import dataclasses
import os
from contextlib import contextmanager
from typing import Any, Callable, Optional, Union
import time

import torch
from vllm.logger import logger
import vllm_kunlun.platforms.envs as xenvs


def find_loaded_library(lib_name) -> Optional[str]:
    """
    According to according to https://man7.org/linux/man-pages/man5/proc_pid_maps.5.html,
    the file `/proc/self/maps` contains the memory maps of the process, which includes the
    shared libraries loaded by the process. We can use this file to find the path of the
    a loaded library.
    """ # noqa
    found_line = None
    with open("/proc/self/maps") as f:
        for line in f:
            if lib_name in line:
                found_line = line
                break
    if found_line is None:
        # the library is not loaded in the current process
        return None
    # if lib_name is libcudart, we need to match a line with:
    # address /path/to/libcudart-hash.so.11.0
    start = found_line.index("/")
    path = found_line[start:].strip()
    filename = path.split("/")[-1]
    assert filename.rpartition(".so")[0].startswith(lib_name), \
        f"Unexpected filename: {filename} for library {lib_name}"
    return path


xpumem_available = False
try:
    if xenvs.VLLM_KUNLUN_ENABLE_VXPU:
        from vllm_kunlun._kunlun_vxpu import (
            init_module,
            create_and_map as py_create_and_map,
            unmap_and_release as py_unmap_and_release,
            my_xpu_memcpy as xpu_memcpy,
            get_mem_info,
            try_lock_gpu,
            unlock_gpu,
        )

        lib_name = find_loaded_library("_kunlun_vxpu")
        xpumem_available = True
    else:
        init_module = None
        py_create_and_map = None
        py_unmap_and_release = None
        xpu_memcpy = None
        get_mem_info = None
        try_lock_gpu = None
        unlock_gpu = None
        lib_name = None
except ImportError as e:
    logger.warning("Failed to import vllm_kunlun._kunlun_vxpu:%s.", e)
    init_module = None
    py_create_and_map = None
    py_unmap_and_release = None
    xpu_memcpy = None
    get_mem_info = None
    try_lock_gpu = None
    unlock_gpu = None
    lib_name = None

# py_device, py_alignedSize, py_d_mem, py_p_memHandle
HandleType = tuple[int, int, int, int]


@dataclasses.dataclass
class AllocationData:
    handle: HandleType
    tag: str
    cpu_backup_tensor: Optional[torch.Tensor] = None


def create_and_map(allocation_handle: HandleType) -> None:
    py_create_and_map(*allocation_handle)


def unmap_and_release(allocation_handle: HandleType) -> None:
    py_unmap_and_release(*allocation_handle)


def get_pluggable_allocator(
    python_malloc_fn: Callable[[tuple[int, int, int, int]], None],
    python_free_func: Callable[[int], tuple[int, int, int, int]],
) -> torch.cuda.memory.CUDAPluggableAllocator:
    current_device = torch.cuda.current_device()
    init_module(python_malloc_fn, python_free_func, current_device)
    new_alloc = torch.cuda.memory.CUDAPluggableAllocator(
        lib_name, 'my_malloc', 'my_free'
    )
    return new_alloc


@contextmanager
def use_memory_pool_with_allocator(
        python_malloc_fn: Callable[[tuple[int, int, int, int]], None],
        python_free_func: Callable[[int], tuple[int, int, int, int]]):
    new_alloc = get_pluggable_allocator(python_malloc_fn, python_free_func)
    mem_pool = torch.cuda.memory.MemPool(new_alloc._allocator)
    with torch.cuda.memory.use_mem_pool(mem_pool):
        yield mem_pool, new_alloc


class XpuMemAllocator:
    """
    A singleton class that manages a memory pool for Kunlun XPU tensors.
    The memory in this pool can be offloaded or discarded when the
    allocator sleeps.
    Inside the `use_memory_pool(tag)` context, all tensors created will
    be allocated in the memory pool, and has the same tag as the
    tag passed to the context.
    When we call `sleep`, all tensors with the specified tag will be
    offloaded to CPU memory, and the rest of the tensors will be discarded.
    When we call `wake_up`, all tensors that are previously offloaded
    will be loaded back to GPU memory, and the rest of the tensors will
    have empty memory.
    Why it needs to be a singleton?
    When allocated tensors are garbage collected, PyTorch will call
    the free callback, which will call the `python_free_callback` method.
    The C-extension uses a global variable to store the function of an
    instance of this class. If we create multiple instances of this class,
    the global variable will be overwritten and the free callback will
    not work as expected.
    """
    nstance = None
    default_tag: str = "default"

    @staticmethod
    def get_instance() -> "XpuMemAllocator":
        """
        XpuMemAllocator is a singleton class.
        We cannot call the constructor directly.
        Call this method to get the instance.
        """
        assert xpumem_available, "xpumem allocator is not available"
        if XpuMemAllocator.nstance is None:
            XpuMemAllocator.nstance = XpuMemAllocator()
        return XpuMemAllocator.nstance

    def __init__(self):
        conf = os.environ.get("PYTORCH_CUDA_ALLOC_CONF", "")
        assert "expandable_segments:True" not in conf, \
            ("Expandable segments are not compatible with memory pool. "
            "Please track https://github.com/pytorch/pytorch/issues/147851 "
            "for the latest updates.")

        self.pointer_to_data: dict[int, AllocationData] = {}
        self.current_tag: str = XpuMemAllocator.default_tag
        self.allocator_and_pools: dict[str, Any] = {}

    def python_malloc_callback(self, allocation_handle: HandleType) -> None:
        """
        Internal method to store the allocation data
        when memory is allocated in the memory pool."""
        py_d_mem = allocation_handle[2]
        self.pointer_to_data[py_d_mem] = AllocationData(
            allocation_handle, self.current_tag)
        return

    def python_free_callback(self, ptr: int) -> HandleType:
        """
        Internal method to look up the allocation data
        when memory is freed in the memory pool."""
        data = self.pointer_to_data.pop(ptr)
        if data.cpu_backup_tensor is not None:
            data.cpu_backup_tensor = None
        return data.handle

    @contextmanager
    def use_memory_pool(self, tag: Optional[str] = None):
        """
        A context manager to use the memory pool.
        All memory allocation created inside the context will be allocated 
        in the memory pool, and has the specified tag.
        :param tag: The tag of the memory allocation. If None, the default tag
            will be used.
        """
        if tag is None:
            tag = XpuMemAllocator.default_tag

        assert isinstance(tag, str)

        old_tag = self.current_tag
        self.current_tag = tag
        with use_memory_pool_with_allocator(self.python_malloc_callback,
                                            self.python_free_callback) as data:
            # start to hit another PyTorch bug in PyTorch 2.6,
            # possibly because of gc-related issue w.r.t. the allocator and
            # the memory pool.
            # to avoid the issue, we keep a reference of the data.
            # see https://github.com/pytorch/pytorch/issues/146431 .
            self.allocator_and_pools[tag] = data
            yield
            # PyTorch's bug, calling torch.cuda.empty_cache() will error
            # when using pluggable allocator, see
            # https://github.com/pytorch/pytorch/issues/145168 .
            # if we have some memory allocated and then freed,
            # the memory will not be released, e.g. in online quantization,
            # where the model is created in higher precision, and then
            # quantized in lower precision.
            # Find all unused allocations and manually release them.
            # TODO: we should expose `empty_cache` method in the memory pool.
            # TODO: ask for help from PyTorch team to expose this method.
            # allocations = data[0].snapshot()
            # for allocation in allocations:
            #     if allocation["allocated_size"] == 0:
            #         handle = self._python_free_callback(allocation["address"])
            #         unmap_and_release(handle)
            self.current_tag = old_tag

    def get_current_usage(self) -> int:
        """
        Get the total number of bytes allocated in the memory pool.
        """
        sum_bytes: int = 0
        for ptr, data in self.pointer_to_data.items():
            handle = data.handle
            sum_bytes += handle[1]
        return sum_bytes

    def vxpu_try_lock_gpu(self) -> tuple[bool, bool]:
        if try_lock_gpu:
            return try_lock_gpu()
        else:
            return False, False

    def _vxpu_lock_gpu(self) -> bool:
        while True:
            success, _ = self.vxpu_try_lock_gpu()
            if success:
                return True
            time.sleep(0.001)

    def vxpu_unlock_gpu(self):
        if unlock_gpu:
            unlock_gpu()

    def get_pool_mem_info(self) -> tuple[int, int]:
        """
        get memory info (available, total) in reserved pool.
        """
        return get_mem_info()

    def offload_vram(
            self,
            offload_tags: Optional[Union[tuple[str, ...],
                                         str]] = None) -> None:
        """
        Put the allocator in sleep mode.
        All data in the memory allocation with the specified tag will be 
        offloaded to CPU memory, and others will be discarded.
        :param offload_tags: The tags of the memory allocation that will be
            offloaded. The rest of the memory allocation will be discarded.
        """
        if offload_tags is None:
            # by default, allocated tensors are offloaded
            # when the allocator sleeps
            offload_tags = (XpuMemAllocator.default_tag,)
        elif isinstance(offload_tags, str):
            offload_tags = (offload_tags,)

        assert isinstance(offload_tags, tuple)

        for ptr, data in self.pointer_to_data.items():
            handle = data.handle
            if data.tag in offload_tags:
                size_in_bytes = handle[1]
                if data.cpu_backup_tensor is None:
                    cpu_backup_tensor = torch.empty(
                        size_in_bytes,
                        dtype=torch.uint8,
                        device='cpu',
                        pin_memory=True)
                    cpu_ptr = cpu_backup_tensor.data_ptr()
                    XPU_DEVICE_TO_HOST = 0
                    xpu_memcpy(cpu_ptr, ptr, size_in_bytes, XPU_DEVICE_TO_HOST)
                    data.cpu_backup_tensor = cpu_backup_tensor
                unmap_and_release(handle)
            else:
                unmap_and_release(handle)

        self.vxpu_unlock_gpu()

    def try_reload_vram(self, tags: Optional[list[str]] = None) -> tuple[bool, bool]:
        succ, prev_is_self = self.vxpu_try_lock_gpu()
        if not succ:
            # not get the lock
            return False, prev_is_self

        if prev_is_self:
            # nothing to do
            return succ, prev_is_self

        for ptr, data in self.pointer_to_data.items():
            handle = data.handle
            if tags is None or data.tag in tags:
                create_and_map(handle)
                if data.cpu_backup_tensor is not None:
                    cpu_backup_tensor = data.cpu_backup_tensor
                    size_in_bytes = (
                        cpu_backup_tensor.numel() * cpu_backup_tensor.element_size()
                    )
                    cpu_ptr = cpu_backup_tensor.data_ptr()
                    XPU_HOST_TO_DEVICE = 1
                    xpu_memcpy(ptr, cpu_ptr, size_in_bytes, XPU_HOST_TO_DEVICE)
                    # data.cpu_backup_tensor = None
        return succ, prev_is_self