Use macro guard CUDA functions for back compatibility in grouped_topk_kernel.cu (vllm-project#25346)

Signed-off-by: Ming Yang <[email protected]>
Signed-off-by: Rahul Tuli <[email protected]>
Co-authored-by: Rahul Tuli <[email protected]>
Co-authored-by: Claude <[email protected]>
Co-authored-by: Wentao Ye <[email protected]>
Co-authored-by: Lu Fang <[email protected]>
Co-authored-by: Ye (Charlotte) Qi <[email protected]>
Signed-off-by: gaojc <[email protected]>

Author: 6 people

csrc/moe/grouped_topk_kernels.cu

@@ -418,6 +418,15 @@ __device__ inline T neg_inf() {
   return cuda_cast<T, float>(-cuda::std::numeric_limits<float>::infinity());
 }
 
+template <typename T>
+__device__ inline bool is_finite(const T val) {
+#if (__CUDACC_VER_MAJOR__ * 10000 + __CUDACC_VER_MINOR__ * 100 >= 120800)
+  return cuda::std::isfinite(val);
+#else
+  return isfinite(cuda_cast<float, T>(val));
+#endif
+}
+
 template <typename T>
 __device__ void topk_with_k2(T* output, T const* input,
                              cg::thread_block_tile<32> const& tile,
@@ -533,7 +542,7 @@ __global__ void group_idx_and_topk_idx_kernel(
     // calculate group_idx
     int32_t target_num_min = WARP_SIZE - n_group + topk_group;
     // The check is necessary to avoid abnormal input
-    if (lane_id < n_group && cuda::std::isfinite(group_scores[lane_id])) {
+    if (lane_id < n_group && is_finite(group_scores[lane_id])) {
       value = group_scores[lane_id];
     }
 
@@ -568,11 +577,10 @@ __global__ void group_idx_and_topk_idx_kernel(
         int32_t offset = i_group * num_experts_per_group;
         for (int32_t i = lane_id; i < align_num_experts_per_group;
              i += WARP_SIZE) {
-          T candidates =
-              (i < num_experts_per_group) &&
-                      cuda::std::isfinite(scores_with_bias[offset + i])
-                  ? scores_with_bias[offset + i]
-                  : neg_inf<T>();
+          T candidates = (i < num_experts_per_group) &&
+                                 is_finite(scores_with_bias[offset + i])
+                             ? scores_with_bias[offset + i]
+                             : neg_inf<T>();
           queue.add(candidates, offset + i);
         }
         if (group_scores[i_group] == topk_group_value) {