minor: cleanup sgl-kernel (#2679)

sgl-kernel/CMakeLists.txt

@@ -28,7 +28,6 @@ find_package(Torch REQUIRED)
 
 # Warp Reduce library
 add_library(_kernels SHARED
-    src/sgl-kernel/csrc/warp_reduce_kernel.cu
     src/sgl-kernel/csrc/trt_reduce_internal.cu
     src/sgl-kernel/csrc/trt_reduce_kernel.cu
     src/sgl-kernel/csrc/moe_align_kernel.cu

sgl-kernel/setup.py

@@ -1,6 +1,3 @@
-import os
-import shutil
-import zipfile
 from pathlib import Path
 
 from setuptools import setup
@@ -16,39 +13,6 @@ def get_version():
                 return line.split("=")[1].strip().strip('"')
 
 
-def rename_wheel():
-    if not os.environ.get("CUDA_VERSION"):
-        return
-    cuda_version = os.environ["CUDA_VERSION"].replace(".", "")
-    base_version = get_version()
-
-    wheel_dir = Path("dist")
-    old_wheel = next(wheel_dir.glob("*.whl"))
-    tmp_dir = wheel_dir / "tmp"
-    tmp_dir.mkdir(exist_ok=True)
-
-    with zipfile.ZipFile(old_wheel, "r") as zip_ref:
-        zip_ref.extractall(tmp_dir)
-
-    old_info = tmp_dir / f"sgl_kernel-{base_version}.dist-info"
-    new_info = tmp_dir / f"sgl_kernel-{base_version}.post0+cu{cuda_version}.dist-info"
-    old_info.rename(new_info)
-
-    platform = "manylinux2014_x86_64"
-    new_wheel = wheel_dir / old_wheel.name.replace("linux_x86_64", platform)
-    new_wheel = wheel_dir / new_wheel.name.replace(
-        base_version, f"{base_version}.post0+cu{cuda_version}"
-    )
-
-    with zipfile.ZipFile(new_wheel, "w", zipfile.ZIP_DEFLATED) as new_zip:
-        for file_path in tmp_dir.rglob("*"):
-            if file_path.is_file():
-                new_zip.write(file_path, file_path.relative_to(tmp_dir))
-
-    old_wheel.unlink()
-    shutil.rmtree(tmp_dir)
-
-
 def update_wheel_platform_tag():
     wheel_dir = Path("dist")
     old_wheel = next(wheel_dir.glob("*.whl"))
@@ -81,7 +45,6 @@ ext_modules = [
     CUDAExtension(
         name="sgl_kernel.ops._kernels",
         sources=[
-            "src/sgl-kernel/csrc/warp_reduce_kernel.cu",
             "src/sgl-kernel/csrc/trt_reduce_internal.cu",
             "src/sgl-kernel/csrc/trt_reduce_kernel.cu",
             "src/sgl-kernel/csrc/moe_align_kernel.cu",

sgl-kernel/src/sgl-kernel/__init__.py

@@ -3,12 +3,10 @@ from sgl_kernel.ops import (
     custom_reduce,
     init_custom_reduce,
     moe_align_block_size,
-    warp_reduce,
 )
 
 __all__ = [
     "moe_align_block_size",
-    "warp_reduce",
     "init_custom_reduce",
     "custom_dispose",
     "custom_reduce",

sgl-kernel/src/sgl-kernel/csrc/sgl_kernel_ops.cu

@@ -1,13 +1,5 @@
 #include "utils.hpp"
 
-// warp_reduce
-torch::Tensor warp_reduce_cuda(torch::Tensor input);
-
-torch::Tensor warp_reduce(torch::Tensor input) {
-  CHECK_CUDA_INPUT(input);
-  return warp_reduce_cuda(input);
-}
-
 // trt_reduce
 using fptr_t = int64_t;
 fptr_t init_custom_ar(int64_t rank_id, int64_t world_size, const std::vector<fptr_t>& buffers,
@@ -21,8 +13,6 @@ void moe_align_block_size(torch::Tensor topk_ids, int64_t num_experts, int64_t b
                           torch::Tensor token_cnts_buffer, torch::Tensor cumsum_buffer);
 
 PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
-  // warp_reduce
-  m.def("reduce", &warp_reduce, "Warp Reduce (CUDA)");
   // trt_reduce
   m.def("init_custom_ar", &init_custom_ar, "init custom allreduce meta (CUDA)");
   m.def("dispose", &dispose, "dispose custom allreduce meta");

sgl-kernel/src/sgl-kernel/csrc/warp_reduce_kernel.cu

@@ -1,90 +0,0 @@
-#include <cuda.h>
-#include <cuda_runtime.h>
-
-#include "utils.hpp"
-
-#define FINAL_MASK 0xffffffff
-#define BLOCK_SIZE 256
-
-template <typename scalar_t>
-__device__ __forceinline__ scalar_t add(scalar_t a, scalar_t b) {
-  return a + b;
-}
-
-template <typename scalar_t>
-__device__ __forceinline__ scalar_t warpReduceSum(scalar_t val) {
-#pragma unroll
-  for (int offset = 16; offset > 0; offset /= 2) {
-    val += __shfl_down_sync(FINAL_MASK, val, offset);
-  }
-  return val;
-}
-
-template <typename scalar_t>
-__device__ __forceinline__ scalar_t blockReduceSum(scalar_t val) {
-  __shared__ scalar_t shared[32];
-  int lane = threadIdx.x % 32;
-  int wid = threadIdx.x / 32;
-
-  val = warpReduceSum(val);  // First reduce within warp
-
-  if (lane == 0) shared[wid] = val;  // Write reduced value to shared memory
-
-  __syncthreads();  // Wait for all partial reductions
-
-  // Read from shared memory only if that warp existed
-  val = (threadIdx.x < (blockDim.x / 32)) ? shared[lane] : 0;
-
-  if (wid == 0) val = warpReduceSum(val);  // Final reduce within first warp
-
-  return val;
-}
-
-template <typename scalar_t>
-__global__ void warp_reduce_cuda_kernel(
-    const torch::PackedTensorAccessor32<scalar_t, 1, torch::RestrictPtrTraits> input,
-    torch::PackedTensorAccessor32<scalar_t, 1, torch::RestrictPtrTraits> output, int N) {
-  scalar_t sum = 0;
-
-  // Grid-stride loop
-  for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < N; i += blockDim.x * gridDim.x) {
-    sum += input[i];
-  }
-
-  // Perform block-wide reduction
-  sum = blockReduceSum(sum);
-
-  // Write result for this block to global memory
-  if (threadIdx.x == 0) {
-    output[blockIdx.x] = sum;
-  }
-}
-
-torch::Tensor warp_reduce_cuda(torch::Tensor input) {
-  // Input validation
-  TORCH_CHECK(input.dim() == 1, "1D tensor expected");
-  TORCH_CHECK(input.is_cuda(), "CUDA tensor expected");
-
-  const auto N = input.size(0);
-
-  // Handle empty tensor
-  if (N == 0) {
-    return torch::zeros({1}, input.options());
-  }
-
-  // Calculate grid dimensions
-  const int threads = BLOCK_SIZE;
-  const int blocks = (N + threads - 1) / threads;
-
-  // Allocate output tensor for partial sums
-  auto output = torch::empty({blocks}, input.options());
-
-  AT_DISPATCH_FLOATING_TYPES(input.scalar_type(), "warp_reduce_cuda", ([&] {
-                               warp_reduce_cuda_kernel<scalar_t><<<blocks, threads>>>(
-                                   input.packed_accessor32<scalar_t, 1, torch::RestrictPtrTraits>(),
-                                   output.packed_accessor32<scalar_t, 1, torch::RestrictPtrTraits>(), N);
-                             }));
-
-  // Sum the partial results
-  return output.sum();
-}

sgl-kernel/src/sgl-kernel/ops/__init__.py

@@ -2,11 +2,6 @@ from sgl_kernel.ops._kernels import all_reduce as _all_reduce
 from sgl_kernel.ops._kernels import dispose as _dispose
 from sgl_kernel.ops._kernels import init_custom_ar as _init_custom_ar
 from sgl_kernel.ops._kernels import moe_align_block_size as _moe_align_block_size
-from sgl_kernel.ops._kernels import reduce as _reduce
-
-
-def warp_reduce(input_tensor):
-    return _reduce(input_tensor)
 
 
 def init_custom_reduce(rank_id, num_devices, buffers, barrier_in, barrier_out):