From db76bed2205fc807e910af8cb783d98eb8bbd169 Mon Sep 17 00:00:00 2001 From: czurnieden Date: Sat, 4 May 2019 18:57:07 +0200 Subject: [PATCH] added autotuning functionality for the Toom-Cook cut-offs --- .travis.yml | 11 + doc/bn.tex | 48 ++- etc/makefile | 43 +-- etc/makefile.icc | 8 +- etc/makefile.msvc | 3 +- etc/tune.c | 821 ++++++++++++++++++++++++++++++++++++++------ etc/tune_it.sh | 129 +++++++ makefile | 4 + makefile.mingw | 4 + makefile.msvc | 4 + makefile.shared | 14 + makefile.unix | 5 + makefile_include.mk | 2 +- testme.sh | 32 +- 14 files changed, 959 insertions(+), 169 deletions(-) create mode 100755 etc/tune_it.sh diff --git a/.travis.yml b/.travis.yml index 56d4a60..bde3d08 100644 --- a/.travis.yml +++ b/.travis.yml @@ -141,6 +141,16 @@ matrix: - env: BUILDOPTIONS='--with-cc=clang-7 --cflags=-DMP_16BIT --with-valgrind' - env: BUILDOPTIONS='--with-cc=clang-7 --cflags=-DMP_32BIT --with-valgrind' + # Test "autotuning", the automatic evaluation and setting of the Toom-Cook cut-offs. + - env: BUILDOPTIONS='--with-cc=gcc-5 --cflags=-DMP_8BIT --with-valgrind --make-option=tune' + - env: BUILDOPTIONS='--with-cc=gcc-5 --cflags=-DMP_16BIT --with-valgrind --make-option=tune' + - env: BUILDOPTIONS='--with-cc=gcc-5 --cflags=-DMP_32BIT --with-valgrind --make-option=tune' + - env: BUILDOPTIONS='--with-cc=gcc-5 --with-valgrind --make-option=tune' + - env: BUILDOPTIONS='--with-cc=clang-7 --cflags=-DMP_8BIT --with-valgrind --make-option=tune' + - env: BUILDOPTIONS='--with-cc=clang-7 --cflags=-DMP_16BIT --with-valgrind --make-option=tune' + - env: BUILDOPTIONS='--with-cc=clang-7 --cflags=-DMP_32BIT --with-valgrind --make-option=tune' + - env: BUILDOPTIONS='--with-cc=clang-7 --with-valgrind --make-option=tune' + # GCC for the x86-64 architecture testing against a different Bigint-implementation # with 333333 different inputs. - env: BUILDOPTIONS='--with-cc=gcc-5 --test-vs-mtest=333333 --with-valgrind' @@ -151,6 +161,7 @@ matrix: - env: BUILDOPTIONS='--with-cc=gcc-5 --test-vs-mtest=333333 --mtest-real-rand --with-valgrind' - env: BUILDOPTIONS='--with-cc=clang-7 --test-vs-mtest=333333 --mtest-real-rand --with-valgrind' + # Notifications go to # An email address is also possible. notifications: diff --git a/doc/bn.tex b/doc/bn.tex index 2e00bee..ed4c2f2 100644 --- a/doc/bn.tex +++ b/doc/bn.tex @@ -110,7 +110,7 @@ ICC does not know all options available for GCC and LibTomMath uses two diagnost icc: command line warning #10148: option '-Wbad-function-cast' not supported icc: command line warning #10148: option '-Wcast-align' not supported \end{alltt} -It is possible to mute this ICC warning with the compiler flag \texttt{-diag-disable=10006}\footnote{It is not recommended to suppress warnings without a very good reason but there is no harm in doing so in this very special case.}. +It is possible to mute this ICC warning with the compiler flag \texttt{-diag-disable=10148}\footnote{It is not recommended to suppress warnings without a very good reason but there is no harm in doing so in this very special case.}. \subsection{Static Libraries} To build as a static library for GCC issue the following @@ -127,6 +127,12 @@ nmake -f makefile.msvc This will build the library and archive the object files in ``tommath.lib''. This has been tested with MSVC version 6.00 with service pack 5. +To run a program to adapt the Toom-Cook cut-off values to your architecture type +\begin{alltt} +make tune +\end{alltt} +This will take some time. + \subsection{Shared Libraries} \subsubsection{GNU based Operating Systems} To build as a shared library for GCC issue the following @@ -137,6 +143,13 @@ This requires the ``libtool'' package (common on most Linux/BSD systems). It wi and static then install (by default) into /usr/lib as well as install the header files in /usr/include. The shared library (resource) will be called ``libtommath.la'' while the static library called ``libtommath.a''. Generally you use libtool to link your application against the shared object. + +To run a program to adapt the Toom-Cook cut-off values to your architecture type +\begin{alltt} +make -f makefile.shared tune +\end{alltt} +This will take some time. + \subsubsection{Microsoft Windows based Operating Systems} There is limited support for making a ``DLL'' in windows via the ``makefile.cygwin\_dll'' makefile. It requires Cygwin to work with since it requires the auto-export/import functionality. The resulting DLL and import library @@ -1366,7 +1379,7 @@ should only be used with very large inputs. This is followed by the Karatsuba m sized inputs. Then followed by the Comba and baseline multipliers. Fortunately for the developer you don't really need to know this unless you really want to fine tune the system. mp\_mul() -will determine on its own\footnote{Some tweaking may be required.} what routine to use automatically when it is called. +will determine on its own\footnote{Some tweaking may be required but \texttt{make tune} will put some reasonable values in \texttt{bncore.c}} what routine to use automatically when it is called. \begin{alltt} int main(void) @@ -1448,34 +1461,17 @@ GCC 3.3.1 and an Athlon XP processor the cutoff point is roughly 110 digits (abo Toom-Cook has incredible overhead and is probably only useful for very large inputs. So far no known cutoff points exist and for the most part I just set the cutoff points very high to make sure they're not called. -A demo program in the ``etc/'' directory of the project called ``tune.c'' can be used to find the cutoff points. This -can be built with GCC as follows +To get reasonable values for the cut-off points for your architecture, type \begin{alltt} -make XXX +make tune \end{alltt} -Where ``XXX'' is one of the following entries from the table \ref{fig:tuning}. -\begin{figure}[h] -\begin{center} -\begin{small} -\begin{tabular}{|l|l|} -\hline \textbf{Value of XXX} & \textbf{Meaning} \\ -\hline tune & Builds portable tuning application \\ -\hline tune86 & Builds x86 (pentium and up) program for COFF \\ -\hline tune86c & Builds x86 program for Cygwin \\ -\hline tune86l & Builds x86 program for Linux (ELF format) \\ -\hline -\end{tabular} -\end{small} -\end{center} -\caption{Build Names for Tuning Programs} -\label{fig:tuning} -\end{figure} +This will run a benchmark, computes the medians, rewrites \texttt{bncore.c}, and recompiles \texttt{bncore.c} and relinks the library. -When the program is running it will output a series of measurements for different cutoff points. It will first find -good Karatsuba squaring and multiplication points. Then it proceeds to find Toom-Cook points. Note that the Toom-Cook -tuning takes a very long time as the cutoff points are likely to be very high. +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. \chapter{Modular Reduction} @@ -1846,7 +1842,7 @@ int mp_sqrt (mp_int * a, mp_digit b, mp_int * c) \chapter{Logarithm} \section{Integer Logarithm} -A logarithm function for positive integer input \texttt{a, base} computing $\floor{\log_bx}$ such that $(\ilog_bx)^b \le x$. +A logarithm function for positive integer input \texttt{a, base} computing $\floor{\log_bx}$ such that $(\log_b x)^b \le x$. \index{mp\_ilogb} \begin{alltt} int mp_ilogb(mp_int *a, mp_digit base, mp_int *c) diff --git a/etc/makefile b/etc/makefile index 99154d8..04a97fa 100644 --- a/etc/makefile +++ b/etc/makefile @@ -1,4 +1,4 @@ -CFLAGS += -Wall -W -Wshadow -O3 -fomit-frame-pointer -funroll-loops -I../ +CFLAGS += -Wall -W -Wextra -Wshadow -O3 -I../ # default lib name (requires install with root) # LIBNAME=-ltommath @@ -8,43 +8,36 @@ LIBNAME=../libtommath.a #provable primes pprime: pprime.o - $(CC) pprime.o $(LIBNAME) -o pprime + $(CC) $(CFLAGS) pprime.o $(LIBNAME) -o pprime # portable [well requires clock()] tuning app tune: tune.o - $(CC) tune.o $(LIBNAME) -o tune - -# 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 + # The actual benchmark program + $(CC) $(CFLAGS) tune.o $(LIBNAME) -o tune + # a small script to run it + /bin/sh tune_it.sh -#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 +test_standalone: tune.o + # The benchmark program works as a testtool, too + $(CC) $(CFLAGS) tune.o $(LIBNAME) -o test # spits out mersenne primes mersenne: mersenne.o - $(CC) mersenne.o $(LIBNAME) -o mersenne + $(CC) $(CFLAGS) mersenne.o $(LIBNAME) -o mersenne -# fines DR safe primes for the given config +# finds DR safe primes for the given config drprime: drprime.o - $(CC) drprime.o $(LIBNAME) -o drprime + $(CC) $(CFLAGS) drprime.o $(LIBNAME) -o drprime -# fines 2k safe primes for the given config +# finds 2k safe primes for the given config 2kprime: 2kprime.o - $(CC) 2kprime.o $(LIBNAME) -o 2kprime + $(CC) $(CFLAGS) 2kprime.o $(LIBNAME) -o 2kprime mont: mont.o - $(CC) mont.o $(LIBNAME) -o mont + $(CC) $(CFLAGS) mont.o $(LIBNAME) -o mont clean: - rm -f *.log *.o *.obj *.exe pprime tune mersenne drprime tune86 tune86l mont 2kprime pprime.dat \ - *.da *.dyn *.dpi *~ + rm -f *.log *.o *.obj *.exe pprime tune mersenne drprime mont 2kprime pprime.dat \ + tuning_list multiplying squaring test *.da *.dyn *.dpi *~ + diff --git a/etc/makefile.icc b/etc/makefile.icc index 8a1ffff..fc1512f 100644 --- a/etc/makefile.icc +++ b/etc/makefile.icc @@ -28,9 +28,11 @@ LIBNAME=../libtommath.a pprime: pprime.o $(CC) pprime.o $(LIBNAME) -o pprime -# portable [well requires clock()] tuning app tune: tune.o - $(CC) tune.o $(LIBNAME) -o tune + # The actual benchmark program + $(CC) $(CFLAGS) tune.o $(LIBNAME) -o tune + # a small script to run it + /bin/sh 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 @@ -64,4 +66,4 @@ mont: mont.o clean: - rm -f *.log *.o *.obj *.exe pprime tune mersenne drprime tune86 tune86l mont 2kprime pprime.dat *.il + rm -f *.log *.o *.obj *.exe pprime tune mersenne drprime tune86 tune86l mont 2kprime pprime.dat *.il tuning_list diff --git a/etc/makefile.msvc b/etc/makefile.msvc index 2833372..592a437 100644 --- a/etc/makefile.msvc +++ b/etc/makefile.msvc @@ -9,10 +9,11 @@ pprime: pprime.obj mersenne: mersenne.obj cl mersenne.obj ../tommath.lib - + tune: tune.obj cl tune.obj ../tommath.lib + mont: mont.obj cl mont.obj ../tommath.lib diff --git a/etc/tune.c b/etc/tune.c index 5232ed7..9038c78 100644 --- a/etc/tune.c +++ b/etc/tune.c @@ -2,141 +2,748 @@ * * Tom St Denis, tstdenis82@gmail.com */ -#include +#include "../tommath.h" +#include "../tommath_private.h" #include +#include +#include +#include +#include -/* how many times todo each size mult. Depends on your computer. For slow computers - * this can be low like 5 or 10. For fast [re: Athlon] should be 25 - 50 or so - */ -#define TIMES (1UL<<14UL) +static uint64_t s_ranval(void); +static void s_raninit(uint64_t seed); +static int s_mp_random(mp_int *a, int limbs); +static uint64_t s_timer_function(void); +static void s_timer_start(void); +static uint64_t s_timer_stop(void); +static uint64_t s_time_mul(int size); +static uint64_t s_time_sqr(int size); +static void s_usage(char *s); -#ifndef X86_TIMER +/* + Please take in mind that both multiplicands are of the same size. The balancing + mechanism in mp_balance works well but has some overhead itself. You can test + the behaviour of it with the option "-o" followed by a (small) positive number 'x' + to generate ratios of the form 1:x. +*/ -/* RDTSC from Scott Duplichan */ -static uint64_t TIMFUNC(void) +/* Bob Jenkins' http://burtleburtle.net/bob/rand/smallprng.html */ +/* Chosen for speed and a good "mix" */ +typedef struct ranctx { + uint64_t a; + uint64_t b; + uint64_t c; + uint64_t d; +} ranctx; + +static ranctx burtle_x; + +# define rot(x,k) (((x)<<(k))|((x)>>(64-(k)))) +static uint64_t s_ranval(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 + uint64_t e = burtle_x.a - rot(burtle_x.b, 7); + burtle_x.a = burtle_x.b ^ rot(burtle_x.c, 13); + burtle_x.b = burtle_x.c + rot(burtle_x.d, 37); + burtle_x.c = burtle_x.d + e; + burtle_x.d = e + burtle_x.a; + return burtle_x.d; +} - /* 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 -# endif - return __getReg(3116); -# else -# error need rdtsc function for this build -# endif +static void s_raninit(uint64_t seed) +{ + uint64_t i; + burtle_x.a = 0xf1ea5eed; + burtle_x.b = burtle_x.c = burtle_x.d = seed; + for (i = 0; i < 20; ++i) { + (void) s_ranval(); + } +} + +/* + The original used LTM's mp_rand which uses the cryptographically secure + source of the OS for its purpose. That is too expensive, too slow and + most important for a benchmark: it is not repeatable. +*/ +static int s_mp_random(mp_int *a, int limbs) +{ + int e = MP_OKAY; + if ((e = mp_grow(a, limbs + 1)) != MP_OKAY) { + goto LTM_ERR; + } + a->used = limbs--; + do { + a->dp[limbs] = (mp_digit)(s_ranval() & MP_MASK); + } while (limbs--); + mp_clamp(a); +LTM_ERR: + return e; +} + +static uint64_t s_timer_function(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); +#else + clock_t t; + t = clock(); + if (t < (clock_t)(0)) { + return (uint64_t)(0); + } + return (uint64_t)(t); +#endif +} + +/* generic ISO C timer */ +static uint64_t s_timer_tmp; +static void s_timer_start(void) +{ + s_timer_tmp = s_timer_function(); +} +static uint64_t s_timer_stop(void) +{ + return s_timer_function() - s_timer_tmp; } -/* *INDENT-OFF* */ -/* generic ISO C timer */ -static uint64_t LBL_T; -static void t_start(void) { LBL_T = TIMFUNC(); } -static uint64_t t_read(void) { return TIMFUNC() - LBL_T; } -/* *INDENT-ON* */ +static int s_check_result; +static int s_number_of_test_loops; +static int s_stabilization_extra; +static int s_offset = 1; -#else -extern void t_start(void); -extern uint64_t t_read(void); -#endif - -static uint64_t time_mult(int size, int s) +#define s_mp_mul(a, b, c) s_mp_mul_digs(a, b, c, (a)->used + (b)->used + 1) +static uint64_t s_time_mul(int size) { - unsigned long x; + int x, e; + mp_int a, b, c, d; + uint64_t t1; + + if ((e = mp_init_multi(&a, &b, &c, &d, NULL)) != MP_OKAY) { + t1 = UINT64_MAX; + goto LTM_ERR; + } + + if ((e = s_mp_random(&a, size * s_offset)) != MP_OKAY) { + t1 = UINT64_MAX; + goto LTM_ERR; + } + if ((e = s_mp_random(&b, size)) != MP_OKAY) { + t1 = UINT64_MAX; + goto LTM_ERR; + } + + s_timer_start(); + for (x = 0; x < s_number_of_test_loops; x++) { + if ((e = mp_mul(&a,&b,&c)) != MP_OKAY) { + t1 = UINT64_MAX; + goto LTM_ERR; + } + if (s_check_result == 1) { + if ((e = s_mp_mul(&a,&b,&d)) != MP_OKAY) { + t1 = UINT64_MAX; + goto LTM_ERR; + } + if (mp_cmp(&c, &d) != MP_EQ) { + /* Time of 0 cannot happen (famous last words?) */ + t1 = 0uLL; + goto LTM_ERR; + } + } + } + + t1 = s_timer_stop(); +LTM_ERR: + mp_clear_multi(&a, &b, &c, &d, NULL); + return t1; +} + +static uint64_t s_time_sqr(int size) +{ + int x, e; mp_int a, b, c; uint64_t t1; - mp_init(&a); - mp_init(&b); - mp_init(&c); - - mp_rand(&a, size); - mp_rand(&b, size); - - if (s == 1) { - KARATSUBA_MUL_CUTOFF = size; - } else { - KARATSUBA_MUL_CUTOFF = 100000; + if ((e = mp_init_multi(&a, &b, &c, NULL)) != MP_OKAY) { + t1 = UINT64_MAX; + goto LTM_ERR; } - t_start(); - for (x = 0; x < TIMES; x++) { - mp_mul(&a,&b,&c); + if ((e = s_mp_random(&a, size)) != MP_OKAY) { + t1 = UINT64_MAX; + goto LTM_ERR; } - t1 = t_read(); - mp_clear(&a); - mp_clear(&b); - mp_clear(&c); + + s_timer_start(); + for (x = 0; x < s_number_of_test_loops; x++) { + if ((e = mp_sqr(&a,&b)) != MP_OKAY) { + t1 = UINT64_MAX; + goto LTM_ERR; + } + if (s_check_result == 1) { + if ((e = s_mp_sqr(&a,&c)) != MP_OKAY) { + t1 = UINT64_MAX; + goto LTM_ERR; + } + if (mp_cmp(&c, &b) != MP_EQ) { + t1 = 0uLL; + goto LTM_ERR; + } + } + } + + t1 = s_timer_stop(); +LTM_ERR: + mp_clear_multi(&a, &b, &c, NULL); return t1; } -static uint64_t time_sqr(int size, int s) +static void s_usage(char *s) { - unsigned long x; - mp_int a, b; - uint64_t t1; - - mp_init(&a); - mp_init(&b); - - mp_rand(&a, size); - - if (s == 1) { - KARATSUBA_SQR_CUTOFF = size; - } else { - KARATSUBA_SQR_CUTOFF = 100000; - } - - t_start(); - for (x = 0; x < TIMES; x++) { - mp_sqr(&a,&b); - } - t1 = t_read(); - mp_clear(&a); - mp_clear(&b); - return t1; + fprintf(stderr,"Usage: %s [TvcpGbtrSLFfMmosh]\n",s); + fprintf(stderr," -T testmode, for use with testme.sh\n"); + fprintf(stderr," -v verbose, print all timings\n"); + fprintf(stderr," -c check results\n"); + fprintf(stderr," -p print benchmark of final cutoffs in files \"multiplying\"\n"); + fprintf(stderr," and \"squaring\"\n"); + fprintf(stderr," -G [string] suffix for the filenames listed above\n"); + fprintf(stderr," Implies '-p'\n"); + fprintf(stderr," -b print benchmark of bncore.c\n"); + fprintf(stderr," -t prints comma separated results\n"); + fprintf(stderr," -r [64] number of rounds\n"); + fprintf(stderr," -S [0xdeadbeef] seed for PRNG\n"); + fprintf(stderr," -L [3] number of negative values accumulated until the result is accepted\n"); + fprintf(stderr," -M [3000] upper limit of T-C tests/prints\n"); + fprintf(stderr," -m [1] increment of T-C tests/prints\n"); + fprintf(stderr," -o [1] multiplier for the second multiplicand\n"); + fprintf(stderr," (Not for computing the cut-offs!)\n"); + fprintf(stderr," -s 'preset' use values in 'preset' for printing.\n"); + fprintf(stderr," 'preset' is a comma separated string with cut-offs for\n"); + fprintf(stderr," ksm, kss, tc3m, tc3s in that order\n"); + fprintf(stderr," ksm = karatsuba multiplication\n"); + fprintf(stderr," kss = karatsuba squaring\n"); + fprintf(stderr," tc3m = Toom-Cook 3-way multiplication\n"); + fprintf(stderr," tc3s = Toom-Cook 3-way squaring\n"); + fprintf(stderr," Implies '-p'\n"); + fprintf(stderr," -h this message\n"); } -int main(void) + + +int main(int argc, char **argv) { uint64_t t1, t2; - int x, y; + int x, i, j; + int count = 0; - for (x = 8; ; x += 2) { - t1 = time_mult(x, 0); - t2 = time_mult(x, 1); - printf("%d: %9llu %9llu, %9llu\n", x, t1, t2, t2 - t1); - if (t2 < t1) break; + int testmode = 0; + int verbose = 0; + int print = 0; + int bncore = 0; + int terse = 0; + + int upper_limit_print = 3000; + int increment_print = 1; + + int printpreset = 0; + /*int preset[8];*/ + int base = 10; + char *endptr, *str; + long val; + + uint64_t seed = 0xdeadbeef; + + int opt; + int ksm, kss, tc3m, tc3s; + + FILE *squaring, *multiplying; + char mullog[256] = "multiplying"; + char sqrlog[256] = "squaring"; + s_number_of_test_loops = 64; + s_stabilization_extra = 3; + + /* Very simple option parser, please treat it nicely. */ + if (argc != 1) { + for (opt = 1; (opt < argc) && (argv[opt][0] == '-'); opt++) { + switch (argv[opt][1]) { + case 'T': + testmode = 1; + s_check_result = 1; + upper_limit_print = 1000; + increment_print = 11; + s_number_of_test_loops = 1; + s_stabilization_extra = 1; + s_offset = 1; + break; + case 'v': + verbose = 1; + break; + case 'c': + s_check_result = 1; + break; + case 'p': + print = 1; + break; + case 'G': + print = 1; + opt++; + if (opt >= argc) { + s_usage(argv[0]); + } + /* manual strcat() */ + for (i = 0; i < 255; i++) { + if (mullog[i] == '\0') { + break; + } + } + for (j = 0; i < 255; j++, i++) { + mullog[i] = argv[opt][j]; + if (argv[opt][j] == '\0') { + break; + } + } + for (i = 0; i < 255; i++) { + if (sqrlog[i] == '\0') { + break; + } + } + for (j = 0; i < 255; j++, i++) { + sqrlog[i] = argv[opt][j]; + if (argv[opt][j] == '\0') { + break; + } + } + break; + case 'b': + bncore = 1; + break; + case 't': + terse = 1; + break; + case 'S': + opt++; + if (opt >= argc) { + s_usage(argv[0]); + } + str = argv[opt]; + errno = 0; + val = strtol(str, &endptr, base); + if ((errno == ERANGE && (val == LONG_MAX || val == LONG_MIN)) + || (errno != 0 && val == 0)) { + fprintf(stderr,"Seed %s not usable\n", argv[opt]); + exit(EXIT_FAILURE); + } + if (endptr == str) { + fprintf(stderr, "No seed given?\n"); + exit(EXIT_FAILURE); + } + seed = (uint64_t)val; + break; + case 'L': + opt++; + if (opt >= argc) { + s_usage(argv[0]); + } + str = argv[opt]; + errno = 0; + val = strtol(str, &endptr, base); + if ((val > INT_MAX || val < 0) || (errno != 0)) { + fprintf(stderr,"Value %s not usable\n", argv[opt]); + exit(EXIT_FAILURE); + } + if (endptr == str) { + fprintf(stderr, "No value for option \"-L\"given\n"); + exit(EXIT_FAILURE); + } + s_stabilization_extra = (int)val; + break; + case 'o': + opt++; + if (opt >= argc) { + s_usage(argv[0]); + } + str = argv[opt]; + errno = 0; + val = strtol(str, &endptr, base); + if ((val > INT_MAX || val < 0) || (errno != 0)) { + fprintf(stderr,"Value %s not usable as an offset\n", argv[opt]); + exit(EXIT_FAILURE); + } + if (endptr == str) { + fprintf(stderr, "No value for the offset given\n"); + exit(EXIT_FAILURE); + } + s_offset = (int)val; + break; + case 'r': + opt++; + if (opt >= argc) { + s_usage(argv[0]); + } + str = argv[opt]; + errno = 0; + val = strtol(str, &endptr, base); + if ((val > INT_MAX || val < 0) || (errno != 0)) { + fprintf(stderr,"Value %s not usable as the number of rounds for \"-r\"\n", argv[opt]); + exit(EXIT_FAILURE); + } + if (endptr == str) { + fprintf(stderr, "No value for the number of rounds given\n"); + exit(EXIT_FAILURE); + } + s_number_of_test_loops = (int)val; + break; + + case 'M': + opt++; + if (opt >= argc) { + s_usage(argv[0]); + } + str = argv[opt]; + errno = 0; + val = strtol(str, &endptr, base); + if ((val > INT_MAX || val < 0) || (errno != 0)) { + fprintf(stderr,"Value %s not usable as the upper limit of T-C tests (\"-M\")\n", argv[opt]); + exit(EXIT_FAILURE); + } + if (endptr == str) { + fprintf(stderr, "No value for the upper limit of T-C tests given\n"); + exit(EXIT_FAILURE); + } + upper_limit_print = (int)val; + break; + case 'm': + opt++; + if (opt >= argc) { + s_usage(argv[0]); + } + str = argv[opt]; + errno = 0; + val = strtol(str, &endptr, base); + if ((val > INT_MAX || val < 0) || (errno != 0)) { + fprintf(stderr,"Value %s not usable as the increment for the T-C tests (\"-m\")\n", argv[opt]); + exit(EXIT_FAILURE); + } + if (endptr == str) { + fprintf(stderr, "No value for the increment for the T-C tests given\n"); + exit(EXIT_FAILURE); + } + increment_print = (int)val; + break; + case 's': + printpreset = 1; + print = 1; + opt++; + if (opt >= argc) { + s_usage(argv[0]); + } + str = argv[opt]; + i = 0; + /* Only the most basic checks */ + errno = 0; + val = strtol(str, &endptr, base); + if ((val > INT_MAX || val < 0) || (errno != 0)) { + fprintf(stderr,"input #%d wrong\n", i+1); + exit(EXIT_FAILURE); + } + if (endptr == str) { + fprintf(stderr, "No input for #%d?\n", i+1); + exit(EXIT_FAILURE); + } + i++; + str = endptr + 1; + KARATSUBA_MUL_CUTOFF = (int)val; + errno = 0; + val = strtol(str, &endptr, base); + if ((val > INT_MAX || val < 0) || (errno != 0)) { + fprintf(stderr,"input #%d wrong\n", i+1); + exit(EXIT_FAILURE); + } + if (endptr == str) { + fprintf(stderr, "No input for #%d?\n", i+1); + exit(EXIT_FAILURE); + } + i++; + str = endptr + 1; + KARATSUBA_SQR_CUTOFF = (int)val; + errno = 0; + val = strtol(str, &endptr, base); + if ((val > INT_MAX || val < 0) || (errno != 0)) { + fprintf(stderr,"input #%d wrong\n", i+1); + exit(EXIT_FAILURE); + } + if (endptr == str) { + fprintf(stderr, "No input for #%d?\n", i+1); + exit(EXIT_FAILURE); + } + i++; + str = endptr + 1; + TOOM_MUL_CUTOFF = (int)val; + errno = 0; + val = strtol(str, &endptr, base); + if ((val > INT_MAX || val < 0) || (errno != 0)) { + fprintf(stderr,"input #%d wrong\n", i+1); + exit(EXIT_FAILURE); + } + if (endptr == str) { + fprintf(stderr, "No input for #%d?\n", i+1); + exit(EXIT_FAILURE); + } + i++; + str = endptr + 1; + TOOM_SQR_CUTOFF = (int)val; + case 'h': + default: + s_usage(argv[0]); + exit(EXIT_FAILURE); + } + } } - y = x; - for (x = 8; ; x += 2) { - t1 = time_sqr(x, 0); - t2 = time_sqr(x, 1); - printf("%d: %9llu %9llu, %9llu\n", x, t1, t2, t2 - t1); - if (t2 < t1) break; + s_raninit(seed); + + ksm = KARATSUBA_MUL_CUTOFF; + kss = KARATSUBA_SQR_CUTOFF; + tc3m = TOOM_MUL_CUTOFF; + tc3s = TOOM_SQR_CUTOFF; + + if ((bncore == 0) && (printpreset == 0)) { + /* Turn all limits from bncore.c to the max */ + KARATSUBA_MUL_CUTOFF = INT_MAX; + KARATSUBA_SQR_CUTOFF = INT_MAX; + TOOM_MUL_CUTOFF = INT_MAX; + TOOM_SQR_CUTOFF = INT_MAX; +#ifdef BN_S_MP_KARATSUBA_MUL_C + /* + The influence of the Comba multiplication cannot be + eradicated programmatically. It depends on the size + of the macro MP_WPARRAY in tommath.h which needs to + be changed manually (to 0 (zero)). + */ + if ((verbose == 1) || (testmode == 1)) { + puts("# Karatsuba multiplication."); + } + for (x = 8; x < upper_limit_print; x += increment_print) { + KARATSUBA_MUL_CUTOFF = INT_MAX; + t1 = s_time_mul(x); + if ((t1 == 0uLL) || (t1 == UINT64_MAX)) { + fprintf(stderr,"Karatsuba multiplication failed at x = INT_MAX (%s)\n", + (t1 == 0uLL)?"wrong result":"internal error"); + exit(EXIT_FAILURE); + } + KARATSUBA_MUL_CUTOFF = x; + t2 = s_time_mul(x); + if ((t2 == 0uLL) || (t2 == UINT64_MAX)) { + fprintf(stderr,"Karatsuba multiplication failed (%s)\n", + (t2 == 0uLL)?"wrong result":"internal error"); + exit(EXIT_FAILURE); + } + if (verbose == 1) { + printf("%d: %9"PRIu64" %9"PRIu64", %9"PRIi64"\n", x, t1, t2, (int64_t)t2 - (int64_t)t1); + } + if (t2 < t1) { + if (count == s_stabilization_extra) { + count = 0; + break; + } else if (count < s_stabilization_extra) { + count++; + } + } else if (count > 0) { + count--; + } + } + KARATSUBA_MUL_CUTOFF = x - s_stabilization_extra * increment_print; +#endif +#ifdef BN_S_MP_KARATSUBA_SQR_C + if ((verbose == 1) || (testmode == 1)) { + puts("# Karatsuba squaring."); + } + for (x = 8; x < upper_limit_print; x += increment_print) { + KARATSUBA_SQR_CUTOFF = INT_MAX; + t1 = s_time_sqr(x); + if ((t1 == 0uLL) || (t1 == UINT64_MAX)) { + fprintf(stderr,"Karatsuba squaring failed at x = INT_MAX (%s)\n", + (t1 == 0uLL)?"wrong result":"internal error"); + exit(EXIT_FAILURE); + } + KARATSUBA_SQR_CUTOFF = x; + t2 = s_time_sqr(x); + if ((t2 == 0uLL) || (t2 == UINT64_MAX)) { + fprintf(stderr,"Karatsuba squaring failed (%s)\n", + (t2 == 0uLL)?"wrong result":"internal error"); + exit(EXIT_FAILURE); + } + if (verbose == 1) { + printf("%d: %9"PRIu64" %9"PRIu64", %9"PRIi64"\n", x, t1, t2, (int64_t)t2 - (int64_t)t1); + } + if (t2 < t1) { + if (count == s_stabilization_extra) { + count = 0; + break; + } else if (count < s_stabilization_extra) { + count++; + } + } else if (count > 0) { + count--; + } + } + KARATSUBA_SQR_CUTOFF = x - s_stabilization_extra * increment_print; +#endif +#ifdef BN_S_MP_TOOM_MUL_C + if ((verbose == 1) || (testmode == 1)) { + puts("# Toom-Cook 3-way multiplying."); + } + for (x = 8; x < upper_limit_print; x += increment_print) { + TOOM_MUL_CUTOFF = INT_MAX; + t1 = s_time_mul(x); + if ((t1 == 0uLL) || (t1 == UINT64_MAX)) { + fprintf(stderr,"Toom-Cook 3-way multiplying failed at x = INT_MAX (%s)\n", + (t1 == 0uLL)?"wrong result":"internal error"); + exit(EXIT_FAILURE); + } + TOOM_MUL_CUTOFF = x; + t2 = s_time_mul(x); + if ((t2 == 0uLL) || (t2 == UINT64_MAX)) { + fprintf(stderr,"Toom-Cook 3-way multiplication failed (%s)\n", + (t2 == 0uLL)?"wrong result":"internal error"); + exit(EXIT_FAILURE); + } + if (verbose == 1) { + printf("%d: %9"PRIu64" %9"PRIu64", %9"PRIi64"\n", x, t1, t2, (int64_t)t2 - (int64_t)t1); + } + if (t2 < t1) { + if (count == s_stabilization_extra) { + count = 0; + break; + } else if (count < s_stabilization_extra) { + count++; + } + } else if (count > 0) { + count--; + } + } + TOOM_MUL_CUTOFF = x - s_stabilization_extra * increment_print; +#endif +#ifdef BN_S_MP_TOOM_SQR_C + if ((verbose == 1) || (testmode == 1)) { + puts("# Toom-Cook 3-way squaring."); + } + for (x = 8; x < upper_limit_print; x += increment_print) { + TOOM_SQR_CUTOFF = INT_MAX; + t1 = s_time_sqr(x); + if ((t1 == 0uLL) || (t1 == UINT64_MAX)) { + fprintf(stderr,"Toom-Cook 3-way squaring failed at x = INT_MAX (%s)\n", + (t1 == 0uLL)?"wrong result":"internal error"); + exit(EXIT_FAILURE); + } + TOOM_SQR_CUTOFF = x; + t2 = s_time_sqr(x); + if ((t2 == 0uLL) || (t2 == UINT64_MAX)) { + fprintf(stderr,"Toom-Cook 3-way squaring failed (%s)\n", + (t2 == 0uLL)?"wrong result":"internal error"); + exit(EXIT_FAILURE); + } + if (verbose == 1) { + printf("%d: %9"PRIu64" %9"PRIu64", %9"PRIi64"\n", x, t1, t2, (int64_t)t2 - (int64_t)t1); + } + if (t2 < t1) { + if (count == s_stabilization_extra) { + count = 0; + break; + } else if (count < s_stabilization_extra) { + count++; + } + } else if (count > 0) { + count--; + } + } + TOOM_SQR_CUTOFF = x - s_stabilization_extra * increment_print; +#endif + } + if (terse == 1) { + printf("%d %d %d %d\n", + KARATSUBA_MUL_CUTOFF, + KARATSUBA_SQR_CUTOFF, + TOOM_MUL_CUTOFF, + TOOM_SQR_CUTOFF); + } else { + printf("KARATSUBA_MUL_CUTOFF = %d\n", KARATSUBA_MUL_CUTOFF); + printf("KARATSUBA_SQR_CUTOFF = %d\n", KARATSUBA_SQR_CUTOFF); + printf("TOOM_MUL_CUTOFF = %d\n", TOOM_MUL_CUTOFF); + printf("TOOM_SQR_CUTOFF = %d\n", TOOM_SQR_CUTOFF); } - printf("KARATSUBA_MUL_CUTOFF = %d\n", y); - printf("KARATSUBA_SQR_CUTOFF = %d\n", x); - return 0; + if (print == 1) { + printf("Printing data for graphing to \"%s\" and \"%s\"\n",mullog, sqrlog); + + multiplying = fopen(mullog, "w+"); + if (multiplying == NULL) { + fprintf(stderr, "Opening file \"%s\" failed\n", mullog); + exit(EXIT_FAILURE); + } + + squaring = fopen(sqrlog, "w+"); + if (squaring == NULL) { + fprintf(stderr, "Opening file \"%s\" failed\n",sqrlog); + exit(EXIT_FAILURE); + } + + for (x = 8; x < upper_limit_print; x += increment_print) { + KARATSUBA_MUL_CUTOFF = INT_MAX; + KARATSUBA_SQR_CUTOFF = INT_MAX; + TOOM_MUL_CUTOFF = INT_MAX; + TOOM_SQR_CUTOFF = INT_MAX; + t1 = s_time_mul(x); + KARATSUBA_MUL_CUTOFF = kss; + KARATSUBA_SQR_CUTOFF = ksm; + TOOM_MUL_CUTOFF = tc3m; + TOOM_SQR_CUTOFF = tc3s; + t2 = s_time_mul(x); + fprintf(multiplying, "%d: %9"PRIu64" %9"PRIu64", %9"PRIi64"\n", x, t1, t2, (int64_t)t2 - (int64_t)t1); + fflush(multiplying); + if (verbose == 1) { + printf("MUL %d: %9"PRIu64" %9"PRIu64", %9"PRIi64"\n", x, t1, t2, (int64_t)t2 - (int64_t)t1); + fflush(stdout); + } + KARATSUBA_MUL_CUTOFF = INT_MAX; + KARATSUBA_SQR_CUTOFF = INT_MAX; + TOOM_MUL_CUTOFF = INT_MAX; + TOOM_SQR_CUTOFF = INT_MAX; + t1 = s_time_sqr(x); + KARATSUBA_MUL_CUTOFF = kss; + KARATSUBA_SQR_CUTOFF = ksm; + TOOM_MUL_CUTOFF = tc3m; + TOOM_SQR_CUTOFF = tc3s; + t2 = s_time_sqr(x); + fprintf(squaring,"%d: %9"PRIu64" %9"PRIu64", %9"PRIi64"\n", x, t1, t2, (int64_t)t2 - (int64_t)t1); + fflush(squaring); + if (verbose == 1) { + printf("SQR %d: %9"PRIu64" %9"PRIu64", %9"PRIi64"\n", x, t1, t2, (int64_t)t2 - (int64_t)t1); + fflush(stdout); + } + } + printf("Finished. Data for graphing in \"%s\" and \"%s\"\n",mullog, sqrlog); + if (verbose == 1) { + KARATSUBA_MUL_CUTOFF = kss; + KARATSUBA_SQR_CUTOFF = ksm; + TOOM_MUL_CUTOFF = tc3m; + TOOM_SQR_CUTOFF = tc3s; + if (terse == 1) { + printf("%d %d %d %d\n", + KARATSUBA_MUL_CUTOFF, + KARATSUBA_SQR_CUTOFF, + TOOM_MUL_CUTOFF, + TOOM_SQR_CUTOFF); + } else { + printf("KARATSUBA_MUL_CUTOFF = %d\n", KARATSUBA_MUL_CUTOFF); + printf("KARATSUBA_SQR_CUTOFF = %d\n", KARATSUBA_SQR_CUTOFF); + printf("TOOM_MUL_CUTOFF = %d\n", TOOM_MUL_CUTOFF); + printf("TOOM_SQR_CUTOFF = %d\n", TOOM_SQR_CUTOFF); + } + } + } + exit(EXIT_SUCCESS); } diff --git a/etc/tune_it.sh b/etc/tune_it.sh new file mode 100755 index 0000000..4a15646 --- /dev/null +++ b/etc/tune_it.sh @@ -0,0 +1,129 @@ +#!/bin/sh + +die() { + echo "$1 failed" + echo "Exiting" + exit $2 +} +# A linear congruential generator is sufficient for the purpose. +SEED=3735928559 +LCG() { + SEED=$(((1103515245 * $SEED + 12345) % 2147483648)) + echo $SEED +} + +median() { + median=0; + flag=1; + for val in $* ; do + if [ $flag -eq 1 ] ; then + flag=$((flag + 1)) + continue + elif [ $flag -eq 2 ] ; then + median=$val + flag=$((flag + 1)) + continue + else + if [ $median -lt $val ] ; then + median=$((median + 1)) + fi + if [ $median -gt $val ] ; then + median=$((median - 1)) + fi + fi + done + echo $median +} + +MPWD=$(pwd) +FILE_NAME="tuning_list" +BNCORE_C="../bncore.c" +BACKUP_SUFFIX=".orig" +RNUM=0; +############################################################################# +# It would be a good idea to isolate these processes (with e.g.: cpuset) # +# # +# It is not a good idea to e.g: watch high resolution videos while this # +# test are running if you do not have enough memory to avoid page faults. # +############################################################################# + +# Number of rounds overall. +LIMIT=100 +# Number of loops for each input. +RLOOPS=10 +# Offset ( > 0 ) . Runs tests with asymmetric input of the form 1:OFFSET +# Please use another destination for BNCORE_C if you change OFFSET, because the numbers +# with an offset different from 1 (one) are not usable as the general cut-off values +# in "bncore.c". +OFFSET=1 +# Number ( >= 3 ) of positive results (TC-is-faster) accumulated until it is accepted. +# Due to the algorithm used to compute the median in this Posix compliant shell script +# the value needs to be 3 (three), not less, to keep the variation small. +LAG=3 +# Keep the temporary file $FILE_NAME. Set to 0 (zero) to remove it at the end. +# The file is in a format fit to feed into R directly. If you do it and find the median +# of this program to be off by more than a couple: please contact the authors and report +# the numbers from this program and R and the standard deviation. This program is known +# to get larger errors if the standard deviation is larger than ~50. +KEEP_TEMP=1 + +echo "You might like to watch the numbers go up to $LIMIT but it will take a long time!" + +# Might not have sufficient rights or disc full. +echo "km ks tc3m tc3s" > $FILE_NAME || die "Writing header to $FILE_NAME" $? +i=1 +while [ $i -le $LIMIT ]; do + RNUM=$(LCG) + echo $i + "$MPWD"/tune -t -r $RLOOPS -L $LAG -S "$RNUM" -o $OFFSET >> $FILE_NAME || die "tune" $? + i=$((i + 1)) +done + +if [ $KEEP_TEMP -eq 0 ]; then + rm -v $FILE_NAME || die "Removing $KEEP_TEMP" $? +fi + +echo "Writing cut-off values to \"bncore.c\"." +echo "In case of failure: a copy of \"bncore.c\" is in \"bncore.c.orig\"" + +cp -v $BNCORE_C $BNCORE_C$BACKUP_SUFFIX || die "Making backup copy of bncore.c" $? + +cat << END_OF_INPUT > $BNCORE_C || die "Writing header to bncore.c" $? +#include "tommath_private.h" +#ifdef BNCORE_C +/* LibTomMath, multiple-precision integer library -- Tom St Denis */ +/* SPDX-License-Identifier: Unlicense */ +/* + Current values evaluated on an AMD A8-6600K (64-bit). + Type "make tune" to optimize them for your machine but + be aware that it may take a long time. It took 2:30 minutes + on the aforementioned machine for example. + */ +END_OF_INPUT + +# The Posix shell does not offer an array data type + +i=1; +TMP="" +TMP=$(cat $FILE_NAME | cut -d' ' -f$i ) +TMP=$(median $TMP ) +echo "int KARATSUBA_MUL_CUTOFF = $TMP;" +echo "int KARATSUBA_MUL_CUTOFF = $TMP;" >> $BNCORE_C || die "(km) Appending to bncore.c" $? +i=$((i + 1)) +TMP=$(cat $FILE_NAME | cut -d' ' -f$i ) +TMP=$(median $TMP ); +echo "int KARATSUBA_SQR_CUTOFF = $TMP;" +echo "int KARATSUBA_SQR_CUTOFF = $TMP;" >> $BNCORE_C || die "(ks) Appending to bncore.c" $? +i=$((i + 1)) +TMP=$(cat $FILE_NAME | cut -d' ' -f$i) +TMP=$(median $TMP ); +echo "int TOOM_MUL_CUTOFF = $TMP;" +echo "int TOOM_MUL_CUTOFF = $TMP;" >> $BNCORE_C || die "(tc3m) Appending to bncore.c" $? +i=$((i + 1)) +TMP=$(cat $FILE_NAME | cut -d' ' -f$i) +TMP=$(median $TMP ); +echo "int TOOM_SQR_CUTOFF = $TMP;" +echo "int TOOM_SQR_CUTOFF = $TMP;" >> $BNCORE_C || die "(tc3s) Appending to bncore.c" $? + +echo "#endif" >> $BNCORE_C || die "(end) Appending to bncore.c" $? + diff --git a/makefile b/makefile index d1d2e23..223b4df 100644 --- a/makefile +++ b/makefile @@ -108,6 +108,10 @@ mtest: timing: $(LIBNAME) demo/timing.c $(CC) $(CFLAGS) -DTIMER demo/timing.c $(LIBNAME) $(LFLAGS) -o timing +tune: $(LIBNAME) + $(MAKE) -C etc tune + $(MAKE) + # You have to create a file .coveralls.yml with the content "repo_token: " # in the base folder to be able to submit to coveralls coveralls: lcov diff --git a/makefile.mingw b/makefile.mingw index 374733a..48d11c9 100644 --- a/makefile.mingw +++ b/makefile.mingw @@ -89,6 +89,10 @@ test_standalone: test.exe all: $(LIBMAIN_S) test_standalone +tune: $(LIBNAME_S) + $(MAKE) -C etc tune + $(MAKE) + clean: @-cmd /c del /Q /S *.o *.a *.exe *.dll 2>nul diff --git a/makefile.msvc b/makefile.msvc index 0b2b212..6d96ad1 100644 --- a/makefile.msvc +++ b/makefile.msvc @@ -75,6 +75,10 @@ test_standalone: test.exe all: $(LIBMAIN_S) test_standalone +tune: $(LIBMAIN_S) + $(MAKE) -C etc tune + $(MAKE) + clean: @-cmd /c del /Q /S *.OBJ *.LIB *.EXE *.DLL 2>nul diff --git a/makefile.shared b/makefile.shared index 8c5d548..36eea7e 100644 --- a/makefile.shared +++ b/makefile.shared @@ -95,3 +95,17 @@ mtest: timing: $(LIBNAME) demo/timing.c $(LTLINK) $(CFLAGS) $(LDFLAGS) -DTIMER demo/timing.c $(LIBNAME) -o timing + +tune: $(LIBNAME) + $(LTCOMPILE) $(CFLAGS) -c etc/tune.c -o etc/tune.o + $(LTLINK) $(LDFLAGS) -o etc/tune etc/tune.o $(LIBNAME) + $(LTCOMPILE) $(CFLAGS) -c etc/statistic_summary_single_column.c -o etc/statistic_summary_single_column.o + $(LTLINK) $(LDFLAGS) -o etc/statistic_summary_single_column etc/statistic_summary_single_column.o + cd etc/ + /bin/sh tune_it.sh + cd .. + $(MAKE) -f makefile.shared + + + + diff --git a/makefile.unix b/makefile.unix index beb7abd..42dfcf9 100644 --- a/makefile.unix +++ b/makefile.unix @@ -87,6 +87,10 @@ test_standalone: test all: $(LIBMAIN_S) test_standalone +tune: $(LIBMAIN_S) + $(MAKE) -C etc tune + $(MAKE) + #NOTE: this makefile works also on cygwin, thus we need to delete *.exe clean: -@rm -f $(OBJECTS) $(LIBMAIN_S) @@ -98,3 +102,4 @@ install: $(LIBMAIN_S) @cp $(LIBMAIN_S) $(DESTDIR)$(LIBPATH)/ @cp $(HEADERS_PUB) $(DESTDIR)$(INCPATH)/ @sed -e 's,^prefix=.*,prefix=$(PREFIX),' -e 's,^Version:.*,Version: $(VERSION),' libtommath.pc.in > $(DESTDIR)$(LIBPATH)/pkgconfig/libtommath.pc + diff --git a/makefile_include.mk b/makefile_include.mk index ccc785f..c5cfc28 100644 --- a/makefile_include.mk +++ b/makefile_include.mk @@ -147,7 +147,7 @@ cleancov-clean: cleancov: cleancov-clean clean clean: - rm -f *.gcda *.gcno *.gcov *.bat *.o *.a *.obj *.lib *.exe *.dll etclib/*.o demo/test.o demo/main.o demo/opponent.o test timing mpitest mtest/mtest mtest/mtest.exe \ + rm -f *.gcda *.gcno *.gcov *.bat *.o *.a *.obj *.lib *.exe *.dll etclib/*.o demo/test.o demo/main.o demo/opponent.o test timing mpitest mtest/mtest mtest/mtest.exe tuning_list\ *.idx *.toc *.log *.aux *.dvi *.lof *.ind *.ilg *.ps *.log *.s mpi.c *.da *.dyn *.dpi tommath.tex `find . -type f | grep [~] | xargs` *.lo *.la rm -rf .libs/ ${MAKE} -C etc/ clean MAKE=${MAKE} diff --git a/testme.sh b/testme.sh index 40ecb74..c8237df 100755 --- a/testme.sh +++ b/testme.sh @@ -118,11 +118,20 @@ _make() _runtest() { make clean > /dev/null - _make "$1" "$2" "test_standalone" local _timeout="" which timeout >/dev/null && _timeout="timeout --foreground 90" - echo -e "\rRun test $1 $2" - $_timeout ./test > test_${suffix}.log || _die "running tests" $? + if [[ "$MAKE_OPTIONS" =~ "tune" ]] + then + # "make tune" will run "tune_it.sh" automatically, hence "autotune", but it cannot + # get switched off without some effort, so we just let it run twice for testing purposes + _make "$1" "$2" "" + echo -e "\rRun autotune $1 $2" + $_timeout ./etc/tune_it.sh > test_${suffix}.log || _die "running autotune" $? + else + _make "$1" "$2" "test_standalone" + echo -e "\rRun test $1 $2" + $_timeout ./test > test_${suffix}.log || _die "running tests" $? + fi } # This is not much more of a C&P of _runtest with a different timeout @@ -131,13 +140,24 @@ _runtest() _runvalgrind() { make clean > /dev/null - _make "$1" "$2" "test_standalone" local _timeout="" # 30 minutes? Yes. Had it at 20 minutes and the Valgrind run needed over 25 minutes. # A bit too close for comfort. which timeout >/dev/null && _timeout="timeout --foreground 1800" - echo -e "\rRun test $1 $2 inside valgrind" - $_timeout $VALGRIND_BIN $VALGRIND_OPTS ./test > test_${suffix}.log || _die "running tests" $? +echo "MAKE_OPTIONS = \"$MAKE_OPTIONS\"" + if [[ "$MAKE_OPTIONS" =~ "tune" ]] + then +echo "autotune branch" + _make "$1" "$2" "" + # The shell used for /bin/sh is DASH 0.5.7-4ubuntu1 on the author's machine which fails valgrind, so + # we just run on instance of etc/tune with the same options as in etc/tune_it.sh + echo -e "\rRun etc/tune $1 $2 once inside valgrind" + $_timeout $VALGRIND_BIN $VALGRIND_OPTS ./etc/tune -t -r 10 -L 3 > test_${suffix}.log || _die "running etc/tune" $? + else + _make "$1" "$2" "test_standalone" + echo -e "\rRun test $1 $2 inside valgrind" + $_timeout $VALGRIND_BIN $VALGRIND_OPTS ./test > test_${suffix}.log || _die "running tests" $? + fi }