From ec7367710e13aaac2ce99c1298c98e062b061eba Mon Sep 17 00:00:00 2001 From: uucoco Date: Tue, 2 Dec 2025 21:04:47 +0800 Subject: [PATCH] finish PELU #40 --- S1/uucoco_#40/PELU_cuda.py | 88 +++++++++++++++++++++++++++++++++++++ S1/uucoco_#40/PELU_torch.py | 29 ++++++++++++ S1/uucoco_#40/prompt.txt | 51 +++++++++++++++++++++ S1/uucoco_#40/run_code.py | 77 ++++++++++++++++++++++++++++++++ 4 files changed, 245 insertions(+) create mode 100644 S1/uucoco_#40/PELU_cuda.py create mode 100644 S1/uucoco_#40/PELU_torch.py create mode 100644 S1/uucoco_#40/prompt.txt create mode 100644 S1/uucoco_#40/run_code.py diff --git a/S1/uucoco_#40/PELU_cuda.py b/S1/uucoco_#40/PELU_cuda.py new file mode 100644 index 0000000..9df0554 --- /dev/null +++ b/S1/uucoco_#40/PELU_cuda.py @@ -0,0 +1,88 @@ +import torch +import torch.nn as nn +from torch.utils.cpp_extension import load_inline + +class ModelNew(nn.Module): + def __init__(self, a=1.0, b=1.0): + super().__init__() + self.a = a + self.b = b + self._compile_cuda_kernel() + + def _compile_cuda_kernel(self): + cpp_source = """ + torch::Tensor pelu_cuda(torch::Tensor x, float a, float b); + """ + + cuda_source = """ + #include + #include + #include + + __device__ __forceinline__ float pelu_op(float x, float a, float b) { + return (x >= 0.0f) ? ((a / b) * x) : (a * expm1f(x / b)); + } + + __global__ void pelu_kernel( + const float* __restrict__ x, + float* __restrict__ output, + const int n_elements, + const float a, + const float b) + { + 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(x); + float4* out_vec = reinterpret_cast(output); + + for (int i = tid; i < vec_loops; i += stride) { + float4 v = __ldg(&x_vec[i]); + float4 r; + + r.x = pelu_op(v.x, a, b); + r.y = pelu_op(v.y, a, b); + r.z = pelu_op(v.z, a, b); + r.w = pelu_op(v.w, a, b); + + out_vec[i] = r; + } + + const int tail_start = vec_loops << 2; + for (int i = tail_start + tid; i < n_elements; i += stride) { + output[i] = pelu_op(x[i], a, b); + } + } + + torch::Tensor pelu_cuda(torch::Tensor x, float a, float b) { + 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); + + pelu_kernel<<>>( + x_c.data_ptr(), + output.data_ptr(), + n_elements, + a, b + ); + + return output; + } + """ + + self.op = load_inline( + name="pelu_v1", + cpp_sources=cpp_source, + cuda_sources=cuda_source, + functions=["pelu_cuda"], + extra_cuda_cflags=["-O3", "--use_fast_math"], + verbose=False + ) + + def forward(self, x): + return self.op.pelu_cuda(x, self.a, self.b) \ No newline at end of file diff --git a/S1/uucoco_#40/PELU_torch.py b/S1/uucoco_#40/PELU_torch.py new file mode 100644 index 0000000..e6eff3c --- /dev/null +++ b/S1/uucoco_#40/PELU_torch.py @@ -0,0 +1,29 @@ +import torch +import torch.nn as nn + + +class Model(nn.Module): + def __init__(self, a=1.0, b=1.0): + super().__init__() + self.a = a + self.b = b + + def forward(self, x: torch.Tensor) -> torch.Tensor: + return torch.where( + x >= 0, + (self.a / self.b) * x, + self.a * torch.expm1(x / self.b) + ) + + +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, 1.0] \ No newline at end of file diff --git a/S1/uucoco_#40/prompt.txt b/S1/uucoco_#40/prompt.txt new file mode 100644 index 0000000..5f784e7 --- /dev/null +++ b/S1/uucoco_#40/prompt.txt @@ -0,0 +1,51 @@ +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 Exponential Linear Unit (PELU) activation function with the following optimizations: +Vectorization: Uses float4memory operations to process 4 elements per thread, significantly increasing memory throughput by leveraging vector loads/stores. +Cache Optimization: Employs __ldg()intrinsic for read-only data to leverage GPU's texture cache and improve memory access patterns. +Memory Coalescing: Accesses contiguous memory blocks through vector operations, optimizing GPU memory bandwidth utilization. +Grid-Stride Loop: Handles arbitrary-sized tensors efficiently by having threads process multiple elements with strided indexing. +Tail Processing: Separately handles non-multiple-of-4 elements after vectorized operations to ensure complete data processing. +Fast Math Optimization: Uses --use_fast_mathcompiler flag and expm1f()intrinsic for optimized exponential-minus-one calculation. +Mathematical Function: Implements PELU activation with learnable parameters: +Positive region​ (x ≥ 0): (a/b) × x(scaled linear) +Negative region​ (x < 0): a × expm1(x/b)(parameterized exponential) +Numerical Stability: Uses expm1f()for accurate computation of exp(x/b) - 1when x/bis near zero. +Parameterized Activation: Supports two learnable parameters (aand b) passed directly to the CUDA kernel, enabling flexible activation shaping. +Branch Prediction: Uses conditional branching for positive/negative regions, efficiently handled by GPU warp schedulers. +Occupancy Optimization: Configures 256 threads per block and dynamically calculates grid size based on vectorized element count (threads × 4) to maximize GPU occupancy. +Compiler Optimizations: Enabled with -O3flag for aggressive performance optimization. +Inlined Device Function: The core piecewise operation is marked with __forceinline__to eliminate function call overhead within the kernel. + +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, a=1.0, b=1.0): + super().__init__() + self.a = a + self.b = b + + def forward(self, x: torch.Tensor) -> torch.Tensor: + return torch.where( + x >= 0, + (self.a / self.b) * x, + self.a * torch.expm1(x / self.b) + ) + + +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, 1.0] \ No newline at end of file diff --git a/S1/uucoco_#40/run_code.py b/S1/uucoco_#40/run_code.py new file mode 100644 index 0000000..1d9de0a --- /dev/null +++ b/S1/uucoco_#40/run_code.py @@ -0,0 +1,77 @@ +########################################################### +# 性能和精度验证程序 +########################################################### +import torch +import torch.nn as nn +import time +from PELU_torch import Model, get_inputs, get_init_inputs +from PELU_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() \ No newline at end of file