Write a custom CUDA kernel to optimize `CRReLU` (Correction Regularized ReLU).

Formula: f(x) = max(0, x) + epsilon * x * exp(-x^2 / 2)

Problem Analysis:
1. Memory Bound & Computationally Heavy: This is an element-wise activation that involves an exponential function, making it more arithmetically complex than simple ReLU.
2. Operator Chaining: The PyTorch implementation chains `relu`, `pow`, `exp`, `mul`, and `add`, creating significant intermediate memory traffic.

Optimization Strategy: Fused Element-wise Kernel with Vectorization

1. One-Thread-per-Element: Map each element to a CUDA thread.

2. Vectorized Loads (float4): Use `float4` to process 128 bits per memory transaction.

3. Fused In-Register Math:
   - For each element `x`:
     `relu_part = fmaxf(x, 0.0f)`
     `correction_part = epsilon * x * __expf(-x * x * 0.5f)`
     `result = relu_part + correction_part`
   - All computations are fused in registers.

4. One-Pass: Fuse all steps into a single read-compute-write kernel.
  
Here's an example to show you the syntax of inline embedding custom CUDA operators in torch: The example given architecture is:   
  
```python
import torch
import torch.nn as nn

BATCH_SIZE = 4096
HIDDEN_DIM = 4096
SHAPE = (BATCH_SIZE, HIDDEN_DIM)

# CRReLU 超参数 epsilon ，论文中建议 0.01
EPSILON_VAL = 0.01

class CRReLU(nn.Module):
    """
    ENTROPY-BASED ACTIVATION FUNCTION OPTIMIZATION: A METHOD ON SEARCHING BETTER ACTIVATION FUNCTIONS
    https://openreview.net/pdf?id=7TZYM6Hm9p

    Formula: f(x) = max(0, x) + epsilon * x * exp(-x^2 / 2)
    """
    def __init__(self, epsilon=0.01):
        super(CRReLU, self).__init__()
        self.epsilon = epsilon

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        relu_part = torch.relu(x)
        correction_part = self.epsilon * x * torch.exp(-x.pow(2) / 2.0)
        return relu_part + correction_part

class Model(nn.Module):
    def __init__(self, epsilon=0.01):
        super(Model, self).__init__()
        self.act = CRReLU(epsilon=epsilon)
    
    def forward(self, x):
        return self.act(x)

def get_inputs():
    input_tensor = torch.randn(SHAPE, dtype=torch.float32) * 2.0
    return [input_tensor.contiguous()]

def get_init_inputs():
    return [EPSILON_VAL]