryRange
* pMemoryRanges
)
948 VkResult
radv_InvalidateMappedMemoryRanges(
950 uint32_t memoryRangeCount
,
951 const VkMappedMemoryRange
* pMemoryRanges
)
956 void radv_GetBufferMemoryRequirements(
959 VkMemoryRequirements
* pMemoryRequirements
)
961 RADV_FROM_HANDLE(radv_buffer
, buffer
, _buffer
);
963 /* The Vulkan spec (git aaed022) says:
965 * memoryTypeBits is a bitfield and contains one bit set for every
966 * supported memory type for the resource. The bit `1<<i` is set if and
967 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
968 * structure for the physical device is supported.
970 * We support exactly one memory type.
972 pMemoryRequirements
->memoryTypeBits
= 0x7;
974 pMemoryRequirements
->size
= buffer
->size
;
975 pMemoryRequirements
->alignment
= 16;
978 void radv_GetImageMemoryRequirements(
981 VkMemoryRequirements
* pMemoryRequirements
)
983 RADV_FROM_HANDLE(radv_image
, image
, _image
);
985 /* The Vulkan spec (git aaed022) says:
987 * memoryTypeBits is a bitfield and contains one bit set for every
988 * supported memory type for the resource. The bit `1<<i` is set if and
989 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
990 * structure for the physical device is supported.
992 * We support exactly one memory type.
994 pMemoryRequirements
->memoryTypeBits
= 0x7;
996 pMemoryRequirements
->size
= image
->size
;
997 pMemoryRequirements
->alignment
= image
->alignment
;
1000 void radv_GetImageSparseMemoryRequirements(
1003 uint32_t* pSparseMemoryRequirementCount
,
1004 VkSparseImageMemoryRequirements
* pSparseMemoryRequirements
)
1009 void radv_GetDeviceMemoryCommitment(
1011 VkDeviceMemory memory
,
1012 VkDeviceSize
* pCommittedMemoryInBytes
)
1014 *pCommittedMemoryInBytes
= 0;
1017 VkResult
radv_BindBufferMemory(
1020 VkDeviceMemory _memory
,
1021 VkDeviceSize memoryOffset
)
1023 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
1024 RADV_FROM_HANDLE(radv_buffer
, buffer
, _buffer
);
1027 buffer
->bo
= mem
->bo
;
1028 buffer
->offset
= memoryOffset
;
1037 VkResult
radv_BindImageMemory(
1040 VkDeviceMemory _memory
,
1041 VkDeviceSize memoryOffset
)
1043 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
1044 RADV_FROM_HANDLE(radv_image
, image
, _image
);
1047 image
->bo
= mem
->bo
;
1048 image
->offset
= memoryOffset
;
1057 VkResult
radv_QueueBindSparse(
1059 uint32_t bindInfoCount
,
1060 const VkBindSparseInfo
* pBindInfo
,
1063 stub_return(VK_ERROR_INCOMPATIBLE_DRIVER
);
1066 VkResult
radv_CreateFence(
1068 const VkFenceCreateInfo
* pCreateInfo
,
1069 const VkAllocationCallbacks
* pAllocator
,
1072 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1073 struct radv_fence
*fence
= vk_alloc2(&device
->alloc
, pAllocator
,
1075 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1078 return VK_ERROR_OUT_OF_HOST_MEMORY
;
1080 memset(fence
, 0, sizeof(*fence
));
1081 fence
->submitted
= false;
1082 fence
->signalled
= !!(pCreateInfo
->flags
& VK_FENCE_CREATE_SIGNALED_BIT
);
1083 fence
->fence
= device
->ws
->create_fence();
1086 *pFence
= radv_fence_to_handle(fence
);
1091 void radv_DestroyFence(
1094 const VkAllocationCallbacks
* pAllocator
)
1096 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1097 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
1101 device
->ws
->destroy_fence(fence
->fence
);
1102 vk_free2(&device
->alloc
, pAllocator
, fence
);
1105 static uint64_t radv_get_absolute_timeout(uint64_t timeout
)
1107 uint64_t current_time
;
1110 clock_gettime(CLOCK_MONOTONIC
, &tv
);
1111 current_time
= tv
.tv_nsec
+ tv
.tv_sec
*1000000000ull;
1113 timeout
= MIN2(UINT64_MAX
- current_time
, timeout
);
1115 return current_time
+ timeout
;
1118 VkResult
radv_WaitForFences(
1120 uint32_t fenceCount
,
1121 const VkFence
* pFences
,
1125 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1126 timeout
= radv_get_absolute_timeout(timeout
);
1128 if (!waitAll
&& fenceCount
> 1) {
1129 fprintf(stderr
, "radv: WaitForFences without waitAll not implemented yet\n");
1132 for (uint32_t i
= 0; i
< fenceCount
; ++i
) {
1133 RADV_FROM_HANDLE(radv_fence
, fence
, pFences
[i
]);
1134 bool expired
= false;
1136 if (fence
->signalled
)
1139 if (!fence
->submitted
)
1142 expired
= device
->ws
->fence_wait(device
->ws
, fence
->fence
, true, timeout
);
1146 fence
->signalled
= true;
1152 VkResult
radv_ResetFences(VkDevice device
,
1153 uint32_t fenceCount
,
1154 const VkFence
*pFences
)
1156 for (unsigned i
= 0; i
< fenceCount
; ++i
) {
1157 RADV_FROM_HANDLE(radv_fence
, fence
, pFences
[i
]);
1158 fence
->submitted
= fence
->signalled
= false;
1164 VkResult
radv_GetFenceStatus(VkDevice _device
, VkFence _fence
)
1166 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1167 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
1169 if (!fence
->submitted
)
1170 return VK_NOT_READY
;
1172 if (!device
->ws
->fence_wait(device
->ws
, fence
->fence
, false, 0))
1173 return VK_NOT_READY
;
1179 // Queue semaphore functions
1181 VkResult
radv_CreateSemaphore(
1183 const VkSemaphoreCreateInfo
* pCreateInfo
,
1184 const VkAllocationCallbacks
* pAllocator
,
1185 VkSemaphore
* pSemaphore
)
1187 /* The DRM execbuffer ioctl always execute in-oder, even between different
1188 * rings. As such, there's nothing to do for the user space semaphore.
1191 *pSemaphore
= (VkSemaphore
)1;
1196 void radv_DestroySemaphore(
1198 VkSemaphore semaphore
,
1199 const VkAllocationCallbacks
* pAllocator
)
1203 VkResult
radv_CreateEvent(
1205 const VkEventCreateInfo
* pCreateInfo
,
1206 const VkAllocationCallbacks
* pAllocator
,
1209 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1210 struct radv_event
*event
= vk_alloc2(&device
->alloc
, pAllocator
,
1212 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1215 return VK_ERROR_OUT_OF_HOST_MEMORY
;
1217 event
->bo
= device
->ws
->buffer_create(device
->ws
, 8, 8,
1219 RADEON_FLAG_CPU_ACCESS
);
1221 vk_free2(&device
->alloc
, pAllocator
, event
);
1222 return VK_ERROR_OUT_OF_DEVICE_MEMORY
;
1225 event
->map
= (uint64_t*)device
->ws
->buffer_map(event
->bo
);
1227 *pEvent
= radv_event_to_handle(event
);
1232 void radv_DestroyEvent(
1235 const VkAllocationCallbacks
* pAllocator
)
1237 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1238 RADV_FROM_HANDLE(radv_event
, event
, _event
);
1242 device
->ws
->buffer_destroy(event
->bo
);
1243 vk_free2(&device
->alloc
, pAllocator
, event
);
1246 VkResult
radv_GetEventStatus(
1250 RADV_FROM_HANDLE(radv_event
, event
, _event
);
1252 if (*event
->map
== 1)
1253 return VK_EVENT_SET
;
1254 return VK_EVENT_RESET
;
1257 VkResult
radv_SetEvent(
1261 RADV_FROM_HANDLE(radv_event
, event
, _event
);
1267 VkResult
radv_ResetEvent(
1271 RADV_FROM_HANDLE(radv_event
, event
, _event
);
1277 VkResult
radv_CreateBuffer(
1279 const VkBufferCreateInfo
* pCreateInfo
,
1280 const VkAllocationCallbacks
* pAllocator
,
1283 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1284 struct radv_buffer
*buffer
;
1286 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
1288 buffer
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*buffer
), 8,
1289 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1291 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1293 buffer
->size
= pCreateInfo
->size
;
1294 buffer
->usage
= pCreateInfo
->usage
;
1298 *pBuffer
= radv_buffer_to_handle(buffer
);
1303 void radv_DestroyBuffer(
1306 const VkAllocationCallbacks
* pAllocator
)
1308 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1309 RADV_FROM_HANDLE(radv_buffer
, buffer
, _buffer
);
1314 vk_free2(&device
->alloc
, pAllocator
, buffer
);
1317 static inline unsigned
1318 si_tile_mode_index(const struct radv_image
*image
, unsigned level
, bool stencil
)
1321 return image
->surface
.stencil_tiling_index
[level
];
1323 return image
->surface
.tiling_index
[level
];
1327 radv_initialise_color_surface(struct radv_device
*device
,
1328 struct radv_color_buffer_info
*cb
,
1329 struct radv_image_view
*iview
)
1331 const struct vk_format_description
*desc
;
1332 unsigned ntype
, format
, swap
, endian
;
1333 unsigned blend_clamp
= 0, blend_bypass
= 0;
1334 unsigned pitch_tile_max
, slice_tile_max
, tile_mode_index
;
1336 const struct radeon_surf
*surf
= &iview
->image
->surface
;
1337 const struct radeon_surf_level
*level_info
= &surf
->level
[iview
->base_mip
];
1339 desc
= vk_format_description(iview
->vk_format
);
1341 memset(cb
, 0, sizeof(*cb
));
1343 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
;
1344 va
+= level_info
->offset
;
1345 cb
->cb_color_base
= va
>> 8;
1347 /* CMASK variables */
1348 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
;
1349 va
+= iview
->image
->cmask
.offset
;
1350 cb
->cb_color_cmask
= va
>> 8;
1351 cb
->cb_color_cmask_slice
= iview
->image
->cmask
.slice_tile_max
;
1353 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
;
1354 va
+= iview
->image
->dcc_offset
;
1355 cb
->cb_dcc_base
= va
>> 8;
1357 cb
->cb_color_view
= S_028C6C_SLICE_START(iview
->base_layer
) |
1358 S_028C6C_SLICE_MAX(iview
->base_layer
+ iview
->extent
.depth
- 1);
1360 cb
->micro_tile_mode
= iview
->image
->surface
.micro_tile_mode
;
1361 pitch_tile_max
= level_info
->nblk_x
/ 8 - 1;
1362 slice_tile_max
= (level_info
->nblk_x
* level_info
->nblk_y
) / 64 - 1;
1363 tile_mode_index
= si_tile_mode_index(iview
->image
, iview
->base_mip
, false);
1365 cb
->cb_color_pitch
= S_028C64_TILE_MAX(pitch_tile_max
);
1366 cb
->cb_color_slice
= S_028C68_TILE_MAX(slice_tile_max
);
1368 /* Intensity is implemented as Red, so treat it that way. */
1369 cb
->cb_color_attrib
= S_028C74_FORCE_DST_ALPHA_1(desc
->swizzle
[3] == VK_SWIZZLE_1
) |
1370 S_028C74_TILE_MODE_INDEX(tile_mode_index
);
1372 if (iview
->image
->samples
> 1) {
1373 unsigned log_samples
= util_logbase2(iview
->image
->samples
);
1375 cb
->cb_color_attrib
|= S_028C74_NUM_SAMPLES(log_samples
) |
1376 S_028C74_NUM_FRAGMENTS(log_samples
);
1379 if (iview
->image
->fmask
.size
) {
1380 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
+ iview
->image
->fmask
.offset
;
1381 if (device
->instance
->physicalDevice
.rad_info
.chip_class
>= CIK
)
1382 cb
->cb_color_pitch
|= S_028C64_FMASK_TILE_MAX(iview
->image
->fmask
.pitch_in_pixels
/ 8 - 1);
1383 cb
->cb_color_attrib
|= S_028C74_FMASK_TILE_MODE_INDEX(iview
->image
->fmask
.tile_mode_index
);
1384 cb
->cb_color_fmask
= va
>> 8;
1385 cb
->cb_color_fmask_slice
= S_028C88_TILE_MAX(iview
->image
->fmask
.slice_tile_max
);
1387 /* This must be set for fast clear to work without FMASK. */
1388 if (device
->instance
->physicalDevice
.rad_info
.chip_class
>= CIK
)
1389 cb
->cb_color_pitch
|= S_028C64_FMASK_TILE_MAX(pitch_tile_max
);
1390 cb
->cb_color_attrib
|= S_028C74_FMASK_TILE_MODE_INDEX(tile_mode_index
);
1391 cb
->cb_color_fmask
= cb
->cb_color_base
;
1392 cb
->cb_color_fmask_slice
= S_028C88_TILE_MAX(slice_tile_max
);
1395 ntype
= radv_translate_color_numformat(iview
->vk_format
,
1397 vk_format_get_first_non_void_channel(iview
->vk_format
));
1398 format
= radv_translate_colorformat(iview
->vk_format
);
1399 if (format
== V_028C70_COLOR_INVALID
|| ntype
== ~0u)
1400 radv_finishme("Illegal color\n");
1401 swap
= radv_translate_colorswap(iview
->vk_format
, FALSE
);
1402 endian
= radv_colorformat_endian_swap(format
);
1404 /* blend clamp should be set for all NORM/SRGB types */
1405 if (ntype
== V_028C70_NUMBER_UNORM
||
1406 ntype
== V_028C70_NUMBER_SNORM
||
1407 ntype
== V_028C70_NUMBER_SRGB
)
1410 /* set blend bypass according to docs if SINT/UINT or
1411 8/24 COLOR variants */
1412 if (ntype
== V_028C70_NUMBER_UINT
|| ntype
== V_028C70_NUMBER_SINT
||
1413 format
== V_028C70_COLOR_8_24
|| format
== V_028C70_COLOR_24_8
||
1414 format
== V_028C70_COLOR_X24_8_32_FLOAT
) {
1419 if ((ntype
== V_028C70_NUMBER_UINT
|| ntype
== V_028C70_NUMBER_SINT
) &&
1420 (format
== V_028C70_COLOR_8
||
1421 format
== V_028C70_COLOR_8_8
||
1422 format
== V_028C70_COLOR_8_8_8_8
))
1423 ->color_is_int8
= true;
1425 cb
->cb_color_info
= S_028C70_FORMAT(format
) |
1426 S_028C70_COMP_SWAP(swap
) |
1427 S_028C70_BLEND_CLAMP(blend_clamp
) |
1428 S_028C70_BLEND_BYPASS(blend_bypass
) |
1429 S_028C70_SIMPLE_FLOAT(1) |
1430 S_028C70_ROUND_MODE(ntype
!= V_028C70_NUMBER_UNORM
&&
1431 ntype
!= V_028C70_NUMBER_SNORM
&&
1432 ntype
!= V_028C70_NUMBER_SRGB
&&
1433 format
!= V_028C70_COLOR_8_24
&&
1434 format
!= V_028C70_COLOR_24_8
) |
1435 S_028C70_NUMBER_TYPE(ntype
) |
1436 S_028C70_ENDIAN(endian
);
1437 if (iview
->image
->samples
> 1)
1438 if (iview
->image
->fmask
.size
)
1439 cb
->cb_color_info
|= S_028C70_COMPRESSION(1);
1441 if (iview
->image
->cmask
.size
&& device
->allow_fast_clears
)
1442 cb
->cb_color_info
|= S_028C70_FAST_CLEAR(1);
1444 if (iview
->image
->surface
.dcc_size
&& level_info
->dcc_enabled
)
1445 cb
->cb_color_info
|= S_028C70_DCC_ENABLE(1);
1447 if (device
->instance
->physicalDevice
.rad_info
.chip_class
>= VI
) {
1448 unsigned max_uncompressed_block_size
= 2;
1449 if (iview
->image
->samples
> 1) {
1450 if (iview
->image
->surface
.bpe
== 1)
1451 max_uncompressed_block_size
= 0;
1452 else if (iview
->image
->surface
.bpe
== 2)
1453 max_uncompressed_block_size
= 1;
1456 cb
->cb_dcc_control
= S_028C78_MAX_UNCOMPRESSED_BLOCK_SIZE(max_uncompressed_block_size
) |
1457 S_028C78_INDEPENDENT_64B_BLOCKS(1);
1460 /* This must be set for fast clear to work without FMASK. */
1461 if (!iview
->image
->fmask
.size
&&
1462 device
->instance
->physicalDevice
.rad_info
.chip_class
== SI
) {
1463 unsigned bankh
= util_logbase2(iview
->image
->surface
.bankh
);
1464 cb
->cb_color_attrib
|= S_028C74_FMASK_BANK_HEIGHT(bankh
);
1469 radv_initialise_ds_surface(struct radv_device
*device
,
1470 struct radv_ds_buffer_info
*ds
,
1471 struct radv_image_view
*iview
)
1473 unsigned level
= iview
->base_mip
;
1475 uint64_t va
, s_offs
, z_offs
;
1476 const struct radeon_surf_level
*level_info
= &iview
->image
->surface
.level
[level
];
1477 memset(ds
, 0, sizeof(*ds
));
1478 switch (iview
->vk_format
) {
1479 case VK_FORMAT_D24_UNORM_S8_UINT
:
1480 case VK_FORMAT_X8_D24_UNORM_PACK32
:
1481 ds
->pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-24);
1482 ds
->offset_scale
= 2.0f
;
1484 case VK_FORMAT_D16_UNORM
:
1485 case VK_FORMAT_D16_UNORM_S8_UINT
:
1486 ds
->pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-16);
1487 ds
->offset_scale
= 4.0f
;
1489 case VK_FORMAT_D32_SFLOAT
:
1490 case VK_FORMAT_D32_SFLOAT_S8_UINT
:
1491 ds
->pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-23) |
1492 S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
1493 ds
->offset_scale
= 1.0f
;
1499 format
= radv_translate_dbformat(iview
->vk_format
);
1500 if (format
== V_028040_Z_INVALID
) {
1501 fprintf(stderr
, "Invalid DB format: %d, disabling DB.\n", iview
->vk_format
);
1504 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
;
1505 s_offs
= z_offs
= va
;
1506 z_offs
+= iview
->image
->surface
.level
[level
].offset
;
1507 s_offs
+= iview
->image
->surface
.stencil_level
[level
].offset
;
1509 ds
->db_depth_view
= S_028008_SLICE_START(iview
->base_layer
) |
1510 S_028008_SLICE_MAX(iview
->base_layer
+ iview
->extent
.depth
- 1);
1511 ds
->db_depth_info
= S_02803C_ADDR5_SWIZZLE_MASK(1);
1512 ds
->db_z_info
= S_028040_FORMAT(format
) | S_028040_ZRANGE_PRECISION(1);
1514 if (iview
->image
->samples
> 1)
1515 ds
->db_z_info
|= S_028040_NUM_SAMPLES(util_logbase2(iview
->image
->samples
));
1517 if (iview
->image
->surface
.flags
& RADEON_SURF_SBUFFER
)
1518 ds
->db_stencil_info
= S_028044_FORMAT(V_028044_STENCIL_8
);
1520 ds
->db_stencil_info
= S_028044_FORMAT(V_028044_STENCIL_INVALID
);
1522 if (device
->instance
->physicalDevice
.rad_info
.chip_class
>= CIK
) {
1523 struct radeon_info
*info
= &device
->instance
->physicalDevice
.rad_info
;
1524 unsigned tiling_index
= iview
->image
->surface
.tiling_index
[level
];
1525 unsigned stencil_index
= iview
->image
->surface
.stencil_tiling_index
[level
];
1526 unsigned macro_index
= iview
->image
->surface
.macro_tile_index
;
1527 unsigned tile_mode
= info
->si_tile_mode_array
[tiling_index
];
1528 unsigned stencil_tile_mode
= info
->si_tile_mode_array
[stencil_index
];
1529 unsigned macro_mode
= info
->cik_macrotile_mode_array
[macro_index
];
1531 ds
->db_depth_info
|=
1532 S_02803C_ARRAY_MODE(G_009910_ARRAY_MODE(tile_mode
)) |
1533 S_02803C_PIPE_CONFIG(G_009910_PIPE_CONFIG(tile_mode
)) |
1534 S_02803C_BANK_WIDTH(G_009990_BANK_WIDTH(macro_mode
)) |
1535 S_02803C_BANK_HEIGHT(G_009990_BANK_HEIGHT(macro_mode
)) |
1536 S_02803C_MACRO_TILE_ASPECT(G_009990_MACRO_TILE_ASPECT(macro_mode
)) |
1537 S_02803C_NUM_BANKS(G_009990_NUM_BANKS(macro_mode
));
1538 ds
->db_z_info
|= S_028040_TILE_SPLIT(G_009910_TILE_SPLIT(tile_mode
));
1539 ds
->db_stencil_info
|= S_028044_TILE_SPLIT(G_009910_TILE_SPLIT(stencil_tile_mode
));
1541 unsigned tile_mode_index
= si_tile_mode_index(iview
->image
, level
, false);
1542 ds
->db_z_info
|= S_028040_TILE_MODE_INDEX(tile_mode_index
);
1543 tile_mode_index
= si_tile_mode_index(iview
->image
, level
, true);
1544 ds
->db_stencil_info
|= S_028044_TILE_MODE_INDEX(tile_mode_index
);
1547 if (iview
->image
->htile
.size
&& !level
) {
1548 ds
->db_z_info
|= S_028040_TILE_SURFACE_ENABLE(1) |
1549 S_028040_ALLOW_EXPCLEAR(1);
1551 if (iview
->image
->surface
.flags
& RADEON_SURF_SBUFFER
) {
1552 /* Workaround: For a not yet understood reason, the
1553 * combination of MSAA, fast stencil clear and stencil
1554 * decompress messes with subsequent stencil buffer
1555 * uses. Problem was reproduced on Verde, Bonaire,
1556 * Tonga, and Carrizo.
1558 * Disabling EXPCLEAR works around the problem.
1560 * Check piglit's arb_texture_multisample-stencil-clear
1561 * test if you want to try changing this.
1563 if (iview
->image
->samples
<= 1)
1564 ds
->db_stencil_info
|= S_028044_ALLOW_EXPCLEAR(1);
1566 /* Use all of the htile_buffer for depth if there's no stencil. */
1567 ds
->db_stencil_info
|= S_028044_TILE_STENCIL_DISABLE(1);
1569 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
+
1570 iview
->image
->htile
.offset
;
1571 ds
->db_htile_data_base
= va
>> 8;
1572 ds
->db_htile_surface
= S_028ABC_FULL_CACHE(1);
1574 ds
->db_htile_data_base
= 0;
1575 ds
->db_htile_surface
= 0;
1578 ds
->db_z_read_base
= ds
->db_z_write_base
= z_offs
>> 8;
1579 ds
->db_stencil_read_base
= ds
->db_stencil_write_base
= s_offs
>> 8;
1581 ds
->db_depth_size
= S_028058_PITCH_TILE_MAX((level_info
->nblk_x
/ 8) - 1) |
1582 S_028058_HEIGHT_TILE_MAX((level_info
->nblk_y
/ 8) - 1);
1583 ds
->db_depth_slice
= S_02805C_SLICE_TILE_MAX((level_info
->nblk_x
* level_info
->nblk_y
) / 64 - 1);
1586 VkResult
radv_CreateFramebuffer(
1588 const VkFramebufferCreateInfo
* pCreateInfo
,
1589 const VkAllocationCallbacks
* pAllocator
,
1590 VkFramebuffer
* pFramebuffer
)
1592 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1593 struct radv_framebuffer
*framebuffer
;
1595 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
1597 size_t size
= sizeof(*framebuffer
) +
1598 sizeof(struct radv_attachment_info
) * pCreateInfo
->attachmentCount
;
1599 framebuffer
= vk_alloc2(&device
->alloc
, pAllocator
, size
, 8,
1600 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1601 if (framebuffer
== NULL
)
1602 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1604 framebuffer
->attachment_count
= pCreateInfo
->attachmentCount
;
1605 for (uint32_t i
= 0; i
< pCreateInfo
->attachmentCount
; i
++) {
1606 VkImageView _iview
= pCreateInfo
->pAttachments
[i
];
1607 struct radv_image_view
*iview
= radv_image_view_from_handle(_iview
);
1608 framebuffer
->attachments
[i
].attachment
= iview
;
1609 if (iview
->aspect_mask
& VK_IMAGE_ASPECT_COLOR_BIT
) {
1610 radv_initialise_color_surface(device
, &framebuffer
->attachments
[i
].cb
, iview
);
1611 } else if (iview
->aspect_mask
& (VK_IMAGE_ASPECT_DEPTH_BIT
| VK_IMAGE_ASPECT_STENCIL_BIT
)) {
1612 radv_initialise_ds_surface(device
, &framebuffer
->attachments
[i
].ds
, iview
);
1616 framebuffer
->width
= pCreateInfo
->width
;
1617 framebuffer
->height
= pCreateInfo
->height
;
1618 framebuffer
->layers
= pCreateInfo
->layers
;
1620 *pFramebuffer
= radv_framebuffer_to_handle(framebuffer
);
1624 void radv_DestroyFramebuffer(
1627 const VkAllocationCallbacks
* pAllocator
)
1629 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1630 RADV_FROM_HANDLE(radv_framebuffer
, fb
, _fb
);
1634 vk_free2(&device
->alloc
, pAllocator
, fb
);
1637 static unsigned radv_tex_wrap(VkSamplerAddressMode address_mode
)
1639 switch (address_mode
) {
1640 case VK_SAMPLER_ADDRESS_MODE_REPEAT
:
1641 return V_008F30_SQ_TEX_WRAP
;
1642 case VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT
:
1643 return V_008F30_SQ_TEX_MIRROR
;
1644 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE
:
1645 return V_008F30_SQ_TEX_CLAMP_LAST_TEXEL
;
1646 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER
:
1647 return V_008F30_SQ_TEX_CLAMP_BORDER
;
1648 case VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE
:
1649 return V_008F30_SQ_TEX_MIRROR_ONCE_LAST_TEXEL
;
1651 unreachable("illegal tex wrap mode");
1657 radv_tex_compare(VkCompareOp op
)
1660 case VK_COMPARE_OP_NEVER
:
1661 return V_008F30_SQ_TEX_DEPTH_COMPARE_NEVER
;
1662 case VK_COMPARE_OP_LESS
:
1663 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESS
;
1664 case VK_COMPARE_OP_EQUAL
:
1665 return V_008F30_SQ_TEX_DEPTH_COMPARE_EQUAL
;
1666 case VK_COMPARE_OP_LESS_OR_EQUAL
:
1667 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESSEQUAL
;
1668 case VK_COMPARE_OP_GREATER
:
1669 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATER
;
1670 case VK_COMPARE_OP_NOT_EQUAL
:
1671 return V_008F30_SQ_TEX_DEPTH_COMPARE_NOTEQUAL
;
1672 case VK_COMPARE_OP_GREATER_OR_EQUAL
:
1673 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATEREQUAL
;
1674 case VK_COMPARE_OP_ALWAYS
:
1675 return V_008F30_SQ_TEX_DEPTH_COMPARE_ALWAYS
;
1677 unreachable("illegal compare mode");
1683 radv_tex_filter(VkFilter filter
, unsigned max_ansio
)
1686 case VK_FILTER_NEAREST
:
1687 return (max_ansio
> 1 ? V_008F38_SQ_TEX_XY_FILTER_ANISO_POINT
:
1688 V_008F38_SQ_TEX_XY_FILTER_POINT
);
1689 case VK_FILTER_LINEAR
:
1690 return (max_ansio
> 1 ? V_008F38_SQ_TEX_XY_FILTER_ANISO_BILINEAR
:
1691 V_008F38_SQ_TEX_XY_FILTER_BILINEAR
);
1692 case VK_FILTER_CUBIC_IMG
:
1694 fprintf(stderr
, "illegal texture filter");
1700 radv_tex_mipfilter(VkSamplerMipmapMode mode
)
1703 case VK_SAMPLER_MIPMAP_MODE_NEAREST
:
1704 return V_008F38_SQ_TEX_Z_FILTER_POINT
;
1705 case VK_SAMPLER_MIPMAP_MODE_LINEAR
:
1706 return V_008F38_SQ_TEX_Z_FILTER_LINEAR
;
1708 return V_008F38_SQ_TEX_Z_FILTER_NONE
;
1713 radv_tex_bordercolor(VkBorderColor bcolor
)
1716 case VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
:
1717 case VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
:
1718 return V_008F3C_SQ_TEX_BORDER_COLOR_TRANS_BLACK
;
1719 case VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
:
1720 case VK_BORDER_COLOR_INT_OPAQUE_BLACK
:
1721 return V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_BLACK
;
1722 case VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
:
1723 case VK_BORDER_COLOR_INT_OPAQUE_WHITE
:
1724 return V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_WHITE
;
1732 radv_init_sampler(struct radv_device
*device
,
1733 struct radv_sampler
*sampler
,
1734 const VkSamplerCreateInfo
*pCreateInfo
)
1736 uint32_t max_aniso
= 0;
1737 uint32_t max_aniso_ratio
= 0;//TODO
1739 is_vi
= (device
->instance
->physicalDevice
.rad_info
.chip_class
>= VI
);
1741 sampler
->state
[0] = (S_008F30_CLAMP_X(radv_tex_wrap(pCreateInfo
->addressModeU
)) |
1742 S_008F30_CLAMP_Y(radv_tex_wrap(pCreateInfo
->addressModeV
)) |
1743 S_008F30_CLAMP_Z(radv_tex_wrap(pCreateInfo
->addressModeW
)) |
1744 S_008F30_MAX_ANISO_RATIO(max_aniso_ratio
) |
1745 S_008F30_DEPTH_COMPARE_FUNC(radv_tex_compare(pCreateInfo
->compareOp
)) |
1746 S_008F30_FORCE_UNNORMALIZED(pCreateInfo
->unnormalizedCoordinates
? 1 : 0) |
1747 S_008F30_DISABLE_CUBE_WRAP(0) |
1748 S_008F30_COMPAT_MODE(is_vi
));
1749 sampler
->state
[1] = (S_008F34_MIN_LOD(S_FIXED(CLAMP(pCreateInfo
->minLod
, 0, 15), 8)) |
1750 S_008F34_MAX_LOD(S_FIXED(CLAMP(pCreateInfo
->maxLod
, 0, 15), 8)));
1751 sampler
->state
[2] = (S_008F38_LOD_BIAS(S_FIXED(CLAMP(pCreateInfo
->mipLodBias
, -16, 16), 8)) |
1752 S_008F38_XY_MAG_FILTER(radv_tex_filter(pCreateInfo
->magFilter
, max_aniso
)) |
1753 S_008F38_XY_MIN_FILTER(radv_tex_filter(pCreateInfo
->minFilter
, max_aniso
)) |
1754 S_008F38_MIP_FILTER(radv_tex_mipfilter(pCreateInfo
->mipmapMode
)) |
1755 S_008F38_MIP_POINT_PRECLAMP(1) |
1756 S_008F38_DISABLE_LSB_CEIL(1) |
1757 S_008F38_FILTER_PREC_FIX(1) |
1758 S_008F38_ANISO_OVERRIDE(is_vi
));
1759 sampler
->state
[3] = (S_008F3C_BORDER_COLOR_PTR(0) |
1760 S_008F3C_BORDER_COLOR_TYPE(radv_tex_bordercolor(pCreateInfo
->borderColor
)));
1763 VkResult
radv_CreateSampler(
1765 const VkSamplerCreateInfo
* pCreateInfo
,
1766 const VkAllocationCallbacks
* pAllocator
,
1767 VkSampler
* pSampler
)
1769 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1770 struct radv_sampler
*sampler
;
1772 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO
);
1774 sampler
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*sampler
), 8,
1775 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1777 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1779 radv_init_sampler(device
, sampler
, pCreateInfo
);
1780 *pSampler
= radv_sampler_to_handle(sampler
);
1785 void radv_DestroySampler(
1788 const VkAllocationCallbacks
* pAllocator
)
1790 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1791 RADV_FROM_HANDLE(radv_sampler
, sampler
, _sampler
);
1795 vk_free2(&device
->alloc
, pAllocator
, sampler
);