forked from ccf-ai-infra/GPUCodeForces
93 lines
2.9 KiB
Python
93 lines
2.9 KiB
Python
import torch
|
|
import torch.nn as nn
|
|
from torch.utils.cpp_extension import load_inline
|
|
|
|
class ModelNew(nn.Module):
|
|
def __init__(self, alpha=1.0, beta=1.0):
|
|
super().__init__()
|
|
self.alpha = alpha
|
|
self.beta = beta
|
|
self._compile_cuda_kernel()
|
|
|
|
def _compile_cuda_kernel(self):
|
|
cpp_source = """
|
|
torch::Tensor ptlu_cuda(torch::Tensor x, float alpha, float beta);
|
|
"""
|
|
|
|
cuda_source = """
|
|
#include <torch/extension.h>
|
|
#include <cuda_runtime.h>
|
|
#include <math.h>
|
|
|
|
__device__ __forceinline__ float ptlu_op(float x, float alpha, float beta) {
|
|
// PTLU: alpha * x if x > 0 else beta * tanh(x)
|
|
if (x > 0.0f) {
|
|
return alpha * x;
|
|
}
|
|
return beta * tanhf(x);
|
|
}
|
|
|
|
__global__ void ptlu_kernel(
|
|
const float* __restrict__ x,
|
|
float* __restrict__ output,
|
|
const int n_elements,
|
|
const float alpha,
|
|
const float beta)
|
|
{
|
|
const int tid = blockIdx.x * blockDim.x + threadIdx.x;
|
|
const int stride = blockDim.x * gridDim.x;
|
|
|
|
const int vec_loops = n_elements >> 2;
|
|
const float4* x_vec = reinterpret_cast<const float4*>(x);
|
|
float4* out_vec = reinterpret_cast<float4*>(output);
|
|
|
|
for (int i = tid; i < vec_loops; i += stride) {
|
|
float4 v = __ldg(&x_vec[i]);
|
|
float4 r;
|
|
|
|
r.x = ptlu_op(v.x, alpha, beta);
|
|
r.y = ptlu_op(v.y, alpha, beta);
|
|
r.z = ptlu_op(v.z, alpha, beta);
|
|
r.w = ptlu_op(v.w, alpha, beta);
|
|
|
|
out_vec[i] = r;
|
|
}
|
|
|
|
const int tail_start = vec_loops << 2;
|
|
for (int i = tail_start + tid; i < n_elements; i += stride) {
|
|
output[i] = ptlu_op(x[i], alpha, beta);
|
|
}
|
|
}
|
|
|
|
torch::Tensor ptlu_cuda(torch::Tensor x, float alpha, float beta) {
|
|
auto x_c = x.contiguous();
|
|
const int n_elements = x_c.numel();
|
|
auto output = torch::empty_like(x_c);
|
|
|
|
const int threads = 256;
|
|
const int max_blocks = 65535;
|
|
const int blocks = std::min((n_elements + threads * 4 - 1) / (threads * 4), max_blocks);
|
|
|
|
ptlu_kernel<<<blocks, threads>>>(
|
|
x_c.data_ptr<float>(),
|
|
output.data_ptr<float>(),
|
|
n_elements,
|
|
alpha,
|
|
beta
|
|
);
|
|
|
|
return output;
|
|
}
|
|
"""
|
|
|
|
self.op = load_inline(
|
|
name="ptlu_v1",
|
|
cpp_sources=cpp_source,
|
|
cuda_sources=cuda_source,
|
|
functions=["ptlu_cuda"],
|
|
extra_cuda_cflags=["-O3", "--use_fast_math"],
|
|
verbose=False
|
|
)
|
|
|
|
def forward(self, x):
|
|
return self.op.ptlu_cuda(x, self.alpha, self.beta) |