[poppler] [PATCH v2 5/5] SplashXPathScanner: Optionally use small_vector from boost

[Stefan Brüns]

Currently, each row in the intersections vector is allocated separately,
when the first intersection is added.

To avoid these allocations for common simple polygons,
boost::container::small_vector<4, T> is used, which stores up to
4 intersections inline. small_vector is a header-only class.

For the documents from fdo#96728 and fdo#78728, the runtime/memory is
significantly reduced (according to /usr/bin/time -v):
(1) $> pdftoppm -r 18 -aa no runsforever-poppler.pdf
(2) $> pdftoppm surf-types.pdf

Before/After
                                  runsforever-poppler |   surf-types
User time (seconds):                2348.08 / 1773.53 |   7.76 /  5.02
Maximum resident set size (kbytes):   46288 /   45896 |  14076 / 13748
---
 CMakeLists.txt               | 10 ++++++++++
 splash/SplashXPathScanner.cc |  2 ++
 splash/SplashXPathScanner.h  |  8 ++++++++
 3 files changed, 20 insertions(+)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index b5a86a5f..463d1e39 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -403,6 +403,11 @@ set(poppler_SRCS
 )
 set(poppler_LIBS ${FREETYPE_LIBRARIES})
 if(ENABLE_SPLASH)
+  find_package(Boost 1.58.0)
+  if(Boost_FOUND)
+    include_directories(${Boost_INCLUDE_DIRS})
+    add_definitions(-DUSE_BOOST_HEADERS)
+  endif()
   set(poppler_SRCS ${poppler_SRCS}
     poppler/SplashOutputDev.cc
     splash/Splash.cc
@@ -726,6 +731,7 @@ show_end_message_yesno("use nss3" ENABLE_NSS3)
 show_end_message_yesno("use curl" ENABLE_LIBCURL)
 show_end_message_yesno("use libopenjpeg2" WITH_OPENJPEG)
 show_end_message_yesno("use lcms2" USE_CMS)
+show_end_message_yesno("use boost" Boost_FOUND)
 show_end_message_yesno("command line utils" ENABLE_UTILS)
 show_end_message("test data dir" ${TESTDATADIR})
 
@@ -757,6 +763,10 @@ if(NOT HAVE_JPX_DECODER)
   message("Warning: You're not compiling any JPX decoder. Some files will fail to display properly.")
 endif()
 
+if(ENABLE_SPLASH AND NOT Boost_FOUND)
+  message("Warning: Use of boost is recommended for better performance.")
+endif()
+
 set(ARCHIVE_NAME ${CMAKE_PROJECT_NAME}-${POPPLER_VERSION})
 add_custom_target(dist
     COMMAND
diff --git a/splash/SplashXPathScanner.cc b/splash/SplashXPathScanner.cc
index 444b8f96..7107586d 100644
--- a/splash/SplashXPathScanner.cc
+++ b/splash/SplashXPathScanner.cc
@@ -337,9 +337,11 @@ GBool SplashXPathScanner::addIntersection(double segYMin, double segYMax,
   }
 
   auto& line = allIntersections[y - yMin];
+#ifndef USE_BOOST_HEADERS
   if (line.empty()) {
       line.reserve(4);
   }
+#endif
   line.push_back(intersect);
 
   return gTrue;
diff --git a/splash/SplashXPathScanner.h b/splash/SplashXPathScanner.h
index 37b85f4b..7405dbd8 100644
--- a/splash/SplashXPathScanner.h
+++ b/splash/SplashXPathScanner.h
@@ -28,6 +28,10 @@
 
 #include "SplashTypes.h"
 
+#ifdef USE_BOOST_HEADERS
+#include <boost/container/small_vector.hpp>
+#endif
+
 #include <vector>
 
 class SplashXPath;
@@ -106,7 +110,11 @@ private:
   int xMin, yMin, xMax, yMax;
   GBool partialClip;
 
+#ifdef USE_BOOST_HEADERS
+  typedef boost::container::small_vector<SplashIntersect, 4> IntersectionLine;
+#else
   typedef std::vector<SplashIntersect> IntersectionLine;
+#endif
   std::vector<IntersectionLine> allIntersections;
 
   unsigned int interIdx;	// current index into <inter> - used by
-- 
2.16.3