Edit
IABSD.fr/xenocara/lib/mesa/src/vulkan/runtime/vk_device.h
jsg
- lib/mesa/src/vulkan/runtime/vk_device.h
-
/* * Copyright © 2020 Intel Corporation * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), * to deal in the Software without restriction, including without limitation * the rights to use, copy, modify, merge, publish, distribute, sublicense, * and/or sell copies of the Software, and to permit persons to whom the * Software is furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice (including the next * paragraph) shall be included in all copies or substantial portions of the * Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS * IN THE SOFTWARE. */ #ifndef VK_DEVICE_H #define VK_DEVICE_H #include "rmv/vk_rmv_common.h" #include "vk_dispatch_table.h" #include "vk_extensions.h" #include "vk_object.h" #include "vk_physical_device_features.h" #include "util/list.h" #include "util/simple_mtx.h" #include "util/u_atomic.h" #ifdef __cplusplus extern "C" { #endif struct vk_acceleration_structure_build_ops; struct vk_command_buffer_ops; struct vk_device_shader_ops; struct vk_sync; enum vk_queue_submit_mode { /** Submits happen immediately * * `vkQueueSubmit()` and `vkQueueBindSparse()` call * ``vk_queue::driver_submit`` directly for all submits and the last call to * ``vk_queue::driver_submit`` will have completed by the time * `vkQueueSubmit()` or `vkQueueBindSparse()` return. */ VK_QUEUE_SUBMIT_MODE_IMMEDIATE, /** Submits may be deferred until a future `vk_queue_flush()` * * Submits are added to the queue and `vk_queue_flush()` is called. * However, any submits with unsatisfied dependencies will be left on the * queue until a future `vk_queue_flush()` call. This is used for * implementing emulated timeline semaphores without threading. */ VK_QUEUE_SUBMIT_MODE_DEFERRED, /** Submits will be added to the queue and handled later by a thread * * This places additional requirements on the vk_sync types used by the * driver: * * 1. All `vk_sync` types which support `VK_SYNC_FEATURE_GPU_WAIT` also * support `VK_SYNC_FEATURE_WAIT_PENDING` so that the threads can * sort out when a given submit has all its dependencies resolved. * * 2. All binary `vk_sync` types which support `VK_SYNC_FEATURE_GPU_WAIT` * also support `VK_SYNC_FEATURE_CPU_RESET` so we can reset * semaphores after waiting on them. * * 3. All vk_sync types used as permanent payloads of semaphores support * ``vk_sync_type::move`` so that it can move the pending signal into a * temporary vk_sync and reset the semaphore. * * This is requied for shared timeline semaphores where we need to handle * wait-before-signal by threading in the driver if we ever see an * unresolve dependency. */ VK_QUEUE_SUBMIT_MODE_THREADED, /** Threaded but only if we need it to resolve dependencies * * This imposes all the same requirements on `vk_sync` types as * `VK_QUEUE_SUBMIT_MODE_THREADED`. */ VK_QUEUE_SUBMIT_MODE_THREADED_ON_DEMAND, }; /** Base struct for VkDevice */ struct vk_device { struct vk_object_base base; /** Allocator used to create this device * * This is used as a fall-back for when a NULL pAllocator is passed into a * device-level create function such as vkCreateImage(). */ VkAllocationCallbacks alloc; /** Pointer to the physical device */ struct vk_physical_device *physical; /** Table of enabled extensions */ struct vk_device_extension_table enabled_extensions; /** Table of enabled features */ struct vk_features enabled_features; /** Device-level dispatch table */ struct vk_device_dispatch_table dispatch_table; /** Command dispatch table * * This is used for emulated secondary command buffer support. To use * emulated (trace/replay) secondary command buffers: * * 1. Provide your "real" command buffer dispatch table here. Because * this doesn't get populated by vk_device_init(), the driver will have * to add the vk_common entrypoints to this table itself. * * 2. Add vk_enqueue_unless_primary_device_entrypoint_table to your device * level dispatch table. */ const struct vk_device_dispatch_table *command_dispatch_table; /** Command buffer vtable when using the common command pool */ const struct vk_command_buffer_ops *command_buffer_ops; /** Shader vtable for VK_EXT_shader_object and common pipelines */ const struct vk_device_shader_ops *shader_ops; /** Acceleration structure build vtable for common BVH building. */ const struct vk_acceleration_structure_build_ops *as_build_ops; /** * Write data to a buffer from the command processor. This is simpler than * setting up a staging buffer and faster for small writes, but is not * meant for larger amounts of data. \p data is owned by the caller and the * driver is expected to write it out directly to the command stream as * part of an immediate write packet. */ void (*write_buffer_cp)(VkCommandBuffer cmdbuf, VkDeviceAddress addr, void *data, uint32_t size); /* Flush data written via write_buffer_cp. Users must use a normal pipeline * barrier in order to read this data, with the appropriate destination * access, but this replaces the source access mask. */ void (*flush_buffer_write_cp)(VkCommandBuffer cmdbuf); /* An unaligned dispatch function. This launches a number of threads that * may not be a multiple of the workgroup size, which may result in partial * workgroups. */ void (*cmd_dispatch_unaligned)(VkCommandBuffer cmdbuf, uint32_t invocations_x, uint32_t invocations_y, uint32_t invocations_z); /* vkCmdFillBuffer but with a device address. */ void (*cmd_fill_buffer_addr)(VkCommandBuffer cmdbuf, VkDeviceAddress devAddr, VkDeviceSize size, uint32_t data); /** Driver provided callback for capturing traces * * Triggers for this callback are: * - Keyboard input (F12) * - Creation of a trigger file * - Reaching the trace frame */ VkResult (*capture_trace)(VkQueue queue); uint32_t current_frame; bool trace_hotkey_trigger; simple_mtx_t trace_mtx; /* For VK_EXT_private_data */ uint32_t private_data_next_index; struct list_head queues; struct { int lost; bool reported; } _lost; /** Checks the status of this device * * This is expected to return either VK_SUCCESS or VK_ERROR_DEVICE_LOST. * It is called before ``vk_queue::driver_submit`` and after every non-trivial * wait operation to ensure the device is still around. This gives the * driver a hook to ask the kernel if its device is still valid. If the * kernel says the device has been lost, it MUST call vk_device_set_lost(). * * This function may be called from any thread at any time. */ VkResult (*check_status)(struct vk_device *device); /* Get the device timestamp in the VK_TIME_DOMAIN_DEVICE_KHR domain */ VkResult (*get_timestamp)(struct vk_device *device, uint64_t *timestamp); /** Host time domain used for timestamp calibration */ VkTimeDomainKHR calibrate_time_domain; /** Period of VK_TIME_DOMAIN_DEVICE_KHR */ uint64_t device_time_domain_period; /** Creates a vk_sync that wraps a memory object * * This is always a one-shot object so it need not track any additional * state. Since it's intended for synchronizing between processes using * implicit synchronization mechanisms, no such tracking would be valid * anyway. * * If `signal_memory` is set, the resulting vk_sync will be used to signal * the memory object from a queue ``via vk_queue_submit::signals``. The common * code guarantees that, by the time vkQueueSubmit() returns, the signal * operation has been submitted to the kernel via the driver's * ``vk_queue::driver_submit`` hook. This means that any vkQueueSubmit() call * which needs implicit synchronization may block. * * If `signal_memory` is not set, it can be assumed that memory object * already has a signal operation pending from some other process and we * need only wait on it. */ VkResult (*create_sync_for_memory)(struct vk_device *device, VkDeviceMemory memory, bool signal_memory, struct vk_sync **sync_out); /* Set by vk_device_set_drm_fd() */ int drm_fd; /** Implicit pipeline cache, or NULL */ struct vk_pipeline_cache *mem_cache; /** An enum describing how timeline semaphores work */ enum vk_device_timeline_mode { /** Timeline semaphores are not supported */ VK_DEVICE_TIMELINE_MODE_NONE, /** Timeline semaphores are emulated with vk_timeline * * In this mode, timeline semaphores are emulated using vk_timeline * which is a collection of binary semaphores, one per time point. * These timeline semaphores cannot be shared because the data structure * exists entirely in userspace. These timelines are virtually * invisible to the driver; all it sees are the binary vk_syncs, one per * time point. * * To handle wait-before-signal, we place all vk_queue_submits in the * queue's submit list in vkQueueSubmit() and call vk_device_flush() at * key points such as the end of vkQueueSubmit() and vkSemaphoreSignal(). * This ensures that, as soon as a given submit's dependencies are fully * resolvable, it gets submitted to the driver. */ VK_DEVICE_TIMELINE_MODE_EMULATED, /** Timeline semaphores are a kernel-assisted emulation * * In this mode, timeline semaphores are still technically an emulation * in the sense that they don't support wait-before-signal natively. * Instead, all GPU-waitable objects support a CPU wait-for-pending * operation which lets the userspace driver wait until a given event * on the (possibly shared) vk_sync is pending. The event is "pending" * if a job has been submitted to the kernel (possibly from a different * process) which will signal it. In vkQueueSubit, we use this wait * mode to detect waits which are not yet pending and, the first time we * do, spawn a thread to manage the queue. That thread waits for each * submit's waits to all be pending before submitting to the driver * queue. * * We have to be a bit more careful about a few things in this mode. * In particular, we can never assume that any given wait operation is * pending. For instance, when we go to export a sync file from a * binary semaphore, we need to first wait for it to be pending. The * spec guarantees that the vast majority of these waits return almost * immediately, but we do need to insert them for correctness. */ VK_DEVICE_TIMELINE_MODE_ASSISTED, /** Timeline semaphores are 100% native * * In this mode, wait-before-signal is natively supported by the * underlying timeline implementation. We can submit-and-forget and * assume that dependencies will get resolved for us by the kernel. * Currently, this isn't supported by any Linux primitives. */ VK_DEVICE_TIMELINE_MODE_NATIVE, } timeline_mode; /** Per-device submit mode * * This represents the device-wide submit strategy which may be different * from the per-queue submit mode. See vk_queue.submit.mode for more * details. */ enum vk_queue_submit_mode submit_mode; struct vk_memory_trace_data memory_trace_data; mtx_t swapchain_private_mtx; struct hash_table *swapchain_private; mtx_t swapchain_name_mtx; struct hash_table *swapchain_name; /* For VK_KHR_pipeline_binary */ bool disable_internal_cache; }; VK_DEFINE_HANDLE_CASTS(vk_device, base, VkDevice, VK_OBJECT_TYPE_DEVICE); /** Initialize a vk_device * * Along with initializing the data structures in `vk_device`, this function * checks that every extension specified by * ``VkInstanceCreateInfo::ppEnabledExtensionNames`` is actually supported by * the physical device and returns `VK_ERROR_EXTENSION_NOT_PRESENT` if an * unsupported extension is requested. It also checks all the feature struct * chained into the `pCreateInfo->pNext` chain against the features returned * by `vkGetPhysicalDeviceFeatures2` and returns * `VK_ERROR_FEATURE_NOT_PRESENT` if an unsupported feature is requested. * * :param device: |out| The device to initialize * :param physical_device: |in| The physical device * :param dispatch_table: |in| Device-level dispatch table * :param pCreateInfo: |in| VkDeviceCreateInfo pointer passed to * `vkCreateDevice()` * :param alloc: |in| Allocation callbacks passed to * `vkCreateDevice()` */ VkResult MUST_CHECK vk_device_init(struct vk_device *device, struct vk_physical_device *physical_device, const struct vk_device_dispatch_table *dispatch_table, const VkDeviceCreateInfo *pCreateInfo, const VkAllocationCallbacks *alloc); static inline void vk_device_set_drm_fd(struct vk_device *device, int drm_fd) { device->drm_fd = drm_fd; } /** Tears down a vk_device * * :param device: |out| The device to tear down */ void vk_device_finish(struct vk_device *device); /** Enables threaded submit on this device * * This doesn't ensure that threaded submit will be used. It just disables * the deferred submit option for emulated timeline semaphores and forces them * to always use the threaded path. It also does some checks that the vk_sync * types used by the driver work for threaded submit. * * This must be called before any queues are created. */ void vk_device_enable_threaded_submit(struct vk_device *device); static inline bool vk_device_supports_threaded_submit(const struct vk_device *device) { return device->submit_mode == VK_QUEUE_SUBMIT_MODE_THREADED || device->submit_mode == VK_QUEUE_SUBMIT_MODE_THREADED_ON_DEMAND; } VkResult vk_device_flush(struct vk_device *device); VkResult PRINTFLIKE(4, 5) _vk_device_set_lost(struct vk_device *device, const char *file, int line, const char *msg, ...); #define vk_device_set_lost(device, ...) \ _vk_device_set_lost(device, __FILE__, __LINE__, __VA_ARGS__) void _vk_device_report_lost(struct vk_device *device); static inline bool vk_device_is_lost_no_report(struct vk_device *device) { return p_atomic_read(&device->_lost.lost) > 0; } static inline bool vk_device_is_lost(struct vk_device *device) { int lost = vk_device_is_lost_no_report(device); if (unlikely(lost && !device->_lost.reported)) _vk_device_report_lost(device); return lost; } static inline VkResult vk_device_check_status(struct vk_device *device) { if (vk_device_is_lost(device)) return VK_ERROR_DEVICE_LOST; if (!device->check_status) return VK_SUCCESS; VkResult result = device->check_status(device); assert(result == VK_SUCCESS || result == VK_ERROR_DEVICE_LOST); if (result == VK_ERROR_DEVICE_LOST) assert(vk_device_is_lost_no_report(device)); return result; } VkResult vk_device_get_timestamp(struct vk_device *device, VkTimeDomainKHR domain, uint64_t *timestamp); #ifndef _WIN32 uint64_t vk_clock_gettime(clockid_t clock_id); #endif //!_WIN32 static inline uint64_t vk_time_max_deviation(uint64_t begin, uint64_t end, uint64_t max_clock_period) { /* * The maximum deviation is the sum of the interval over which we * perform the sampling and the maximum period of any sampled * clock. That's because the maximum skew between any two sampled * clock edges is when the sampled clock with the largest period is * sampled at the end of that period but right at the beginning of the * sampling interval and some other clock is sampled right at the * beginning of its sampling period and right at the end of the * sampling interval. Let's assume the GPU has the longest clock * period and that the application is sampling GPU and monotonic: * * s e * w x y z 0 1 2 3 4 5 6 7 8 9 a b c d e f * Raw -_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_- * * g * 0 1 2 3 * GPU -----_____-----_____-----_____-----_____ * * m * x y z 0 1 2 3 4 5 6 7 8 9 a b c * Monotonic -_-_-_-_-_-_-_-_-_-_-_-_-_-_-_- * * Interval <-----------------> * Deviation <--------------------------> * * s = read(raw) 2 * g = read(GPU) 1 * m = read(monotonic) 2 * e = read(raw) b * * We round the sample interval up by one tick to cover sampling error * in the interval clock */ uint64_t sample_interval = end - begin + 1; return sample_interval + max_clock_period; } PFN_vkVoidFunction vk_device_get_proc_addr(const struct vk_device *device, const char *name); #ifdef __cplusplus } #endif #endif /* VK_DEVICE_H */