radv: implement VK_EXT_external_memory_host

Ported from the radeonsi GL_AMD_pinned_memory implementation.

Signed-off-by: Fredrik Höglund <fredrik@kde.org>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

src/amd/vulkan/radv_device.c

@@ -810,6 +810,12 @@ void radv_GetPhysicalDeviceProperties2KHR(
 			properties->maxDiscardRectangles = MAX_DISCARD_RECTANGLES;
 			break;
 		}
+		case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_MEMORY_HOST_PROPERTIES_EXT: {
+			VkPhysicalDeviceExternalMemoryHostPropertiesEXT *properties =
+			    (VkPhysicalDeviceExternalMemoryHostPropertiesEXT *) ext;
+			properties->minImportedHostPointerAlignment = 4096;
+			break;
+		}
 		default:
 			break;
 		}
@@ -923,6 +929,33 @@ void radv_GetPhysicalDeviceMemoryProperties2KHR(
 						      &pMemoryProperties->memoryProperties);
 }
 
+VkResult radv_GetMemoryHostPointerPropertiesEXT(
+	VkDevice                                    _device,
+	VkExternalMemoryHandleTypeFlagBitsKHR       handleType,
+	const void                                 *pHostPointer,
+	VkMemoryHostPointerPropertiesEXT           *pMemoryHostPointerProperties)
+{
+	RADV_FROM_HANDLE(radv_device, device, _device);
+
+	switch (handleType)
+	{
+	case VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT: {
+		const struct radv_physical_device *physical_device = device->physical_device;
+		uint32_t memoryTypeBits = 0;
+		for (int i = 0; i < physical_device->memory_properties.memoryTypeCount; i++) {
+			if (physical_device->mem_type_indices[i] == RADV_MEM_TYPE_GTT_CACHED) {
+				memoryTypeBits = (1 << i);
+				break;
+			}
+		}
+		pMemoryHostPointerProperties->memoryTypeBits = memoryTypeBits;
+		return VK_SUCCESS;
+	}
+	default:
+		return VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR;
+	}
+}
+
 static enum radeon_ctx_priority
 radv_get_queue_global_priority(const VkDeviceQueueGlobalPriorityCreateInfoEXT *pObj)
 {
@@ -2246,6 +2279,8 @@ static VkResult radv_alloc_memory(struct radv_device *device,
 		vk_find_struct_const(pAllocateInfo->pNext, MEMORY_DEDICATED_ALLOCATE_INFO_KHR);
 	const VkExportMemoryAllocateInfoKHR *export_info =
 		vk_find_struct_const(pAllocateInfo->pNext, EXPORT_MEMORY_ALLOCATE_INFO_KHR);
+	const VkImportMemoryHostPointerInfoEXT *host_ptr_info =
+		vk_find_struct_const(pAllocateInfo->pNext, IMPORT_MEMORY_HOST_POINTER_INFO_EXT);
 
 	const struct wsi_memory_allocate_info *wsi_info =
 		vk_find_struct_const(pAllocateInfo->pNext, WSI_MEMORY_ALLOCATE_INFO_MESA);
@@ -2266,6 +2301,8 @@ static VkResult radv_alloc_memory(struct radv_device *device,
 		mem->buffer = NULL;
 	}
 
+	mem->user_ptr = NULL;
+
 	if (import_info) {
 		assert(import_info->handleType ==
 		       VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR ||
@@ -2282,6 +2319,20 @@ static VkResult radv_alloc_memory(struct radv_device *device,
 		}
 	}
 
+	if (host_ptr_info) {
+		assert(host_ptr_info->handleType == VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT);
+		assert(mem_type_index == RADV_MEM_TYPE_GTT_CACHED);
+		mem->bo = device->ws->buffer_from_ptr(device->ws, host_ptr_info->pHostPointer,
+		                                      pAllocateInfo->allocationSize);
+		if (!mem->bo) {
+			result = VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR;
+			goto fail;
+		} else {
+			mem->user_ptr = host_ptr_info->pHostPointer;
+			goto out_success;
+		}
+	}
+
 	uint64_t alloc_size = align_u64(pAllocateInfo->allocationSize, 4096);
 	if (mem_type_index == RADV_MEM_TYPE_GTT_WRITE_COMBINE ||
 	    mem_type_index == RADV_MEM_TYPE_GTT_CACHED)
@@ -2362,7 +2413,11 @@ VkResult radv_MapMemory(
 		return VK_SUCCESS;
 	}
 
-	*ppData = device->ws->buffer_map(mem->bo);
+	if (mem->user_ptr)
+		*ppData = mem->user_ptr;
+	else
+		*ppData = device->ws->buffer_map(mem->bo);
+
 	if (*ppData) {
 		*ppData += offset;
 		return VK_SUCCESS;
@@ -2381,7 +2436,8 @@ void radv_UnmapMemory(
 	if (mem == NULL)
 		return;
 
-	device->ws->buffer_unmap(mem->bo);
+	if (mem->user_ptr == NULL)
+		device->ws->buffer_unmap(mem->bo);
 }
 
 VkResult radv_FlushMappedMemoryRanges(

src/amd/vulkan/radv_extensions.py

@@ -85,6 +85,7 @@ EXTENSIONS = [
     Extension('VK_EXT_debug_report',                      9, True),
     Extension('VK_EXT_discard_rectangles',                1, True),
     Extension('VK_EXT_external_memory_dma_buf',           1, True),
+    Extension('VK_EXT_external_memory_host',              1, 'device->rad_info.has_userptr'),
     Extension('VK_EXT_global_priority',                   1, 'device->rad_info.has_ctx_priority'),
     Extension('VK_AMD_draw_indirect_count',               1, True),
     Extension('VK_AMD_rasterization_order',               1, 'device->rad_info.chip_class >= VI && device->rad_info.max_se >= 2'),

src/amd/vulkan/radv_formats.c

@@ -1177,16 +1177,28 @@ VkResult radv_GetPhysicalDeviceImageFormatProperties(
 
 static void
 get_external_image_format_properties(const VkPhysicalDeviceImageFormatInfo2KHR *pImageFormatInfo,
+				     VkExternalMemoryHandleTypeFlagBitsKHR handleType,
 				     VkExternalMemoryPropertiesKHR *external_properties)
 {
 	VkExternalMemoryFeatureFlagBitsKHR flags = 0;
 	VkExternalMemoryHandleTypeFlagsKHR export_flags = 0;
 	VkExternalMemoryHandleTypeFlagsKHR compat_flags = 0;
-	switch (pImageFormatInfo->type) {
+	switch (handleType) {
-	case VK_IMAGE_TYPE_2D:
+	case VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR:
-		flags = VK_EXTERNAL_MEMORY_FEATURE_DEDICATED_ONLY_BIT_KHR|VK_EXTERNAL_MEMORY_FEATURE_EXPORTABLE_BIT_KHR|VK_EXTERNAL_MEMORY_FEATURE_IMPORTABLE_BIT_KHR;
+	case VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT:
-		compat_flags = export_flags = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR |
+		switch (pImageFormatInfo->type) {
-					      VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT;
+		case VK_IMAGE_TYPE_2D:
+			flags = VK_EXTERNAL_MEMORY_FEATURE_DEDICATED_ONLY_BIT_KHR|VK_EXTERNAL_MEMORY_FEATURE_EXPORTABLE_BIT_KHR|VK_EXTERNAL_MEMORY_FEATURE_IMPORTABLE_BIT_KHR;
+			compat_flags = export_flags = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR |
+						      VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT;
+			break;
+		default:
+			break;
+		}
+		break;
+	case VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT:
+		flags = VK_EXTERNAL_MEMORY_FEATURE_IMPORTABLE_BIT_KHR;
+		compat_flags = VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT;
 		break;
 	default:
 		break;
@@ -1246,7 +1258,9 @@ VkResult radv_GetPhysicalDeviceImageFormatProperties2KHR(
 		switch (external_info->handleType) {
 		case VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR:
 		case VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT:
-			get_external_image_format_properties(base_info, &external_props->externalMemoryProperties);
+		case VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT:
+			get_external_image_format_properties(base_info, external_info->handleType,
+			                                     &external_props->externalMemoryProperties);
 			break;
 		default:
 			/* From the Vulkan 1.0.42 spec:
@@ -1320,6 +1334,10 @@ void radv_GetPhysicalDeviceExternalBufferPropertiesKHR(
 		compat_flags = export_flags = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR |
 					      VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT;
 		break;
+	case VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT:
+		flags = VK_EXTERNAL_MEMORY_FEATURE_IMPORTABLE_BIT_KHR;
+		compat_flags = VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT;
+		break;
 	default:
 		break;
 	}

src/amd/vulkan/radv_private.h

@@ -646,6 +646,7 @@ struct radv_device_memory {
 	uint32_t                                     type_index;
 	VkDeviceSize                                 map_size;
 	void *                                       map;
+	void *                                       user_ptr;
 };
 
 

src/amd/vulkan/radv_radeon_winsys.h

@@ -200,6 +200,10 @@ struct radeon_winsys {
 	void (*buffer_destroy)(struct radeon_winsys_bo *bo);
 	void *(*buffer_map)(struct radeon_winsys_bo *bo);
 
+	struct radeon_winsys_bo *(*buffer_from_ptr)(struct radeon_winsys *ws,
+						    void *pointer,
+						    uint64_t size);
+
 	struct radeon_winsys_bo *(*buffer_from_fd)(struct radeon_winsys *ws,
 						   int fd,
 						   unsigned *stride, unsigned *offset);

src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.c

@@ -402,6 +402,54 @@ radv_amdgpu_winsys_bo_unmap(struct radeon_winsys_bo *_bo)
 	amdgpu_bo_cpu_unmap(bo->bo);
 }
 
+static struct radeon_winsys_bo *
+radv_amdgpu_winsys_bo_from_ptr(struct radeon_winsys *_ws,
+                               void *pointer,
+                               uint64_t size)
+{
+	struct radv_amdgpu_winsys *ws = radv_amdgpu_winsys(_ws);
+	amdgpu_bo_handle buf_handle;
+	struct radv_amdgpu_winsys_bo *bo;
+	uint64_t va;
+	amdgpu_va_handle va_handle;
+
+	bo = CALLOC_STRUCT(radv_amdgpu_winsys_bo);
+	if (!bo)
+		return NULL;
+
+	if (amdgpu_create_bo_from_user_mem(ws->dev, pointer, size, &buf_handle))
+		goto error;
+
+	if (amdgpu_va_range_alloc(ws->dev, amdgpu_gpu_va_range_general,
+	                          size, 1 << 12, 0, &va, &va_handle, 0))
+		goto error_va_alloc;
+
+	if (amdgpu_bo_va_op(buf_handle, 0, size, va, 0, AMDGPU_VA_OP_MAP))
+		goto error_va_map;
+
+	/* Initialize it */
+	bo->base.va = va;
+	bo->va_handle = va_handle;
+	bo->size = size;
+	bo->ref_count = 1;
+	bo->ws = ws;
+	bo->bo = buf_handle;
+	bo->initial_domain = RADEON_DOMAIN_GTT;
+
+	radv_amdgpu_add_buffer_to_global_list(bo);
+	return (struct radeon_winsys_bo *)bo;
+
+error_va_map:
+	amdgpu_va_range_free(va_handle);
+
+error_va_alloc:
+	amdgpu_bo_free(buf_handle);
+
+error:
+	FREE(bo);
+	return NULL;
+}
+
 static struct radeon_winsys_bo *
 radv_amdgpu_winsys_bo_from_fd(struct radeon_winsys *_ws,
 			      int fd, unsigned *stride,
@@ -540,6 +588,7 @@ void radv_amdgpu_bo_init_functions(struct radv_amdgpu_winsys *ws)
 	ws->base.buffer_destroy = radv_amdgpu_winsys_bo_destroy;
 	ws->base.buffer_map = radv_amdgpu_winsys_bo_map;
 	ws->base.buffer_unmap = radv_amdgpu_winsys_bo_unmap;
+	ws->base.buffer_from_ptr = radv_amdgpu_winsys_bo_from_ptr;
 	ws->base.buffer_from_fd = radv_amdgpu_winsys_bo_from_fd;
 	ws->base.buffer_get_fd = radv_amdgpu_winsys_get_fd;
 	ws->base.buffer_set_metadata = radv_amdgpu_winsys_bo_set_metadata;