Merge branch 'master' into dev

README.md

@@ -1,3 +1,74 @@
-# libpll-2
+# # Libpll-2
+
+libpll-2 is the new official fork of libpll (https://github.com/xflouris/libpll/). It implements site repeats to speed up computations.
+
+
+Please read the wiki for more information.
+
+
+
+# Projects that are already using libpll-2
+ List of projects already using libpll-2 and site repeats, and reported speedups compared with the tip pattern optimization:
+ * [RAxML-NG](https://github.com/amkozlov/raxml-ng): speedup ranges between 1.2 and 1.5 
+ * [ModelTest-NG](https://github.com/ddarriba/modeltest): speedup around 1.3
+ * [EPA-ng](https://github.com/Pbdas/epa-ng): memory footprint reduced by 30%.
+
+
+# Compilation instructions
+
+Currently, `libpll` requires that [GNU Bison](http://www.gnu.org/software/bison/)
+and [Flex](http://flex.sourceforge.net/) are installed on the target system. On
+a Debian-based Linux system, the two packages can be installed using the command
+
+`apt-get install flex bison`
+
+The library also requires that a GNU system is available as it uses several
+functions (e.g. `asprintf`) which are not present in the POSIX standard.
+This, however will change in the future in order to have a more portable
+and cross-platform library.
+
+The library can be compiled using either of the following two ways.
+
+**Cloning the repo** Clone the repo and bild the executable and documentation
+using the following commands.
+
+```bash
+git clone https://github.com/xflouris/libpll.git
+cd libpll
+./autogen.sh
+./configure
+make
+make install    # as root, otherwise run: sudo make install
+```
+
+When using the cloned repository version, you will also need
+[autoconf](https://www.gnu.org/software/autoconf/autoconf.html),
+[automake](https://www.gnu.org/software/automake/) and
+[libtool](https://www.gnu.org/software/libtool/) installed. On a Debian-based
+Linux system, the packages can be installed using the command
+
+```bash
+sudo apt-get install autotools-dev autoconf libtool
+```
+
+The library will be installed on the operating system's standard paths.  For
+some GNU/Linux distributions it might be necessary to add that standard path
+(typically `/usr/local/lib`) to `/etc/ld.so.conf` and run `ldconfig`.
+
+Microsoft Windows compatibility was tested with a cross-compiler and seems to
+work out-of-the-box using [MingW](http://www.mingw.org/).
+
+# libpll-2 license and third party licenses
+
+The libpll-2 code is currently licensed under the
+[GNU Affero General Public License version 3](http://www.gnu.org/licenses/agpl-3.0.en.html).
+Please see LICENSE.txt for details.
+
+libpll-2 includes code from several other projects. We would like to thank the
+authors for making their source code available.
+
+libpll includes code from GNU Compiler Collection distributed under the GNU
+General Public License.
+
+
 
-The future line of libpll...

src/CMakeLists.txt

@@ -11,10 +11,11 @@ endif()
 set (SSE_FLAGS "-msse3")
 set (AVX_FLAGS "-mavx")
 set (AVX2_FLAGS "-mfma -mavx2")
+set (NEON_FLAGS "-march=armv8-a+fp+simd")
 
 find_package(BISON)
 find_package(FLEX)
-set(LIBPLL_BISON_FLAGS "-y -d -p pll_utree_")
+set(LIBPLL_BISON_FLAGS "-d -p pll_utree_")
 set(LIBPLL_FLEX_FLAGS "-P pll_utree_")
 BISON_TARGET(parse_utree_t 
   ${CMAKE_CURRENT_SOURCE_DIR}/parse_utree.y ${CMAKE_CURRENT_BINARY_DIR}/parse_utree.c
@@ -23,7 +24,7 @@ FLEX_TARGET(lex_utree_t
   ${CMAKE_CURRENT_SOURCE_DIR}/lex_utree.l ${CMAKE_CURRENT_BINARY_DIR}/lex_utree.c
   COMPILE_FLAGS ${LIBPLL_FLEX_FLAGS})
 ADD_FLEX_BISON_DEPENDENCY(lex_utree_t parse_utree_t)
-set(LIBPLL_BISON_FLAGS "-y -d -p pll_rtree_")
+set(LIBPLL_BISON_FLAGS "-d -p pll_rtree_")
 set(LIBPLL_FLEX_FLAGS "-P pll_rtree_")
 BISON_TARGET(parse_rtree_t 
   ${CMAKE_CURRENT_SOURCE_DIR}/parse_rtree.y ${CMAKE_CURRENT_BINARY_DIR}/parse_rtree.c
@@ -104,8 +105,11 @@ endif ()
 if (NOT DEFINED ENABLE_AVX2)
   SET(ENABLE_AVX2 "True")
 endif ()
+if (NOT DEFINED ENABLE_SSE2NEON)
+  SET(ENABLE_SSE2NEON "True")
+endif ()
 
-# check simd installed 
+# check simd supported by the compiler 
 if (ENABLE_SSE)
   SET(_code " #include <immintrin.h>
   int main() {__m128d a = _mm_setzero_pd();  return 1;}")
@@ -142,7 +146,18 @@ if (ENABLE_AVX2)
     set(ENABLE_AVX2 "False")
   endif()
 endif()
-
+if (ENABLE_SSE2NEON)
+  SET(_code " #include <sse2neon.h>
+  int main() {__m128d a = _mm_setzero_pd();  return 1;}")
+  SET(_file ${CMAKE_CURRENT_BINARY_DIR}/testsse2neon.c)
+  FILE(WRITE "${_file}" "${_code}")
+  TRY_COMPILE(SSE2NEON_COMPILED ${CMAKE_CURRENT_BINARY_DIR} ${_file}
+    COMPILE_DEFINITIONS "${NEON_FLAGS} -I${CMAKE_CURRENT_SOURCE_DIR}")
+  if (NOT SSE2NEON_COMPILED)
+    message(STATUS "Disable sse2neon simd, because not supported")
+    set(ENABLE_SSE2NEON "False")
+  endif()
+endif()
 
 # set simd flags
 if (ENABLE_SSE)
@@ -151,7 +166,15 @@ if (ENABLE_SSE)
   message(STATUS "SSE enabled. To disable it, run cmake with -DENABLE_SSE=false")
   set(LIBPLL_SOURCES ${LIBPLL_SOURCES} ${LIBPLL_SSE_SOURCES})
   SET_SOURCE_FILES_PROPERTIES( ${LIBPLL_SSE_SOURCES} PROPERTIES COMPILE_FLAGS ${SSE_FLAGS} )
+elseif (ENABLE_SSE2NEON)
+  # atm either SSE or SSE2NEON kernels can be built, since the share the same function and object names
+  add_definitions(-DHAVE_SSE3 -DHAVE_SSE2NEON)
+  set(SIMD_FLAGS "${SIMD_FLAGS} ${NEON_FLAGS}")
+  message(STATUS "SSE2NEON enabled. To disable it, run cmake with -DENABLE_SSE2NEON=false")
+  set(LIBPLL_SOURCES ${LIBPLL_SOURCES} ${LIBPLL_SSE_SOURCES})
+  SET_SOURCE_FILES_PROPERTIES( ${LIBPLL_SSE_SOURCES} PROPERTIES COMPILE_FLAGS ${NEON_FLAGS} )
 endif ()
+
 if (ENABLE_AVX)
   add_definitions(-DHAVE_AVX)
   set(SIMD_FLAGS "${SIMD_FLAGS} ${AVX_FLAGS}")
@@ -167,7 +190,9 @@ if (ENABLE_AVX2)
   SET_SOURCE_FILES_PROPERTIES( ${LIBPLL_AVX2_SOURCES} PROPERTIES COMPILE_FLAGS ${AVX2_FLAGS} )
 endif ()
 
-add_definitions(-DHAVE_X86INTRIN_H)
+if (ENABLE_SSE OR ENABLE_AVX OR ENABLE_AVX2)
+  add_definitions(-DHAVE_X86INTRIN_H)
+endif()
 
 set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${LIBPLL_BASE_FLAGS}")
 

src/hardware.c

@@ -28,7 +28,7 @@
     https://github.com/xflouris/libpll/issues/138
 
 */
-#if (defined(__APPLE__)) || \
+#if (defined(__APPLE__) && !defined(__aarch64__)) || \
     (!defined(__clang__) && defined(__GNUC__) && (__GNUC__ < 4 || \
       (__GNUC__ == 4 && __GNUC_MINOR__ < 8))) || \
     (defined(__clang__) && (__clang_major__ < 3 || \
@@ -112,6 +112,13 @@ static void cpu_features_detect()
   pll_hardware.init            = 1;
 #if defined(__PPC__)
   pll_hardware.altivec_present = __builtin_cpu_supports("altivec");
+#elif defined(__aarch64__) && defined(HAVE_SSE2NEON)
+  pll_hardware.sse_present     = 1;
+  pll_hardware.sse2_present    = 1;
+  pll_hardware.sse3_present    = 1;
+  pll_hardware.ssse3_present   = 1;
+  pll_hardware.sse41_present   = 1;
+  pll_hardware.sse42_present   = 1;
 #elif defined(__x86_64__) || defined(__i386__)
   pll_hardware.mmx_present     = __builtin_cpu_supports("mmx");
   pll_hardware.sse_present     = __builtin_cpu_supports("sse");