issue/367 - Fix compile bug on cuda 13.0

* fix compile bug on cuda 13.0

* issue/367 - clang format code on ubuntu

---------

Co-authored-by: root <root@Crapromer>

Author: crapromer

src/infiniop/ops/logsoftmax/cuda/kernel.cuh

@@ -54,7 +54,11 @@ __device__ void logSoftmaxKernel(
             }
         }
     }
+#if CUDART_VERSION >= 12090
+    max_val = BlockReduce(temp_storage).Reduce(max_val, ::cuda::maximum());
+#else
     max_val = BlockReduce(temp_storage).Reduce(max_val, cub::Max());
+#endif
     if (tid == 0) {
         shared_max_val = max_val;
     }

src/infiniop/ops/random_sample/nvidia/random_sample_kernel.cuh

@@ -89,9 +89,13 @@ utils::Result<size_t> calculateWorkspace(size_t n_) {
         nullptr, size_inclusive_sum,
         nullptr, n,
         nullptr));
+#if CUDART_VERSION >= 12090
+    size_random += ::cuda::maximum()(size_radix_sort, size_inclusive_sum);
+    return utils::Result<size_t>(::cuda::maximum()(argmax, size_random));
+#else
     size_random += cub::Max()(size_radix_sort, size_inclusive_sum);
-
     return utils::Result<size_t>(cub::Max()(argmax, size_random));
+#endif
 }
 
 // ↑↑↑ 计算 workspace
@@ -161,8 +165,13 @@ static __global__ void randomSampleKernel(
     const Tidx *__restrict__ indices_out,
     size_t n,
     float random, float topp, size_t topk) {
+#if CUDART_VERSION >= 12090
+    topk = ::cuda::minimum()(topk, n);
+    auto p = (Tval)(random * ::cuda::minimum()(topp * (float)sorted[n - 1], (float)sorted[topk - 1]));
+#else
     topk = cub::Min()(topk, n);
     auto p = (Tval)(random * cub::Min()(topp * (float)sorted[n - 1], (float)sorted[topk - 1]));
+#endif
     for (size_t i = 0;; ++i) {
         if ((sorted[i]) >= p) {
             *result = indices_out[i];
@@ -228,8 +237,11 @@ struct Algo {
 
         workspace_ = reinterpret_cast<void *>(workspace);
         workspace_size = workspace_end - workspace;
-
+#if CUDART_VERSION >= 12090
+        auto block = ::cuda::minimum()((size_t)block_size, n);
+#else
         auto block = cub::Min()((size_t)block_size, n);
+#endif
         auto grid = (n + block - 1) / block;
         // sort
         fillIndices<<<static_cast<unsigned int>(grid), static_cast<unsigned int>(block), 0, stream>>>(indices, static_cast<int>(n));

src/infiniop/ops/topksoftmax/cuda/kernel.cuh

@@ -55,7 +55,11 @@ __global__ void softmax_topk_row_kernel(float *values_topk, // 输出数据, 形
 
     {
         __shared__ typename BlockReduce::TempStorage temp_storage_max;
+#if CUDART_VERSION >= 12090
+        T value_max = BlockReduce(temp_storage_max).Reduce(thread_max, ::cuda::maximum());
+#else
         T value_max = BlockReduce(temp_storage_max).Reduce(thread_max, cub::Max());
+#endif
         if (tid == 0) {
             shared_max = value_max;
         }

src/infiniop/reduce/cuda/reduce.cuh

@@ -54,8 +54,12 @@ __device__ __forceinline__ Tdata max(const Tdata *data_ptr, size_t count) {
     for (size_t i = threadIdx.x; i < count; i += BLOCK_SIZE) {
 #ifdef ENABLE_HYGON_API
         max_ = (data_ptr[i] > max_) ? data_ptr[i] : max_;
+#else
+#if CUDART_VERSION >= 12090
+        max_ = ::cuda::maximum()(max_, data_ptr[i]);
 #else
         max_ = cub::Max()(max_, data_ptr[i]);
+#endif
 #endif
     }
 
@@ -65,9 +69,13 @@ __device__ __forceinline__ Tdata max(const Tdata *data_ptr, size_t count) {
 #ifdef ENABLE_HYGON_API
     return BlockReduce(temp_storage).Reduce(
         max_, [](const Tdata &a, const Tdata &b) { return (a > b) ? a : b; }, BLOCK_SIZE);
+#else
+#if CUDART_VERSION >= 12090
+    return BlockReduce(temp_storage).Reduce(max_, ::cuda::maximum(), BLOCK_SIZE);
 #else
     return BlockReduce(temp_storage).Reduce(max_, cub::Max(), BLOCK_SIZE);
 #endif
+#endif
 }
 
 } // namespace op::common_cuda::reduce_op