refactor: 统一硬件相关头文件引用

将分散在各文件中的CUDA/HIP/MUSA硬件相关头文件引用统一到vendors目录下的对应头文件中，提高代码可维护性。移除重复的头文件引用，优化构建配置。

csrc/quantization/w8a8/cutlass/c3x/cutlass_gemm_caller.cuh

@@ -2,9 +2,9 @@
 
 // clang-format will break include orders
 // clang-format off
-#include <torch/all.h>
+#include "../../../../vendors/functions.h"
+
 
-#include <ATen/cuda/CUDAContext.h>
 
 #include "cutlass/cutlass.h"
 

csrc/quantization/w8a8/cutlass/c3x/scaled_mm_helper.hpp

@@ -1,4 +1,4 @@
-#include <torch/all.h>
+#include "../../../../vendors/functions.h"
 #include "cuda_utils.h"
 #include "cutlass_extensions/common.hpp"
 

csrc/quantization/w8a8/cutlass/c3x/scaled_mm_kernels.hpp

@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/all.h>
+#include "../../../../vendors/functions.h"
 
 namespace vllm {
 

csrc/quantization/w8a8/cutlass/moe/blockwise_scaled_group_mm_sm100.cu

@@ -1,11 +1,10 @@
 #include "core/registration.h"
 
-#include <torch/all.h>
-#include <cutlass/arch/arch.h>
+// #include <cutlass/arch/arch.h>
+
+#include "../../../../vendors/functions.h"
+
 
-#include <ATen/cuda/CUDAContext.h>
-#include <c10/cuda/CUDAGuard.h>
-#include <c10/cuda/CUDAStream.h>
 
 #include "cute/tensor.hpp"
 #include "cutlass/tensor_ref.h"

csrc/quantization/w8a8/cutlass/moe/get_group_starts.cuh

@@ -1,8 +1,6 @@
 #pragma once
 
-#include <cuda.h>
-#include <torch/all.h>
-#include <c10/cuda/CUDAStream.h>
+#include "../../../../vendors/functions.h"
 
 #include "core/scalar_type.hpp"
 #include "cutlass/bfloat16.h"

csrc/quantization/w8a8/cutlass/moe/grouped_mm_c3x.cuh

@@ -1,5 +1,7 @@
 #pragma once
 
+#include "../../../../vendors/functions.h"
+
 #include "cutlass/cutlass.h"
 
 #include "cutlass/gemm/collective/collective_builder.hpp"

csrc/quantization/w8a8/cutlass/moe/grouped_mm_c3x_sm100.cu

@@ -1,7 +1,5 @@
-#include <cudaTypedefs.h>
 
-#include <c10/cuda/CUDAGuard.h>
-#include <torch/all.h>
+#include "../../../../vendors/functions.h"
 
 #include "cutlass/cutlass.h"
 #include "grouped_mm_c3x.cuh"

csrc/quantization/w8a8/cutlass/moe/grouped_mm_c3x_sm90.cu

@@ -1,7 +1,5 @@
-#include <cudaTypedefs.h>
 
-#include <c10/cuda/CUDAGuard.h>
-#include <torch/all.h>
+#include "../../../../vendors/functions.h"
 
 #include "cutlass/cutlass.h"
 #include "grouped_mm_c3x.cuh"

csrc/quantization/w8a8/cutlass/moe/moe_data.cu

@@ -1,7 +1,4 @@
-#include <cudaTypedefs.h>
-
-#include <c10/cuda/CUDAGuard.h>
-#include <torch/all.h>
+#include "../../../../vendors/functions.h"
 
 #include <iostream>
 

csrc/quantization/w8a8/cutlass/scaled_mm_c2x.cuh

@@ -2,7 +2,7 @@
 #include <stddef.h>
 #include <torch/all.h>
 
-#include <ATen/cuda/CUDAContext.h>
+#include "../../../../vendors/functions.h"
 
 // clang-format will break include orders
 // clang-format off

csrc/quantization/w8a8/cutlass/scaled_mm_entry.cu

@@ -1,7 +1,7 @@
-#include <cudaTypedefs.h>
 
-#include <c10/cuda/CUDAGuard.h>
-#include <torch/all.h>
+#include "../../../vendors/functions.h"
+
+
 
 #include "cutlass_extensions/common.hpp"
 

csrc/quantization/w8a8/fp8/common.cu

@@ -2,8 +2,9 @@
 #include "dispatch_utils.h"
 #include "cub_helpers.h"
 #include "quantization/vectorization_utils.cuh"
-#include <c10/cuda/CUDAGuard.h>
-#include <ATen/cuda/Exceptions.h>
+#include "../../../../vendors/functions.h"
+
+
 
 namespace vllm {
 

csrc/quantization/w8a8/fp8/per_token_group_quant.cu

@@ -1,5 +1,4 @@
-#include <ATen/cuda/CUDAContext.h>
-
+#include "../../../vendors/functions.h"
 #include "quantization/w8a8/per_token_group_quant_8bit.h"
 
 #include <cmath>

csrc/quantization/w8a8/int8/per_token_group_quant.cu

@@ -1,6 +1,4 @@
-#include <ATen/cuda/CUDAContext.h>
-#include <torch/all.h>
-
+#include "../../../vendors/functions.h"
 #include "quantization/w8a8/per_token_group_quant_8bit.h"
 
 void per_token_group_quant_int8(const torch::Tensor& input,

csrc/quantization/w8a8/int8/scaled_quant.cu

@@ -1,6 +1,6 @@
-#include <ATen/cuda/CUDAContext.h>
-#include <torch/all.h>
-#include <c10/cuda/CUDAGuard.h>
+#include "../../../vendors/functions.h"
+
+
 
 #include <cmath>
 

csrc/quantization/w8a8/per_token_group_quant_8bit.h

@@ -1,6 +1,5 @@
 #pragma once
-#include <torch/all.h>
-
+#include "../../vendors/functions.h"
 // 8-bit per-token-group quantization helper used by both FP8 and INT8
 void per_token_group_quant_8bit(const torch::Tensor& input,
                                 torch::Tensor& output_q,