[lld] r314192 - Speed up SymbolTable::insert().

[Rafael Avila de Espindola via llvm-commits]

I can reproduce the speedup on various benchmarks.

But note that it is not just a faster algorithm. The old version would
produce a different result for foo at bar@@zed. That is fine for us, but we
couldn't do the same as a special case for small needles in StringRef.

Thanks,
Rafael

Rui Ueyama via llvm-commits <llvm-commits at lists.llvm.org> writes:

> Author: ruiu
> Date: Mon Sep 25 21:17:13 2017
> New Revision: 314192
>
> URL: http://llvm.org/viewvc/llvm-project?rev=314192&view=rev
> Log:
> Speed up SymbolTable::insert().
>
> SymbolTable::insert() is a hot path function. When linking a clang debug
> build, the function is called 3.7 million times. The total amount of "Name"
> string contents is 300 MiB. That means this `Name.find("@@")` scans almost
> 300 MiB of data. That's far from negligible.
>
> StringRef::find(StringRef) uses a sophisticated algorithm, but the
> function is slow for a short needle. This patch replaces it with
> StringRef::find(char).
>
> This patch alone speeds up a clang debug build link time by 0.5 seconds
> from 8.2s to 7.7s. That's 6% speed up. It seems too good for this tiny
> change, but looks like it's real.
>
> Modified:
>     lld/trunk/ELF/SymbolTable.cpp
>
> Modified: lld/trunk/ELF/SymbolTable.cpp
> URL: http://llvm.org/viewvc/llvm-project/lld/trunk/ELF/SymbolTable.cpp?rev=314192&r1=314191&r2=314192&view=diff
> ==============================================================================
> --- lld/trunk/ELF/SymbolTable.cpp (original)
> +++ lld/trunk/ELF/SymbolTable.cpp Mon Sep 25 21:17:13 2017
> @@ -206,8 +206,12 @@ static uint8_t getMinVisibility(uint8_t
>  std::pair<Symbol *, bool> SymbolTable::insert(StringRef Name) {
>    // <name>@@<version> means the symbol is the default version. In that
>    // case <name>@@<version> will be used to resolve references to <name>.
> -  size_t Pos = Name.find("@@");
> -  if (Pos != StringRef::npos)
> +  //
> +  // Since this is a hot path, the following string search code is
> +  // optimized for speed. StringRef::find(char) is much faster than
> +  // StringRef::find(StringRef).
> +  size_t Pos = Name.find('@');
> +  if (Pos != StringRef::npos && Pos + 1 < Name.size() && Name[Pos + 1] == '@')
>      Name = Name.take_front(Pos);
>  
>    auto P = Symtab.insert(
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits