import torch
import torch.nn as nn
import torch.nn.functional as F
import math

# From https://github.com/karpathy/minGPT/blob/master/mingpt/model.py

class Model(nn.Module):
    """
    Implementation of the GELU activation function currently in Google BERT repo (identical to OpenAI GPT).
    Reference: Gaussian Error Linear Units (GELU) paper: https://arxiv.org/abs/1606.08415
    """
    def __init__(self):
        super(Model, self).__init__()
    
    def forward(self, x):
        return 0.5 * x * (1.0 + torch.tanh(math.sqrt(2.0 / math.pi) * (x + 0.044715 * torch.pow(x, 3.0))))

batch_size = 8192
dim = 8192

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

def get_init_inputs():
    return []

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

cpp_src = r"""
#include <torch/extension.h>
torch::Tensor gelu_forward(torch::Tensor x);
"""

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

__device__ __forceinline__ float gelu_one(float x) {
    const float kAlpha = 0.7978845608028654f;
    const float kBeta = 0.044715f;
    float x2 = x * x;
    float u = kAlpha * (x + kBeta * x * x2);
    float t = tanhf(u);
    return 0.5f * x * (1.0f + t);
}

__global__ void gelu_float4_kernel(const float* __restrict__ x, float* __restrict__ y, int n4) {
    int idx = blockIdx.x * blockDim.x + threadIdx.x;
    int stride = blockDim.x * gridDim.x;
    const float4* __restrict__ x4 = reinterpret_cast<const float4*>(x);
    float4* __restrict__ y4 = reinterpret_cast<float4*>(y);
    for (int i = idx; i < n4; i += stride) {
        float4 v = x4[i];
        v.x = gelu_one(v.x);
        v.y = gelu_one(v.y);
        v.z = gelu_one(v.z);
        v.w = gelu_one(v.w);
        y4[i] = v;
    }
}

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

torch::Tensor gelu_forward(torch::Tensor x) {
    auto y = torch::empty_like(x);
    long n = x.numel();
    if (n == 0) return y;
    const int threads = 256;
    cudaStream_t stream = at::cuda::getCurrentCUDAStream();
    uintptr_t xp = reinterpret_cast<uintptr_t>(x.data_ptr<float>());
    uintptr_t yp = reinterpret_cast<uintptr_t>(y.data_ptr<float>());
    if ((n % 4 == 0) && ((xp & 15) == 0) && ((yp & 15) == 0)) {
        int n4 = (int)(n / 4);
        int blocks = (n4 + threads - 1) / threads;
        if (blocks > 65535) blocks = 65535;
        gelu_float4_kernel<<<blocks, threads, 0, stream>>>(x.data_ptr<float>(), y.data_ptr<float>(), n4);
    } else {
        int blocks = (int)((n + threads - 1) / threads);
        if (blocks > 65535) blocks = 65535;
        gelu_scalar_kernel<<<blocks, threads, 0, stream>>>(x.data_ptr<float>(), y.data_ptr<float>(), n);
    }
    return y;
}
"""

_gelu_ext = load_inline(
    name="gelu_approx_ext_final",
    cpp_sources=cpp_src,
    cuda_sources=cuda_src,
    functions=["gelu_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):
        return _gelu_ext.gelu_forward(x)