[PyTorch] Fix MobileDebugInfo vector copy (#64030)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/64030

ghstack-source-id: 137566816

Test Plan:
Pixel 3 before:  https://our.intern.facebook.com/intern/aibench/details/320277034999340
Pixel 3 after: https://our.intern.facebook.com/intern/aibench/details/724509739115867

can see the vector copy disappear in the flame graph. Overall mean decreased from 354 ms to 348 ms (though I'm not sure if this is outside usual noise).

Reviewed By: raziel

Differential Revision: D30559032

fbshipit-source-id: 6d8bb5396d3449cc63023ee7acf694b5d146ddc1

diff --git a/torch/csrc/jit/mobile/debug_info.cpp b/torch/csrc/jit/mobile/debug_info.cpp
index a75ffe1..50bfd8a 100644 (file)
--- a/torch/csrc/jit/mobile/debug_info.cpp
+++ b/torch/csrc/jit/mobile/debug_info.cpp
@@ -122,13 +122,14 @@ MobileDebugTable::MobileDebugTable(
       size_t debug_size{0};
       std::tie(debug_data, debug_size) = reader->getRecord(record_name);
       auto ivalues =
-          jit::unpickle(
-              reinterpret_cast<const char*>(debug_data.get()), debug_size)
-              .toTuple()
-              ->elements();
+          std::move(
+              *jit::unpickle(
+                   reinterpret_cast<const char*>(debug_data.get()), debug_size)
+                   .toTuple())
+              .elements();
       SourceRangeDeserializer deserializer;
       for (auto& val : ivalues) {
-        auto tup_elems = val.toTuple()->elements();
+        auto tup_elems = std::move(*std::move(val).toTuple()).elements();
         // For BC we decode only tuples with 3 elements
         // assuming it contains
         // byte_offset, debug_handle (=source range tag), source range