import torch
import torch.nn as nn

class Model(nn.Module):
    """
    Simple model that performs a SELU activation.
    """
    def __init__(self):
        super(Model, self).__init__()
    
    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """
        Applies SELU activation to the input tensor.

        Args:
            x (torch.Tensor): Input tensor of any shape.

        Returns:
            torch.Tensor: Output tensor with SELU applied, same shape as input.
        """
        return torch.selu(x)
    
batch_size = 4096
dim = 393216

def get_inputs():
    x = torch.rand(batch_size, dim)
    return [x]

def get_init_inputs():
    return []  # No special initialization inputs needed

import torch
import torch.nn as nn
from torch.utils.cpp_extension import load_inline

_cpp_src = r'''
#include <torch/extension.h>
torch::Tensor selu_forward(torch::Tensor x);
'''

_cuda_src = r'''
#include <ATen/ATen.h>
#include <ATen/cuda/CUDAContext.h>
#include <c10/cuda/CUDAGuard.h>
#include <c10/cuda/CUDAException.h>
#include <cuda_runtime.h>
#include <stdint.h>

constexpr float SELU_SCALE = 1.050700987355480493f;
constexpr float SELU_SA = 1.7580993408473766f;

__device__ __forceinline__ float selu_val(float v) {
    return v > 0.0f ? SELU_SCALE * v : SELU_SA * (expf(v) - 1.0f);
}

__global__ void selu_vec4_unroll_kernel(const float4* __restrict__ x, float4* __restrict__ y, int64_t n4) {
    const int U = 4;
    int64_t base0 = (int64_t)blockIdx.x * blockDim.x * U + threadIdx.x;
    int64_t stride = (int64_t)blockDim.x * gridDim.x * U;
    for (int64_t base = base0; base < n4; base += stride) {
        #pragma unroll
        for (int k = 0; k < U; ++k) {
            int64_t i = base + (int64_t)k * blockDim.x;
            if (i < n4) {
                float4 v = x[i];
                float4 o;
                o.x = selu_val(v.x);
                o.y = selu_val(v.y);
                o.z = selu_val(v.z);
                o.w = selu_val(v.w);
                y[i] = o;
            }
        }
    }
}

__global__ void selu_scalar_kernel(const float* __restrict__ x, float* __restrict__ y, int64_t n) {
    int64_t idx = (int64_t)blockIdx.x * blockDim.x + threadIdx.x;
    int64_t stride = (int64_t)blockDim.x * gridDim.x;
    for (int64_t i = idx; i < n; i += stride) {
        y[i] = selu_val(x[i]);
    }
}

torch::Tensor selu_forward(torch::Tensor x) {
    c10::cuda::CUDAGuard device_guard(x.device());
    auto y = torch::empty_like(x);
    int64_t n = x.numel();
    if (n == 0) return y;

    const int threads = 512;
    cudaStream_t stream = at::cuda::getCurrentCUDAStream();
    const float* xp = x.data_ptr<float>();
    float* yp = y.data_ptr<float>();
    uintptr_t ax = reinterpret_cast<uintptr_t>(xp);
    uintptr_t ay = reinterpret_cast<uintptr_t>(yp);

    if (((ax | ay) & 15) == 0 && (n % 4 == 0)) {
        int64_t n4 = n / 4;
        int blocks = (int)((n4 + (int64_t)threads * 4 - 1) / ((int64_t)threads * 4));
        if (blocks > 65535) blocks = 65535;
        selu_vec4_unroll_kernel<<<blocks, threads, 0, stream>>>(
            reinterpret_cast<const float4*>(xp), reinterpret_cast<float4*>(yp), n4);
    } else {
        int blocks = (int)((n + threads - 1) / threads);
        if (blocks > 65535) blocks = 65535;
        selu_scalar_kernel<<<blocks, threads, 0, stream>>>(xp, yp, n);
    }
    C10_CUDA_KERNEL_LAUNCH_CHECK();
    return y;
}
'''

_ext = load_inline(
    name='selu_ext_final',
    cpp_sources=[_cpp_src],
    cuda_sources=[_cuda_src],
    functions=['selu_forward'],
    extra_cflags=['-O3'],
    extra_cuda_cflags=['-O3', '--use_fast_math'],
    verbose=False,
)

class ModelNew(nn.Module):
    def __init__(self):
        super(ModelNew, self).__init__()

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        return _ext.selu_forward(x)