Don't use pinned memory when unnecessary (#17832)

Reports of excessive GPU memory usage and errors
from cudaHostRegister. There are some cases where pinning is
not required.

perf/src/cuda_runtime.rs

@@ -146,6 +146,10 @@ impl<'a, T: Clone + Send + Sync + Default + Sized> IntoParallelIterator for &'a
 }
 
 impl<T: Clone + Default + Sized> PinnedVec<T> {
+    pub fn reserve(&mut self, size: usize) {
+        self.x.reserve(size);
+    }
+
     pub fn reserve_and_pin(&mut self, size: usize) {
         if self.x.capacity() < size {
             if self.pinned {

perf/src/packet.rs

@@ -28,11 +28,22 @@ impl Packets {
         Packets { packets }
     }
 
+    pub fn new_unpinned_with_recycler(
+        recycler: PacketsRecycler,
+        size: usize,
+        name: &'static str,
+    ) -> Self {
+        let mut packets = recycler.allocate(name);
+        packets.reserve(size);
+        Packets { packets }
+    }
+
     pub fn new_with_recycler(recycler: PacketsRecycler, size: usize, name: &'static str) -> Self {
         let mut packets = recycler.allocate(name);
         packets.reserve_and_pin(size);
         Packets { packets }
     }
+
     pub fn new_with_recycler_data(
         recycler: &PacketsRecycler,
         name: &'static str,
@@ -43,6 +54,16 @@ impl Packets {
         vec
     }
 
+    pub fn new_unpinned_with_recycler_data(
+        recycler: &PacketsRecycler,
+        name: &'static str,
+        mut packets: Vec<Packet>,
+    ) -> Self {
+        let mut vec = Self::new_unpinned_with_recycler(recycler.clone(), packets.len(), name);
+        vec.packets.append(&mut packets);
+        vec
+    }
+
     pub fn set_addr(&mut self, addr: &SocketAddr) {
         for m in self.packets.iter_mut() {
             m.meta.set_addr(&addr);
@@ -76,7 +97,7 @@ pub fn to_packets_with_destination<T: Serialize>(
     recycler: PacketsRecycler,
     dests_and_data: &[(SocketAddr, T)],
 ) -> Packets {
-    let mut out = Packets::new_with_recycler(
+    let mut out = Packets::new_unpinned_with_recycler(
         recycler,
         dests_and_data.len(),
         "to_packets_with_destination",