- added tuning and graphics to cmake

 - replaced RTDSC timers with a more standard one
 - made gnuplot generated plots available via tune_it.sh
 - moved timing points in tune.c to avoid elimination by
   optimization

.gitignore

@@ -82,6 +82,8 @@ tuning_list
 etc/tune
 2kprime.1
 drprimes.txt
+etc/multiplying*
+etc/squaring*
 
 # ignore stuff generated by "make manual" and "make poster"
 *.aux
@@ -134,3 +136,20 @@ build*/
 # kdevelop section
 .kdev4/
 *.kdev4
+
+# ignore cmake files
+CMakeFiles
+Makefile
+cmake_install.cmake
+
+
+
+
+
+
+
+
+
+
+
+

CMakeLists.txt

@@ -32,7 +32,8 @@ include(sources.cmake)
 # Options
 #-----------------------------------------------------------------------------
 option(BUILD_SHARED_LIBS "Build shared library and only the shared library if \"ON\", default is static" OFF)
-
+option(BUILD_TUNING "Run a tuning program for the fast multiplication/squaring algorithms if \"ON\"" OFF)
+option(BUILD_GRAPHS "Run a benchmark of the fast multiplication/squaring algorithms and make graphics if \"ON\"" OFF)
 #-----------------------------------------------------------------------------
 # Compose CFLAGS
 #-----------------------------------------------------------------------------
@@ -137,6 +138,14 @@ if(BUILD_TESTING)
     add_subdirectory(demo)
 endif()
 
+#-----------------------------------------------------------------------------
+# tuning and benchmark targets
+#-----------------------------------------------------------------------------
+
+if(BUILD_TUNING OR BUILD_GRAPHS)
+    add_subdirectory(etc ${CMAKE_CURRENT_SOURCE_DIR}/etc)
+endif()
+
 #-----------------------------------------------------------------------------
 # Install/export targets and files
 #-----------------------------------------------------------------------------

demo/timing.c

@@ -55,42 +55,35 @@ static unsigned int lbit(void)
    }
 }
 
-/* RDTSC from Scott Duplichan */
-static uint64_t TIMFUNC(void)
-{
-#if defined __GNUC__
-#if defined(__i386__) || defined(__x86_64__)
-   /* version from http://www.mcs.anl.gov/~kazutomo/rdtsc.html
-    * the old code always got a warning issued by gcc, clang did not complain...
-    */
-   unsigned hi, lo;
-   __asm__ __volatile__("rdtsc" : "=a"(lo), "=d"(hi));
-   return ((uint64_t)lo)|(((uint64_t)hi)<<32);
-#else /* gcc-IA64 version */
-   unsigned long result;
-   __asm__ __volatile__("mov %0=ar.itc":"=r"(result)::"memory");
-
-   while (__builtin_expect((int) result == -1, 0))
-      __asm__ __volatile__("mov %0=ar.itc":"=r"(result)::"memory");
-
-   return result;
-#endif
 
-   /* Microsoft and Intel Windows compilers */
-#elif defined _M_IX86
-   __asm rdtsc
-#elif defined _M_AMD64
-   return __rdtsc();
-#elif defined _M_IA64
-#if defined __INTEL_COMPILER
-#include <ia64intrin.h>
+#if defined(_WIN32)
+#  include <windows.h>
 #endif
-   return __getReg(3116);
+
+static uint64_t TIMFUNC(void)
+{
+#if _POSIX_C_SOURCE >= 199309L
+#define LTM_BILLION 1000000000
+   struct timespec ts;
+
+   /* TODO: Sets errno in case of error. Use? */
+   clock_gettime(CLOCK_MONOTONIC, &ts);
+   return (((uint64_t)ts.tv_sec) * LTM_BILLION + (uint64_t)ts.tv_nsec);
+#elif defined(_WIN32)
+   LARGE_INTEGER ticks;
+   QueryPerformanceCounter(&ticks);
+   return (uint64_t)ticks.QuadPart;
 #else
-#error need rdtsc function for this build
+   clock_t t;
+   t = clock();
+   if (t < (clock_t)(0)) {
+      return (uint64_t)(0);
+   }
+   return (uint64_t)(t);
 #endif
 }
 
+
 #define DO2(x) do { mp_err err = x; err = x; (void)err; }while(0)
 #define DO4(x) DO2(x); DO2(x)
 #define DO8(x) DO4(x); DO4(x)
@@ -141,6 +134,12 @@ int main(int argc, char **argv)
    int n, cnt, ix, old_kara_m, old_kara_s, old_toom_m, old_toom_s;
    unsigned rr;
 
+#ifdef _WIN32
+   LARGE_INTEGER Frequency;
+#else
+   struct timespec ts;
+#endif
+
    CHECK_OK(mp_init(&a));
    CHECK_OK(mp_init(&b));
    CHECK_OK(mp_init(&c));
@@ -150,10 +149,21 @@ int main(int argc, char **argv)
 
    srand(LTM_TIMING_RAND_SEED);
 
-
+#ifdef _WIN32
+   QueryPerformanceFrequency(&Frequency);
+   CLK_PER_SEC = (uint64) Frequency;
+#elif _POSIX_C_SOURCE >= 199309L
+   /* returns -1 for an error and 0 for okay, sets errno (not used here) */
+   if (clock_getres(CLOCK_MONOTONIC, &ts)) {
+      fprintf(stderr, "%d, clock_getres failed\n", __LINE__);
+      exit(EXIT_FAILURE);
+   }
+   CLK_PER_SEC = LTM_BILLION / ts.tv_nsec;
+#else
    CLK_PER_SEC = TIMFUNC();
    sleep(1);
    CLK_PER_SEC = TIMFUNC() - CLK_PER_SEC;
+#endif
 
    printf("CLK_PER_SEC == %" PRIu64 "\n", CLK_PER_SEC);
 

doc/bn.tex

@@ -102,6 +102,7 @@ \section{License}
 
 \section{Building LibTomMath}
 
+
 LibTomMath is meant to be very ``GCC friendly'' as it comes with a makefile well suited for GCC.
 However, the library will also build in MSVC, Borland C out of the box.  For any other ISO C
 compiler a makefile will have to be made by the end
@@ -270,6 +271,53 @@ \subsection{Testing}
 test was invoked.  If an error is detected the program will exit with a dump of the relevant
 numbers it was working with.
 
+\subsection{CMake}
+Some of the options above are also available with CMake.
+
+\subsubsection{Shared Library}
+The default is a static library. To produce a shared library use the CMake option
+\begin{alltt}
+-DBUILD_SHARED_LIBS=ON
+\end{alltt}
+
+\subsubsection{Testing}
+To run the testsuite use option
+\begin{alltt}
+-DBUILD_TESTING=ON
+\end{alltt}
+
+\subsubsection{Tuning}
+To run the tuning itself use option
+\begin{alltt}
+-DBUILD_TUNING=ON
+\end{alltt}
+
+To run a benchmark with the tuned library and print plots of the benchmark tables use option
+\begin{alltt}
+-DBUILD_GRAPHS=ON
+\end{alltt}
+
+To compile with LTO (Link Time Optimization) use option
+\begin{alltt}
+-DCOMPILE_LTO=ON
+\end{alltt}
+
+There are several build types available:
+\begin{description}
+\item[Debug] Build a library with debugging symbols (\texttt{-g3}) and no extra optimization
+\item[Release] Build the normal release version (\texttt{-O3 -funroll-loops -fomit-frame-pointer}) (default)
+\item[RelWithDebInfo] Build a library with debugging symbols (\texttt{-g3 -O2}) and a bit of optimization
+\item[MinSizeRel] Build a small sized library (\texttt{-Os})
+\end{description}
+The build types are case-sensitive!
+
+Choose one with:
+\begin{alltt}
+-DCMAKE_BUILD_TYPE=buildtype
+\end{alltt}
+
+
+
 \section{Build Configuration}
 LibTomMath can configured at build time in two phases we shall call ``depends'' and
 ``trims''. Each phase changes how the library is built and they are applied one after another
@@ -1600,13 +1648,35 @@ \section{Tuning Polynomial Basis Routines}
 make tune
 \end{alltt}
 
-This will run a benchmark, computes the medians, rewrites \texttt{bncore.c}, and recompiles
-\texttt{bncore.c} and relinks the library.
+With CMake
+\begin{alltt}
+cmake --build /path/to/build/dir -DBUILD_TUNING=ON
+\end{alltt}
+
+
+This will run a benchmark, computes the medians, rewrites \texttt{tommath\_cutoffs.h}, recompiles
+\texttt{cutoffs.c}, and relinks the library.
 
 The benchmark itself can be fine--tuned in the file \texttt{etc/tune\_it.sh}.
 
 The program \texttt{etc/tune} is also able to print a list of values for printing curves with e.g.:
-\texttt{gnuplot}. type \texttt{./etc/tune -h} to get a list of all available options.
+\texttt{gnuplot}. Type \texttt{./etc/tune -h} to get a list of all the available options. There
+are a lot.
+
+To get some nice plots in \texttt{etc} try
+
+\begin{alltt}
+make graphs
+\end{alltt}
+
+With CMake
+\begin{alltt}
+cmake --build /path/to/build/dir -DBUILD_GRAPHS=ON
+\end{alltt}
+
+This will run a benchmark, computes the medians, rewrites \texttt{tommath\_cutoffs.h}, recompiles
+\texttt{cutoffs.c}, relinks the library and runs gnuplot to print plots in the PNG format. The size
+of the images is fixed in the file \texttt{etc/plot\_graphs.gp} and has to be changed manually.
 
 \chapter{Modular Reduction}
 

etc/CMakeLists.txt

@@ -0,0 +1,51 @@
+# SPDX-License-Identifier: Unlicense
+#
+# LibTomMath, a free open source portable number theoretic multiple-precision
+# integer (MPI) library written entirely in C.
+#
+
+cmake_minimum_required(VERSION 3.10)
+
+set(LTM_TUNE tune-ltm)
+
+# This file can be included from the top level or used stand-alone
+if(PROJECT_NAME)
+    set(LIBRARY_NAME ${PROJECT_NAME})
+else()
+    # Define an independent project and all the necessary stuff around
+    project(${LTM_TUNE}
+        LANGUAGES C)
+    set(LIBRARY_NAME libtommath)
+    find_package(${LIBRARY_NAME})
+    if(NOT CMAKE_BUILD_TYPE AND NOT CMAKE_CONFIGURATION_TYPES)
+        set(CMAKE_BUILD_TYPE "Release")
+    endif()
+endif()
+
+add_executable(tune
+    ${CMAKE_CURRENT_SOURCE_DIR}/tune.c
+)
+
+target_include_directories(tune PRIVATE
+    ${CMAKE_CURRENT_SOURCE_DIR}
+    ${CMAKE_CURRENT_SOURCE_DIR}/..
+)
+
+target_link_libraries(tune PRIVATE
+    ${LIBRARY_NAME}
+)
+
+target_compile_options(tune PRIVATE
+    ${LTM_C_FLAGS}
+)
+target_link_options(tune BEFORE PUBLIC
+    ${LTM_LD_FLAGS}
+)
+
+if(BUILD_GRAPHS)
+   # used in tune_it.sh
+   find_program(GNUPLOT gnuplot)
+   add_custom_command(TARGET tune POST_BUILD COMMAND  ${CMAKE_CURRENT_SOURCE_DIR}/tune_it.sh 1000 WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} VERBATIM)
+else()
+   add_custom_command(TARGET tune POST_BUILD COMMAND  ${CMAKE_CURRENT_SOURCE_DIR}/tune_it.sh WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} VERBATIM)
+endif()

etc/makefile

@@ -6,7 +6,7 @@ LTM_TUNE_CFLAGS = $(CFLAGS) $(LTM_CFLAGS) -Wall -W -Wextra -Wshadow -O3 -I../
 # libname when you can't install the lib with install
 LIBNAME=../libtommath.a
 
-all: pprime tune test_standalone mersenne drprime 2kprime mont
+all: pprime tune test_standalone mersenne drprime 2kprime mont getlimbsize graph
 
 #provable primes
 pprime: pprime.o
@@ -36,10 +36,15 @@ drprime: drprime.o
 mont: mont.o
 	$(CC) $(LTM_TUNE_CFLAGS) mont.o $(LIBNAME) -o mont
 
+# Make pretty pictures (1000 is the maximum number of limbs to print for mul/sqr)
+# "tune" runs twice because it runs automatically when build.
+graphs: tune
+	./tune_it.sh 1000
 
 clean:
 	rm -f *.log *.o *.obj *.exe pprime tune mersenne drprime mont 2kprime pprime.dat \
-        tuning_list multiplying squaring test *.da *.dyn *.dpi *~
+        tuning_list get_limbsize out *.da *.dyn *.dpi *~ cmake_install.cmake Makefile
 	rm -rf .libs
+	rm -rf CMakeFiles
 
 .PHONY: tune

etc/makefile.icc

@@ -32,20 +32,10 @@ tune: tune.o
 	$(CC) $(CFLAGS) tune.o $(LIBNAME) -o tune
 	./tune_it.sh
 
-# same app but using RDTSC for higher precision [requires 80586+], coff based gcc installs [e.g. ming, cygwin, djgpp]
-tune86: tune.c
-	nasm -f coff timer.asm
-	$(CC) -DX86_TIMER $(CFLAGS) tune.c timer.o  $(LIBNAME) -o tune86
-
-# for cygwin
-tune86c: tune.c
-	nasm -f gnuwin32 timer.asm
-	$(CC) -DX86_TIMER $(CFLAGS) tune.c timer.o  $(LIBNAME) -o tune86
-
-#make tune86 for linux or any ELF format
-tune86l: tune.c
-	nasm -f elf -DUSE_ELF timer.asm
-	$(CC) -DX86_TIMER $(CFLAGS) tune.c timer.o $(LIBNAME) -o tune86l
+# Make pretty pictures (1000 is the maximum number of limbs to print for mul/sqr)
+# "tune" runs twice because it runs automatically when build.
+graphs: tune
+	./tune_it.sh 1000
 
 # spits out mersenne primes
 mersenne: mersenne.o
@@ -64,4 +54,4 @@ mont: mont.o
 
 
 clean:
-	rm -f *.log *.o *.obj *.exe pprime tune mersenne drprime tune86 tune86l mont 2kprime pprime.dat *.il tuning_list
+	rm -f *.log *.o *.obj *.exe pprime tune mersenne drprime mont 2kprime pprime.dat get_limbsize *.il tuning_list

etc/plot_graphs.gp

@@ -0,0 +1,19 @@
+set term pngcairo size 720,540
+# Good for most colorblinds
+set colorsequence podo
+
+set key top left;
+
+set ylabel "Time"
+set xlabel "Operand size (limbs)"
+
+set output "multiplying".ARG1.".png";
+set title "Comparing fast and slow multiplying [".ARG1." bits limbsize]";
+plot "multiplying".ARG1."" using 1:2 w lines t "slow", "multiplying".ARG1."" using 1:3 w lines t "fast"
+
+set output "squaring".ARG1.".png";
+set title "Comparing fast and slow squaring [".ARG1." bits limbsize]";
+plot "squaring".ARG1."" using 1:2 w lines t "slow", "squaring".ARG1."" using 1:3 w lines t "fast"
+
+
+