diff --git a/elf/icf.cc b/elf/icf.cc
index ac32319868..ea8160ee05 100644
--- a/elf/icf.cc
+++ b/elf/icf.cc
@@ -249,7 +249,7 @@ static Digest compute_digest(Context<E> &ctx, InputSection<E> &isec) {
       hash((u64)&sym);
     } else if (Subsection<E> *subsec = sym.get_subsec()) {
       hash('2');
-      hash_string(subsec->data);
+      hash((u64)subsec);
     } else if (!isec) {
       hash('3');
     } else if (isec->leader) {
@@ -298,7 +298,7 @@ static Digest compute_digest(Context<E> &ctx, InputSection<E> &isec) {
       SubsectionRef<E> &ref = isec.rel_subsections[subsec_idx++];
       hash('a');
       isec.get_addend(rel);
-      hash_string(ref.subsec->data);
+      hash((u64)ref.subsec);
     } else {
       hash_symbol(*isec.file.symbols[rel.r_sym]);
     }
diff --git a/elf/mold.h b/elf/mold.h
index d1a7265cb0..7f667d9fe2 100644
--- a/elf/mold.h
+++ b/elf/mold.h
@@ -73,18 +73,15 @@ std::ostream &operator<<(std::ostream &out, const Symbol<E> &sym);
 
 template <typename E>
 struct Subsection {
-  Subsection(MergedSection<E> *sec, std::string_view data)
-    : output_section(*sec), data(data) {}
+  Subsection(MergedSection<E> *sec) : output_section(*sec) {}
 
   Subsection(const Subsection &other)
-    : output_section(other.output_section), data(other.data),
-      offset(other.offset), alignment(other.alignment.load()),
-      is_alive(other.is_alive.load()) {}
+    : output_section(other.output_section), offset(other.offset),
+      alignment(other.alignment.load()), is_alive(other.is_alive.load()) {}
 
   inline u64 get_addr(Context<E> &ctx) const;
 
   MergedSection<E> &output_section;
-  std::string_view data;
   u32 offset = -1;
   std::atomic_uint16_t alignment = 1;
   std::atomic_bool is_alive = false;
diff --git a/elf/output-chunks.cc b/elf/output-chunks.cc
index 6e5f6825c9..2a592b9e22 100644
--- a/elf/output-chunks.cc
+++ b/elf/output-chunks.cc
@@ -1175,7 +1175,7 @@ MergedSection<E>::insert(std::string_view data, u64 hash, i64 alignment) {
 
   Subsection<E> *subsec;
   bool inserted;
-  std::tie(subsec, inserted) = map.insert(data, hash, Subsection(this, data));
+  std::tie(subsec, inserted) = map.insert(data, hash, Subsection(this));
   assert(subsec);
 
   for (u16 cur = subsec->alignment; cur < alignment;)
@@ -1193,32 +1193,38 @@ void MergedSection<E>::assign_offsets(Context<E> &ctx) {
   i64 shard_size = map.nbuckets / map.NUM_SHARDS;
 
   tbb::parallel_for((i64)0, map.NUM_SHARDS, [&](i64 i) {
-    std::vector<Subsection<E> *> subsections;
+    struct KeyVal {
+      std::string_view key;
+      Subsection<E> *val;
+    };
+
+    std::vector<KeyVal> subsections;
     subsections.reserve(shard_size);
 
     for (i64 j = shard_size * i; j < shard_size * (i + 1); j++)
       if (Subsection<E> &subsec = map.values[j]; subsec.is_alive)
-        subsections.push_back(&subsec);
+        subsections.push_back({{map.keys[j], map.sizes[j]}, &subsec});
 
     // Sort subsections to make output deterministic.
     tbb::parallel_sort(subsections.begin(), subsections.end(),
-                       [](Subsection<E> *a, Subsection<E> *b) {
-      if (a->alignment != b->alignment)
-        return a->alignment < b->alignment;
-      if (a->data.size() != b->data.size())
-        return a->data.size() < b->data.size();
-      return a->data < b->data;
+                       [](const KeyVal &a, const KeyVal &b) {
+      if (a.val->alignment != b.val->alignment)
+        return a.val->alignment < b.val->alignment;
+      if (a.key.size() != b.key.size())
+        return a.key.size() < b.key.size();
+      return a.key < b.key;
     });
 
     // Assign offsets.
     i64 offset = 0;
     i64 max_alignment = 0;
 
-    for (Subsection<E> *subsec : subsections) {
-      offset = align_to(offset, subsec->alignment);
-      subsec->offset = offset;
-      offset += subsec->data.size();
-      max_alignment = std::max<i64>(max_alignment, subsec->alignment);
+    for (KeyVal &kv : subsections) {
+      Subsection<E> &subsec = *kv.val;
+      offset = align_to(offset, subsec.alignment);
+      subsec.offset = offset;
+      offset += kv.key.size();
+      max_alignment = std::max<i64>(max_alignment, subsec.alignment);
     }
 
     sizes[i] = offset;
@@ -1260,7 +1266,7 @@ void MergedSection<E>::write_to(Context<E> &ctx, u8 *buf) {
 
     for (i64 j = shard_size * i; j < shard_size * (i + 1); j++)
       if (Subsection<E> &subsec = map.values[j]; subsec.is_alive)
-        memcpy(buf + subsec.offset, subsec.data.data(), subsec.data.size());
+        memcpy(buf + subsec.offset, map.keys[j], map.sizes[j]);
   });
 }