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"
36 #include <amdgpu_drm.h>
37 #include "amdgpu_id.h"
38 #include "winsys/amdgpu/radv_amdgpu_winsys_public.h"
39 #include "ac_llvm_util.h"
40 #include "vk_format.h"
42 #include "radv_timestamp.h"
43 #include "util/debug.h"
44 struct radv_dispatch_table dtable
;
47 struct radeon_winsys_fence
*fence
;
53 radv_physical_device_init(struct radv_physical_device
*device
,
54 struct radv_instance
*instance
,
60 fd
= open(path
, O_RDWR
| O_CLOEXEC
);
62 return vk_errorf(VK_ERROR_INCOMPATIBLE_DRIVER
,
63 "failed to open %s: %m", path
);
65 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
66 device
->instance
= instance
;
67 assert(strlen(path
) < ARRAY_SIZE(device
->path
));
68 strncpy(device
->path
, path
, ARRAY_SIZE(device
->path
));
70 device
->ws
= radv_amdgpu_winsys_create(fd
);
72 result
= VK_ERROR_INCOMPATIBLE_DRIVER
;
75 device
->ws
->query_info(device
->ws
, &device
->rad_info
);
76 result
= radv_init_wsi(device
);
77 if (result
!= VK_SUCCESS
)
80 fprintf(stderr
, "WARNING: radv is not a conformant vulkan implementation, testing use only.\n");
81 device
->name
= device
->rad_info
.name
;
90 radv_physical_device_finish(struct radv_physical_device
*device
)
92 radv_finish_wsi(device
);
93 device
->ws
->destroy(device
->ws
);
96 static const VkExtensionProperties global_extensions
[] = {
98 .extensionName
= VK_KHR_SURFACE_EXTENSION_NAME
,
101 #ifdef VK_USE_PLATFORM_XCB_KHR
103 .extensionName
= VK_KHR_XCB_SURFACE_EXTENSION_NAME
,
107 #ifdef VK_USE_PLATFORM_WAYLAND_KHR
109 .extensionName
= VK_KHR_WAYLAND_SURFACE_EXTENSION_NAME
,
115 static const VkExtensionProperties device_extensions
[] = {
117 .extensionName
= VK_KHR_SWAPCHAIN_EXTENSION_NAME
,
123 default_alloc_func(void *pUserData
, size_t size
, size_t align
,
124 VkSystemAllocationScope allocationScope
)
130 default_realloc_func(void *pUserData
, void *pOriginal
, size_t size
,
131 size_t align
, VkSystemAllocationScope allocationScope
)
133 return realloc(pOriginal
, size
);
137 default_free_func(void *pUserData
, void *pMemory
)
142 static const VkAllocationCallbacks default_alloc
= {
144 .pfnAllocation
= default_alloc_func
,
145 .pfnReallocation
= default_realloc_func
,
146 .pfnFree
= default_free_func
,
149 VkResult
radv_CreateInstance(
150 const VkInstanceCreateInfo
* pCreateInfo
,
151 const VkAllocationCallbacks
* pAllocator
,
152 VkInstance
* pInstance
)
154 struct radv_instance
*instance
;
156 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO
);
158 uint32_t client_version
;
159 if (pCreateInfo
->pApplicationInfo
&&
160 pCreateInfo
->pApplicationInfo
->apiVersion
!= 0) {
161 client_version
= pCreateInfo
->pApplicationInfo
->apiVersion
;
163 client_version
= VK_MAKE_VERSION(1, 0, 0);
166 if (VK_MAKE_VERSION(1, 0, 0) > client_version
||
167 client_version
> VK_MAKE_VERSION(1, 0, 0xfff)) {
168 return vk_errorf(VK_ERROR_INCOMPATIBLE_DRIVER
,
169 "Client requested version %d.%d.%d",
170 VK_VERSION_MAJOR(client_version
),
171 VK_VERSION_MINOR(client_version
),
172 VK_VERSION_PATCH(client_version
));
175 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
177 for (uint32_t j
= 0; j
< ARRAY_SIZE(global_extensions
); j
++) {
178 if (strcmp(pCreateInfo
->ppEnabledExtensionNames
[i
],
179 global_extensions
[j
].extensionName
) == 0) {
185 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
188 instance
= radv_alloc2(&default_alloc
, pAllocator
, sizeof(*instance
), 8,
189 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE
);
191 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
193 instance
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
196 instance
->alloc
= *pAllocator
;
198 instance
->alloc
= default_alloc
;
200 instance
->apiVersion
= client_version
;
201 instance
->physicalDeviceCount
= -1;
205 VG(VALGRIND_CREATE_MEMPOOL(instance
, 0, false));
207 *pInstance
= radv_instance_to_handle(instance
);
212 void radv_DestroyInstance(
213 VkInstance _instance
,
214 const VkAllocationCallbacks
* pAllocator
)
216 RADV_FROM_HANDLE(radv_instance
, instance
, _instance
);
218 if (instance
->physicalDeviceCount
> 0) {
219 /* We support at most one physical device. */
220 assert(instance
->physicalDeviceCount
== 1);
221 radv_physical_device_finish(&instance
->physicalDevice
);
224 VG(VALGRIND_DESTROY_MEMPOOL(instance
));
228 radv_free(&instance
->alloc
, instance
);
231 VkResult
radv_EnumeratePhysicalDevices(
232 VkInstance _instance
,
233 uint32_t* pPhysicalDeviceCount
,
234 VkPhysicalDevice
* pPhysicalDevices
)
236 RADV_FROM_HANDLE(radv_instance
, instance
, _instance
);
239 if (instance
->physicalDeviceCount
< 0) {
241 for (unsigned i
= 0; i
< 8; i
++) {
242 snprintf(path
, sizeof(path
), "/dev/dri/renderD%d", 128 + i
);
243 result
= radv_physical_device_init(&instance
->physicalDevice
,
245 if (result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
249 if (result
== VK_ERROR_INCOMPATIBLE_DRIVER
) {
250 instance
->physicalDeviceCount
= 0;
251 } else if (result
== VK_SUCCESS
) {
252 instance
->physicalDeviceCount
= 1;
258 /* pPhysicalDeviceCount is an out parameter if pPhysicalDevices is NULL;
259 * otherwise it's an inout parameter.
261 * The Vulkan spec (git aaed022) says:
263 * pPhysicalDeviceCount is a pointer to an unsigned integer variable
264 * that is initialized with the number of devices the application is
265 * prepared to receive handles to. pname:pPhysicalDevices is pointer to
266 * an array of at least this many VkPhysicalDevice handles [...].
268 * Upon success, if pPhysicalDevices is NULL, vkEnumeratePhysicalDevices
269 * overwrites the contents of the variable pointed to by
270 * pPhysicalDeviceCount with the number of physical devices in in the
271 * instance; otherwise, vkEnumeratePhysicalDevices overwrites
272 * pPhysicalDeviceCount with the number of physical handles written to
275 if (!pPhysicalDevices
) {
276 *pPhysicalDeviceCount
= instance
->physicalDeviceCount
;
277 } else if (*pPhysicalDeviceCount
>= 1) {
278 pPhysicalDevices
[0] = radv_physical_device_to_handle(&instance
->physicalDevice
);
279 *pPhysicalDeviceCount
= 1;
281 *pPhysicalDeviceCount
= 0;
287 void radv_GetPhysicalDeviceFeatures(
288 VkPhysicalDevice physicalDevice
,
289 VkPhysicalDeviceFeatures
* pFeatures
)
291 // RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
293 memset(pFeatures
, 0, sizeof(*pFeatures
));
295 *pFeatures
= (VkPhysicalDeviceFeatures
) {
296 .robustBufferAccess
= true,
297 .fullDrawIndexUint32
= true,
298 .imageCubeArray
= true,
299 .independentBlend
= true,
300 .geometryShader
= false,
301 .tessellationShader
= false,
302 .sampleRateShading
= false,
303 .dualSrcBlend
= true,
305 .multiDrawIndirect
= true,
306 .drawIndirectFirstInstance
= true,
308 .depthBiasClamp
= true,
309 .fillModeNonSolid
= true,
314 .multiViewport
= false,
315 .samplerAnisotropy
= false, /* FINISHME */
316 .textureCompressionETC2
= false,
317 .textureCompressionASTC_LDR
= false,
318 .textureCompressionBC
= true,
319 .occlusionQueryPrecise
= true,
320 .pipelineStatisticsQuery
= false,
321 .vertexPipelineStoresAndAtomics
= true,
322 .fragmentStoresAndAtomics
= true,
323 .shaderTessellationAndGeometryPointSize
= true,
324 .shaderImageGatherExtended
= false,
325 .shaderStorageImageExtendedFormats
= false,
326 .shaderStorageImageMultisample
= false,
327 .shaderUniformBufferArrayDynamicIndexing
= true,
328 .shaderSampledImageArrayDynamicIndexing
= true,
329 .shaderStorageBufferArrayDynamicIndexing
= true,
330 .shaderStorageImageArrayDynamicIndexing
= true,
331 .shaderStorageImageReadWithoutFormat
= false,
332 .shaderStorageImageWriteWithoutFormat
= true,
333 .shaderClipDistance
= true,
334 .shaderCullDistance
= true,
335 .shaderFloat64
= false,
336 .shaderInt64
= false,
337 .shaderInt16
= false,
339 .variableMultisampleRate
= false,
340 .inheritedQueries
= false,
345 radv_device_get_cache_uuid(void *uuid
)
347 memset(uuid
, 0, VK_UUID_SIZE
);
348 snprintf(uuid
, VK_UUID_SIZE
, "radv-%s", RADV_TIMESTAMP
);
351 void radv_GetPhysicalDeviceProperties(
352 VkPhysicalDevice physicalDevice
,
353 VkPhysicalDeviceProperties
* pProperties
)
355 RADV_FROM_HANDLE(radv_physical_device
, pdevice
, physicalDevice
);
356 VkSampleCountFlags sample_counts
= 0xf;
357 VkPhysicalDeviceLimits limits
= {
358 .maxImageDimension1D
= (1 << 14),
359 .maxImageDimension2D
= (1 << 14),
360 .maxImageDimension3D
= (1 << 11),
361 .maxImageDimensionCube
= (1 << 14),
362 .maxImageArrayLayers
= (1 << 11),
363 .maxTexelBufferElements
= 128 * 1024 * 1024,
364 .maxUniformBufferRange
= UINT32_MAX
,
365 .maxStorageBufferRange
= UINT32_MAX
,
366 .maxPushConstantsSize
= MAX_PUSH_CONSTANTS_SIZE
,
367 .maxMemoryAllocationCount
= UINT32_MAX
,
368 .maxSamplerAllocationCount
= 64 * 1024,
369 .bufferImageGranularity
= 64, /* A cache line */
370 .sparseAddressSpaceSize
= 0,
371 .maxBoundDescriptorSets
= MAX_SETS
,
372 .maxPerStageDescriptorSamplers
= 64,
373 .maxPerStageDescriptorUniformBuffers
= 64,
374 .maxPerStageDescriptorStorageBuffers
= 64,
375 .maxPerStageDescriptorSampledImages
= 64,
376 .maxPerStageDescriptorStorageImages
= 64,
377 .maxPerStageDescriptorInputAttachments
= 64,
378 .maxPerStageResources
= 128,
379 .maxDescriptorSetSamplers
= 256,
380 .maxDescriptorSetUniformBuffers
= 256,
381 .maxDescriptorSetUniformBuffersDynamic
= 256,
382 .maxDescriptorSetStorageBuffers
= 256,
383 .maxDescriptorSetStorageBuffersDynamic
= 256,
384 .maxDescriptorSetSampledImages
= 256,
385 .maxDescriptorSetStorageImages
= 256,
386 .maxDescriptorSetInputAttachments
= 256,
387 .maxVertexInputAttributes
= 32,
388 .maxVertexInputBindings
= 32,
389 .maxVertexInputAttributeOffset
= 2047,
390 .maxVertexInputBindingStride
= 2048,
391 .maxVertexOutputComponents
= 128,
392 .maxTessellationGenerationLevel
= 0,
393 .maxTessellationPatchSize
= 0,
394 .maxTessellationControlPerVertexInputComponents
= 0,
395 .maxTessellationControlPerVertexOutputComponents
= 0,
396 .maxTessellationControlPerPatchOutputComponents
= 0,
397 .maxTessellationControlTotalOutputComponents
= 0,
398 .maxTessellationEvaluationInputComponents
= 0,
399 .maxTessellationEvaluationOutputComponents
= 0,
400 .maxGeometryShaderInvocations
= 32,
401 .maxGeometryInputComponents
= 64,
402 .maxGeometryOutputComponents
= 128,
403 .maxGeometryOutputVertices
= 256,
404 .maxGeometryTotalOutputComponents
= 1024,
405 .maxFragmentInputComponents
= 128,
406 .maxFragmentOutputAttachments
= 8,
407 .maxFragmentDualSrcAttachments
= 2,
408 .maxFragmentCombinedOutputResources
= 8,
409 .maxComputeSharedMemorySize
= 32768,
410 .maxComputeWorkGroupCount
= { 65535, 65535, 65535 },
411 .maxComputeWorkGroupInvocations
= 16 * 1024,
412 .maxComputeWorkGroupSize
= {
413 16 * 1024/*devinfo->max_cs_threads*/,
417 .subPixelPrecisionBits
= 4 /* FIXME */,
418 .subTexelPrecisionBits
= 4 /* FIXME */,
419 .mipmapPrecisionBits
= 4 /* FIXME */,
420 .maxDrawIndexedIndexValue
= UINT32_MAX
,
421 .maxDrawIndirectCount
= UINT32_MAX
,
422 .maxSamplerLodBias
= 16,
423 .maxSamplerAnisotropy
= 16,
424 .maxViewports
= MAX_VIEWPORTS
,
425 .maxViewportDimensions
= { (1 << 14), (1 << 14) },
426 .viewportBoundsRange
= { INT16_MIN
, INT16_MAX
},
427 .viewportSubPixelBits
= 13, /* We take a float? */
428 .minMemoryMapAlignment
= 4096, /* A page */
429 .minTexelBufferOffsetAlignment
= 1,
430 .minUniformBufferOffsetAlignment
= 4,
431 .minStorageBufferOffsetAlignment
= 4,
432 .minTexelOffset
= -8,
434 .minTexelGatherOffset
= -8,
435 .maxTexelGatherOffset
= 7,
436 .minInterpolationOffset
= 0, /* FIXME */
437 .maxInterpolationOffset
= 0, /* FIXME */
438 .subPixelInterpolationOffsetBits
= 0, /* FIXME */
439 .maxFramebufferWidth
= (1 << 14),
440 .maxFramebufferHeight
= (1 << 14),
441 .maxFramebufferLayers
= (1 << 10),
442 .framebufferColorSampleCounts
= sample_counts
,
443 .framebufferDepthSampleCounts
= sample_counts
,
444 .framebufferStencilSampleCounts
= sample_counts
,
445 .framebufferNoAttachmentsSampleCounts
= sample_counts
,
446 .maxColorAttachments
= MAX_RTS
,
447 .sampledImageColorSampleCounts
= sample_counts
,
448 .sampledImageIntegerSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
449 .sampledImageDepthSampleCounts
= sample_counts
,
450 .sampledImageStencilSampleCounts
= sample_counts
,
451 .storageImageSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
452 .maxSampleMaskWords
= 1,
453 .timestampComputeAndGraphics
= false,
454 .timestampPeriod
= 100000.0 / pdevice
->rad_info
.clock_crystal_freq
,
455 .maxClipDistances
= 8,
456 .maxCullDistances
= 8,
457 .maxCombinedClipAndCullDistances
= 8,
458 .discreteQueuePriorities
= 1,
459 .pointSizeRange
= { 0.125, 255.875 },
460 .lineWidthRange
= { 0.0, 7.9921875 },
461 .pointSizeGranularity
= (1.0 / 8.0),
462 .lineWidthGranularity
= (1.0 / 128.0),
463 .strictLines
= false, /* FINISHME */
464 .standardSampleLocations
= true,
465 .optimalBufferCopyOffsetAlignment
= 128,
466 .optimalBufferCopyRowPitchAlignment
= 128,
467 .nonCoherentAtomSize
= 64,
470 *pProperties
= (VkPhysicalDeviceProperties
) {
471 .apiVersion
= VK_MAKE_VERSION(1, 0, 5),
474 .deviceID
= pdevice
->rad_info
.pci_id
,
475 .deviceType
= VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU
,
477 .sparseProperties
= {0}, /* Broadwell doesn't do sparse. */
480 strcpy(pProperties
->deviceName
, pdevice
->name
);
481 radv_device_get_cache_uuid(pProperties
->pipelineCacheUUID
);
484 void radv_GetPhysicalDeviceQueueFamilyProperties(
485 VkPhysicalDevice physicalDevice
,
487 VkQueueFamilyProperties
* pQueueFamilyProperties
)
489 if (pQueueFamilyProperties
== NULL
) {
493 assert(*pCount
>= 1);
495 *pQueueFamilyProperties
= (VkQueueFamilyProperties
) {
496 .queueFlags
= VK_QUEUE_GRAPHICS_BIT
|
497 VK_QUEUE_COMPUTE_BIT
|
498 VK_QUEUE_TRANSFER_BIT
,
500 .timestampValidBits
= 64,
501 .minImageTransferGranularity
= (VkExtent3D
) { 1, 1, 1 },
505 void radv_GetPhysicalDeviceMemoryProperties(
506 VkPhysicalDevice physicalDevice
,
507 VkPhysicalDeviceMemoryProperties
* pMemoryProperties
)
509 RADV_FROM_HANDLE(radv_physical_device
, physical_device
, physicalDevice
);
511 pMemoryProperties
->memoryTypeCount
= 3;
512 pMemoryProperties
->memoryTypes
[0] = (VkMemoryType
) {
513 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
,
516 pMemoryProperties
->memoryTypes
[1] = (VkMemoryType
) {
517 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
518 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
519 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
,
522 pMemoryProperties
->memoryTypes
[2] = (VkMemoryType
) {
523 .propertyFlags
= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
524 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
|
525 VK_MEMORY_PROPERTY_HOST_CACHED_BIT
,
529 pMemoryProperties
->memoryHeapCount
= 2;
530 pMemoryProperties
->memoryHeaps
[0] = (VkMemoryHeap
) {
531 .size
= physical_device
->rad_info
.vram_size
,
532 .flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
,
534 pMemoryProperties
->memoryHeaps
[1] = (VkMemoryHeap
) {
535 .size
= physical_device
->rad_info
.gart_size
,
541 radv_queue_init(struct radv_device
*device
, struct radv_queue
*queue
)
543 queue
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
544 queue
->device
= device
;
550 radv_queue_finish(struct radv_queue
*queue
)
554 VkResult
radv_CreateDevice(
555 VkPhysicalDevice physicalDevice
,
556 const VkDeviceCreateInfo
* pCreateInfo
,
557 const VkAllocationCallbacks
* pAllocator
,
560 RADV_FROM_HANDLE(radv_physical_device
, physical_device
, physicalDevice
);
562 struct radv_device
*device
;
564 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
566 for (uint32_t j
= 0; j
< ARRAY_SIZE(device_extensions
); j
++) {
567 if (strcmp(pCreateInfo
->ppEnabledExtensionNames
[i
],
568 device_extensions
[j
].extensionName
) == 0) {
574 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
577 device
= radv_alloc2(&physical_device
->instance
->alloc
, pAllocator
,
579 VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
581 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
583 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
584 device
->instance
= physical_device
->instance
;
586 device
->ws
= physical_device
->ws
;
588 device
->alloc
= *pAllocator
;
590 device
->alloc
= physical_device
->instance
->alloc
;
592 device
->hw_ctx
= device
->ws
->ctx_create(device
->ws
);
593 if (!device
->hw_ctx
) {
594 result
= VK_ERROR_OUT_OF_HOST_MEMORY
;
598 radv_queue_init(device
, &device
->queue
);
600 result
= radv_device_init_meta(device
);
601 if (result
!= VK_SUCCESS
) {
602 device
->ws
->ctx_destroy(device
->hw_ctx
);
605 device
->allow_fast_clears
= env_var_as_boolean("RADV_FAST_CLEARS", false);
606 device
->allow_dcc
= !env_var_as_boolean("RADV_DCC_DISABLE", false);
608 if (device
->allow_fast_clears
&& device
->allow_dcc
)
609 radv_finishme("DCC fast clears have not been tested\n");
611 radv_device_init_msaa(device
);
612 device
->empty_cs
= device
->ws
->cs_create(device
->ws
, RING_GFX
);
613 radeon_emit(device
->empty_cs
, PKT3(PKT3_CONTEXT_CONTROL
, 1, 0));
614 radeon_emit(device
->empty_cs
, CONTEXT_CONTROL_LOAD_ENABLE(1));
615 radeon_emit(device
->empty_cs
, CONTEXT_CONTROL_SHADOW_ENABLE(1));
616 device
->ws
->cs_finalize(device
->empty_cs
);
617 *pDevice
= radv_device_to_handle(device
);
620 radv_free(&device
->alloc
, device
);
624 void radv_DestroyDevice(
626 const VkAllocationCallbacks
* pAllocator
)
628 RADV_FROM_HANDLE(radv_device
, device
, _device
);
630 device
->ws
->ctx_destroy(device
->hw_ctx
);
631 radv_queue_finish(&device
->queue
);
632 radv_device_finish_meta(device
);
634 radv_free(&device
->alloc
, device
);
637 VkResult
radv_EnumerateInstanceExtensionProperties(
638 const char* pLayerName
,
639 uint32_t* pPropertyCount
,
640 VkExtensionProperties
* pProperties
)
643 if (pProperties
== NULL
) {
644 *pPropertyCount
= ARRAY_SIZE(global_extensions
);
648 for (i
= 0; i
< *pPropertyCount
; i
++)
649 memcpy(&pProperties
[i
], &global_extensions
[i
], sizeof(VkExtensionProperties
));
652 if (i
< ARRAY_SIZE(global_extensions
))
653 return VK_INCOMPLETE
;
658 VkResult
radv_EnumerateDeviceExtensionProperties(
659 VkPhysicalDevice physicalDevice
,
660 const char* pLayerName
,
661 uint32_t* pPropertyCount
,
662 VkExtensionProperties
* pProperties
)
666 if (pProperties
== NULL
) {
667 *pPropertyCount
= ARRAY_SIZE(device_extensions
);
671 for (i
= 0; i
< *pPropertyCount
; i
++)
672 memcpy(&pProperties
[i
], &device_extensions
[i
], sizeof(VkExtensionProperties
));
675 if (i
< ARRAY_SIZE(device_extensions
))
676 return VK_INCOMPLETE
;
680 VkResult
radv_EnumerateInstanceLayerProperties(
681 uint32_t* pPropertyCount
,
682 VkLayerProperties
* pProperties
)
684 if (pProperties
== NULL
) {
689 /* None supported at this time */
690 return vk_error(VK_ERROR_LAYER_NOT_PRESENT
);
693 VkResult
radv_EnumerateDeviceLayerProperties(
694 VkPhysicalDevice physicalDevice
,
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 void radv_GetDeviceQueue(
709 uint32_t queueNodeIndex
,
713 RADV_FROM_HANDLE(radv_device
, device
, _device
);
715 assert(queueIndex
== 0);
717 *pQueue
= radv_queue_to_handle(&device
->queue
);
720 VkResult
radv_QueueSubmit(
722 uint32_t submitCount
,
723 const VkSubmitInfo
* pSubmits
,
726 RADV_FROM_HANDLE(radv_queue
, queue
, _queue
);
727 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
728 struct radeon_winsys_fence
*base_fence
= fence
? fence
->fence
: NULL
;
729 struct radeon_winsys_ctx
*ctx
= queue
->device
->hw_ctx
;
732 for (uint32_t i
= 0; i
< submitCount
; i
++) {
733 struct radeon_winsys_cs
**cs_array
;
734 bool can_patch
= true;
736 if (!pSubmits
[i
].commandBufferCount
)
739 cs_array
= malloc(sizeof(struct radeon_winsys_cs
*) *
740 pSubmits
[i
].commandBufferCount
);
742 for (uint32_t j
= 0; j
< pSubmits
[i
].commandBufferCount
; j
++) {
743 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
,
744 pSubmits
[i
].pCommandBuffers
[j
]);
745 assert(cmd_buffer
->level
== VK_COMMAND_BUFFER_LEVEL_PRIMARY
);
747 cs_array
[j
] = cmd_buffer
->cs
;
748 if ((cmd_buffer
->usage_flags
& VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT
))
751 ret
= queue
->device
->ws
->cs_submit(ctx
, cs_array
,
752 pSubmits
[i
].commandBufferCount
,
753 can_patch
, base_fence
);
755 radv_loge("failed to submit CS %d\n", i
);
761 ret
= queue
->device
->ws
->cs_submit(ctx
, &queue
->device
->empty_cs
,
762 1, false, base_fence
);
764 fence
->submitted
= true;
770 VkResult
radv_QueueWaitIdle(
773 RADV_FROM_HANDLE(radv_queue
, queue
, _queue
);
775 queue
->device
->ws
->ctx_wait_idle(queue
->device
->hw_ctx
);
779 VkResult
radv_DeviceWaitIdle(
782 RADV_FROM_HANDLE(radv_device
, device
, _device
);
784 device
->ws
->ctx_wait_idle(device
->hw_ctx
);
788 PFN_vkVoidFunction
radv_GetInstanceProcAddr(
792 return radv_lookup_entrypoint(pName
);
795 /* The loader wants us to expose a second GetInstanceProcAddr function
796 * to work around certain LD_PRELOAD issues seen in apps.
799 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
804 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
808 return radv_GetInstanceProcAddr(instance
, pName
);
811 PFN_vkVoidFunction
radv_GetDeviceProcAddr(
815 return radv_lookup_entrypoint(pName
);
818 VkResult
radv_AllocateMemory(
820 const VkMemoryAllocateInfo
* pAllocateInfo
,
821 const VkAllocationCallbacks
* pAllocator
,
822 VkDeviceMemory
* pMem
)
824 RADV_FROM_HANDLE(radv_device
, device
, _device
);
825 struct radv_device_memory
*mem
;
827 enum radeon_bo_domain domain
;
829 assert(pAllocateInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO
);
831 if (pAllocateInfo
->allocationSize
== 0) {
832 /* Apparently, this is allowed */
833 *pMem
= VK_NULL_HANDLE
;
837 mem
= radv_alloc2(&device
->alloc
, pAllocator
, sizeof(*mem
), 8,
838 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
840 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
842 uint64_t alloc_size
= align_u64(pAllocateInfo
->allocationSize
, 4096);
843 if (pAllocateInfo
->memoryTypeIndex
== 2)
844 domain
= RADEON_DOMAIN_GTT
;
846 domain
= RADEON_DOMAIN_VRAM
;
848 if (pAllocateInfo
->memoryTypeIndex
== 0)
849 flags
|= RADEON_FLAG_NO_CPU_ACCESS
;
851 flags
|= RADEON_FLAG_CPU_ACCESS
;
852 mem
->bo
= device
->ws
->buffer_create(device
->ws
, alloc_size
, 32768,
856 result
= VK_ERROR_OUT_OF_DEVICE_MEMORY
;
859 mem
->type_index
= pAllocateInfo
->memoryTypeIndex
;
861 *pMem
= radv_device_memory_to_handle(mem
);
866 radv_free2(&device
->alloc
, pAllocator
, mem
);
871 void radv_FreeMemory(
874 const VkAllocationCallbacks
* pAllocator
)
876 RADV_FROM_HANDLE(radv_device
, device
, _device
);
877 RADV_FROM_HANDLE(radv_device_memory
, mem
, _mem
);
882 device
->ws
->buffer_destroy(mem
->bo
);
885 radv_free2(&device
->alloc
, pAllocator
, mem
);
888 VkResult
radv_MapMemory(
890 VkDeviceMemory _memory
,
893 VkMemoryMapFlags flags
,
896 RADV_FROM_HANDLE(radv_device
, device
, _device
);
897 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
904 *ppData
= device
->ws
->buffer_map(mem
->bo
);
910 return VK_ERROR_MEMORY_MAP_FAILED
;
913 void radv_UnmapMemory(
915 VkDeviceMemory _memory
)
917 RADV_FROM_HANDLE(radv_device
, device
, _device
);
918 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
923 device
->ws
->buffer_unmap(mem
->bo
);
926 VkResult
radv_FlushMappedMemoryRanges(
928 uint32_t memoryRangeCount
,
929 const VkMappedMemoryRange
* pMemoryRanges
)
934 VkResult
radv_InvalidateMappedMemoryRanges(
936 uint32_t memoryRangeCount
,
937 const VkMappedMemoryRange
* pMemoryRanges
)
942 void radv_GetBufferMemoryRequirements(
945 VkMemoryRequirements
* pMemoryRequirements
)
947 RADV_FROM_HANDLE(radv_buffer
, buffer
, _buffer
);
949 /* The Vulkan spec (git aaed022) says:
951 * memoryTypeBits is a bitfield and contains one bit set for every
952 * supported memory type for the resource. The bit `1<<i` is set if and
953 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
954 * structure for the physical device is supported.
956 * We support exactly one memory type.
958 pMemoryRequirements
->memoryTypeBits
= 0x7;
960 pMemoryRequirements
->size
= buffer
->size
;
961 pMemoryRequirements
->alignment
= 16;
964 void radv_GetImageMemoryRequirements(
967 VkMemoryRequirements
* pMemoryRequirements
)
969 RADV_FROM_HANDLE(radv_image
, image
, _image
);
971 /* The Vulkan spec (git aaed022) says:
973 * memoryTypeBits is a bitfield and contains one bit set for every
974 * supported memory type for the resource. The bit `1<<i` is set if and
975 * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
976 * structure for the physical device is supported.
978 * We support exactly one memory type.
980 pMemoryRequirements
->memoryTypeBits
= 0x7;
982 pMemoryRequirements
->size
= image
->size
;
983 pMemoryRequirements
->alignment
= image
->alignment
;
986 void radv_GetImageSparseMemoryRequirements(
989 uint32_t* pSparseMemoryRequirementCount
,
990 VkSparseImageMemoryRequirements
* pSparseMemoryRequirements
)
995 void radv_GetDeviceMemoryCommitment(
997 VkDeviceMemory memory
,
998 VkDeviceSize
* pCommittedMemoryInBytes
)
1000 *pCommittedMemoryInBytes
= 0;
1003 VkResult
radv_BindBufferMemory(
1006 VkDeviceMemory _memory
,
1007 VkDeviceSize memoryOffset
)
1009 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
1010 RADV_FROM_HANDLE(radv_buffer
, buffer
, _buffer
);
1013 buffer
->bo
= mem
->bo
;
1014 buffer
->offset
= memoryOffset
;
1023 VkResult
radv_BindImageMemory(
1026 VkDeviceMemory _memory
,
1027 VkDeviceSize memoryOffset
)
1029 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
1030 RADV_FROM_HANDLE(radv_image
, image
, _image
);
1033 image
->bo
= mem
->bo
;
1034 image
->offset
= memoryOffset
;
1043 VkResult
radv_QueueBindSparse(
1045 uint32_t bindInfoCount
,
1046 const VkBindSparseInfo
* pBindInfo
,
1049 stub_return(VK_ERROR_INCOMPATIBLE_DRIVER
);
1052 VkResult
radv_CreateFence(
1054 const VkFenceCreateInfo
* pCreateInfo
,
1055 const VkAllocationCallbacks
* pAllocator
,
1058 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1059 struct radv_fence
*fence
= radv_alloc2(&device
->alloc
, pAllocator
,
1061 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1064 return VK_ERROR_OUT_OF_HOST_MEMORY
;
1066 memset(fence
, 0, sizeof(*fence
));
1067 fence
->submitted
= false;
1068 fence
->signalled
= !!(pCreateInfo
->flags
& VK_FENCE_CREATE_SIGNALED_BIT
);
1069 fence
->fence
= device
->ws
->create_fence();
1072 *pFence
= radv_fence_to_handle(fence
);
1077 void radv_DestroyFence(
1080 const VkAllocationCallbacks
* pAllocator
)
1082 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1083 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
1087 device
->ws
->destroy_fence(fence
->fence
);
1088 radv_free2(&device
->alloc
, pAllocator
, fence
);
1091 static uint64_t radv_get_absolute_timeout(uint64_t timeout
)
1093 uint64_t current_time
;
1096 clock_gettime(CLOCK_MONOTONIC
, &tv
);
1097 current_time
= tv
.tv_nsec
+ tv
.tv_sec
*1000000000ull;
1099 timeout
= MIN2(UINT64_MAX
- current_time
, timeout
);
1101 return current_time
+ timeout
;
1104 VkResult
radv_WaitForFences(
1106 uint32_t fenceCount
,
1107 const VkFence
* pFences
,
1111 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1112 timeout
= radv_get_absolute_timeout(timeout
);
1114 if (!waitAll
&& fenceCount
> 1) {
1115 fprintf(stderr
, "radv: WaitForFences without waitAll not implemented yet\n");
1118 for (uint32_t i
= 0; i
< fenceCount
; ++i
) {
1119 RADV_FROM_HANDLE(radv_fence
, fence
, pFences
[i
]);
1120 bool expired
= false;
1122 if (!fence
->submitted
)
1125 if (fence
->signalled
)
1128 expired
= device
->ws
->fence_wait(device
->ws
, fence
->fence
, true, timeout
);
1132 fence
->signalled
= true;
1138 VkResult
radv_ResetFences(VkDevice device
,
1139 uint32_t fenceCount
,
1140 const VkFence
*pFences
)
1142 for (unsigned i
= 0; i
< fenceCount
; ++i
) {
1143 RADV_FROM_HANDLE(radv_fence
, fence
, pFences
[i
]);
1144 fence
->submitted
= fence
->signalled
= false;
1150 VkResult
radv_GetFenceStatus(VkDevice _device
, VkFence _fence
)
1152 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1153 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
1155 if (!fence
->submitted
)
1156 return VK_NOT_READY
;
1158 if (!device
->ws
->fence_wait(device
->ws
, fence
->fence
, false, 0))
1159 return VK_NOT_READY
;
1165 // Queue semaphore functions
1167 VkResult
radv_CreateSemaphore(
1169 const VkSemaphoreCreateInfo
* pCreateInfo
,
1170 const VkAllocationCallbacks
* pAllocator
,
1171 VkSemaphore
* pSemaphore
)
1173 /* The DRM execbuffer ioctl always execute in-oder, even between different
1174 * rings. As such, there's nothing to do for the user space semaphore.
1177 *pSemaphore
= (VkSemaphore
)1;
1182 void radv_DestroySemaphore(
1184 VkSemaphore semaphore
,
1185 const VkAllocationCallbacks
* pAllocator
)
1189 VkResult
radv_CreateEvent(
1191 const VkEventCreateInfo
* pCreateInfo
,
1192 const VkAllocationCallbacks
* pAllocator
,
1195 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1196 struct radv_event
*event
= radv_alloc2(&device
->alloc
, pAllocator
,
1198 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1201 return VK_ERROR_OUT_OF_HOST_MEMORY
;
1203 event
->bo
= device
->ws
->buffer_create(device
->ws
, 8, 8,
1205 RADEON_FLAG_CPU_ACCESS
);
1207 radv_free2(&device
->alloc
, pAllocator
, event
);
1208 return VK_ERROR_OUT_OF_DEVICE_MEMORY
;
1211 event
->map
= (uint64_t*)device
->ws
->buffer_map(event
->bo
);
1213 *pEvent
= radv_event_to_handle(event
);
1218 void radv_DestroyEvent(
1221 const VkAllocationCallbacks
* pAllocator
)
1223 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1224 RADV_FROM_HANDLE(radv_event
, event
, _event
);
1228 device
->ws
->buffer_destroy(event
->bo
);
1229 radv_free2(&device
->alloc
, pAllocator
, event
);
1232 VkResult
radv_GetEventStatus(
1236 RADV_FROM_HANDLE(radv_event
, event
, _event
);
1238 if (*event
->map
== 1)
1239 return VK_EVENT_SET
;
1240 return VK_EVENT_RESET
;
1243 VkResult
radv_SetEvent(
1247 RADV_FROM_HANDLE(radv_event
, event
, _event
);
1253 VkResult
radv_ResetEvent(
1257 RADV_FROM_HANDLE(radv_event
, event
, _event
);
1263 VkResult
radv_CreateBuffer(
1265 const VkBufferCreateInfo
* pCreateInfo
,
1266 const VkAllocationCallbacks
* pAllocator
,
1269 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1270 struct radv_buffer
*buffer
;
1272 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
1274 buffer
= radv_alloc2(&device
->alloc
, pAllocator
, sizeof(*buffer
), 8,
1275 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1277 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1279 buffer
->size
= pCreateInfo
->size
;
1280 buffer
->usage
= pCreateInfo
->usage
;
1284 *pBuffer
= radv_buffer_to_handle(buffer
);
1289 void radv_DestroyBuffer(
1292 const VkAllocationCallbacks
* pAllocator
)
1294 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1295 RADV_FROM_HANDLE(radv_buffer
, buffer
, _buffer
);
1300 radv_free2(&device
->alloc
, pAllocator
, buffer
);
1303 static inline unsigned
1304 si_tile_mode_index(const struct radv_image
*image
, unsigned level
, bool stencil
)
1307 return image
->surface
.stencil_tiling_index
[level
];
1309 return image
->surface
.tiling_index
[level
];
1313 radv_initialise_color_surface(struct radv_device
*device
,
1314 struct radv_color_buffer_info
*cb
,
1315 struct radv_image_view
*iview
)
1317 const struct vk_format_description
*desc
;
1318 unsigned ntype
, format
, swap
, endian
;
1319 unsigned blend_clamp
= 0, blend_bypass
= 0;
1320 unsigned pitch_tile_max
, slice_tile_max
, tile_mode_index
;
1322 const struct radeon_surf
*surf
= &iview
->image
->surface
;
1323 const struct radeon_surf_level
*level_info
= &surf
->level
[iview
->base_mip
];
1325 desc
= vk_format_description(iview
->vk_format
);
1327 memset(cb
, 0, sizeof(*cb
));
1329 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
;
1330 va
+= level_info
->offset
;
1331 cb
->cb_color_base
= va
>> 8;
1333 /* CMASK variables */
1334 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
;
1335 va
+= iview
->image
->cmask
.offset
;
1336 cb
->cb_color_cmask
= va
>> 8;
1337 cb
->cb_color_cmask_slice
= iview
->image
->cmask
.slice_tile_max
;
1339 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
;
1340 va
+= iview
->image
->dcc_offset
;
1341 cb
->cb_dcc_base
= va
>> 8;
1343 cb
->cb_color_view
= S_028C6C_SLICE_START(iview
->base_layer
) |
1344 S_028C6C_SLICE_MAX(iview
->base_layer
+ iview
->extent
.depth
- 1);
1346 cb
->micro_tile_mode
= iview
->image
->surface
.micro_tile_mode
;
1347 pitch_tile_max
= level_info
->nblk_x
/ 8 - 1;
1348 slice_tile_max
= (level_info
->nblk_x
* level_info
->nblk_y
) / 64 - 1;
1349 tile_mode_index
= si_tile_mode_index(iview
->image
, iview
->base_mip
, false);
1351 cb
->cb_color_pitch
= S_028C64_TILE_MAX(pitch_tile_max
);
1352 cb
->cb_color_slice
= S_028C68_TILE_MAX(slice_tile_max
);
1354 /* Intensity is implemented as Red, so treat it that way. */
1355 cb
->cb_color_attrib
= S_028C74_FORCE_DST_ALPHA_1(desc
->swizzle
[3] == VK_SWIZZLE_1
) |
1356 S_028C74_TILE_MODE_INDEX(tile_mode_index
);
1358 if (iview
->image
->samples
> 1) {
1359 unsigned log_samples
= util_logbase2(iview
->image
->samples
);
1361 cb
->cb_color_attrib
|= S_028C74_NUM_SAMPLES(log_samples
) |
1362 S_028C74_NUM_FRAGMENTS(log_samples
);
1365 if (iview
->image
->fmask
.size
) {
1366 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
+ iview
->image
->fmask
.offset
;
1367 if (device
->instance
->physicalDevice
.rad_info
.chip_class
>= CIK
)
1368 cb
->cb_color_pitch
|= S_028C64_FMASK_TILE_MAX(iview
->image
->fmask
.pitch_in_pixels
/ 8 - 1);
1369 cb
->cb_color_attrib
|= S_028C74_FMASK_TILE_MODE_INDEX(iview
->image
->fmask
.tile_mode_index
);
1370 cb
->cb_color_fmask
= va
>> 8;
1371 cb
->cb_color_fmask_slice
= S_028C88_TILE_MAX(iview
->image
->fmask
.slice_tile_max
);
1373 /* This must be set for fast clear to work without FMASK. */
1374 if (device
->instance
->physicalDevice
.rad_info
.chip_class
>= CIK
)
1375 cb
->cb_color_pitch
|= S_028C64_FMASK_TILE_MAX(pitch_tile_max
);
1376 cb
->cb_color_attrib
|= S_028C74_FMASK_TILE_MODE_INDEX(tile_mode_index
);
1377 cb
->cb_color_fmask
= cb
->cb_color_base
;
1378 cb
->cb_color_fmask_slice
= S_028C88_TILE_MAX(slice_tile_max
);
1381 ntype
= radv_translate_color_numformat(iview
->vk_format
,
1383 vk_format_get_first_non_void_channel(iview
->vk_format
));
1384 format
= radv_translate_colorformat(iview
->vk_format
);
1385 if (format
== V_028C70_COLOR_INVALID
|| ntype
== ~0u)
1386 radv_finishme("Illegal color\n");
1387 swap
= radv_translate_colorswap(iview
->vk_format
, FALSE
);
1388 endian
= radv_colorformat_endian_swap(format
);
1390 /* blend clamp should be set for all NORM/SRGB types */
1391 if (ntype
== V_028C70_NUMBER_UNORM
||
1392 ntype
== V_028C70_NUMBER_SNORM
||
1393 ntype
== V_028C70_NUMBER_SRGB
)
1396 /* set blend bypass according to docs if SINT/UINT or
1397 8/24 COLOR variants */
1398 if (ntype
== V_028C70_NUMBER_UINT
|| ntype
== V_028C70_NUMBER_SINT
||
1399 format
== V_028C70_COLOR_8_24
|| format
== V_028C70_COLOR_24_8
||
1400 format
== V_028C70_COLOR_X24_8_32_FLOAT
) {
1405 if ((ntype
== V_028C70_NUMBER_UINT
|| ntype
== V_028C70_NUMBER_SINT
) &&
1406 (format
== V_028C70_COLOR_8
||
1407 format
== V_028C70_COLOR_8_8
||
1408 format
== V_028C70_COLOR_8_8_8_8
))
1409 ->color_is_int8
= true;
1411 cb
->cb_color_info
= S_028C70_FORMAT(format
) |
1412 S_028C70_COMP_SWAP(swap
) |
1413 S_028C70_BLEND_CLAMP(blend_clamp
) |
1414 S_028C70_BLEND_BYPASS(blend_bypass
) |
1415 S_028C70_SIMPLE_FLOAT(1) |
1416 S_028C70_ROUND_MODE(ntype
!= V_028C70_NUMBER_UNORM
&&
1417 ntype
!= V_028C70_NUMBER_SNORM
&&
1418 ntype
!= V_028C70_NUMBER_SRGB
&&
1419 format
!= V_028C70_COLOR_8_24
&&
1420 format
!= V_028C70_COLOR_24_8
) |
1421 S_028C70_NUMBER_TYPE(ntype
) |
1422 S_028C70_ENDIAN(endian
);
1423 if (iview
->image
->samples
> 1)
1424 if (iview
->image
->fmask
.size
)
1425 cb
->cb_color_info
|= S_028C70_COMPRESSION(1);
1427 if (iview
->image
->cmask
.size
&& device
->allow_fast_clears
)
1428 cb
->cb_color_info
|= S_028C70_FAST_CLEAR(1);
1430 if (iview
->image
->surface
.dcc_size
&& level_info
->dcc_enabled
)
1431 cb
->cb_color_info
|= S_028C70_DCC_ENABLE(1);
1433 if (device
->instance
->physicalDevice
.rad_info
.chip_class
>= VI
) {
1434 unsigned max_uncompressed_block_size
= 2;
1435 if (iview
->image
->samples
> 1) {
1436 if (iview
->image
->surface
.bpe
== 1)
1437 max_uncompressed_block_size
= 0;
1438 else if (iview
->image
->surface
.bpe
== 2)
1439 max_uncompressed_block_size
= 1;
1442 cb
->cb_dcc_control
= S_028C78_MAX_UNCOMPRESSED_BLOCK_SIZE(max_uncompressed_block_size
) |
1443 S_028C78_INDEPENDENT_64B_BLOCKS(1);
1446 /* This must be set for fast clear to work without FMASK. */
1447 if (!iview
->image
->fmask
.size
&&
1448 device
->instance
->physicalDevice
.rad_info
.chip_class
== SI
) {
1449 unsigned bankh
= util_logbase2(iview
->image
->surface
.bankh
);
1450 cb
->cb_color_attrib
|= S_028C74_FMASK_BANK_HEIGHT(bankh
);
1455 radv_initialise_ds_surface(struct radv_device
*device
,
1456 struct radv_ds_buffer_info
*ds
,
1457 struct radv_image_view
*iview
)
1459 unsigned level
= iview
->base_mip
;
1461 uint64_t va
, s_offs
, z_offs
;
1462 const struct radeon_surf_level
*level_info
= &iview
->image
->surface
.level
[level
];
1463 memset(ds
, 0, sizeof(*ds
));
1464 switch (iview
->vk_format
) {
1465 case VK_FORMAT_D24_UNORM_S8_UINT
:
1466 case VK_FORMAT_X8_D24_UNORM_PACK32
:
1467 ds
->pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-24);
1468 ds
->offset_scale
= 2.0f
;
1470 case VK_FORMAT_D16_UNORM
:
1471 case VK_FORMAT_D16_UNORM_S8_UINT
:
1472 ds
->pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-16);
1473 ds
->offset_scale
= 4.0f
;
1475 case VK_FORMAT_D32_SFLOAT
:
1476 case VK_FORMAT_D32_SFLOAT_S8_UINT
:
1477 ds
->pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-23) |
1478 S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
1479 ds
->offset_scale
= 1.0f
;
1485 format
= radv_translate_dbformat(iview
->vk_format
);
1486 if (format
== V_028040_Z_INVALID
) {
1487 fprintf(stderr
, "Invalid DB format: %d, disabling DB.\n", iview
->vk_format
);
1490 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
;
1491 s_offs
= z_offs
= va
;
1492 z_offs
+= iview
->image
->surface
.level
[level
].offset
;
1493 s_offs
+= iview
->image
->surface
.stencil_level
[level
].offset
;
1495 ds
->db_depth_view
= S_028008_SLICE_START(iview
->base_layer
) |
1496 S_028008_SLICE_MAX(iview
->base_layer
+ iview
->extent
.depth
- 1);
1497 ds
->db_depth_info
= S_02803C_ADDR5_SWIZZLE_MASK(1);
1498 ds
->db_z_info
= S_028040_FORMAT(format
) | S_028040_ZRANGE_PRECISION(1);
1500 if (iview
->image
->samples
> 1)
1501 ds
->db_z_info
|= S_028040_NUM_SAMPLES(util_logbase2(iview
->image
->samples
));
1503 if (iview
->image
->surface
.flags
& RADEON_SURF_SBUFFER
)
1504 ds
->db_stencil_info
= S_028044_FORMAT(V_028044_STENCIL_8
);
1506 ds
->db_stencil_info
= S_028044_FORMAT(V_028044_STENCIL_INVALID
);
1508 if (device
->instance
->physicalDevice
.rad_info
.chip_class
>= CIK
) {
1509 struct radeon_info
*info
= &device
->instance
->physicalDevice
.rad_info
;
1510 unsigned tiling_index
= iview
->image
->surface
.tiling_index
[level
];
1511 unsigned stencil_index
= iview
->image
->surface
.stencil_tiling_index
[level
];
1512 unsigned macro_index
= iview
->image
->surface
.macro_tile_index
;
1513 unsigned tile_mode
= info
->si_tile_mode_array
[tiling_index
];
1514 unsigned stencil_tile_mode
= info
->si_tile_mode_array
[stencil_index
];
1515 unsigned macro_mode
= info
->cik_macrotile_mode_array
[macro_index
];
1517 ds
->db_depth_info
|=
1518 S_02803C_ARRAY_MODE(G_009910_ARRAY_MODE(tile_mode
)) |
1519 S_02803C_PIPE_CONFIG(G_009910_PIPE_CONFIG(tile_mode
)) |
1520 S_02803C_BANK_WIDTH(G_009990_BANK_WIDTH(macro_mode
)) |
1521 S_02803C_BANK_HEIGHT(G_009990_BANK_HEIGHT(macro_mode
)) |
1522 S_02803C_MACRO_TILE_ASPECT(G_009990_MACRO_TILE_ASPECT(macro_mode
)) |
1523 S_02803C_NUM_BANKS(G_009990_NUM_BANKS(macro_mode
));
1524 ds
->db_z_info
|= S_028040_TILE_SPLIT(G_009910_TILE_SPLIT(tile_mode
));
1525 ds
->db_stencil_info
|= S_028044_TILE_SPLIT(G_009910_TILE_SPLIT(stencil_tile_mode
));
1527 unsigned tile_mode_index
= si_tile_mode_index(iview
->image
, level
, false);
1528 ds
->db_z_info
|= S_028040_TILE_MODE_INDEX(tile_mode_index
);
1529 tile_mode_index
= si_tile_mode_index(iview
->image
, level
, true);
1530 ds
->db_stencil_info
|= S_028044_TILE_MODE_INDEX(tile_mode_index
);
1533 if (iview
->image
->htile
.size
&& !level
) {
1534 ds
->db_z_info
|= S_028040_TILE_SURFACE_ENABLE(1) |
1535 S_028040_ALLOW_EXPCLEAR(1);
1537 if (iview
->image
->surface
.flags
& RADEON_SURF_SBUFFER
) {
1538 /* Workaround: For a not yet understood reason, the
1539 * combination of MSAA, fast stencil clear and stencil
1540 * decompress messes with subsequent stencil buffer
1541 * uses. Problem was reproduced on Verde, Bonaire,
1542 * Tonga, and Carrizo.
1544 * Disabling EXPCLEAR works around the problem.
1546 * Check piglit's arb_texture_multisample-stencil-clear
1547 * test if you want to try changing this.
1549 if (iview
->image
->samples
<= 1)
1550 ds
->db_stencil_info
|= S_028044_ALLOW_EXPCLEAR(1);
1552 /* Use all of the htile_buffer for depth if there's no stencil. */
1553 ds
->db_stencil_info
|= S_028044_TILE_STENCIL_DISABLE(1);
1555 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
+
1556 iview
->image
->htile
.offset
;
1557 ds
->db_htile_data_base
= va
>> 8;
1558 ds
->db_htile_surface
= S_028ABC_FULL_CACHE(1);
1560 ds
->db_htile_data_base
= 0;
1561 ds
->db_htile_surface
= 0;
1564 ds
->db_z_read_base
= ds
->db_z_write_base
= z_offs
>> 8;
1565 ds
->db_stencil_read_base
= ds
->db_stencil_write_base
= s_offs
>> 8;
1567 ds
->db_depth_size
= S_028058_PITCH_TILE_MAX((level_info
->nblk_x
/ 8) - 1) |
1568 S_028058_HEIGHT_TILE_MAX((level_info
->nblk_y
/ 8) - 1);
1569 ds
->db_depth_slice
= S_02805C_SLICE_TILE_MAX((level_info
->nblk_x
* level_info
->nblk_y
) / 64 - 1);
1572 VkResult
radv_CreateFramebuffer(
1574 const VkFramebufferCreateInfo
* pCreateInfo
,
1575 const VkAllocationCallbacks
* pAllocator
,
1576 VkFramebuffer
* pFramebuffer
)
1578 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1579 struct radv_framebuffer
*framebuffer
;
1581 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
1583 size_t size
= sizeof(*framebuffer
) +
1584 sizeof(struct radv_attachment_info
) * pCreateInfo
->attachmentCount
;
1585 framebuffer
= radv_alloc2(&device
->alloc
, pAllocator
, size
, 8,
1586 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1587 if (framebuffer
== NULL
)
1588 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1590 framebuffer
->attachment_count
= pCreateInfo
->attachmentCount
;
1591 for (uint32_t i
= 0; i
< pCreateInfo
->attachmentCount
; i
++) {
1592 VkImageView _iview
= pCreateInfo
->pAttachments
[i
];
1593 struct radv_image_view
*iview
= radv_image_view_from_handle(_iview
);
1594 framebuffer
->attachments
[i
].attachment
= iview
;
1595 if (iview
->aspect_mask
& VK_IMAGE_ASPECT_COLOR_BIT
) {
1596 radv_initialise_color_surface(device
, &framebuffer
->attachments
[i
].cb
, iview
);
1597 } else if (iview
->aspect_mask
& (VK_IMAGE_ASPECT_DEPTH_BIT
| VK_IMAGE_ASPECT_STENCIL_BIT
)) {
1598 radv_initialise_ds_surface(device
, &framebuffer
->attachments
[i
].ds
, iview
);
1602 framebuffer
->width
= pCreateInfo
->width
;
1603 framebuffer
->height
= pCreateInfo
->height
;
1604 framebuffer
->layers
= pCreateInfo
->layers
;
1606 *pFramebuffer
= radv_framebuffer_to_handle(framebuffer
);
1610 void radv_DestroyFramebuffer(
1613 const VkAllocationCallbacks
* pAllocator
)
1615 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1616 RADV_FROM_HANDLE(radv_framebuffer
, fb
, _fb
);
1620 radv_free2(&device
->alloc
, pAllocator
, fb
);
1623 static unsigned radv_tex_wrap(VkSamplerAddressMode address_mode
)
1625 switch (address_mode
) {
1626 case VK_SAMPLER_ADDRESS_MODE_REPEAT
:
1627 return V_008F30_SQ_TEX_WRAP
;
1628 case VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT
:
1629 return V_008F30_SQ_TEX_MIRROR
;
1630 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE
:
1631 return V_008F30_SQ_TEX_CLAMP_LAST_TEXEL
;
1632 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER
:
1633 return V_008F30_SQ_TEX_CLAMP_BORDER
;
1634 case VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE
:
1635 return V_008F30_SQ_TEX_MIRROR_ONCE_LAST_TEXEL
;
1637 unreachable("illegal tex wrap mode");
1643 radv_tex_compare(VkCompareOp op
)
1646 case VK_COMPARE_OP_NEVER
:
1647 return V_008F30_SQ_TEX_DEPTH_COMPARE_NEVER
;
1648 case VK_COMPARE_OP_LESS
:
1649 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESS
;
1650 case VK_COMPARE_OP_EQUAL
:
1651 return V_008F30_SQ_TEX_DEPTH_COMPARE_EQUAL
;
1652 case VK_COMPARE_OP_LESS_OR_EQUAL
:
1653 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESSEQUAL
;
1654 case VK_COMPARE_OP_GREATER
:
1655 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATER
;
1656 case VK_COMPARE_OP_NOT_EQUAL
:
1657 return V_008F30_SQ_TEX_DEPTH_COMPARE_NOTEQUAL
;
1658 case VK_COMPARE_OP_GREATER_OR_EQUAL
:
1659 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATEREQUAL
;
1660 case VK_COMPARE_OP_ALWAYS
:
1661 return V_008F30_SQ_TEX_DEPTH_COMPARE_ALWAYS
;
1663 unreachable("illegal compare mode");
1669 radv_tex_filter(VkFilter filter
, unsigned max_ansio
)
1672 case VK_FILTER_NEAREST
:
1673 return (max_ansio
> 1 ? V_008F38_SQ_TEX_XY_FILTER_ANISO_POINT
:
1674 V_008F38_SQ_TEX_XY_FILTER_POINT
);
1675 case VK_FILTER_LINEAR
:
1676 return (max_ansio
> 1 ? V_008F38_SQ_TEX_XY_FILTER_ANISO_BILINEAR
:
1677 V_008F38_SQ_TEX_XY_FILTER_BILINEAR
);
1678 case VK_FILTER_CUBIC_IMG
:
1680 fprintf(stderr
, "illegal texture filter");
1686 radv_tex_mipfilter(VkSamplerMipmapMode mode
)
1689 case VK_SAMPLER_MIPMAP_MODE_NEAREST
:
1690 return V_008F38_SQ_TEX_Z_FILTER_POINT
;
1691 case VK_SAMPLER_MIPMAP_MODE_LINEAR
:
1692 return V_008F38_SQ_TEX_Z_FILTER_LINEAR
;
1694 return V_008F38_SQ_TEX_Z_FILTER_NONE
;
1699 radv_tex_bordercolor(VkBorderColor bcolor
)
1702 case VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
:
1703 case VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
:
1704 return V_008F3C_SQ_TEX_BORDER_COLOR_TRANS_BLACK
;
1705 case VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
:
1706 case VK_BORDER_COLOR_INT_OPAQUE_BLACK
:
1707 return V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_BLACK
;
1708 case VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
:
1709 case VK_BORDER_COLOR_INT_OPAQUE_WHITE
:
1710 return V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_WHITE
;
1718 radv_init_sampler(struct radv_device
*device
,
1719 struct radv_sampler
*sampler
,
1720 const VkSamplerCreateInfo
*pCreateInfo
)
1722 uint32_t max_aniso
= 0;
1723 uint32_t max_aniso_ratio
= 0;//TODO
1725 is_vi
= (device
->instance
->physicalDevice
.rad_info
.chip_class
>= VI
);
1727 sampler
->state
[0] = (S_008F30_CLAMP_X(radv_tex_wrap(pCreateInfo
->addressModeU
)) |
1728 S_008F30_CLAMP_Y(radv_tex_wrap(pCreateInfo
->addressModeV
)) |
1729 S_008F30_CLAMP_Z(radv_tex_wrap(pCreateInfo
->addressModeW
)) |
1730 S_008F30_MAX_ANISO_RATIO(max_aniso_ratio
) |
1731 S_008F30_DEPTH_COMPARE_FUNC(radv_tex_compare(pCreateInfo
->compareOp
)) |
1732 S_008F30_FORCE_UNNORMALIZED(pCreateInfo
->unnormalizedCoordinates
? 1 : 0) |
1733 S_008F30_DISABLE_CUBE_WRAP(0) |
1734 S_008F30_COMPAT_MODE(is_vi
));
1735 sampler
->state
[1] = (S_008F34_MIN_LOD(S_FIXED(CLAMP(pCreateInfo
->minLod
, 0, 15), 8)) |
1736 S_008F34_MAX_LOD(S_FIXED(CLAMP(pCreateInfo
->maxLod
, 0, 15), 8)));
1737 sampler
->state
[2] = (S_008F38_LOD_BIAS(S_FIXED(CLAMP(pCreateInfo
->mipLodBias
, -16, 16), 8)) |
1738 S_008F38_XY_MAG_FILTER(radv_tex_filter(pCreateInfo
->magFilter
, max_aniso
)) |
1739 S_008F38_XY_MIN_FILTER(radv_tex_filter(pCreateInfo
->minFilter
, max_aniso
)) |
1740 S_008F38_MIP_FILTER(radv_tex_mipfilter(pCreateInfo
->mipmapMode
)) |
1741 S_008F38_MIP_POINT_PRECLAMP(1) |
1742 S_008F38_DISABLE_LSB_CEIL(1) |
1743 S_008F38_FILTER_PREC_FIX(1) |
1744 S_008F38_ANISO_OVERRIDE(is_vi
));
1745 sampler
->state
[3] = (S_008F3C_BORDER_COLOR_PTR(0) |
1746 S_008F3C_BORDER_COLOR_TYPE(radv_tex_bordercolor(pCreateInfo
->borderColor
)));
1749 VkResult
radv_CreateSampler(
1751 const VkSamplerCreateInfo
* pCreateInfo
,
1752 const VkAllocationCallbacks
* pAllocator
,
1753 VkSampler
* pSampler
)
1755 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1756 struct radv_sampler
*sampler
;
1758 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO
);
1760 sampler
= radv_alloc2(&device
->alloc
, pAllocator
, sizeof(*sampler
), 8,
1761 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1763 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1765 radv_init_sampler(device
, sampler
, pCreateInfo
);
1766 *pSampler
= radv_sampler_to_handle(sampler
);
1771 void radv_DestroySampler(
1774 const VkAllocationCallbacks
* pAllocator
)
1776 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1777 RADV_FROM_HANDLE(radv_sampler
, sampler
, _sampler
);
1781 radv_free2(&device
->alloc
, pAllocator
, sampler
);