Fix inline compilation issue with LL (#1806)

src/device/prims_ll.h

@@ -431,7 +431,7 @@ private:
   }
 
   template <int RECV, int SEND, int SrcBuf, int DstBuf>
-  __device__ void LLGenericOp(intptr_t srcIx, intptr_t dstIx, int nelem, bool postOp) {
+  __device__ __attribute__((noinline)) void LLGenericOp(intptr_t srcIx, intptr_t dstIx, int nelem, bool postOp) {
     constexpr int SRC = SrcBuf != -1 ? 1 : 0;
     constexpr int DST = DstBuf != -1 ? 1 : 0;
     T *srcElts = SrcBuf == -1 ? nullptr : userBufs[SrcBuf] + srcIx;