From 26b17702fc285bb8e624e4d18cc4394b5859fd57 Mon Sep 17 00:00:00 2001
From: Alexander Kruppa <akruppa@gmail.com>
Date: Tue, 25 Jun 2024 16:11:20 +0200
Subject: [PATCH] Allow mmap-ing s data
Adds a new configure option: --enable-mmap
and, if mmap is enabled, two new command line options: -bsavems, -bloadms
The rationale is that when multiple ecm processes run on the same machine,
all doing ECM with batch stage 1 and the same B1, they all have their own
batch product s in memory which contains the same data for all processes.
As the batch product can become quite large with large B1, this may cause the
machine to run out of memory with many cores. By using a file-backed mmap
instead, the operating system can use shared memory between all processes.
The file format for the s save files has changed. It has a header now, with
magic number, endianness, B1 value and a checksum. The file format is
different between mmap-ed and non-mmap-ed data; the mmap-ed one uses the
host's endianness for mp_limb_t (as the disk file is just a memory image)
while the non-mmap-ed one uses mpz_out_raw() (as before) which always
writes in big-endian.
---
configure.ac | 14 +++-
ecm-ecm.h | 6 +-
main.c | 51 ++++++++++-
resume.c | 233 +++++++++++++++++++++++++++++++++++++++++++++++++--
4 files changed, 290 insertions(+), 14 deletions(-)
diff --git a/configure.ac b/configure.ac
index fcd56a50..7cdb7088 100644
--- a/configure.ac
+++ b/configure.ac
@@ -90,6 +90,9 @@ if test "x$enable_mulredc_svoboda" = xyes; then
GMP_DEFINE([MULREDC_SVOBODA], 1)
fi
+AC_ARG_ENABLE([mmap],
+[AS_HELP_STRING([--enable-mmap], [use mmap() to load batch s files (EXPERIMENTAL. Expect bugs. Please report them.)])])
+
AC_ARG_ENABLE([valgrind-client],
[AS_HELP_STRING([--enable-valgrind-client], [enable Valgrind client check requests [[default=no]]])],[],[])
if test "x$enable_valgrind" = xyes; then
@@ -379,7 +382,7 @@ AC_FUNC_ALLOCA
m4_version_prereq([2.70],[AC_CHECK_INCLUDES_DEFAULT],[AC_HEADER_STDC])
AC_PROG_EGREP
-AC_CHECK_HEADERS([math.h limits.h malloc.h strings.h sys/time.h unistd.h io.h signal.h fcntl.h])
+AC_CHECK_HEADERS([math.h limits.h malloc.h strings.h stdint.h inttypes.h endian.h sys/time.h unistd.h io.h signal.h fcntl.h])
AC_CHECK_HEADERS([windows.h psapi.h], [], [],
[[#ifdef HAVE_WINDOWS_H
# include <windows.h>
@@ -387,6 +390,11 @@ AC_CHECK_HEADERS([windows.h psapi.h], [], [],
]])
AC_CHECK_HEADERS([ctype.h sys/types.h sys/resource.h aio.h])
+if test "x$enable_mmap" = xyes; then
+ AC_CHECK_HEADERS([sys/mman.h])
+fi
+
+
dnl Checks for library functions that are not in GMP
AC_FUNC_STRTOD
@@ -405,6 +413,10 @@ AC_CHECK_FUNCS([isspace isdigit isxdigit], [], [AC_MSG_ERROR([required function
AC_CHECK_FUNCS([time ctime], [], [AC_MSG_ERROR([required function missing])])
AC_CHECK_FUNCS([gethostname gettimeofday getrusage memmove signal fcntl fileno setvbuf fallocate aio_read aio_init])
+if test "x$enable_mmap" = xyes; then
+ AC_CHECK_FUNCS([mmap])
+fi
+
dnl Test for some Windows-specific functions that are available under MinGW
dnl FIXME: which win32 library contains these functions?
dnl AC_CHECK_FUNCS([GetCurrentProcess GetProcessTimes])
diff --git a/ecm-ecm.h b/ecm-ecm.h
index c44126c1..54e26918 100644
--- a/ecm-ecm.h
+++ b/ecm-ecm.h
@@ -31,6 +31,7 @@ http://www.gnu.org/licenses/ or write to the Free Software Foundation, Inc.,
#define ASSERT(expr) do {} while (0)
#endif
+#include <stdint.h>
#include "ecm.h"
/* Structure for candidate usage. This is much more powerful than using a
@@ -121,8 +122,9 @@ int read_resumefile_line (int *, mpz_t, mpz_t, mpcandi_t *,
int write_resumefile (char *, int, ecm_params params,
mpcandi_t *, const mpz_t, const mpz_t, const mpz_t,
const char *);
-int write_s_in_file (char *, mpz_t);
-int read_s_from_file (mpz_t, char *, double);
+int write_s_in_file (const char *, mpz_t, int, uint64_t);
+int read_s_from_file (mpz_t, const char *, int, double);
+void free_s_data(int, mpz_t);
/* main.c */
int kbnc_z (double *k, unsigned long *b, unsigned long *n, signed long *c,
diff --git a/main.c b/main.c
index 0a7d62e1..5613bd3e 100644
--- a/main.c
+++ b/main.c
@@ -135,6 +135,10 @@ usage (void)
printf (" -bsaves file With -param 1-3, save stage 1 exponent in file.\n");
printf (" -bloads file With -param 1-3, load stage 1 exponent from file.\n");
+#ifdef HAVE_MMAP
+ printf (" -bsavems file With -param 1-3, save stage 1 exponent in file (file format for -bloadms).\n");
+ printf (" -bloadms file With -param 1-3, mmap stage 1 exponent from file.\n");
+#endif
#ifdef WITH_GPU
printf (" -gpu Use CGBN for computations stage 1.\n");
printf (" -gpudevice n Use device n to execute GPU code (by default, "
@@ -398,6 +402,8 @@ main (int argc, char *argv[])
int param = ECM_PARAM_DEFAULT; /* automatic choice */
char *savefile_s = NULL;
char *loadfile_s = NULL;
+ int save_s_mmap = 0, /* Do we write s in mmap file format? */
+ load_s_mmap = 0; /* Do we mmap s data from the file? */
#ifdef HAVE_GWNUM
double gw_k = 0.0; /* set default values for gwnum poly k*b^n+c */
unsigned long gw_b = 0; /* set default values for gwnum poly k*b^n+c */
@@ -540,16 +546,50 @@ main (int argc, char *argv[])
}
else if ((argc > 2) && (strcmp (argv[1], "-bsaves") == 0))
{
+ if (savefile_s != NULL) {
+ fprintf (stderr, "Multiple -bsaves or -bsavems options were given\n");
+ exit (EXIT_FAILURE);
+ }
savefile_s = argv[2];
+ save_s_mmap = 0;
argv += 2;
argc -= 2;
}
else if ((argc > 2) && (strcmp (argv[1], "-bloads") == 0))
{
+ if (loadfile_s != NULL) {
+ fprintf (stderr, "Multiple -bloads or -bloadms options were given\n");
+ exit (EXIT_FAILURE);
+ }
loadfile_s = argv[2];
+ load_s_mmap = 0;
argv += 2;
argc -= 2;
}
+#ifdef HAVE_MMAP
+ else if ((argc > 2) && (strcmp (argv[1], "-bsavems") == 0))
+ {
+ if (savefile_s != NULL) {
+ fprintf (stderr, "Multiple -bsaves or -bsavems options were given\n");
+ exit (EXIT_FAILURE);
+ }
+ savefile_s = argv[2];
+ save_s_mmap = 1;
+ argv += 2;
+ argc -= 2;
+ }
+ else if ((argc > 2) && (strcmp (argv[1], "-bloadms") == 0))
+ {
+ if (loadfile_s != NULL) {
+ fprintf (stderr, "Multiple -bloads or -bloadms options were given\n");
+ exit (EXIT_FAILURE);
+ }
+ loadfile_s = argv[2];
+ load_s_mmap = 1;
+ argv += 2;
+ argc -= 2;
+ }
+#endif
else if (strcmp (argv[1], "-h") == 0 || strcmp (argv[1], "--help") == 0)
{
usage ();
@@ -1389,7 +1429,7 @@ main (int argc, char *argv[])
exit (EXIT_FAILURE);
}
params->batch_last_B1_used = B1;
- if (read_s_from_file (params->batch_s, loadfile_s, B1))
+ if (read_s_from_file (params->batch_s, loadfile_s, load_s_mmap, B1))
{
fprintf (stderr, "Error while reading s from file\n");
exit (EXIT_FAILURE);
@@ -1595,7 +1635,12 @@ main (int argc, char *argv[])
/* Save the batch exponent s if requested */
if (savefile_s != NULL)
{
- int ret = write_s_in_file (savefile_s, params->batch_s);
+ int ret = write_s_in_file (savefile_s, params->batch_s, save_s_mmap,
+ (uint64_t) B1);
+ if (ret == 0) {
+ fprintf(stderr, "Error writing s to file %s\n", savefile_s);
+ exit(EXIT_FAILURE);
+ }
if (verbose >= OUTPUT_VERBOSE && ret > 0)
printf ("Saved batch product (of %u bytes) in %s\n", ret,
savefile_s);
@@ -1634,6 +1679,8 @@ main (int argc, char *argv[])
mpz_clear (sigma);
mpgocandi_t_free (&go);
+ if (loadfile_s != NULL)
+ free_s_data (load_s_mmap, params->batch_s);
ecm_clear (params);
/* exit 0 if a factor was found for the last input, except if we exit due
diff --git a/resume.c b/resume.c
index 0eac2921..986fffa7 100644
--- a/resume.c
+++ b/resume.c
@@ -23,11 +23,18 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
*/
+#include "config.h"
+
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <math.h>
#include <string.h>
+#include <endian.h>
+#include <inttypes.h>
+#ifdef HAVE_MMAP
+#include <sys/mman.h>
+#endif
#if !defined (_MSC_VER)
#include <unistd.h>
#endif
@@ -47,6 +54,11 @@ SOFTWARE.
Returns the number of matching characters that were read.
*/
+#define S_FILE_MAGIC 0xe2ad5e97UL
+/* This is a prime p with 2 a generator of (Z/pZ)* */
+#define S_FILE_CHECKSUM_MODULUS 4294967291UL
+
+
static int
facceptstr (FILE *fd, char *s)
{
@@ -654,12 +666,74 @@ write_resumefile (char *fn, int method, ecm_params params,
return 0;
}
+typedef struct {
+ uint32_t magic;
+ unsigned char version,
+ mmapped, /* mmapped data uses host endianness */
+ is_le, /* Is this machine little_endian? */
+ dummy; /* Always 0 for word alignment */
+ uint32_t checksum; /* s % S_FILE_CHECKSUM_MODULUS */
+ uint64_t B1;
+} s_file_header_t;
+
+/* Wrapper around fread(). Prints error message on error. If close_on_error
+ * is non-zero, also closes the stream.
+ * Returns 0 on success and 1 on error.
+ */
+int
+fread_perror(void * restrict data,
+ size_t size, size_t nmemb,
+ FILE *restrict stream,
+ const char *restrict fn,
+ const char *restrict data_description,
+ int close_on_error)
+{
+ size_t members_read = fread(data, size, nmemb, stream);
+ if (members_read != nmemb) {
+ fprintf(stderr, "Could not read %s from file %s\n", data_description, fn);
+ perror("");
+ if (close_on_error)
+ fclose(stream);
+ return 1;
+ }
+ return 0;
+}
+
+
+/* Wrapper around fwrite(). Prints error message on error. If close_on_error
+ * is non-zero, also closes the stream.
+ * Returns 0 on success and 1 on error.
+ */
+int
+fwrite_perror(const void * restrict data,
+ size_t size, size_t nmemb,
+ FILE *restrict stream,
+ const char *restrict fn,
+ const char *restrict data_description,
+ int close_on_error)
+{
+ size_t members_written = fwrite(data, size, nmemb, stream);
+ if (members_written != nmemb) {
+ fprintf(stderr, "Could not write %s to file %s\n", data_description, fn);
+ perror("");
+ if (close_on_error)
+ fclose(stream);
+ return 1;
+ }
+ return 0;
+}
+
+size_t
+compute_filesize(const size_t nr_limbs)
+{
+ return sizeof(s_file_header_t) + sizeof(int) + nr_limbs * sizeof(mp_limb_t);
+}
/* For the batch mode */
/* Write the batch exponent s in a file */
/* Return the number of bytes written */
int
-write_s_in_file (char *fn, mpz_t s)
+write_s_in_file (const char *fn, mpz_t s, int want_mmap, uint64_t B1)
{
FILE *file;
int ret = 0;
@@ -671,6 +745,20 @@ write_s_in_file (char *fn, mpz_t s)
exit (EXIT_FAILURE);
}
#endif
+#ifndef HAVE_MMAP
+ if (want_mmap != 0) {
+ fprintf(stderr, "Refusing to write batch s data to file %s in mmap() "
+ "format because mmap() is not supported on this system. Please "
+ "check the the configure log.", fn);
+ return 0;
+ }
+#endif
+
+ if (mpz_sgn(s) <= 0) {
+ fprintf(stderr, "Error, s is %s. This should never happen.",
+ mpz_sgn(s) == 0 ? "zero" : "negative");
+ return 0;
+ }
file = fopen (fn, "wb");
if (file == NULL)
@@ -679,8 +767,26 @@ write_s_in_file (char *fn, mpz_t s)
return 0;
}
- ret = mpz_out_raw (file, s);
+ const uint32_t magic = htole32(S_FILE_MAGIC);
+ const unsigned char mmapped = want_mmap ? 1 : 0;
+ const unsigned char is_le = htole32(1) == 1 ? 1 : 0;
+ const uint32_t checksum = mpz_fdiv_ui(s, S_FILE_CHECKSUM_MODULUS);
+ const uint32_t B1le = htole64(B1);
+ const s_file_header_t header = {magic, 1, mmapped, is_le, 0, checksum, B1le};
+ if (fwrite_perror(&header, sizeof(s_file_header_t), 1, file, fn, "header", 1))
+ return 0;
+ if (want_mmap) {
+ if (fwrite_perror(&s->_mp_size, sizeof(int), 1, file, fn, "size", 1))
+ return 0;
+ if (fwrite_perror(s->_mp_d, sizeof(mp_limb_t), s->_mp_size, file, fn, "limbs", 1))
+ return 0;
+ ret = compute_filesize(s->_mp_size);
+ } else {
+ ret = mpz_out_raw (file, s);
+ }
+ /* gmp_printf("Wrote %Zx to %s\n", s, fn); */
+
fclose (file);
return ret;
}
@@ -688,7 +794,7 @@ write_s_in_file (char *fn, mpz_t s)
/* For the batch mode */
/* read the batch exponent s from a file */
int
-read_s_from_file (mpz_t s, char *fn, double B1)
+read_s_from_file (mpz_t s, const char *fn, int want_mmap, double B1)
{
FILE *file;
mpz_t tmp, tmp2;
@@ -703,22 +809,117 @@ read_s_from_file (mpz_t s, char *fn, double B1)
}
#endif
+#ifndef HAVE_MMAP
+ if (want_mmap != 0) {
+ fprintf(stderr, "Cannot mmap() batch s data from file %s because mmap() "
+ "is not supported on this system. Please check the the configure "
+ "log.", fn);
+ }
+#endif
+
file = fopen (fn, "rb");
if (file == NULL)
{
fprintf (stderr, "Could not open file %s for reading\n", fn);
return 1;
}
-
- ret = mpz_inp_raw (s, file);
- if (ret == 0)
- {
- fprintf (stderr, "read_s_from_file: 0 bytes read from %s\n", fn);
+
+ const unsigned char host_is_le = htole32(1) == 1 ? 1 : 0;
+ s_file_header_t header;
+ if (fread_perror(&header, sizeof(s_file_header_t), 1, file, fn, "header", 1))
+ return 1;
+ header.magic = le32toh(header.magic);
+ header.B1 = le64toh(header.B1);
+ if (header.magic != S_FILE_MAGIC) {
+ fprintf(stderr, "Wrong magic number in file %s; is this a batch "
+ "product save file? Note that the file format changed after "
+ "version 7.0; if this is an old file, please re-create it.\n",
+ fn);
+ fclose(file);
+ return 1;
+ }
+ if (header.mmapped != want_mmap) {
+ fprintf(stderr, "File %s uses %smmapped data but we want to read "
+ "%smmaped data.\n",
+ fn, header.mmapped ? "" : "non-", want_mmap ? "" : "non-");
+ fclose(file);
return 1;
+ }
+ if (header.B1 != (unsigned long) B1) {
+ fprintf(stderr, "Wrong B1 value in file %s; it has stored data for B1 = %"
+ PRIu64 " but this run uses B1 = %" PRIu64 "\n",
+ fn, header.B1, (uint64_t) B1);
+ fclose(file);
+ return 1;
+ }
+
+#ifdef HAVE_MMAP
+ if (want_mmap != 0) {
+ /* printf("Using mmap() for s data\n"); */
+
+ if (header.is_le != host_is_le) {
+ fprintf(stderr, "Cannot mmap because file %s has wrong endianness for"
+ "this system. File uses %s, but host is %s\n",
+ fn, header.is_le ? "little-endian" : "big-endian",
+ host_is_le ? "little-endian" : "big-endian");
+ fclose(file);
+ return 1;
+ }
+
+ int fd = fileno(file);
+ if (fd == -1) {
+ fprintf (stderr, "Could not get file descriptor for file %s\n", fn);
+ fclose(file);
+ return 1;
+ }
+
+ int nr_limbs;
+ if (fread_perror(&nr_limbs, sizeof(int), 1, file, fn, "size", 1))
+ return 1;
+ if (nr_limbs < 0) {
+ fprintf(stderr, "Error, nr_limbs in file %s is negative (%d). "
+ "This should never happen.", fn, nr_limbs);
}
+ size_t filesize = compute_filesize(nr_limbs);
+ const int prot = PROT_READ;
+ const int flags = MAP_SHARED;
+ const off_t offset = 0;
+ char *s_data = mmap(NULL, filesize, prot, flags, fd, offset);
+ if (s_data == MAP_FAILED) {
+ fprintf (stderr, "mmap(NULL, length=%zu, prot=%d, flags=%d, fd=%d, offset=%zu) failed for file %s\n",
+ (size_t) filesize, prot, flags, fd, (size_t) offset, fn);
+ perror("");
+ fclose(file);
+ return 1;
+ }
+ mpz_clear (s);
+ s->_mp_size = nr_limbs;
+ s->_mp_alloc = s->_mp_size;
+ s->_mp_d = (mp_limb_t *) (s_data + sizeof(s_file_header_t) + sizeof(int));
+ }
+#endif
+
+ if (want_mmap == 0) {
+ ret = mpz_inp_raw (s, file);
+ if (ret == 0)
+ {
+ fprintf (stderr, "read_s_from_file: 0 bytes read from %s\n", fn);
+ fclose(file);
+ return 1;
+ }
+ }
fclose (file);
-
+
+ /* gmp_printf("Read %Zx from %s\n", s, fn); */
+ const uint32_t checksum = mpz_fdiv_ui(s, S_FILE_CHECKSUM_MODULUS);
+ if (checksum != header.checksum) {
+ fprintf(stderr, "Checksum in file %s is wrong. File header has %" PRIu32
+ " but checksum over file data is %" PRIu32 "\n",
+ fn, header.checksum, checksum);
+ return 1;
+ }
+
/* Some elementaty check that it correspond to the actual B1 */
mpz_init (tmp);
mpz_init (tmp2);
@@ -760,3 +961,17 @@ read_s_from_file (mpz_t s, char *fn, double B1)
return 0;
}
+void
+free_s_data(int want_mmap, mpz_t s)
+{
+#ifdef HAVE_MMAP
+ const int nr_limbs = s->_mp_size;
+ const size_t filesize = sizeof(s_file_header_t) +
+ sizeof(int) +
+ (size_t)nr_limbs * sizeof(mp_limb_t);
+ if (want_mmap) {
+ munmap(s->_mp_d, filesize);
+ mpz_init(s);
+ }
+#endif
+}
--
GitLab