finish FunnelActivationforVisualRecognition #46

S1/gsd123_#46/FunnelActivationforVisualRecognition_cuda.py

@@ -0,0 +1,87 @@
+import torch
+import torch.nn as nn
+from torch.utils.cpp_extension import load_inline
+
+class ModelNew(nn.Module):
+    def __init__(self, tau=0.0):
+        super().__init__()
+        self.tau = tau
+        self._compile_cuda_kernel()
+
+    def _compile_cuda_kernel(self):
+        cpp_source = """
+        torch::Tensor frlu_cuda(torch::Tensor x, float tau);
+        """
+
+        cuda_source = """
+        #include <torch/extension.h>
+        #include <cuda_runtime.h>
+        #include <math.h>
+
+        __device__ __forceinline__ float frlu_op(float x, float tau) {
+            // f(x) = max(x, tau)
+            return fmaxf(x, tau);
+        }
+
+        __global__ void frlu_kernel(
+            const float* __restrict__ x,
+            float* __restrict__ output,
+            const int n_elements,
+            const float tau) 
+        {
+            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 = frlu_op(v.x, tau);
+                r.y = frlu_op(v.y, tau);
+                r.z = frlu_op(v.z, tau);
+                r.w = frlu_op(v.w, tau);
+
+                out_vec[i] = r;
+            }
+
+            const int tail_start = vec_loops << 2;
+            for (int i = tail_start + tid; i < n_elements; i += stride) {
+                output[i] = frlu_op(x[i], tau);
+            }
+        }
+
+        torch::Tensor frlu_cuda(torch::Tensor x, float tau) {
+            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);
+
+            frlu_kernel<<<blocks, threads>>>(
+                x_c.data_ptr<float>(),
+                output.data_ptr<float>(),
+                n_elements,
+                tau
+            );
+
+            return output;
+        }
+        """
+
+        self.op = load_inline(
+            name="frlu_v1",
+            cpp_sources=cpp_source,
+            cuda_sources=cuda_source,
+            functions=["frlu_cuda"],
+            extra_cuda_cflags=["-O3", "--use_fast_math"],
+            verbose=False
+        )
+
+    def forward(self, x):
+        return self.op.frlu_cuda(x, self.tau)

S1/gsd123_#46/FunnelActivationforVisualRecognition_torch.py

@@ -0,0 +1,24 @@
+import torch
+import torch.nn as nn
+
+
+class Model(nn.Module):
+    def __init__(self, tau=0.0):
+        super().__init__()
+        self.tau = tau
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return torch.max(x, torch.tensor(self.tau, dtype=x.dtype, device=x.device))
+
+
+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 [0.0]

S1/gsd123_#46/prompt.txt

@@ -0,0 +1,53 @@
+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.
+
+Just-In-Time (JIT) Compilation: CUDA kernel compiled at runtime via load_inline.
+
+Vectorized Memory Access: Uses float4 for reading/writing 4 floats per instruction (coalesced memory).
+
+Memory Coalescing: Contiguous memory access (x.contiguous()).
+
+Kernel Grid/Block Optimization: Fixed 256 threads per block, grid size capped at 65535 blocks.
+
+Fast Math Compiler Flags: --use_fast_math for faster approximate math (fmaxf).
+
+Restrict Pointers: __restrict__ to avoid pointer aliasing.
+
+Read-Only Caching: __ldg() for cached constant memory reads.
+
+Tail Processing: Handles leftover elements after vectorized loops.
+
+Element-wise Custom Operator: FReLU (Funnel ReLU) function f(x) = max(x, tau).
+
+Inline Helper Function: frlu_op marked __forceinline__.
+
+Kernel Parameters: Passes tau as a constant to 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, tau=0.0):
+        super().__init__()
+        self.tau = tau
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return torch.max(x, torch.tensor(self.tau, dtype=x.dtype, device=x.device))
+
+
+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 [0.0]

S1/gsd123_#46/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 FunnelActivationforVisualRecognition_torch import Model, get_inputs, get_init_inputs
+from FunnelActivationforVisualRecognition_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()