Optimize fdstream so performance is on par with fstream

3 files changed, 263 insertions, 5 deletions

diff --git a/butl/fdstream b/butl/fdstream
index bb5364b..49b3d4d 100644
--- a/butl/fdstream
+++ b/butl/fdstream
@@ -86,13 +86,16 @@ namespace butl
     virtual int
     sync ();
 
+    virtual std::streamsize
+    xsputn (const char_type*, std::streamsize);
+
   private:
     bool
     save ();
 
   private:
     int fd_ = -1;
-    char buf_[4096];
+    char buf_[8192];
   };
 
   // File stream mode.
diff --git a/butl/fdstream.cxx b/butl/fdstream.cxx
index cff9b59..ac5aa79 100644
--- a/butl/fdstream.cxx
+++ b/butl/fdstream.cxx
@@ -6,7 +6,8 @@
 
 #ifndef _WIN32
 #  include <fcntl.h>     // open(), O_*
-#  include <unistd.h>    // close(), read(), write(), lseek()
+#  include <unistd.h>    // close(), read(), write(), lseek(), ssize_t
+#  include <sys/uio.h>   // writev(), iovec
 #  include <sys/stat.h>  // S_I*
 #  include <sys/types.h> // off_t
 #else
@@ -24,6 +25,7 @@
 #include <new>          // bad_alloc
 #include <limits>       // numeric_limits
 #include <cassert>
+#include <cstring>      // memcpy(), memmove()
 #include <exception>    // uncaught_exception()
 #include <stdexcept>    // invalid_argument
 #include <type_traits>
@@ -178,6 +180,124 @@ namespace butl
     return true;
   }
 
+  streamsize fdbuf::
+  xsputn (const char_type* s, streamsize sn)
+  {
+    // To avoid futher 'signed/unsigned comparison' compiler warnings.
+    //
+    size_t n (static_cast<size_t> (sn));
+
+    // Buffer the data if there is enough space.
+    //
+    size_t an (epptr () - pptr ()); // Amount of free space in the buffer.
+    if (n <= an)
+    {
+      memcpy (pptr (), s, n);
+      pbump (n);
+      return n;
+    }
+
+    size_t bn (pptr () - pbase ()); // Buffered data size.
+
+#ifndef _WIN32
+
+    ssize_t r;
+    if (bn > 0)
+    {
+      // Write both buffered and new data with a single system call.
+      //
+      iovec iov[2] = {{pbase (), bn}, {const_cast<char*> (s), n}};
+      r = writev (fd_, iov, 2);
+    }
+    else
+      r = write (fd_, s, n);
+
+    if (r == -1)
+      throw_ios_failure (errno);
+
+    size_t m (static_cast<size_t> (r));
+
+    // If the buffered data wasn't fully written then move the unwritten part
+    // to the beginning of the buffer.
+    //
+    if (m < bn)
+    {
+      memmove (pbase (), pbase () + m, bn - m);
+      pbump (-m); // Note that pbump() accepts negatives.
+      return 0;
+    }
+
+    setp (buf_, buf_ + sizeof (buf_) - 1);
+    return m - bn;
+
+#else
+
+    // On Windows there is no writev() available so sometimes we will make two
+    // system calls. Fill and flush the buffer, then try to fit the data tail
+    // into the empty buffer. If the data tail is too long then just write it
+    // to the file and keep the buffer empty.
+    //
+    // We will end up with two _write() calls if the total data size to be
+    // written exceeds double the buffer size. In this case the buffer filling
+    // is redundant so let's pretend there is no free space in the buffer, and
+    // so buffered and new data will be written separatelly.
+    //
+    if (bn + n > 2 * (bn + an))
+      an = 0;
+    else
+    {
+      memcpy (pptr (), s, an);
+      pbump (an);
+    }
+
+    // Flush the buffer.
+    //
+    size_t wn (bn + an);
+    int r (wn > 0 ? _write (fd_, buf_, wn) : 0);
+
+    if (r == -1)
+      throw_ios_failure (errno);
+
+    size_t m (static_cast<size_t> (r));
+
+    // If the buffered data wasn't fully written then move the unwritten part
+    // to the beginning of the buffer.
+    //
+    if (m < wn)
+    {
+      memmove (pbase (), pbase () + m, wn - m);
+      pbump (-m); // Note that pbump() accepts negatives.
+      return m < bn ? 0 : m - bn;
+    }
+
+    setp (buf_, buf_ + sizeof (buf_) - 1);
+
+    // Now 'an' holds the size of the data portion written as a part of the
+    // buffer flush.
+    //
+    s += an;
+    n -= an;
+
+    // Buffer the data tail if it fits the buffer.
+    //
+    if (n <= static_cast<size_t> (epptr () - pbase ()))
+    {
+      memcpy (pbase (), s, n);
+      pbump (n);
+      return sn;
+    }
+
+    // The data tail doesn't fit the buffer so write it to the file.
+    //
+    r = _write (fd_, s, n);
+
+    if (r == -1)
+      throw_ios_failure (errno);
+
+    return an + r;
+#endif
+  }
+
   // fdstream_base
   //
   fdstream_base::
@@ -477,9 +597,9 @@ namespace butl
     if (mode (fdopen_mode::at_end))
     {
 #ifndef _WIN32
-      bool r (lseek(fd, 0, SEEK_END) != static_cast<off_t>(-1));
+      bool r (lseek (fd, 0, SEEK_END) != static_cast<off_t> (-1));
 #else
-      bool r (_lseek(fd, 0, SEEK_END) != -1);
+      bool r (_lseek (fd, 0, SEEK_END) != -1);
 #endif
 
       // Note that in the case of an error we don't delete the newly created
diff --git a/tests/fdstream/driver.cxx b/tests/fdstream/driver.cxx
index 03eab80..2e3bac5 100644
--- a/tests/fdstream/driver.cxx
+++ b/tests/fdstream/driver.cxx
@@ -4,12 +4,17 @@
 
 #include <ios>
 #include <string>
+#include <vector>
+#include <iomanip>
 #include <cassert>
 #include <sstream>
+#include <fstream>
+#include <iostream>
 #include <exception>
 
 #include <butl/path>
 #include <butl/fdstream>
+#include <butl/timestamp>
 #include <butl/filesystem>
 
 using namespace std;
@@ -38,9 +43,60 @@ to_file (const path& f, const string& s, fdopen_mode m = fdopen_mode::none)
   ofs.close ();
 }
 
+template <typename S, typename T>
+static duration
+write_time (const path& p, const T& s, size_t n)
+{
+  timestamp t (timestamp::clock::now ());
+  S os (p.string (), ofstream::out);
+  os.exceptions (S::failbit | S::badbit);
+
+  for (size_t i (0); i < n; ++i)
+  {
+    if (i > 0)
+      os << '\n'; // Important not to use endl as it syncs a stream.
+
+    os << s;
+  }
+
+  os.close ();
+  return timestamp::clock::now () - t;
+}
+
+template <typename S, typename T>
+static duration
+read_time (const path& p, const T& s, size_t n)
+{
+  vector<T> v (n);
+
+  timestamp t (timestamp::clock::now ());
+  S is (p.string (), ofstream::in);
+  is.exceptions (S::failbit | S::badbit);
+
+  for (auto& ve: v)
+    is >> ve;
+
+  assert (is.eof ());
+
+  is.close ();
+  duration d (timestamp::clock::now () - t);
+
+  for (const auto& ve: v)
+    assert (ve == s);
+
+  return d;
+}
+
 int
-main ()
+main (int argc, const char* argv[])
 {
+  if (!(argc == 1 || (argc == 2 && argv[1] == string ("-v"))))
+  {
+    cerr << "usage: " << argv[0] << " [-v]" << endl;
+    return 1;
+  }
+
+  bool v (argc == 2);
   dir_path td (dir_path::temp_directory () / dir_path ("butl-fdstream"));
 
   // Recreate the temporary directory (that possibly exists from the previous
@@ -307,5 +363,84 @@ main ()
 
 #endif
 
+  // Compare fdstream and fstream operations performance.
+  //
+  duration fwd (0);
+  duration dwd (0);
+  duration frd (0);
+  duration drd (0);
+
+  path ff (td / path ("fstream"));
+  path fd (td / path ("fdstream"));
+
+  // Make several measurements with different ordering for each benchmark to
+  // level fluctuations.
+  //
+  // Write/read ~10M-size files by 100, 1000, 10 000, 100 000 byte-length
+  // strings.
+  //
+  size_t sz (100);
+  for (size_t i (0); i < 4; ++i)
+  {
+    string s;
+    s.reserve (sz);
+
+    // Fill string with characters from '0' to 'z'.
+    //
+    for (size_t i (0); i < sz; ++i)
+      s.push_back ('0' + i % (123 - 48));
+
+    size_t n (10 * 1024 * 1024 / sz);
+
+    for (size_t i (0); i < 4; ++i)
+    {
+      if (i % 2 == 0)
+      {
+        fwd += write_time<ofstream>  (ff, s, n);
+        dwd += write_time<ofdstream> (fd, s, n);
+        frd += read_time<ifstream>   (ff, s, n);
+        drd += read_time<ifdstream>  (fd, s, n);
+      }
+      else
+      {
+        dwd += write_time<ofdstream> (fd, s, n);
+        fwd += write_time<ofstream>  (ff, s, n);
+        drd += read_time<ifdstream>  (fd, s, n);
+        frd += read_time<ifstream>   (ff, s, n);
+      }
+    }
+
+    sz *= 10;
+  }
+
+  // Write/read ~10M-size files by 64-bit integers.
+  //
+  uint64_t u (0x1234567890123456);
+  size_t n (10 * 1024 * 1024 / sizeof (u));
+
+  for (size_t i (0); i < 4; ++i)
+  {
+    if (i % 2 == 0)
+    {
+      fwd += write_time<ofstream>  (ff, u, n);
+      dwd += write_time<ofdstream> (fd, u, n);
+      frd += read_time<ifstream>   (ff, u, n);
+      drd += read_time<ifdstream>  (fd, u, n);
+    }
+    else
+    {
+      dwd += write_time<ofdstream> (fd, u, n);
+      fwd += write_time<ofstream>  (ff, u, n);
+      drd += read_time<ifdstream>  (fd, u, n);
+      frd += read_time<ifstream>   (ff, u, n);
+    }
+  }
+
+  if (v)
+    cerr << "fdstream/fstream write and read duration ratios are "
+         << fixed << setprecision (2)
+         << static_cast<double> (dwd.count ()) / fwd.count () << " and "
+         << static_cast<double> (drd.count ()) / frd.count () << endl;
+
   rmdir_r (td);
 }