import torch
import torch.nn as nn

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

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

        Returns:
            torch.Tensor: Output tensor with Softsign applied, same shape as input.
        """
        return x / (1 + torch.abs(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 softsign_forward(torch::Tensor x);
"""

_cuda_src = r"""
#include <torch/extension.h>
#include <cuda.h>
#include <cuda_runtime.h>
#include <stdint.h>

__device__ __forceinline__ float softsign_val(float v) {
    return __fdividef(v, 1.0f + fabsf(v));
}

__global__ void softsign_vec4_kernel(const float* __restrict__ x, float* __restrict__ y, long long n4) {
    long long idx = (long long)blockIdx.x * blockDim.x + threadIdx.x;
    long long stride = (long long)blockDim.x * gridDim.x;
    const float4* __restrict__ x4 = reinterpret_cast<const float4*>(x);
    float4* __restrict__ y4 = reinterpret_cast<float4*>(y);
    for (long long i = idx; i < n4; i += stride) {
        float4 v = x4[i];
        float4 o;
        o.x = softsign_val(v.x);
        o.y = softsign_val(v.y);
        o.z = softsign_val(v.z);
        o.w = softsign_val(v.w);
        y4[i] = o;
    }
}

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

torch::Tensor softsign_forward(torch::Tensor x) {
    auto y = torch::empty_like(x);
    long long n = x.numel();
    if (n == 0) return y;
    const int threads = 256;
    const float* xp = x.data_ptr<float>();
    float* yp = y.data_ptr<float>();
    if ((n % 4 == 0) && (((uintptr_t)xp & 15) == 0) && (((uintptr_t)yp & 15) == 0)) {
        long long n4 = n / 4;
        long long need = (n4 + threads - 1) / threads;
        int blocks = (int)(need < 65535LL ? need : 65535LL);
        softsign_vec4_kernel<<<blocks, threads>>>(xp, yp, n4);
    } else {
        long long need = (n + threads - 1) / threads;
        int blocks = (int)(need < 65535LL ? need : 65535LL);
        softsign_scalar_kernel<<<blocks, threads>>>(xp, yp, n);
    }
    return y;
}
"""

_ext = load_inline(name="softsign_ext_v3", cpp_sources=[_cpp_src], cuda_sources=[_cuda_src], functions=["softsign_forward"], extra_cuda_cflags=["-O3", "--use_fast_math"], extra_cflags=["-O3"], verbose=False)

class ModelNew(nn.Module):
    def __init__(self):
        super(ModelNew, self).__init__()
    def forward(self, x: torch.Tensor) -> torch.Tensor:
        return _ext.softsign_forward(x)