finish dot_mse_tanh #121

S1/uucoco_#121/dot_mse_tanh_cuda.py

@@ -0,0 +1,111 @@
+import torch
+import torch.nn as nn
+from torch.utils.cpp_extension import load_inline
+
+cuda_source = """
+#include <torch/extension.h>
+#include <cuda_runtime.h>
+#include <math.h>
+
+__inline__ __device__ float warp_reduce(float val) {
+    for (int offset = 16; offset > 0; offset /= 2)
+        val += __shfl_down_sync(0xffffffff, val, offset);
+    return val;
+}
+
+__global__ void dot_mse_tanh_kernel(
+    const float* __restrict__ x, 
+    const float* __restrict__ target, 
+    float* __restrict__ y, 
+    int batch_size, 
+    int width) 
+{
+    int row = blockIdx.x;
+    int tid = threadIdx.x;
+
+    if (row >= batch_size) return;
+
+    const float* row_x = x + row * width;
+
+    float sum_dot = 0.0f;
+    float sum_sq_diff = 0.0f;
+
+    for (int i = tid; i < width; i += blockDim.x) {
+        float val_x = row_x[i];
+        float val_t = target[i];
+
+        sum_dot += val_x * val_t;
+
+        float diff = val_x - val_t;
+        sum_sq_diff += diff * diff;
+    }
+
+    sum_dot = warp_reduce(sum_dot);
+    sum_sq_diff = warp_reduce(sum_sq_diff);
+
+    static __shared__ float shared_dot[32];
+    static __shared__ float shared_sq_diff[32];
+
+    int lane = tid % 32;
+    int wid = tid / 32;
+
+    if (lane == 0) {
+        shared_dot[wid] = sum_dot;
+        shared_sq_diff[wid] = sum_sq_diff;
+    }
+    __syncthreads();
+
+    sum_dot = (tid < blockDim.x / 32) ? shared_dot[lane] : 0.0f;
+    sum_sq_diff = (tid < blockDim.x / 32) ? shared_sq_diff[lane] : 0.0f;
+
+    if (wid == 0) {
+        sum_dot = warp_reduce(sum_dot);
+        sum_sq_diff = warp_reduce(sum_sq_diff);
+    }
+
+    if (tid == 0) {
+        float mse = sum_sq_diff / (float)width;
+        y[row] = tanhf(sum_dot - mse);
+    }
+}
+
+torch::Tensor launch_dot_mse_tanh(torch::Tensor x, torch::Tensor target) {
+    auto batch_size = x.size(0);
+    auto width = x.size(1);
+    auto y = torch::empty({batch_size}, x.options());
+
+    const int threads = 256;
+    const int blocks = batch_size;
+
+    dot_mse_tanh_kernel<<<blocks, threads>>>(
+        x.data_ptr<float>(), 
+        target.data_ptr<float>(), 
+        y.data_ptr<float>(), 
+        batch_size, 
+        width
+    );
+    return y;
+}
+"""
+
+cpp_source = """
+torch::Tensor launch_dot_mse_tanh(torch::Tensor x, torch::Tensor target);
+"""
+
+dot_mse_tanh_module = load_inline(
+    name='dot_mse_tanh_op',
+    cpp_sources=cpp_source,
+    cuda_sources=cuda_source,
+    functions=['launch_dot_mse_tanh'],
+    verbose=False
+)
+
+
+class ModelNew(nn.Module):
+    def __init__(self, target):
+        super(ModelNew, self).__init__()
+        self.target = nn.Parameter(target)
+        self.op = dot_mse_tanh_module
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.op.launch_dot_mse_tanh(x.contiguous(), self.target.contiguous())

S1/uucoco_#121/dot_mse_tanh_torch.py

@@ -0,0 +1,23 @@
+import torch
+import torch.nn as nn
+
+class Model(nn.Module):
+    def __init__(self, target):
+        super(Model, self).__init__()
+        self.target = nn.Parameter(target)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        dot = torch.sum(x * self.target, dim=-1)
+        mse = torch.mean((x - self.target) ** 2, dim=-1)
+        return torch.tanh(dot - mse)
+
+batch_size = 128
+input_dim = 1024
+
+def get_inputs():
+    x = torch.randn(batch_size, input_dim)
+    return [x]
+
+def get_init_inputs():
+    target = torch.randn(input_dim)
+    return [target]

S1/uucoco_#121/prompt.txt

@@ -0,0 +1,46 @@
+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 code implements dot product + MSE + tanh activation with CUDA optimizations:
+
+Parallel reduction - Warp shuffle for two simultaneous sums: dot product and squared differences.
+
+Dual shared memory buffers - Separate buffers for dot and MSE sums to avoid bank conflicts.
+
+Fused operations - Combines dot product, MSE calculation, and tanh activation in one kernel.
+
+Grid-stride loop - Threads process multiple elements for load balancing.
+
+Single-pass computation - Computes both dot product and squared differences in one memory traversal.
+
+Memory coalescing - Contiguous tensor access patterns.
+
+Batch parallelism - One CUDA block per input row.
+
+
+
+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, target):
+        super(Model, self).__init__()
+        self.target = nn.Parameter(target)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        dot = torch.sum(x * self.target, dim=-1)
+        mse = torch.mean((x - self.target) ** 2, dim=-1)
+        return torch.tanh(dot - mse)
+
+batch_size = 128
+input_dim = 1024
+
+def get_inputs():
+    x = torch.randn(batch_size, input_dim)
+    return [x]
+
+def get_init_inputs():
+    target = torch.randn(input_dim)
+    return [target]

S1/uucoco_#121/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 dot_mse_tanh_torch import Model, get_inputs, get_init_inputs
+from dot_mse_tanh_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()