import torch
import torch.nn as nn

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

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

        Returns:
            torch.Tensor: Output tensor with ReLU applied, same shape as input.
        """
        return torch.relu(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 relu_forward(torch::Tensor x);
"""

cuda_src = r"""
#include <torch/extension.h>
#include <ATen/cuda/CUDAContext.h>
#include <cuda.h>
#include <cuda_runtime.h>
#include <cstdint>

__global__ void relu_inplace4_kernel(float* __restrict__ x, long long n4) {
    long long idx = (long long)blockIdx.x * blockDim.x + threadIdx.x;
    long long stride = (long long)blockDim.x * gridDim.x;
    float4* __restrict__ x4 = reinterpret_cast<float4*>(x);
    for (long long i = idx; i < n4; i += stride) {
        float4 v = x4[i];
        bool any = false;
        if (v.x < 0.0f) { v.x = 0.0f; any = true; }
        if (v.y < 0.0f) { v.y = 0.0f; any = true; }
        if (v.z < 0.0f) { v.z = 0.0f; any = true; }
        if (v.w < 0.0f) { v.w = 0.0f; any = true; }
        if (any) x4[i] = v;
    }
}

__global__ void relu_inplace_kernel(float* __restrict__ x, 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];
        if (v < 0.0f) x[i] = 0.0f;
    }
}

torch::Tensor relu_forward(torch::Tensor x) {
    long long n = x.numel();
    if (n == 0) return x;
    const int threads = 256;
    cudaStream_t stream = at::cuda::getCurrentCUDAStream();
    float* xp = x.data_ptr<float>();
    uintptr_t ax = reinterpret_cast<uintptr_t>(xp);
    if ((ax & 15) == 0 && (n % 4 == 0)) {
        long long n4 = n / 4;
        int blocks = (int)(((n4 + threads - 1) / threads) < 65535LL ? ((n4 + threads - 1) / threads) : 65535LL);
        relu_inplace4_kernel<<<blocks, threads, 0, stream>>>(xp, n4);
    } else {
        int blocks = (int)(((n + threads - 1) / threads) < 65535LL ? ((n + threads - 1) / threads) : 65535LL);
        relu_inplace_kernel<<<blocks, threads, 0, stream>>>(xp, n);
    }
    return x;
}
"""

_ext = load_inline(name="relu_inplace_ext_final", cpp_sources=cpp_src, cuda_sources=cuda_src, functions=["relu_forward"], extra_cuda_cflags=["-O3"], 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.relu_forward(x)