forked from ccf-ai-infra/GPUCodeForces
Merge pull request 'feat add a hardmish #8' (#236) from zizi05/GPUCodeForces:hardmish into main
This commit is contained in:
commit
4df3024dad
|
|
@ -0,0 +1,76 @@
|
|||
import torch
|
||||
import torch.nn as nn
|
||||
from torch.utils.cpp_extension import load_inline
|
||||
|
||||
hardmish_source = """
|
||||
#include <torch/extension.h>
|
||||
#include <cuda_runtime.h>
|
||||
|
||||
__device__ __forceinline__ float hardmish_impl(float x) {
|
||||
if (x <= -3.0f) {
|
||||
return 0.0f;
|
||||
} else if (x >= 3.0f) {
|
||||
return x;
|
||||
} else {
|
||||
return x * (x + 3.0f) / 6.0f;
|
||||
}
|
||||
}
|
||||
|
||||
__global__ void hardmish_kernel(
|
||||
const float* __restrict__ input,
|
||||
float* __restrict__ output,
|
||||
int size
|
||||
) {
|
||||
int idx = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
if (idx >= size) return;
|
||||
output[idx] = hardmish_impl(input[idx]);
|
||||
}
|
||||
|
||||
torch::Tensor hardmish_cuda(torch::Tensor input) {
|
||||
TORCH_CHECK(input.is_cuda(), "input must be CUDA tensor");
|
||||
TORCH_CHECK(input.dtype() == torch::kFloat32, "input must be float32");
|
||||
|
||||
input = input.contiguous();
|
||||
int total_size = input.numel();
|
||||
auto output = torch::empty_like(input);
|
||||
|
||||
const int threads_per_block = 256;
|
||||
const int blocks = (total_size + threads_per_block - 1) / threads_per_block;
|
||||
|
||||
hardmish_kernel<<<blocks, threads_per_block>>>(
|
||||
input.data_ptr<float>(),
|
||||
output.data_ptr<float>(),
|
||||
total_size
|
||||
);
|
||||
|
||||
cudaError_t err = cudaGetLastError();
|
||||
if (err != cudaSuccess) {
|
||||
throw std::runtime_error("CUDA error: " + std::string(cudaGetErrorString(err)));
|
||||
}
|
||||
return output;
|
||||
}
|
||||
"""
|
||||
|
||||
hardmish_cpp_source = """
|
||||
torch::Tensor hardmish_cuda(torch::Tensor input);
|
||||
"""
|
||||
|
||||
hardmish_module = load_inline(
|
||||
name="hardmish_final",
|
||||
cpp_sources=hardmish_cpp_source,
|
||||
cuda_sources=hardmish_source,
|
||||
functions=["hardmish_cuda"],
|
||||
extra_cuda_cflags=["-O2"],
|
||||
verbose=False
|
||||
)
|
||||
|
||||
class ModelNew(nn.Module):
|
||||
def __init__(self, in_features):
|
||||
super().__init__()
|
||||
torch.manual_seed(42)
|
||||
self.linear = nn.Linear(in_features, in_features)
|
||||
self.hardmish = hardmish_module.hardmish_cuda
|
||||
|
||||
def forward(self, x):
|
||||
x = self.linear(x)
|
||||
return self.hardmish(x)
|
||||
|
|
@ -0,0 +1,30 @@
|
|||
import torch
|
||||
import torch.nn as nn
|
||||
|
||||
class Model(nn.Module):
|
||||
def __init__(self, in_features):
|
||||
super().__init__()
|
||||
torch.manual_seed(42)
|
||||
self.linear = nn.Linear(in_features, in_features)
|
||||
|
||||
def forward(self, x):
|
||||
x = self.linear(x)
|
||||
# 分段计算HardMish
|
||||
return torch.where(
|
||||
x <= -3,
|
||||
torch.zeros_like(x),
|
||||
torch.where(
|
||||
x >= 3,
|
||||
x,
|
||||
x * (x + 3) / 6
|
||||
)
|
||||
)
|
||||
|
||||
def get_inputs():
|
||||
batch_size = 8192
|
||||
in_features = 256
|
||||
x = torch.randn(batch_size, in_features)
|
||||
return [x]
|
||||
|
||||
def get_init_inputs():
|
||||
return [256]
|
||||
|
|
@ -0,0 +1,118 @@
|
|||
You write custom CUDA kernels to replace the pytorch operators in the given architecture to get speedups.
|
||||
|
||||
You have complete freedom to choose the set of operators you want to replace. You may make the decision to replace some operators with custom CUDA kernels and leave others unchanged. You may replace multiple operators with custom implementations, consider operator fusion opportunities (combining multiple operators into a single kernel, for example, combining matmul+relu), or algorithmic changes (such as online softmax). You are only limited by your imagination.
|
||||
|
||||
Here's an example to show you the syntax of inline embedding custom CUDA operators in torch: The example given architecture is:
|
||||
|
||||
Given hardmish Architecture (Base PyTorch Implementation)
|
||||
python
|
||||
运行
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
|
||||
class Model(nn.Module):
|
||||
def __init__(self, in_features):
|
||||
super().__init__()
|
||||
torch.manual_seed(42)
|
||||
self.linear = nn.Linear(in_features, in_features)
|
||||
|
||||
def forward(self, x):
|
||||
x = self.linear(x)
|
||||
# 分段计算HardMish
|
||||
return torch.where(
|
||||
x <= -3,
|
||||
torch.zeros_like(x),
|
||||
torch.where(
|
||||
x >= 3,
|
||||
x,
|
||||
x * (x + 3) / 6
|
||||
)
|
||||
)
|
||||
|
||||
def get_inputs():
|
||||
batch_size = 8192
|
||||
in_features = 256
|
||||
x = torch.randn(batch_size, in_features)
|
||||
return [x]
|
||||
|
||||
def get_init_inputs():
|
||||
return [256]
|
||||
New Architecture with Custom CUDA Kernels (hardmish Optimization)
|
||||
python
|
||||
运行
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
from torch.utils.cpp_extension import load_inline
|
||||
|
||||
hardmish_source = """
|
||||
#include <torch/extension.h>
|
||||
#include <cuda_runtime.h>
|
||||
|
||||
__device__ __forceinline__ float hardmish_impl(float x) {
|
||||
if (x <= -3.0f) {
|
||||
return 0.0f;
|
||||
} else if (x >= 3.0f) {
|
||||
return x;
|
||||
} else {
|
||||
return x * (x + 3.0f) / 6.0f;
|
||||
}
|
||||
}
|
||||
|
||||
__global__ void hardmish_kernel(
|
||||
const float* __restrict__ input,
|
||||
float* __restrict__ output,
|
||||
int size
|
||||
) {
|
||||
int idx = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
if (idx >= size) return;
|
||||
output[idx] = hardmish_impl(input[idx]);
|
||||
}
|
||||
|
||||
torch::Tensor hardmish_cuda(torch::Tensor input) {
|
||||
TORCH_CHECK(input.is_cuda(), "input must be CUDA tensor");
|
||||
TORCH_CHECK(input.dtype() == torch::kFloat32, "input must be float32");
|
||||
|
||||
input = input.contiguous();
|
||||
int total_size = input.numel();
|
||||
auto output = torch::empty_like(input);
|
||||
|
||||
const int threads_per_block = 256;
|
||||
const int blocks = (total_size + threads_per_block - 1) / threads_per_block;
|
||||
|
||||
hardmish_kernel<<<blocks, threads_per_block>>>(
|
||||
input.data_ptr<float>(),
|
||||
output.data_ptr<float>(),
|
||||
total_size
|
||||
);
|
||||
|
||||
cudaError_t err = cudaGetLastError();
|
||||
if (err != cudaSuccess) {
|
||||
throw std::runtime_error("CUDA error: " + std::string(cudaGetErrorString(err)));
|
||||
}
|
||||
return output;
|
||||
}
|
||||
"""
|
||||
|
||||
hardmish_cpp_source = """
|
||||
torch::Tensor hardmish_cuda(torch::Tensor input);
|
||||
"""
|
||||
|
||||
hardmish_module = load_inline(
|
||||
name="hardmish_final",
|
||||
cpp_sources=hardmish_cpp_source,
|
||||
cuda_sources=hardmish_source,
|
||||
functions=["hardmish_cuda"],
|
||||
extra_cuda_cflags=["-O2"],
|
||||
verbose=False
|
||||
)
|
||||
|
||||
class ModelNew(nn.Module):
|
||||
def __init__(self, in_features):
|
||||
super().__init__()
|
||||
torch.manual_seed(42)
|
||||
self.linear = nn.Linear(in_features, in_features)
|
||||
self.hardmish = hardmish_module.hardmish_cuda
|
||||
|
||||
def forward(self, x):
|
||||
x = self.linear(x)
|
||||
return self.hardmish(x)
|
||||
|
|
@ -0,0 +1,74 @@
|
|||
###########################################################
|
||||
# 性能和精度验证程序
|
||||
###########################################################
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
import time
|
||||
from hardmish_torchcode import Model,get_inputs,get_init_inputs
|
||||
from hardmish_cudacode import ModelNew
|
||||
|
||||
def run_benchmark():
|
||||
# 检查 CUDA 是否可用
|
||||
if not torch.cuda.is_available():
|
||||
print("CUDA 不可用，请确保您有可用的 NVIDIA GPU 并已正确安装 PyTorch CUDA 版本。")
|
||||
return
|
||||
else:
|
||||
device = torch.device("cuda")
|
||||
|
||||
# 初始化模型
|
||||
init_inputs = get_init_inputs()
|
||||
init_inputs = [
|
||||
x.cuda(device=device) if isinstance(x, torch.Tensor) else x for x in init_inputs
|
||||
]
|
||||
inputs = get_inputs()
|
||||
inputs = [
|
||||
x.cuda(device=device) if isinstance(x, torch.Tensor) else x for x in inputs
|
||||
]
|
||||
|
||||
torch_model = Model(*init_inputs).cuda()
|
||||
cuda_model = ModelNew(*init_inputs).cuda()
|
||||
|
||||
torch_model.eval()
|
||||
cuda_model.eval()
|
||||
|
||||
print("-------------------- 精度对齐验证 --------------------")
|
||||
with torch.no_grad():
|
||||
output_torch = torch_model( *inputs)
|
||||
output_cuda = cuda_model(*inputs)
|
||||
|
||||
precision_flag = torch.allclose(output_torch, output_cuda,rtol=1e-03, atol=1e-4)
|
||||
if precision_flag:
|
||||
print("✅ 精度对齐：两个模型的输出结果非常接近。")
|
||||
else:
|
||||
print("❌ 精度不一致！")
|
||||
|
||||
print("\n-------------------- 性能加速比测试 --------------------")
|
||||
num_iterations = 100
|
||||
|
||||
# PyTorch 模型计时
|
||||
torch.cuda.synchronize()
|
||||
start_time = time.time()
|
||||
for _ in range(num_iterations):
|
||||
_ = torch_model(*inputs)
|
||||
torch.cuda.synchronize()
|
||||
torch_time = (time.time() - start_time) / num_iterations
|
||||
|
||||
# 自定义 CUDA 内核计时
|
||||
torch.cuda.synchronize()
|
||||
start_time = time.time()
|
||||
for _ in range(num_iterations):
|
||||
_ = cuda_model(*inputs)
|
||||
torch.cuda.synchronize()
|
||||
cuda_time = (time.time() - start_time) / num_iterations
|
||||
|
||||
print(f"PyTorch hardwish 平均执行时间: {torch_time:.6f} 秒")
|
||||
print(f"自定义 CUDA 内核 平均执行时间: {cuda_time:.6f} 秒")
|
||||
speedup = 0
|
||||
if cuda_time > 0:
|
||||
speedup = torch_time / cuda_time
|
||||
print(f"加速比 (Speedup): {speedup:.2f}x")
|
||||
else:
|
||||
print("CUDA 内核执行时间为0，无法计算加速比。")
|
||||
return precision_flag,speedup
|
||||
if __name__ == "__main__":
|
||||
precision_flag,speedup = run_benchmark()
|
||||
Loading…
Reference in New Issue