radv/wsi: drop allocate memory special case
[mesa.git] / src / amd / vulkan / radv_device.c
index ae9baf590d584678b09be766dc7f3a005ffe3e2a..1b7cd355938516794506d18ba7407c447ac959fb 100644 (file)
@@ -39,7 +39,6 @@
 #include <xf86drm.h>
 #include <amdgpu.h>
 #include <amdgpu_drm.h>
-#include "amdgpu_id.h"
 #include "winsys/amdgpu/radv_amdgpu_winsys_public.h"
 #include "ac_llvm_util.h"
 #include "vk_format.h"
@@ -76,190 +75,6 @@ radv_get_device_uuid(struct radeon_info *info, void *uuid)
        ac_compute_device_uuid(info, uuid, VK_UUID_SIZE);
 }
 
-static const VkExtensionProperties instance_extensions[] = {
-       {
-               .extensionName = VK_KHR_SURFACE_EXTENSION_NAME,
-               .specVersion = 25,
-       },
-#ifdef VK_USE_PLATFORM_XCB_KHR
-       {
-               .extensionName = VK_KHR_XCB_SURFACE_EXTENSION_NAME,
-               .specVersion = 6,
-       },
-#endif
-#ifdef VK_USE_PLATFORM_XLIB_KHR
-       {
-               .extensionName = VK_KHR_XLIB_SURFACE_EXTENSION_NAME,
-               .specVersion = 6,
-       },
-#endif
-#ifdef VK_USE_PLATFORM_WAYLAND_KHR
-       {
-               .extensionName = VK_KHR_WAYLAND_SURFACE_EXTENSION_NAME,
-               .specVersion = 6,
-       },
-#endif
-       {
-               .extensionName = VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_EXTERNAL_MEMORY_CAPABILITIES_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_EXTERNAL_SEMAPHORE_CAPABILITIES_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-};
-
-static const VkExtensionProperties common_device_extensions[] = {
-       {
-               .extensionName = VK_KHR_DESCRIPTOR_UPDATE_TEMPLATE_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_INCREMENTAL_PRESENT_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_MAINTENANCE1_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_SAMPLER_MIRROR_CLAMP_TO_EDGE_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_SWAPCHAIN_EXTENSION_NAME,
-               .specVersion = 68,
-       },
-       {
-               .extensionName = VK_AMD_DRAW_INDIRECT_COUNT_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_SHADER_DRAW_PARAMETERS_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_GET_MEMORY_REQUIREMENTS_2_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_EXTERNAL_MEMORY_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_EXTERNAL_MEMORY_FD_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_STORAGE_BUFFER_STORAGE_CLASS_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_VARIABLE_POINTERS_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_IMAGE_FORMAT_LIST_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_BIND_MEMORY_2_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-};
-
-static const VkExtensionProperties rasterization_order_extension[] ={
-       {
-               .extensionName = VK_AMD_RASTERIZATION_ORDER_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-};
-
-static const VkExtensionProperties ext_sema_device_extensions[] = {
-       {
-               .extensionName = VK_KHR_EXTERNAL_SEMAPHORE_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHR_EXTERNAL_SEMAPHORE_FD_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-       {
-               .extensionName = VK_KHX_MULTIVIEW_EXTENSION_NAME,
-               .specVersion = 1,
-       },
-};
-
-static VkResult
-radv_extensions_register(struct radv_instance *instance,
-                       struct radv_extensions *extensions,
-                       const VkExtensionProperties *new_ext,
-                       uint32_t num_ext)
-{
-       size_t new_size;
-       VkExtensionProperties *new_ptr;
-
-       assert(new_ext && num_ext > 0);
-
-       if (!new_ext)
-               return VK_ERROR_INITIALIZATION_FAILED;
-
-       new_size = (extensions->num_ext + num_ext) * sizeof(VkExtensionProperties);
-       new_ptr = vk_realloc(&instance->alloc, extensions->ext_array,
-                               new_size, 8, VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
-
-       /* Old array continues to be valid, update nothing */
-       if (!new_ptr)
-               return VK_ERROR_OUT_OF_HOST_MEMORY;
-
-       memcpy(&new_ptr[extensions->num_ext], new_ext,
-               num_ext * sizeof(VkExtensionProperties));
-       extensions->ext_array = new_ptr;
-       extensions->num_ext += num_ext;
-
-       return VK_SUCCESS;
-}
-
-static void
-radv_extensions_finish(struct radv_instance *instance,
-                       struct radv_extensions *extensions)
-{
-       assert(extensions);
-
-       if (!extensions)
-               radv_loge("Attemted to free invalid extension struct\n");
-
-       if (extensions->ext_array)
-               vk_free(&instance->alloc, extensions->ext_array);
-}
-
-static bool
-is_extension_enabled(const VkExtensionProperties *extensions,
-                       size_t num_ext,
-                       const char *name)
-{
-       assert(extensions && name);
-
-       for (uint32_t i = 0; i < num_ext; i++) {
-               if (strcmp(name, extensions[i].extensionName) == 0)
-                       return true;
-       }
-
-       return false;
-}
-
 static const char *
 get_chip_name(enum radeon_family family)
 {
@@ -288,6 +103,75 @@ get_chip_name(enum radeon_family family)
        }
 }
 
+static void
+radv_physical_device_init_mem_types(struct radv_physical_device *device)
+{
+       STATIC_ASSERT(RADV_MEM_HEAP_COUNT <= VK_MAX_MEMORY_HEAPS);
+       uint64_t visible_vram_size = MIN2(device->rad_info.vram_size,
+                                         device->rad_info.vram_vis_size);
+
+       int vram_index = -1, visible_vram_index = -1, gart_index = -1;
+       device->memory_properties.memoryHeapCount = 0;
+       if (device->rad_info.vram_size - visible_vram_size > 0) {
+               vram_index = device->memory_properties.memoryHeapCount++;
+               device->memory_properties.memoryHeaps[vram_index] = (VkMemoryHeap) {
+                       .size = device->rad_info.vram_size - visible_vram_size,
+                       .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
+               };
+       }
+       if (visible_vram_size) {
+               visible_vram_index = device->memory_properties.memoryHeapCount++;
+               device->memory_properties.memoryHeaps[visible_vram_index] = (VkMemoryHeap) {
+                       .size = visible_vram_size,
+                       .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
+               };
+       }
+       if (device->rad_info.gart_size > 0) {
+               gart_index = device->memory_properties.memoryHeapCount++;
+               device->memory_properties.memoryHeaps[gart_index] = (VkMemoryHeap) {
+                       .size = device->rad_info.gart_size,
+                       .flags = 0,
+               };
+       }
+
+       STATIC_ASSERT(RADV_MEM_TYPE_COUNT <= VK_MAX_MEMORY_TYPES);
+       unsigned type_count = 0;
+       if (vram_index >= 0) {
+               device->mem_type_indices[type_count] = RADV_MEM_TYPE_VRAM;
+               device->memory_properties.memoryTypes[type_count++] = (VkMemoryType) {
+                       .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
+                       .heapIndex = vram_index,
+               };
+       }
+       if (gart_index >= 0) {
+               device->mem_type_indices[type_count] = RADV_MEM_TYPE_GTT_WRITE_COMBINE;
+               device->memory_properties.memoryTypes[type_count++] = (VkMemoryType) {
+                       .propertyFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
+                       VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
+                       .heapIndex = gart_index,
+               };
+       }
+       if (visible_vram_index >= 0) {
+               device->mem_type_indices[type_count] = RADV_MEM_TYPE_VRAM_CPU_ACCESS;
+               device->memory_properties.memoryTypes[type_count++] = (VkMemoryType) {
+                       .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
+                       VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
+                       VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
+                       .heapIndex = visible_vram_index,
+               };
+       }
+       if (gart_index >= 0) {
+               device->mem_type_indices[type_count] = RADV_MEM_TYPE_GTT_CACHED;
+               device->memory_properties.memoryTypes[type_count++] = (VkMemoryType) {
+                       .propertyFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
+                       VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
+                       VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
+                       .heapIndex = gart_index,
+               };
+       }
+       device->memory_properties.memoryTypeCount = type_count;
+}
+
 static VkResult
 radv_physical_device_init(struct radv_physical_device *device,
                          struct radv_instance *instance,
@@ -300,7 +184,7 @@ radv_physical_device_init(struct radv_physical_device *device,
 
        fd = open(path, O_RDWR | O_CLOEXEC);
        if (fd < 0)
-               return VK_ERROR_INCOMPATIBLE_DRIVER;
+               return vk_error(VK_ERROR_INCOMPATIBLE_DRIVER);
 
        version = drmGetVersion(fd);
        if (!version) {
@@ -330,47 +214,29 @@ radv_physical_device_init(struct radv_physical_device *device,
 
        device->local_fd = fd;
        device->ws->query_info(device->ws, &device->rad_info);
-       result = radv_init_wsi(device);
-       if (result != VK_SUCCESS) {
-               device->ws->destroy(device->ws);
-               goto fail;
-       }
+
+       device->name = get_chip_name(device->rad_info.family);
 
        if (radv_device_get_cache_uuid(device->rad_info.family, device->cache_uuid)) {
-               radv_finish_wsi(device);
                device->ws->destroy(device->ws);
                result = vk_errorf(VK_ERROR_INITIALIZATION_FAILED,
                                   "cannot generate UUID");
                goto fail;
        }
 
-       result = radv_extensions_register(instance,
-                                       &device->extensions,
-                                       common_device_extensions,
-                                       ARRAY_SIZE(common_device_extensions));
-       if (result != VK_SUCCESS)
-               goto fail;
-
-       if (device->rad_info.chip_class >= VI && device->rad_info.max_se >= 2) {
-               result = radv_extensions_register(instance,
-                                               &device->extensions,
-                                               rasterization_order_extension,
-                                               ARRAY_SIZE(rasterization_order_extension));
-               if (result != VK_SUCCESS)
-                       goto fail;
-       }
+       /* These flags affect shader compilation. */
+       uint64_t shader_env_flags =
+               (device->instance->perftest_flags & RADV_PERFTEST_SISCHED ? 0x1 : 0) |
+               (device->instance->debug_flags & RADV_DEBUG_UNSAFE_MATH ? 0x2 : 0);
 
-       if (device->rad_info.has_syncobj) {
-               result = radv_extensions_register(instance,
-                                                 &device->extensions,
-                                                 ext_sema_device_extensions,
-                                                 ARRAY_SIZE(ext_sema_device_extensions));
-               if (result != VK_SUCCESS)
-                       goto fail;
-       }
+       /* The gpu id is already embeded in the uuid so we just pass "radv"
+        * when creating the cache.
+        */
+       char buf[VK_UUID_SIZE * 2 + 1];
+       disk_cache_format_hex_id(buf, device->cache_uuid, VK_UUID_SIZE * 2);
+       device->disk_cache = disk_cache_create(device->name, buf, shader_env_flags);
 
        fprintf(stderr, "WARNING: radv is not a conformant vulkan implementation, testing use only.\n");
-       device->name = get_chip_name(device->rad_info.family);
 
        radv_get_driver_uuid(&device->device_uuid);
        radv_get_device_uuid(&device->rad_info, &device->device_uuid);
@@ -381,6 +247,19 @@ radv_physical_device_init(struct radv_physical_device *device,
                device->rbplus_allowed = device->rad_info.family == CHIP_STONEY;
        }
 
+       /* The mere presense of CLEAR_STATE in the IB causes random GPU hangs
+        * on SI.
+        */
+       device->has_clear_state = device->rad_info.chip_class >= CIK;
+
+       radv_physical_device_init_mem_types(device);
+
+       result = radv_init_wsi(device);
+       if (result != VK_SUCCESS) {
+               device->ws->destroy(device->ws);
+               goto fail;
+       }
+
        return VK_SUCCESS;
 
 fail:
@@ -391,9 +270,9 @@ fail:
 static void
 radv_physical_device_finish(struct radv_physical_device *device)
 {
-       radv_extensions_finish(device->instance, &device->extensions);
        radv_finish_wsi(device);
        device->ws->destroy(device->ws);
+       disk_cache_destroy(device->disk_cache);
        close(device->local_fd);
 }
 
@@ -439,6 +318,7 @@ static const struct debug_control radv_debug_options[] = {
        {"vmfaults", RADV_DEBUG_VM_FAULTS},
        {"zerovram", RADV_DEBUG_ZERO_VRAM},
        {"syncshaders", RADV_DEBUG_SYNC_SHADERS},
+       {"nosisched", RADV_DEBUG_NO_SISCHED},
        {NULL, 0}
 };
 
@@ -462,6 +342,24 @@ radv_get_perftest_option_name(int id)
        return radv_perftest_options[id].string;
 }
 
+static void
+radv_handle_per_app_options(struct radv_instance *instance,
+                           const VkApplicationInfo *info)
+{
+       const char *name = info ? info->pApplicationName : NULL;
+
+       if (!name)
+               return;
+
+       if (!strcmp(name, "Talos - Linux - 32bit") ||
+           !strcmp(name, "Talos - Linux - 64bit")) {
+               /* Force enable LLVM sisched for Talos because it looks safe
+                * and it gives few more FPS.
+                */
+               instance->perftest_flags |= RADV_PERFTEST_SISCHED;
+       }
+}
+
 VkResult radv_CreateInstance(
        const VkInstanceCreateInfo*                 pCreateInfo,
        const VkAllocationCallbacks*                pAllocator,
@@ -489,19 +387,16 @@ VkResult radv_CreateInstance(
        }
 
        for (uint32_t i = 0; i < pCreateInfo->enabledExtensionCount; i++) {
-               if (!is_extension_enabled(instance_extensions,
-                                       ARRAY_SIZE(instance_extensions),
-                                       pCreateInfo->ppEnabledExtensionNames[i]))
+               const char *ext_name = pCreateInfo->ppEnabledExtensionNames[i];
+               if (!radv_instance_extension_supported(ext_name))
                        return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);
        }
 
-       instance = vk_alloc2(&default_alloc, pAllocator, sizeof(*instance), 8,
-                              VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
+       instance = vk_zalloc2(&default_alloc, pAllocator, sizeof(*instance), 8,
+                             VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
        if (!instance)
                return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 
-       memset(instance, 0, sizeof(*instance));
-
        instance->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
 
        if (pAllocator)
@@ -522,6 +417,16 @@ VkResult radv_CreateInstance(
        instance->perftest_flags = parse_debug_string(getenv("RADV_PERFTEST"),
                                                   radv_perftest_options);
 
+       radv_handle_per_app_options(instance, pCreateInfo->pApplicationInfo);
+
+       if (instance->debug_flags & RADV_DEBUG_NO_SISCHED) {
+               /* Disable sisched when the user requests it, this is mostly
+                * useful when the driver force-enable sisched for the given
+                * application.
+                */
+               instance->perftest_flags &= ~RADV_PERFTEST_SISCHED;
+       }
+
        *pInstance = radv_instance_to_handle(instance);
 
        return VK_SUCCESS;
@@ -559,7 +464,7 @@ radv_enumerate_devices(struct radv_instance *instance)
 
        max_devices = drmGetDevices2(0, devices, ARRAY_SIZE(devices));
        if (max_devices < 1)
-               return VK_ERROR_INCOMPATIBLE_DRIVER;
+               return vk_error(VK_ERROR_INCOMPATIBLE_DRIVER);
 
        for (unsigned i = 0; i < (unsigned)max_devices; i++) {
                if (devices[i]->available_nodes & 1 << DRM_NODE_RENDER &&
@@ -612,8 +517,6 @@ void radv_GetPhysicalDeviceFeatures(
        VkPhysicalDevice                            physicalDevice,
        VkPhysicalDeviceFeatures*                   pFeatures)
 {
-       RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
-       bool is_gfx9 = pdevice->rad_info.chip_class >= GFX9;
        memset(pFeatures, 0, sizeof(*pFeatures));
 
        *pFeatures = (VkPhysicalDeviceFeatures) {
@@ -621,8 +524,8 @@ void radv_GetPhysicalDeviceFeatures(
                .fullDrawIndexUint32                      = true,
                .imageCubeArray                           = true,
                .independentBlend                         = true,
-               .geometryShader                           = !is_gfx9,
-               .tessellationShader                       = !is_gfx9,
+               .geometryShader                           = true,
+               .tessellationShader                       = true,
                .sampleRateShading                        = true,
                .dualSrcBlend                             = true,
                .logicOp                                  = true,
@@ -825,7 +728,7 @@ void radv_GetPhysicalDeviceProperties(
        };
 
        *pProperties = (VkPhysicalDeviceProperties) {
-               .apiVersion = VK_MAKE_VERSION(1, 0, 42),
+               .apiVersion = radv_physical_device_api_version(pdevice),
                .driverVersion = vk_get_driver_version(),
                .vendorID = ATI_VENDOR_ID,
                .deviceID = pdevice->rad_info.pci_id,
@@ -974,49 +877,7 @@ void radv_GetPhysicalDeviceMemoryProperties(
 {
        RADV_FROM_HANDLE(radv_physical_device, physical_device, physicalDevice);
 
-       STATIC_ASSERT(RADV_MEM_TYPE_COUNT <= VK_MAX_MEMORY_TYPES);
-
-       pMemoryProperties->memoryTypeCount = RADV_MEM_TYPE_COUNT;
-       pMemoryProperties->memoryTypes[RADV_MEM_TYPE_VRAM] = (VkMemoryType) {
-               .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
-               .heapIndex = RADV_MEM_HEAP_VRAM,
-       };
-       pMemoryProperties->memoryTypes[RADV_MEM_TYPE_GTT_WRITE_COMBINE] = (VkMemoryType) {
-               .propertyFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
-               VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
-               .heapIndex = RADV_MEM_HEAP_GTT,
-       };
-       pMemoryProperties->memoryTypes[RADV_MEM_TYPE_VRAM_CPU_ACCESS] = (VkMemoryType) {
-               .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
-               VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
-               VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
-               .heapIndex = RADV_MEM_HEAP_VRAM_CPU_ACCESS,
-       };
-       pMemoryProperties->memoryTypes[RADV_MEM_TYPE_GTT_CACHED] = (VkMemoryType) {
-               .propertyFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
-               VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
-               VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
-               .heapIndex = RADV_MEM_HEAP_GTT,
-       };
-
-       STATIC_ASSERT(RADV_MEM_HEAP_COUNT <= VK_MAX_MEMORY_HEAPS);
-       uint64_t visible_vram_size = MIN2(physical_device->rad_info.vram_size,
-                                         physical_device->rad_info.vram_vis_size);
-
-       pMemoryProperties->memoryHeapCount = RADV_MEM_HEAP_COUNT;
-       pMemoryProperties->memoryHeaps[RADV_MEM_HEAP_VRAM] = (VkMemoryHeap) {
-               .size = physical_device->rad_info.vram_size -
-                               visible_vram_size,
-               .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
-       };
-       pMemoryProperties->memoryHeaps[RADV_MEM_HEAP_VRAM_CPU_ACCESS] = (VkMemoryHeap) {
-               .size = visible_vram_size,
-               .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
-       };
-       pMemoryProperties->memoryHeaps[RADV_MEM_HEAP_GTT] = (VkMemoryHeap) {
-               .size = physical_device->rad_info.gart_size,
-               .flags = 0,
-       };
+       *pMemoryProperties = physical_device->memory_properties;
 }
 
 void radv_GetPhysicalDeviceMemoryProperties2KHR(
@@ -1027,18 +888,42 @@ void radv_GetPhysicalDeviceMemoryProperties2KHR(
                                                      &pMemoryProperties->memoryProperties);
 }
 
+static enum radeon_ctx_priority
+radv_get_queue_global_priority(const VkDeviceQueueGlobalPriorityCreateInfoEXT *pObj)
+{
+       /* Default to MEDIUM when a specific global priority isn't requested */
+       if (!pObj)
+               return RADEON_CTX_PRIORITY_MEDIUM;
+
+       switch(pObj->globalPriority) {
+       case VK_QUEUE_GLOBAL_PRIORITY_REALTIME_EXT:
+               return RADEON_CTX_PRIORITY_REALTIME;
+       case VK_QUEUE_GLOBAL_PRIORITY_HIGH_EXT:
+               return RADEON_CTX_PRIORITY_HIGH;
+       case VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_EXT:
+               return RADEON_CTX_PRIORITY_MEDIUM;
+       case VK_QUEUE_GLOBAL_PRIORITY_LOW_EXT:
+               return RADEON_CTX_PRIORITY_LOW;
+       default:
+               unreachable("Illegal global priority value");
+               return RADEON_CTX_PRIORITY_INVALID;
+       }
+}
+
 static int
 radv_queue_init(struct radv_device *device, struct radv_queue *queue,
-               int queue_family_index, int idx)
+               uint32_t queue_family_index, int idx,
+               const VkDeviceQueueGlobalPriorityCreateInfoEXT *global_priority)
 {
        queue->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
        queue->device = device;
        queue->queue_family_index = queue_family_index;
        queue->queue_idx = idx;
+       queue->priority = radv_get_queue_global_priority(global_priority);
 
-       queue->hw_ctx = device->ws->ctx_create(device->ws);
+       queue->hw_ctx = device->ws->ctx_create(device->ws, queue->priority);
        if (!queue->hw_ctx)
-               return VK_ERROR_OUT_OF_HOST_MEMORY;
+               return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 
        return VK_SUCCESS;
 }
@@ -1114,11 +999,15 @@ VkResult radv_CreateDevice(
        VkResult result;
        struct radv_device *device;
 
+       bool keep_shader_info = false;
+
        for (uint32_t i = 0; i < pCreateInfo->enabledExtensionCount; i++) {
-               if (!is_extension_enabled(physical_device->extensions.ext_array,
-                                       physical_device->extensions.num_ext,
-                                       pCreateInfo->ppEnabledExtensionNames[i]))
+               const char *ext_name = pCreateInfo->ppEnabledExtensionNames[i];
+               if (!radv_physical_device_extension_supported(physical_device, ext_name))
                        return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);
+
+               if (strcmp(ext_name, VK_AMD_SHADER_INFO_EXTENSION_NAME) == 0)
+                       keep_shader_info = true;
        }
 
        /* Check enabled features */
@@ -1134,20 +1023,16 @@ VkResult radv_CreateDevice(
                }
        }
 
-       device = vk_alloc2(&physical_device->instance->alloc, pAllocator,
-                            sizeof(*device), 8,
-                            VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
+       device = vk_zalloc2(&physical_device->instance->alloc, pAllocator,
+                           sizeof(*device), 8,
+                           VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
        if (!device)
                return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 
-       memset(device, 0, sizeof(*device));
-
        device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
        device->instance = physical_device->instance;
        device->physical_device = physical_device;
 
-       device->debug_flags = device->instance->debug_flags;
-
        device->ws = physical_device->ws;
        if (pAllocator)
                device->alloc = *pAllocator;
@@ -1160,6 +1045,10 @@ VkResult radv_CreateDevice(
        for (unsigned i = 0; i < pCreateInfo->queueCreateInfoCount; i++) {
                const VkDeviceQueueCreateInfo *queue_create = &pCreateInfo->pQueueCreateInfos[i];
                uint32_t qfi = queue_create->queueFamilyIndex;
+               const VkDeviceQueueGlobalPriorityCreateInfoEXT *global_priority =
+                       vk_find_struct_const(queue_create->pNext, DEVICE_QUEUE_GLOBAL_PRIORITY_CREATE_INFO_EXT);
+
+               assert(!global_priority || device->physical_device->rad_info.has_ctx_priority);
 
                device->queues[qfi] = vk_alloc(&device->alloc,
                                               queue_create->queueCount * sizeof(struct radv_queue), 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
@@ -1173,7 +1062,7 @@ VkResult radv_CreateDevice(
                device->queue_count[qfi] = queue_create->queueCount;
 
                for (unsigned q = 0; q < queue_create->queueCount; q++) {
-                       result = radv_queue_init(device, &device->queues[qfi][q], qfi, q);
+                       result = radv_queue_init(device, &device->queues[qfi][q], qfi, q, global_priority);
                        if (result != VK_SUCCESS)
                                goto fail;
                }
@@ -1210,10 +1099,14 @@ VkResult radv_CreateDevice(
                device->physical_device->rad_info.max_se >= 2;
 
        if (getenv("RADV_TRACE_FILE")) {
+               keep_shader_info = true;
+
                if (!radv_init_trace(device))
                        goto fail;
        }
 
+       device->keep_shader_info = keep_shader_info;
+
        result = radv_device_init_meta(device);
        if (result != VK_SUCCESS)
                goto fail;
@@ -1307,47 +1200,6 @@ void radv_DestroyDevice(
        vk_free(&device->alloc, device);
 }
 
-VkResult radv_EnumerateInstanceExtensionProperties(
-       const char*                                 pLayerName,
-       uint32_t*                                   pPropertyCount,
-       VkExtensionProperties*                      pProperties)
-{
-       if (pProperties == NULL) {
-               *pPropertyCount = ARRAY_SIZE(instance_extensions);
-               return VK_SUCCESS;
-       }
-
-       *pPropertyCount = MIN2(*pPropertyCount, ARRAY_SIZE(instance_extensions));
-       typed_memcpy(pProperties, instance_extensions, *pPropertyCount);
-
-       if (*pPropertyCount < ARRAY_SIZE(instance_extensions))
-               return VK_INCOMPLETE;
-
-       return VK_SUCCESS;
-}
-
-VkResult radv_EnumerateDeviceExtensionProperties(
-       VkPhysicalDevice                            physicalDevice,
-       const char*                                 pLayerName,
-       uint32_t*                                   pPropertyCount,
-       VkExtensionProperties*                      pProperties)
-{
-       RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
-
-       if (pProperties == NULL) {
-               *pPropertyCount = pdevice->extensions.num_ext;
-               return VK_SUCCESS;
-       }
-
-       *pPropertyCount = MIN2(*pPropertyCount, pdevice->extensions.num_ext);
-       typed_memcpy(pProperties, pdevice->extensions.ext_array, *pPropertyCount);
-
-       if (*pPropertyCount < pdevice->extensions.num_ext)
-               return VK_INCOMPLETE;
-
-       return VK_SUCCESS;
-}
-
 VkResult radv_EnumerateInstanceLayerProperties(
        uint32_t*                                   pPropertyCount,
        VkLayerProperties*                          pProperties)
@@ -1604,6 +1456,7 @@ radv_get_preamble_cs(struct radv_queue *queue,
        unsigned tess_factor_ring_size = 0, tess_offchip_ring_size = 0;
        unsigned max_offchip_buffers;
        unsigned hs_offchip_param = 0;
+       uint32_t ring_bo_flags = RADEON_FLAG_NO_CPU_ACCESS | RADEON_FLAG_NO_INTERPROCESS_SHARING;
        if (!queue->has_tess_rings) {
                if (needs_tess_rings)
                        add_tess_rings = true;
@@ -1637,7 +1490,7 @@ radv_get_preamble_cs(struct radv_queue *queue,
                                                              scratch_size,
                                                              4096,
                                                              RADEON_DOMAIN_VRAM,
-                                                             RADEON_FLAG_NO_CPU_ACCESS);
+                                                             ring_bo_flags);
                if (!scratch_bo)
                        goto fail;
        } else
@@ -1648,7 +1501,7 @@ radv_get_preamble_cs(struct radv_queue *queue,
                                                                      compute_scratch_size,
                                                                      4096,
                                                                      RADEON_DOMAIN_VRAM,
-                                                                     RADEON_FLAG_NO_CPU_ACCESS);
+                                                                     ring_bo_flags);
                if (!compute_scratch_bo)
                        goto fail;
 
@@ -1660,7 +1513,7 @@ radv_get_preamble_cs(struct radv_queue *queue,
                                                                esgs_ring_size,
                                                                4096,
                                                                RADEON_DOMAIN_VRAM,
-                                                               RADEON_FLAG_NO_CPU_ACCESS);
+                                                               ring_bo_flags);
                if (!esgs_ring_bo)
                        goto fail;
        } else {
@@ -1673,7 +1526,7 @@ radv_get_preamble_cs(struct radv_queue *queue,
                                                                gsvs_ring_size,
                                                                4096,
                                                                RADEON_DOMAIN_VRAM,
-                                                               RADEON_FLAG_NO_CPU_ACCESS);
+                                                               ring_bo_flags);
                if (!gsvs_ring_bo)
                        goto fail;
        } else {
@@ -1686,14 +1539,14 @@ radv_get_preamble_cs(struct radv_queue *queue,
                                                                       tess_factor_ring_size,
                                                                       256,
                                                                       RADEON_DOMAIN_VRAM,
-                                                                      RADEON_FLAG_NO_CPU_ACCESS);
+                                                                      ring_bo_flags);
                if (!tess_factor_ring_bo)
                        goto fail;
                tess_offchip_ring_bo = queue->device->ws->buffer_create(queue->device->ws,
                                                                       tess_offchip_ring_size,
                                                                       256,
                                                                       RADEON_DOMAIN_VRAM,
-                                                                      RADEON_FLAG_NO_CPU_ACCESS);
+                                                                       ring_bo_flags);
                if (!tess_offchip_ring_bo)
                        goto fail;
        } else {
@@ -1720,7 +1573,7 @@ radv_get_preamble_cs(struct radv_queue *queue,
                                                                 size,
                                                                 4096,
                                                                 RADEON_DOMAIN_VRAM,
-                                                                RADEON_FLAG_CPU_ACCESS);
+                                                                RADEON_FLAG_CPU_ACCESS|RADEON_FLAG_NO_INTERPROCESS_SHARING);
                if (!descriptor_bo)
                        goto fail;
        } else
@@ -1736,22 +1589,22 @@ radv_get_preamble_cs(struct radv_queue *queue,
                dest_cs[i] = cs;
 
                if (scratch_bo)
-                       queue->device->ws->cs_add_buffer(cs, scratch_bo, 8);
+                       radv_cs_add_buffer(queue->device->ws, cs, scratch_bo, 8);
 
                if (esgs_ring_bo)
-                       queue->device->ws->cs_add_buffer(cs, esgs_ring_bo, 8);
+                       radv_cs_add_buffer(queue->device->ws, cs, esgs_ring_bo, 8);
 
                if (gsvs_ring_bo)
-                       queue->device->ws->cs_add_buffer(cs, gsvs_ring_bo, 8);
+                       radv_cs_add_buffer(queue->device->ws, cs, gsvs_ring_bo, 8);
 
                if (tess_factor_ring_bo)
-                       queue->device->ws->cs_add_buffer(cs, tess_factor_ring_bo, 8);
+                       radv_cs_add_buffer(queue->device->ws, cs, tess_factor_ring_bo, 8);
 
                if (tess_offchip_ring_bo)
-                       queue->device->ws->cs_add_buffer(cs, tess_offchip_ring_bo, 8);
+                       radv_cs_add_buffer(queue->device->ws, cs, tess_offchip_ring_bo, 8);
 
                if (descriptor_bo)
-                       queue->device->ws->cs_add_buffer(cs, descriptor_bo, 8);
+                       radv_cs_add_buffer(queue->device->ws, cs, descriptor_bo, 8);
 
                if (descriptor_bo != queue->descriptor_bo) {
                        uint32_t *map = (uint32_t*)queue->device->ws->buffer_map(descriptor_bo);
@@ -1817,19 +1670,31 @@ radv_get_preamble_cs(struct radv_queue *queue,
                }
 
                if (descriptor_bo) {
-                       uint32_t regs[] = {R_00B030_SPI_SHADER_USER_DATA_PS_0,
-                                          R_00B130_SPI_SHADER_USER_DATA_VS_0,
-                                          R_00B230_SPI_SHADER_USER_DATA_GS_0,
-                                          R_00B330_SPI_SHADER_USER_DATA_ES_0,
-                                          R_00B430_SPI_SHADER_USER_DATA_HS_0,
-                                          R_00B530_SPI_SHADER_USER_DATA_LS_0};
-
                        uint64_t va = radv_buffer_get_va(descriptor_bo);
-
-                       for (int i = 0; i < ARRAY_SIZE(regs); ++i) {
-                               radeon_set_sh_reg_seq(cs, regs[i], 2);
-                               radeon_emit(cs, va);
-                               radeon_emit(cs, va >> 32);
+                       if (queue->device->physical_device->rad_info.chip_class >= GFX9) {
+                               uint32_t regs[] = {R_00B030_SPI_SHADER_USER_DATA_PS_0,
+                                               R_00B130_SPI_SHADER_USER_DATA_VS_0,
+                                               R_00B208_SPI_SHADER_USER_DATA_ADDR_LO_GS,
+                                               R_00B408_SPI_SHADER_USER_DATA_ADDR_LO_HS};
+
+                               for (int i = 0; i < ARRAY_SIZE(regs); ++i) {
+                                       radeon_set_sh_reg_seq(cs, regs[i], 2);
+                                       radeon_emit(cs, va);
+                                       radeon_emit(cs, va >> 32);
+                               }
+                       } else {
+                               uint32_t regs[] = {R_00B030_SPI_SHADER_USER_DATA_PS_0,
+                                               R_00B130_SPI_SHADER_USER_DATA_VS_0,
+                                               R_00B230_SPI_SHADER_USER_DATA_GS_0,
+                                               R_00B330_SPI_SHADER_USER_DATA_ES_0,
+                                               R_00B430_SPI_SHADER_USER_DATA_HS_0,
+                                               R_00B530_SPI_SHADER_USER_DATA_LS_0};
+
+                               for (int i = 0; i < ARRAY_SIZE(regs); ++i) {
+                                       radeon_set_sh_reg_seq(cs, regs[i], 2);
+                                       radeon_emit(cs, va);
+                                       radeon_emit(cs, va >> 32);
+                               }
                        }
                }
 
@@ -1838,7 +1703,7 @@ radv_get_preamble_cs(struct radv_queue *queue,
                        uint32_t rsrc1 = S_008F04_BASE_ADDRESS_HI(scratch_va >> 32) |
                                         S_008F04_SWIZZLE_ENABLE(1);
 
-                       queue->device->ws->cs_add_buffer(cs, compute_scratch_bo, 8);
+                       radv_cs_add_buffer(queue->device->ws, cs, compute_scratch_bo, 8);
 
                        radeon_set_sh_reg_seq(cs, R_00B900_COMPUTE_USER_DATA_0, 2);
                        radeon_emit(cs, scratch_va);
@@ -1958,7 +1823,7 @@ fail:
                queue->device->ws->buffer_destroy(tess_factor_ring_bo);
        if (tess_offchip_ring_bo && tess_offchip_ring_bo != queue->tess_offchip_ring_bo)
                queue->device->ws->buffer_destroy(tess_offchip_ring_bo);
-       return VK_ERROR_OUT_OF_DEVICE_MEMORY;
+       return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
 }
 
 static VkResult radv_alloc_sem_counts(struct radv_winsys_sem_counts *counts,
@@ -1982,14 +1847,14 @@ static VkResult radv_alloc_sem_counts(struct radv_winsys_sem_counts *counts,
        if (counts->syncobj_count) {
                counts->syncobj = (uint32_t *)malloc(sizeof(uint32_t) * counts->syncobj_count);
                if (!counts->syncobj)
-                       return VK_ERROR_OUT_OF_HOST_MEMORY;
+                       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
        }
 
        if (counts->sem_count) {
                counts->sem = (struct radeon_winsys_sem **)malloc(sizeof(struct radeon_winsys_sem *) * counts->sem_count);
                if (!counts->sem) {
                        free(counts->syncobj);
-                       return VK_ERROR_OUT_OF_HOST_MEMORY;
+                       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
                }
        }
 
@@ -1998,10 +1863,6 @@ static VkResult radv_alloc_sem_counts(struct radv_winsys_sem_counts *counts,
 
                if (sem->temp_syncobj) {
                        counts->syncobj[syncobj_idx++] = sem->temp_syncobj;
-                       if (reset_temp) {
-                               /* after we wait on a temp import - drop it */
-                               sem->temp_syncobj = 0;
-                       }
                }
                else if (sem->syncobj)
                        counts->syncobj[syncobj_idx++] = sem->syncobj;
@@ -2022,6 +1883,21 @@ void radv_free_sem_info(struct radv_winsys_sem_info *sem_info)
        free(sem_info->signal.sem);
 }
 
+
+static void radv_free_temp_syncobjs(struct radv_device *device,
+                                   int num_sems,
+                                   const VkSemaphore *sems)
+{
+       for (uint32_t i = 0; i < num_sems; i++) {
+               RADV_FROM_HANDLE(radv_semaphore, sem, sems[i]);
+
+               if (sem->temp_syncobj) {
+                       device->ws->destroy_syncobj(device->ws, sem->temp_syncobj);
+                       sem->temp_syncobj = 0;
+               }
+       }
+}
+
 VkResult radv_alloc_sem_info(struct radv_winsys_sem_info *sem_info,
                             int num_wait_sems,
                             const VkSemaphore *wait_sems,
@@ -2160,6 +2036,9 @@ VkResult radv_QueueSubmit(
                        }
                }
 
+               radv_free_temp_syncobjs(queue->device,
+                                       pSubmits[i].waitSemaphoreCount,
+                                       pSubmits[i].pWaitSemaphores);
                radv_free_sem_info(&sem_info);
                free(cs_array);
        }
@@ -2247,17 +2126,16 @@ bool radv_get_memory_fd(struct radv_device *device,
                                         pFD);
 }
 
-VkResult radv_AllocateMemory(
-       VkDevice                                    _device,
-       const VkMemoryAllocateInfo*                 pAllocateInfo,
-       const VkAllocationCallbacks*                pAllocator,
-       VkDeviceMemory*                             pMem)
+static VkResult radv_alloc_memory(struct radv_device *device,
+                                 const VkMemoryAllocateInfo*     pAllocateInfo,
+                                 const VkAllocationCallbacks*    pAllocator,
+                                 VkDeviceMemory*                 pMem)
 {
-       RADV_FROM_HANDLE(radv_device, device, _device);
        struct radv_device_memory *mem;
        VkResult result;
        enum radeon_bo_domain domain;
        uint32_t flags = 0;
+       enum radv_mem_type mem_type_index = device->physical_device->mem_type_indices[pAllocateInfo->memoryTypeIndex];
 
        assert(pAllocateInfo->sType == VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO);
 
@@ -2272,11 +2150,17 @@ VkResult radv_AllocateMemory(
        const VkMemoryDedicatedAllocateInfoKHR *dedicate_info =
                vk_find_struct_const(pAllocateInfo->pNext, MEMORY_DEDICATED_ALLOCATE_INFO_KHR);
 
+       const struct wsi_memory_allocate_info *wsi_info =
+               vk_find_struct_const(pAllocateInfo->pNext, WSI_MEMORY_ALLOCATE_INFO_MESA);
+
        mem = vk_alloc2(&device->alloc, pAllocator, sizeof(*mem), 8,
                          VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
        if (mem == NULL)
                return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 
+       if (wsi_info && wsi_info->implicit_sync)
+               flags |= RADEON_FLAG_IMPLICIT_SYNC;
+
        if (dedicate_info) {
                mem->image = radv_image_from_handle(dedicate_info->image);
                mem->buffer = radv_buffer_from_handle(dedicate_info->buffer);
@@ -2287,7 +2171,9 @@ VkResult radv_AllocateMemory(
 
        if (import_info) {
                assert(import_info->handleType ==
-                      VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR);
+                      VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR ||
+                      import_info->handleType ==
+                      VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT);
                mem->bo = device->ws->buffer_from_fd(device->ws, import_info->fd,
                                                     NULL, NULL);
                if (!mem->bo) {
@@ -2300,20 +2186,23 @@ VkResult radv_AllocateMemory(
        }
 
        uint64_t alloc_size = align_u64(pAllocateInfo->allocationSize, 4096);
-       if (pAllocateInfo->memoryTypeIndex == RADV_MEM_TYPE_GTT_WRITE_COMBINE ||
-           pAllocateInfo->memoryTypeIndex == RADV_MEM_TYPE_GTT_CACHED)
+       if (mem_type_index == RADV_MEM_TYPE_GTT_WRITE_COMBINE ||
+           mem_type_index == RADV_MEM_TYPE_GTT_CACHED)
                domain = RADEON_DOMAIN_GTT;
        else
                domain = RADEON_DOMAIN_VRAM;
 
-       if (pAllocateInfo->memoryTypeIndex == RADV_MEM_TYPE_VRAM)
+       if (mem_type_index == RADV_MEM_TYPE_VRAM)
                flags |= RADEON_FLAG_NO_CPU_ACCESS;
        else
                flags |= RADEON_FLAG_CPU_ACCESS;
 
-       if (pAllocateInfo->memoryTypeIndex == RADV_MEM_TYPE_GTT_WRITE_COMBINE)
+       if (mem_type_index == RADV_MEM_TYPE_GTT_WRITE_COMBINE)
                flags |= RADEON_FLAG_GTT_WC;
 
+       if (!dedicate_info && !import_info)
+               flags |= RADEON_FLAG_NO_INTERPROCESS_SHARING;
+
        mem->bo = device->ws->buffer_create(device->ws, alloc_size, device->physical_device->rad_info.max_alignment,
                                               domain, flags);
 
@@ -2321,7 +2210,7 @@ VkResult radv_AllocateMemory(
                result = VK_ERROR_OUT_OF_DEVICE_MEMORY;
                goto fail;
        }
-       mem->type_index = pAllocateInfo->memoryTypeIndex;
+       mem->type_index = mem_type_index;
 out_success:
        *pMem = radv_device_memory_to_handle(mem);
 
@@ -2333,6 +2222,16 @@ fail:
        return result;
 }
 
+VkResult radv_AllocateMemory(
+       VkDevice                                    _device,
+       const VkMemoryAllocateInfo*                 pAllocateInfo,
+       const VkAllocationCallbacks*                pAllocator,
+       VkDeviceMemory*                             pMem)
+{
+       RADV_FROM_HANDLE(radv_device, device, _device);
+       return radv_alloc_memory(device, pAllocateInfo, pAllocator, pMem);
+}
+
 void radv_FreeMemory(
        VkDevice                                    _device,
        VkDeviceMemory                              _mem,
@@ -2372,7 +2271,7 @@ VkResult radv_MapMemory(
                return VK_SUCCESS;
        }
 
-       return VK_ERROR_MEMORY_MAP_FAILED;
+       return vk_error(VK_ERROR_MEMORY_MAP_FAILED);
 }
 
 void radv_UnmapMemory(
@@ -2405,13 +2304,14 @@ VkResult radv_InvalidateMappedMemoryRanges(
 }
 
 void radv_GetBufferMemoryRequirements(
-       VkDevice                                    device,
+       VkDevice                                    _device,
        VkBuffer                                    _buffer,
        VkMemoryRequirements*                       pMemoryRequirements)
 {
+       RADV_FROM_HANDLE(radv_device, device, _device);
        RADV_FROM_HANDLE(radv_buffer, buffer, _buffer);
 
-       pMemoryRequirements->memoryTypeBits = (1u << RADV_MEM_TYPE_COUNT) - 1;
+       pMemoryRequirements->memoryTypeBits = (1u << device->physical_device->memory_properties.memoryTypeCount) - 1;
 
        if (buffer->flags & VK_BUFFER_CREATE_SPARSE_BINDING_BIT)
                pMemoryRequirements->alignment = 4096;
@@ -2428,13 +2328,13 @@ void radv_GetBufferMemoryRequirements2KHR(
 {
        radv_GetBufferMemoryRequirements(device, pInfo->buffer,
                                         &pMemoryRequirements->memoryRequirements);
-
+       RADV_FROM_HANDLE(radv_buffer, buffer, pInfo->buffer);
        vk_foreach_struct(ext, pMemoryRequirements->pNext) {
                switch (ext->sType) {
                case VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS_KHR: {
                        VkMemoryDedicatedRequirementsKHR *req =
                                       (VkMemoryDedicatedRequirementsKHR *) ext;
-                       req->requiresDedicatedAllocation = false;
+                       req->requiresDedicatedAllocation = buffer->shareable;
                        req->prefersDedicatedAllocation = req->requiresDedicatedAllocation;
                        break;
                }
@@ -2445,13 +2345,14 @@ void radv_GetBufferMemoryRequirements2KHR(
 }
 
 void radv_GetImageMemoryRequirements(
-       VkDevice                                    device,
+       VkDevice                                    _device,
        VkImage                                     _image,
        VkMemoryRequirements*                       pMemoryRequirements)
 {
+       RADV_FROM_HANDLE(radv_device, device, _device);
        RADV_FROM_HANDLE(radv_image, image, _image);
 
-       pMemoryRequirements->memoryTypeBits = (1u << RADV_MEM_TYPE_COUNT) - 1;
+       pMemoryRequirements->memoryTypeBits = (1u << device->physical_device->memory_properties.memoryTypeCount) - 1;
 
        pMemoryRequirements->size = image->size;
        pMemoryRequirements->alignment = image->alignment;
@@ -2685,15 +2586,14 @@ VkResult radv_CreateFence(
                                               VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 
        if (!fence)
-               return VK_ERROR_OUT_OF_HOST_MEMORY;
+               return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 
-       memset(fence, 0, sizeof(*fence));
        fence->submitted = false;
        fence->signalled = !!(pCreateInfo->flags & VK_FENCE_CREATE_SIGNALED_BIT);
        fence->fence = device->ws->create_fence();
        if (!fence->fence) {
                vk_free2(&device->alloc, pAllocator, fence);
-               return VK_ERROR_OUT_OF_HOST_MEMORY;
+               return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
        }
 
        *pFence = radv_fence_to_handle(fence);
@@ -2809,7 +2709,7 @@ VkResult radv_CreateSemaphore(
                                               sizeof(*sem), 8,
                                               VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
        if (!sem)
-               return VK_ERROR_OUT_OF_HOST_MEMORY;
+               return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 
        sem->temp_syncobj = 0;
        /* create a syncobject if we are going to export this semaphore */
@@ -2819,14 +2719,14 @@ VkResult radv_CreateSemaphore(
                int ret = device->ws->create_syncobj(device->ws, &sem->syncobj);
                if (ret) {
                        vk_free2(&device->alloc, pAllocator, sem);
-                       return VK_ERROR_OUT_OF_HOST_MEMORY;
+                       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
                }
                sem->sem = NULL;
        } else {
                sem->sem = device->ws->create_sem(device->ws);
                if (!sem->sem) {
                        vk_free2(&device->alloc, pAllocator, sem);
-                       return VK_ERROR_OUT_OF_HOST_MEMORY;
+                       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
                }
                sem->syncobj = 0;
        }
@@ -2864,14 +2764,14 @@ VkResult radv_CreateEvent(
                                               VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 
        if (!event)
-               return VK_ERROR_OUT_OF_HOST_MEMORY;
+               return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 
        event->bo = device->ws->buffer_create(device->ws, 8, 8,
                                              RADEON_DOMAIN_GTT,
-                                             RADEON_FLAG_VA_UNCACHED | RADEON_FLAG_CPU_ACCESS);
+                                             RADEON_FLAG_VA_UNCACHED | RADEON_FLAG_CPU_ACCESS | RADEON_FLAG_NO_INTERPROCESS_SHARING);
        if (!event->bo) {
                vk_free2(&device->alloc, pAllocator, event);
-               return VK_ERROR_OUT_OF_DEVICE_MEMORY;
+               return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
        }
 
        event->map = (uint64_t*)device->ws->buffer_map(event->bo);
@@ -2948,6 +2848,9 @@ VkResult radv_CreateBuffer(
        buffer->offset = 0;
        buffer->flags = pCreateInfo->flags;
 
+       buffer->shareable = vk_find_struct_const(pCreateInfo->pNext,
+                                                EXTERNAL_MEMORY_BUFFER_CREATE_INFO_KHR) != NULL;
+
        if (pCreateInfo->flags & VK_BUFFER_CREATE_SPARSE_BINDING_BIT) {
                buffer->bo = device->ws->buffer_create(device->ws,
                                                       align64(buffer->size, 4096),
@@ -3047,7 +2950,6 @@ radv_initialise_color_surface(struct radv_device *device,
                cb->cb_color_cmask_slice = iview->image->cmask.slice_tile_max;
 
                cb->cb_color_attrib |= S_028C74_TILE_MODE_INDEX(tile_mode_index);
-               cb->micro_tile_mode = iview->image->surface.micro_tile_mode;
 
                if (iview->image->fmask.size) {
                        if (device->physical_device->rad_info.chip_class >= CIK)
@@ -3143,7 +3045,7 @@ radv_initialise_color_surface(struct radv_device *device,
        }
 
        if (iview->image->cmask.size &&
-           !(device->debug_flags & RADV_DEBUG_NO_FAST_CLEARS))
+           !(device->instance->debug_flags & RADV_DEBUG_NO_FAST_CLEARS))
                cb->cb_color_info |= S_028C70_FAST_CLEAR(1);
 
        if (radv_vi_dcc_enabled(iview->image, iview->base_mip))
@@ -3179,9 +3081,6 @@ radv_initialise_color_surface(struct radv_device *device,
                cb->cb_color_attrib2 = S_028C68_MIP0_WIDTH(iview->extent.width - 1) |
                        S_028C68_MIP0_HEIGHT(iview->extent.height - 1) |
                        S_028C68_MAX_MIP(iview->image->info.levels - 1);
-
-               cb->gfx9_epitch = S_0287A0_EPITCH(iview->image->surface.u.gfx9.surf.epitch);
-
        }
 }
 
@@ -3251,8 +3150,7 @@ radv_initialise_ds_surface(struct radv_device *device,
                ds->db_depth_size = S_02801C_X_MAX(iview->image->info.width - 1) |
                        S_02801C_Y_MAX(iview->image->info.height - 1);
 
-               /* Only use HTILE for the first level. */
-               if (iview->image->surface.htile_size && !level) {
+               if (radv_htile_enabled(iview->image, level)) {
                        ds->db_z_info |= S_028038_TILE_SURFACE_ENABLE(1);
 
                        if (iview->image->tc_compatible_htile) {
@@ -3327,7 +3225,7 @@ radv_initialise_ds_surface(struct radv_device *device,
                        S_028058_HEIGHT_TILE_MAX((level_info->nblk_y / 8) - 1);
                ds->db_depth_slice = S_02805C_SLICE_TILE_MAX((level_info->nblk_x * level_info->nblk_y) / 64 - 1);
 
-               if (iview->image->surface.htile_size && !level) {
+               if (radv_htile_enabled(iview->image, level)) {
                        ds->db_z_info |= S_028040_TILE_SURFACE_ENABLE(1);
 
                        if (!iview->image->surface.has_stencil &&
@@ -3546,7 +3444,7 @@ radv_init_sampler(struct radv_device *device,
                             S_008F38_XY_MIN_FILTER(radv_tex_filter(pCreateInfo->minFilter, max_aniso)) |
                             S_008F38_MIP_FILTER(radv_tex_mipfilter(pCreateInfo->mipmapMode)) |
                             S_008F38_MIP_POINT_PRECLAMP(0) |
-                            S_008F38_DISABLE_LSB_CEIL(1) |
+                            S_008F38_DISABLE_LSB_CEIL(device->physical_device->rad_info.chip_class <= VI) |
                             S_008F38_FILTER_PREC_FIX(1) |
                             S_008F38_ANISO_OVERRIDE(is_vi));
        sampler->state[3] = (S_008F3C_BORDER_COLOR_PTR(0) |
@@ -3641,13 +3539,15 @@ VkResult radv_GetMemoryFdKHR(VkDevice _device,
 
        assert(pGetFdInfo->sType == VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR);
 
-       /* We support only one handle type. */
+       /* At the moment, we support only the below handle types. */
        assert(pGetFdInfo->handleType ==
-              VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR);
+              VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR ||
+              pGetFdInfo->handleType ==
+              VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT);
 
        bool ret = radv_get_memory_fd(device, memory, pFD);
        if (ret == false)
-               return VK_ERROR_OUT_OF_DEVICE_MEMORY;
+               return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
        return VK_SUCCESS;
 }
 
@@ -3656,13 +3556,21 @@ VkResult radv_GetMemoryFdPropertiesKHR(VkDevice _device,
                                       int fd,
                                       VkMemoryFdPropertiesKHR *pMemoryFdProperties)
 {
-   /* The valid usage section for this function says:
-    *
-    *    "handleType must not be one of the handle types defined as opaque."
-    *
-    * Since we only handle opaque handles for now, there are no FD properties.
-    */
-   return VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR;
+   switch (handleType) {
+   case VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR:
+      pMemoryFdProperties->memoryTypeBits = (1 << RADV_MEM_TYPE_COUNT) - 1;
+      return VK_SUCCESS;
+
+   default:
+      /* The valid usage section for this function says:
+       *
+       *    "handleType must not be one of the handle types defined as
+       *    opaque."
+       *
+       * So opaque handle types fall into the default "unsupported" case.
+       */
+      return vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR);
+   }
 }
 
 VkResult radv_ImportSemaphoreFdKHR(VkDevice _device,
@@ -3671,17 +3579,23 @@ VkResult radv_ImportSemaphoreFdKHR(VkDevice _device,
        RADV_FROM_HANDLE(radv_device, device, _device);
        RADV_FROM_HANDLE(radv_semaphore, sem, pImportSemaphoreFdInfo->semaphore);
        uint32_t syncobj_handle = 0;
+       uint32_t *syncobj_dst = NULL;
        assert(pImportSemaphoreFdInfo->handleType == VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR);
 
        int ret = device->ws->import_syncobj(device->ws, pImportSemaphoreFdInfo->fd, &syncobj_handle);
        if (ret != 0)
-               return VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR;
+               return vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR);
 
        if (pImportSemaphoreFdInfo->flags & VK_SEMAPHORE_IMPORT_TEMPORARY_BIT_KHR) {
-               sem->temp_syncobj = syncobj_handle;
+               syncobj_dst = &sem->temp_syncobj;
        } else {
-               sem->syncobj = syncobj_handle;
+               syncobj_dst = &sem->syncobj;
        }
+
+       if (*syncobj_dst)
+               device->ws->destroy_syncobj(device->ws, *syncobj_dst);
+
+       *syncobj_dst = syncobj_handle;
        close(pImportSemaphoreFdInfo->fd);
        return VK_SUCCESS;
 }