forked from ccf-ai-infra/GPUCodeForces
finish parametricSigMoid #63
This commit is contained in:
parent
10eed82956
commit
ff3d59ff4a
|
|
@ -0,0 +1,90 @@
|
|||
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=0.0):
|
||||
super().__init__()
|
||||
self.alpha = alpha
|
||||
self.beta = beta
|
||||
self._compile_cuda_kernel()
|
||||
|
||||
def _compile_cuda_kernel(self):
|
||||
cpp_source = """
|
||||
torch::Tensor parametric_sigmoid_cuda(torch::Tensor x, float alpha, float beta);
|
||||
"""
|
||||
|
||||
cuda_source = """
|
||||
#include <torch/extension.h>
|
||||
#include <cuda_runtime.h>
|
||||
#include <math.h>
|
||||
|
||||
__device__ __forceinline__ float parametric_sigmoid_op(float x, float alpha, float beta) {
|
||||
// f(x) = 1 / (1 + exp(-(alpha * x + beta)))
|
||||
float arg = -(alpha * x + beta);
|
||||
return 1.0f / (1.0f + expf(arg));
|
||||
}
|
||||
|
||||
__global__ void parametric_sigmoid_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 = parametric_sigmoid_op(v.x, alpha, beta);
|
||||
r.y = parametric_sigmoid_op(v.y, alpha, beta);
|
||||
r.z = parametric_sigmoid_op(v.z, alpha, beta);
|
||||
r.w = parametric_sigmoid_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] = parametric_sigmoid_op(x[i], alpha, beta);
|
||||
}
|
||||
}
|
||||
|
||||
torch::Tensor parametric_sigmoid_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);
|
||||
|
||||
parametric_sigmoid_kernel<<<blocks, threads>>>(
|
||||
x_c.data_ptr<float>(),
|
||||
output.data_ptr<float>(),
|
||||
n_elements,
|
||||
alpha, beta
|
||||
);
|
||||
|
||||
return output;
|
||||
}
|
||||
"""
|
||||
|
||||
self.op = load_inline(
|
||||
name="parametric_sigmoid_v1",
|
||||
cpp_sources=cpp_source,
|
||||
cuda_sources=cuda_source,
|
||||
functions=["parametric_sigmoid_cuda"],
|
||||
extra_cuda_cflags=["-O3", "--use_fast_math"],
|
||||
verbose=False
|
||||
)
|
||||
|
||||
def forward(self, x):
|
||||
return self.op.parametric_sigmoid_cuda(x, self.alpha, self.beta)
|
||||
|
|
@ -0,0 +1,25 @@
|
|||
import torch
|
||||
import torch.nn as nn
|
||||
|
||||
|
||||
class Model(nn.Module):
|
||||
def __init__(self, alpha=1.0, beta=0.0):
|
||||
super().__init__()
|
||||
self.alpha = alpha
|
||||
self.beta = beta
|
||||
|
||||
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
||||
return torch.sigmoid(self.alpha * x + self.beta)
|
||||
|
||||
|
||||
batch_size = 128
|
||||
feature_dim = 512
|
||||
|
||||
|
||||
def get_inputs():
|
||||
x = torch.randn(batch_size, feature_dim, dtype=torch.float32)
|
||||
return [x]
|
||||
|
||||
|
||||
def get_init_inputs():
|
||||
return [1.0, 0.0]
|
||||
|
|
@ -0,0 +1,42 @@
|
|||
You write custom CUDA kernels to replace the pytorch operators in the given GeGLU 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 chunk+gelu+elementwise_mul), or algorithmic changes (such as optimized memory access patterns). You are only limited by your imagination.
|
||||
|
||||
|
||||
This CUDA kernel implements a parametric sigmoid activation with three key optimizations:
|
||||
|
||||
Vectorized Memory Operations: Uses float4 loads/stores to process 4 elements per instruction, improving memory bandwidth utilization.
|
||||
|
||||
Coalesced Memory Access: Threads access contiguous memory locations via vectorized operations, enabling efficient memory coalescing.
|
||||
|
||||
Fast Math & Loop Unrolling: Compiler flags enable fast approximate math (expf) and implicit loop unrolling improves instruction-level parallelism.
|
||||
|
||||
|
||||
|
||||
|
||||
Here's an example to show you the syntax of inline embedding custom CUDA operators in torch: The example given architecture is:
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
|
||||
|
||||
class Model(nn.Module):
|
||||
def __init__(self, alpha=1.0, beta=0.0):
|
||||
super().__init__()
|
||||
self.alpha = alpha
|
||||
self.beta = beta
|
||||
|
||||
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
||||
return torch.sigmoid(self.alpha * x + self.beta)
|
||||
|
||||
|
||||
batch_size = 128
|
||||
feature_dim = 512
|
||||
|
||||
|
||||
def get_inputs():
|
||||
x = torch.randn(batch_size, feature_dim, dtype=torch.float32)
|
||||
return [x]
|
||||
|
||||
|
||||
def get_init_inputs():
|
||||
return [1.0, 0.0]
|
||||
|
|
@ -0,0 +1,77 @@
|
|||
###########################################################
|
||||
# 性能和精度验证程序
|
||||
###########################################################
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
import time
|
||||
from ParametricSigmoid_torch import Model, get_inputs, get_init_inputs
|
||||
from ParametricSigmoid_cuda 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)
|
||||
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 torch.relu 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