# Copyright (c) Huawei Technologies Co., Ltd. 2025. All rights reserved.
#
# 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 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.

import triton
import triton.language as tl
import numpy as np
import torch
import logging
import pytest
import test_common
from test_common import TestUtils


def torch_invert(x0, ddtype):
    if 'float' in str(ddtype):
        x0 = x0.to(torch.int32)
        y_ref = ~x0
        y_ref = y_ref.to(ddtype)
    else:
        y_ref = ~x0
    return y_ref


@triton.jit
def triton_sub(output_ptr, x_ptr, y_ptr, z_ptr, XB: tl.constexpr, YB: tl.constexpr, ZB: tl.constexpr,
               XNUMEL: tl.constexpr,
               YNUMEL: tl.constexpr, ZNUMEL: tl.constexpr):
    xoffs = tl.program_id(0) * XB
    yoffs = tl.program_id(1) * YB
    zoffs = tl.program_id(2) * ZB

    xidx = tl.arange(0, XB) + xoffs
    yidx = tl.arange(0, YB) + yoffs
    zidx = tl.arange(0, ZB) + zoffs

    idx = xidx[:, None, None] * YNUMEL * ZNUMEL + yidx[None, :, None] * ZNUMEL + zidx[None, None, :]

    X = tl.load(x_ptr + idx)
    Y = tl.load(y_ptr + idx)

    ret = X - Y
    tl.debug_barrier()
    tl.store(output_ptr + idx, ret)


@triton.jit
def triton_invert_4d_5d(
        output_ptr, x_ptr,
        BLOCK_0: tl.constexpr, BLOCK_1: tl.constexpr, BLOCK_2: tl.constexpr, BLOCK_3: tl.constexpr,
        BLOCK_4: tl.constexpr,
        SHAPE_0: tl.constexpr, SHAPE_1: tl.constexpr, SHAPE_2: tl.constexpr, SHAPE_3: tl.constexpr,
        SHAPE_4: tl.constexpr,
        STRIDE_0: tl.constexpr, STRIDE_1: tl.constexpr, STRIDE_2: tl.constexpr, STRIDE_3: tl.constexpr,
        STRIDE_4: tl.constexpr
):
    offsets = tl.program_id(0)

    offsets = offsets + tl.arange(0, BLOCK_0) * STRIDE_0
    masks = tl.arange(0, BLOCK_0) < SHAPE_0
    if (BLOCK_1 * BLOCK_2 * BLOCK_3 * BLOCK_4) > 1:
        offsets = offsets[:, None] + tl.arange(0, BLOCK_1)[None, :] * STRIDE_1
        masks = masks[:, None] & (tl.arange(0, BLOCK_1)[None, :] < SHAPE_1)
    if (BLOCK_2 * BLOCK_3 * BLOCK_4) > 1:
        offsets = offsets[:, :, None] + tl.arange(0, BLOCK_2)[None, None, :] * STRIDE_2
        masks = masks[:, :, None] & (tl.arange(0, BLOCK_2)[None, None, :] < SHAPE_2)
    if (BLOCK_3 * BLOCK_4) > 1:
        offsets = offsets[:, :, :, None] + tl.arange(0, BLOCK_3)[None, None, None, :] * STRIDE_3
        masks = masks[:, :, :, None] & (tl.arange(0, BLOCK_3)[None, None, None, :] < SHAPE_3)
    if BLOCK_4 > 1:
        offsets = offsets[:, :, :, :, None] + tl.arange(0, BLOCK_4)[None, None, None, None, :] * STRIDE_4
        masks = masks[:, :, :, :, None] & (tl.arange(0, BLOCK_4)[None, None, None, None, :] < SHAPE_4)

    x_val = tl.load(x_ptr + offsets, masks)
    ret = ~x_val
    tl.debug_barrier()
    tl.store(output_ptr + offsets, ret, mask=masks)


test_shape_1d_2d_3d = [(1,), (2,), (1, 1), (3, 13), (1, 1, 1), (4, 3, 8)]
test_shape_4_5d = [(1, 1, 1, 1), (2, 2, 2, 2), (1, 1, 1, 1, 1), (2, 2, 2, 2, 1)]


@pytest.mark.parametrize('shape', test_shape_1d_2d_3d)
@pytest.mark.parametrize('dtype', ['int8', 'int16', 'int32', 'int64', 'float16', 'float32', 'bfloat16'])
def test_sub(shape, dtype):
    logging.log(logging.DEBUG, f"shape = {shape}")
    x = test_common.generate_tensor(shape, dtype).npu()
    y = test_common.generate_tensor(shape, dtype).npu()
    z = test_common.generate_tensor(shape, dtype).npu()

    new_shape = shape

    output = torch.randint(1, new_shape, dtype=eval('torch.' + dtype)).npu()

    logging.log(logging.DEBUG, f"output.dtype={output.dtype}")

    ans = x - y

    if len(shape) == 1:
        XB = 1
        xnumel = 1
        YB = 1
        ynumel = 1
        ZB = shape[0]
        znumel = shape[0]
    elif len(shape) == 2:
        XB = 1
        xnumel = 1
        YB = shape[0]
        ynumel = shape[0]
        ZB = shape[1]
        znumel = shape[1]
    else:
        XB = shape[0]
        xnumel = shape[0]
        YB = shape[1]
        ynumel = shape[1]
        ZB = shape[2]
        znumel = shape[2]

    grid = (1, 1, 1)
    if dtype == 'int8':
        if x.numel() * x.element_size() >= 512:
            grid = (1, 1, ZB)
            ZB = 1
    else:
        if x.numel() * x.element_size() >= 8192:
            grid = (1, 1, ZB)
            ZB = 1

    triton_sub[grid](output, x, y, z, XB, YB, ZB, xnumel, ynumel, znumel)

    test_common.validate_cmp(dtype, ans, output)


@pytest.mark.parametrize('shape', test_shape_1d_2d_3d + test_shape_4_5d)
@pytest.mark.parametrize('dtype', ['bool'])
def test_invert_4d_5d(shape, dtype):
    logging.log(logging.DEBUG, f"shape = {shape}")
    x = test_common.generate_tensor(shape, dtype).npu()

    output = torch.randint(1, shape, dtype=eval('torch.' + dtype)).npu()

    logging.log(logging.DEBUG, f"output.dtype={output.dtype}")

    ans = torch_invert(x, eval('torch.' + dtype))

    blocks = list(x.size())
    strides = list(x.stride())
    while len(blocks) < 5:
        blocks.append(1)
        strides.append(1)

    grid = (1,)
    triton_invert_4d_5d[grid](output, x, *blocks, *blocks, *strides)

    test_common.validate_cmp(dtype, ans, output)