Home My Page Projects MPFR
Summary Activity Forums Tracker Lists Tasks Docs News SCM Files

Activity

Activity:
Start Date:
End Date:
Time Activity By
2016-07-01
00:56:37 Source Code scm commit: [src/add1sp.c] Corrected indentation. (r10538) Vincent Lefèvre
00:51:59 Source Code scm commit: [src/sub1sp.c] Added comments. (r10537) Vincent Lefèvre
00:47:19 Source Code scm commit: [src/add1sp.c] Added comments. (r10536) Vincent Lefèvre
2016-06-30
07:57:08 Source Code scm commit: optimize mpfr_add (and mpfr_sub, not yet finished) for 1 limb (r10535) Paul Zimmermann
2016-06-29
12:17:49 Source Code scm commit: Renamed macro MPFR_EXP_LESS_P to MPFR_UBF_EXP_LESS_P. (r10534) Vincent Lefèvre
12:16:16 Source Code scm commit: [src/add1sp.c] Fixed bug introduced in r10530 when MPFR_WANT_ASSERT >= 2 (--enable-assert=full). (r10533) Vincent Lefèvre
12:11:44 Source Code scm commit: [src/add1.c] Re-added the MPFR_ASSERTD. The culprit seems to be r10530. (r10532) Vincent Lefèvre
12:09:05 Source Code scm commit: [src/add1.c] Reverted r10529, which now fails. That's strange. (r10531) Vincent Lefèvre
12:00:06 Source Code scm commit: mpfr_add1sp does not assume any more EXP(b) >= EXP(c) (r10530) Paul Zimmermann
11:49:30 Source Code scm commit: [src/add1.c] Added a MPFR_ASSERTD. (r10529) Vincent Lefèvre
09:56:17 Source Code scm commit: still faster code for 1 limb (r10528) Paul Zimmermann
09:14:57 Source Code scm commit: merged r10522 from faithful branch + other special-code for 1 limb: the worst case for mpfr_add in 53 limbs decreases from 70 to 60 cycles (r10527) Paul Zimmermann
08:39:06 Source Code scm commit: removed leftover debug statement (r10526) Paul Zimmermann
08:08:35 Source Code scm commit: added comment that EXP(b) >= EXP(c) (r10525) Paul Zimmermann
07:42:36 Source Code scm commit: mpfr_add1sp no longer assumes EXP(b) >= EXP(c): gives small speedup (r10524) Paul Zimmermann
2016-06-27
17:08:25 Source Code scm commit: [doc/sum.txt] Changed "hole" to "gap". (r10523) Vincent Lefèvre
08:31:19 Source Code scm commit: added special code for 1 limb (r10522) Paul Zimmermann
2016-06-25
15:17:34 Source Code scm commit: C++ compatibility. (r10521) Vincent Lefèvre
14:43:20 Source Code scm commit: Merged the latest changes from the trunk. (r10520) Vincent Lefèvre
00:47:35 Source Code scm commit: [autogen.sh] Treat autoreconf warnings as errors. (r10519) Vincent Lefèvre
2016-06-24
23:55:21 Source Code scm commit: [src/vasprintf.c] Some improvements. In particular, avoid two successive calls to mpfr_cmpabs on the same arguments (the compiler doesn't know that it could optimize, as the function is not pure). (r10518) Vincent Lefèvre
23:40:49 Source Code scm commit: [src/vasprintf.c] Fixed an assertion failure. (r10517) Vincent Lefèvre
20:28:29 Source Code scm commit: fixed reuse with --enable-assert (r10516) Paul Zimmermann
15:57:54 Source Code scm commit: [sum-timings/sum-new.c] Merged the latest changes from trunk with svn merge '^/trunk/src/sum.c' sum-new.c and updated the note. (r10515) Vincent Lefèvre
15:28:14 Source Code scm commit: small speedup of mpfr_add for RNDF (r10514) Paul Zimmermann
15:02:03 Source Code scm commit: [tests/tsum.c] Correction: r10484 didn't take the latest version from the trunk in the conflict resolution. (r10513) Vincent Lefèvre
14:51:00 Source Code scm commit: [src/sum.c] A more efficient way to support faithful rounding, avoiding the TMD completely and an assertion failure at the same time. (r10512) Vincent Lefèvre
14:30:55 Source Code scm commit: Avoid failures in mpfr_mul for MPFR_RNDF with MPFR_WANT_ASSERT >= 2 (--enable-assert=full) by disabling the comparison with the old implementation, which doesn't make sense for faithful rounding. (r10511) Vincent Lefèvre
14:26:16 Source Code scm commit: Avoid failures in mpfr_add1sp / mpfr_sub1sp for MPFR_RNDF with MPFR_WANT_ASSERT >= 2 (--enable-assert=full) by disabling the comparison with mpfr_add1 / mpfr_sub1, which doesn't make sense for faithful rounding. (r10510) Vincent Lefèvre
14:21:13 Source Code scm commit: Cosmetic change (whitespace). (r10509) Vincent Lefèvre
14:10:45 Source Code scm commit: Merged the latest changes from the trunk. (r10508) Vincent Lefèvre
14:09:35 Source Code scm commit: Removed trailing spaces. (r10507) Vincent Lefèvre
14:06:17 Source Code scm commit: [src/add1sp.c] For MPFR_WANT_ASSERT >= 2 (--enable-assert=full), test the flags too and replaced "mpfr_cmp" by "! mpfr_equal_p". [src/sub1sp.c] Make it similar to add1sp.c for MPFR_WANT_ASSERT >= 2. (r10506) Vincent Lefèvre
14:00:22 Source Code scm commit: use macro for the rounding mode used in the tests (r10505) Paul Zimmermann
13:44:21 Source Code scm commit: [doc/mpfr.texi] Updated the month. (r10504) Vincent Lefèvre
12:17:25 Source Code scm commit: [src/sum.c] Made the code a bit more natural: change of pos to neg (an indirect consequence of the support of reused arguments). Some other minor related changes. (r10503) Vincent Lefèvre
10:59:59 Source Code scm commit: added more tests for RNDF (r10502) Paul Zimmermann
10:17:40 Source Code scm commit: The support of reused arguments in mpfr_sum is now complete, without any visible regressions in the timings (which are unfortunately too inaccurate to see anything). * doc/mpfr.texi: removed sentences about the previous limitation. * src/sum.c: removed the corresponding FIXME/TODO comment. (r10501) Vincent Lefèvre
10:11:10 Source Code scm commit: [tests/tsum.c] Test the support of reused arguments. (r10500) Vincent Lefèvre
08:34:19 Source Code scm commit: disable mpfr_all_div() for RNDF (r10499) Paul Zimmermann
08:24:01 Source Code scm commit: replaced mpfr_cmp by !mpfr_equal_p (r10498) Paul Zimmermann
08:20:41 Source Code scm commit: added exhaustive tests for add,sub,mul,div,sqrt (r10497) Paul Zimmermann
07:54:46 Source Code scm commit: fixed corner case with RNDF (r10496) Paul Zimmermann
2016-06-23
12:20:19 Source Code scm commit: [www/pub.html] Removed a TODO. (r10495) Vincent Lefèvre
11:58:08 Source Code scm commit: [www/pub.html] Added PhD theses. (r10494) Vincent Lefèvre
11:17:51 Source Code scm commit: [src/mpfr-impl.h] Added a comment for MPFR_RNDRAW_GEN about the use of the exponent field of the destination. (r10493) Vincent Lefèvre
11:09:48 Source Code scm commit: [src/add1sp.c] For MPFR_RNDF, set inexact = 0 (like what is done in sub1.c) instead of 1, though this doesn't really matter. (r10492) Vincent Lefèvre
11:07:45 Source Code scm commit: [src/sub1.c] Fix: the conflict resolution done in r10484 was incorrect. (r10491) Vincent Lefèvre
10:59:00 Source Code scm commit: [src/add1sp.c] Avoid the use of uninitialized values with MPFR_RNDF. (r10490) Vincent Lefèvre
10:41:48 Source Code scm commit: [www/pub.html] Update about PhD theses. (r10489) Vincent Lefèvre
10:31:18 Source Code scm commit: [www/pub.html] Update about PhD theses. (r10488) Vincent Lefèvre
10:17:16 Source Code scm commit: [www/pub.html] Updated a PhD thesis. (r10487) Vincent Lefèvre
10:15:28 Source Code scm commit: fix for RNDF (r10486) Paul Zimmermann
10:11:57 Source Code scm commit: [www/pub.html] Added an article. (r10485) Vincent Lefèvre
09:52:19 Source Code scm commit: merged changed from trunk with svn merge '^/trunk' (resolved conflict for sub1.c; copied tests/tsum.c from trunk and re-incorporated changes from faithful branch) (r10484) Paul Zimmermann
09:12:45 Source Code scm commit: src/li2.c: better like this (r10483) Paul Zimmermann
08:53:42 Source Code scm commit: src/li2.c: revision 10379 was ok (r10482) Paul Zimmermann
08:36:27 Source Code scm commit: fixed mpfr_all_div() for RNDF (r10481) Paul Zimmermann
2016-06-22
21:13:45 Source Code scm commit: more work on branch faithful (r10480) Paul Zimmermann
2016-06-18
22:41:01 Source Code scm commit: [sum-timings/sum-timings.c] Updated a comment. (r10479) Vincent Lefèvre
22:32:22 Source Code scm commit: [sum-timings/sum-timings.c] Improvement: Each test is done 5 times and one keeps the median. This allows one to avoid most big differences between two invocations of the same test, but not completely; a factor larger than 3 can still be obtained. (r10478) Vincent Lefèvre
2016-06-17
15:38:51 Source Code scm commit: [src/mpfr.h] Coding style: added spaces. (r10477) Vincent Lefèvre
15:17:34 Source Code scm commit: [TODO] Suggest the use the keyword "static" in array indices of parameter declarations with C99 compilers (6.7.5.3p7) when the pointer is expected not to be null. (r10476) Vincent Lefèvre
13:36:48 Source Code scm commit: [sum-timings/sum-timings.c] Minor improvements. (r10475) Vincent Lefèvre
2016-06-16
08:18:49 Source Code scm commit: [sum-timings/sum-timings.c] Added a comment: warning about the inaccuracy of these timings in practice. (r10474) Vincent Lefèvre
2016-06-15
14:24:54 Source Code scm commit: [sum-timings/cmp-timings] Differences above some given number of percents can also be output. (r10473) Vincent Lefèvre
13:35:53 Source Code scm commit: Added sum-timings/cmp-timings Perl script to compare the timings from two directories and store the results in a third directory. (r10472) Vincent Lefèvre
11:41:25 Source Code scm commit: [sum-timings] * sum-new.c: merged the latest changes from trunk with svn merge '^/trunk/src/sum.c' sum-new.c and updated the note. * ReadMe, sum-timings.c: one can now test some mpfr_sum only. (r10471) Vincent Lefèvre
11:31:22 Source Code scm commit: [sum-timings] Copied sum-new.c into results-20160316 as this is the version used for these results (the sum-new.c file in sum-timings may be modified later for new tests, whereas results-20160316/sum-new.c will never be modified). (r10470) Vincent Lefèvre
11:27:03 Source Code scm commit: In the sum-timings directory, renamed "results" to "results-20160316" to avoid confusion, since these are not the results for the latest mpfr_sum version. (r10469) Vincent Lefèvre
09:46:29 Source Code scm commit: [doc/README.dev] Update (GCC trunk → GCC 5). (r10468) Vincent Lefèvre
2016-06-13
09:06:29 Source Code scm commit: [src/sum.c] Copy the significand to the destination after resolving the TMD in order to support reused arguments. (r10467) Vincent Lefèvre
06:40:25 Source Code scm commit: [src/sum.c] Improved a comment. (r10466) Vincent Lefèvre
2016-06-10
13:58:34 Source Code scm commit: [src/sum.c] Corrected a MPFR_LOG_MSG modified in the latest change. (r10465) Vincent Lefèvre
13:47:29 Source Code scm commit: [src/sum.c] For the future support of reused arguments: TMD resolution is now done in a specific TMD accumulator, allocated at the same time as the main accumulator. This TMD accumulator currently just takes at most 2 limbs in practice, so that's not a problem. [doc/sum.txt] Added TODO for things that will change. (r10464) Vincent Lefèvre
2016-06-09
14:30:32 Source Code scm commit: [src/mpfr-impl.h] Define mpfr_ueexp_t (needed for r10462). (r10463) Vincent Lefèvre
14:09:11 Source Code scm commit: [tests/tversion.c] Do not use "%zu" with printf (added in r10460). (r10462) Vincent Lefèvre
12:41:16 Source Code scm commit: [doc/README.dev] Document environment variables that affect the tests. (r10461) Vincent Lefèvre
12:15:14 Source Code scm commit: Other changes concerning the memory limit for the tests. * Export the tests_memory_limit variable so that it can be accessed in test programs (e.g. read by tversion as mentioned below, or modified by a test program when executed with particular arguments). * In tversion, print a warning when the memory limit has been modified (with the MPFR_TESTS_MEMORY_LIMIT environment variable). (r10460) Vincent Lefèvre
11:51:52 Source Code scm commit: [tests/memory.c] The MPFR_TESTS_MEMORY_LIMIT environment variable can now contain an integer specifying the memory limit for the tests, or 0 for unlimited, the default still being 2^22 = 4 MB. (r10459) Vincent Lefèvre
2016-06-08
15:21:59 Source Code scm commit: [vl-tests/mpfrtests.{data,sh}] Update. (r10458) Vincent Lefèvre
12:28:43 Source Code scm commit: [tests/tsum.c] Improved check3: * Also do the tests with output precision MPFR_PREC_MIN; this completes the value coverage for tsum. * Also compare the flags. (r10457) Vincent Lefèvre
12:06:57 Source Code scm commit: Value coverage for tsum: After r9984 to differentiate sq > MPFR_PREC_MIN and sq == MPFR_PREC_MIN, 42 tests were not done for sq == MPFR_PREC_MIN on a 64-bit machine. In the check4 test, changed a 2 to MPFR_PREC_MIN. This reduces to 12 tests that are not done. (r10456) Vincent Lefèvre
2016-06-07
17:29:02 Source Code scm commit: Removed ubf branch (since it has been merged into the trunk). (r10455) Vincent Lefèvre
17:27:04 Source Code scm commit: Merged the "ubf" branch into the trunk: * limited implementation of "unbounded floats" (numbers with a mpz_t exponent), which could be completed in the future, in particular to handle intermediate overflows/underflows in polynomial evaluation; * rewrote mpfr_fmma and mpfr_fmms to use them, and added tests; * restored the MPFR 3.1- exponent range for compatibility; * restored some tests that were based on this exponent range. The choice for the UBF implementation was to use a compatible structure: the same as the mpfr_t one, with an additional field at the end, used only when the exponent field _mpfr_exp has some special value. Thus some functions such as addition and subtraction just needed minor changes to support UBF, and no code was duplicated. This means that on MPFR numbers (the usual case), one has a few more tests than before, but with known branch probabilities. In practice, this should not be visible, at least in most applications. This was more or less confirmed by timings (even with many additions/subtractions in low precision, this was difficult to say, as timings were influenced by various parameters that we cannot control). (r10454) Vincent Lefèvre
10:22:49 Source Code scm commit: Merged the latest change from the trunk (r10452). (r10453) Vincent Lefèvre
10:22:03 Source Code scm commit: [tools/mbench/Makefile] Added multiarch support for GMP. (r10452) Vincent Lefèvre
08:31:31 Source Code scm commit: [src/mpfr-impl.h] Updated a comment (removing a FIXME). (r10451) Vincent Lefèvre
07:32:09 Source Code scm commit: Merged the latest change from the trunk (r10449). (r10450) Vincent Lefèvre
07:26:52 Source Code scm commit: [tests/tprintf.c] Removed tests of native %'g and %'f (from r8292) as the ' flag is an extension from Single UNIX Specification and in particular, they fail with MinGW under Wine. (r10449) Vincent Lefèvre
00:44:56 Source Code scm commit: Merged the latest changes from the trunk. (r10448) Vincent Lefèvre
00:44:17 Source Code scm commit: [tests/tlgamma.c] Typo in comment. (r10447) Vincent Lefèvre
00:23:13 Source Code scm commit: [src/uceil_log2.c] Added a comment. Cosmetic changes. (r10446) Vincent Lefèvre
00:18:47 Source Code scm commit: [tests/tlgamma.c] Corrected updated comment from r10444. (r10445) Vincent Lefèvre
2016-06-06
23:45:31 Source Code scm commit: [tests/tlgamma.c] Updated comment of the test added in r10442. (r10444) Vincent Lefèvre
23:43:44 Source Code scm commit: [src/uceil_log2.c] Correction in __gmpfr_ceil_log2, avoiding an incorrect result with tcc: x.s.exp is declared as an unsigned bit-field, so that tcc considers that x.s.exp - 1023 is unsigned. However, since all the values of x.s.exp are representable in an int, according to the integer promotion rules, x.s.exp should be converted to an int, so that the subtraction is signed. So, this appears to be a bug in tcc. (r10443) Vincent Lefèvre
23:09:17 Source Code scm commit: [tests/tlgamma.c] Added a test causing a "too much memory" error with tcc 0.9.27~git20151227.933c223-1 (there's already one in special(), but this one is a simpler, standalone test). (r10442) Vincent Lefèvre
16:20:38 Source Code scm commit: Merged the latest change from the trunk (r10440). (r10441) Vincent Lefèvre
16:19:21 Source Code scm commit: [acinclude.m4] When checking if __float128 is available, we now also check whether C99 constants (in particular the __float128 ones, such as 0x1.fp+16383q) are supported since this is now required with the __float128 support. (r10440) Vincent Lefèvre
15:57:16 Source Code scm commit: Merged the latest changes from the trunk. (r10439) Vincent Lefèvre
13:15:40 Source Code scm commit: [src/add1.c] Improved a comment about faithful rounding. (r10438) Vincent Lefèvre
12:53:15 Source Code scm commit: [src/add1.c] The code for faithful rounding (MPFR_RNDF) was not avoiding some useless loops needing to know on which side of a breakpoint the result is. Quick fix. (r10437) Vincent Lefèvre
12:20:09 Source Code scm commit: Merged the latest changes from the trunk. (r10436) Vincent Lefèvre
12:15:56 Source Code scm commit: [src/add1.c] Simplified code: removed fb = 1 since it was already 1 in the "if (fb > 0)" branch. (r10435) Vincent Lefèvre
10:59:02 Source Code scm commit: [src/set_float128.c] Fix: removed the divisions by zero, replacing them by comparisons with the maximum finite binary128 number. Changes to be similar to the handling of infinities in "src/set_ld.c". (r10434) Vincent Lefèvre
10:07:47 Source Code scm commit: [TODO] Faithful rounding: small correction. (r10433) Vincent Lefèvre
09:28:17 Source Code scm commit: [TODO] Added "faithful rounding", with information about its definition and implementation. (r10432) Vincent Lefèvre
2016-06-03
15:51:24 Source Code scm commit: [src/{fits_intmax.c,fits_s.h}] In comments, fit → fits. (r10431) Vincent Lefèvre
15:30:53 Source Code scm commit: [src/set_float128.c] Added FIXME comment: The mpfr_set_float128 code generates a divide-by-zero exception. Suggest a solution. Note: this is detected with "-DMPFR_TESTS_DIVBYZERO -DMPFR_ERRDIVZERO". (r10430) Vincent Lefèvre
15:02:05 Source Code scm commit: [tests/tset_float128.c] Take into account failures in division by 0 and absence of signed zeros. Also fixed the code checking the sign of 0 (mpfr_sgn cannot be used for that since it returns 0 for ±0). (r10429) Vincent Lefèvre
15:00:22 Source Code scm commit: RNDF: fixed the fits functions (r10428) Paul Zimmermann
14:20:52 Source Code scm commit: Merged the latest change from the trunk (r10426 fix). (r10427) Vincent Lefèvre
14:19:33 Source Code scm commit: [src/set_d.c] Added some assertions and fixed a 32-bit shift count when GMP_NUMB_BITS == 32. (r10426) Vincent Lefèvre
13:55:12 Source Code scm commit: Merged the latest changes from the trunk (C++ compatibility). (r10425) Vincent Lefèvre
13:54:29 Source Code scm commit: [tests/tset_float128.c] C++ compatibility. (r10424) Vincent Lefèvre
13:50:17 Source Code scm commit: [tests/tlog.c] Added a cast to mpfr_rnd_t for C++ compatibility. (r10423) Vincent Lefèvre
13:28:45 Source Code scm commit: [tests/tfmma.c] Forgot a cast to mpfr_rnd_t for C++ compatibility. (r10422) Vincent Lefèvre
13:05:57 Source Code scm commit: [src/add1.c] Completed UBF support. Note: due to the restriction on the exponent values, diff_exp does not need to be the unsigned integer type mpfr_uexp_t (mpfr_exp_t is sufficient). (r10421) Vincent Lefèvre
13:03:52 Source Code scm commit: [tests/tfmma.c] Correction. (r10420) Vincent Lefèvre
12:47:12 Source Code scm commit: [tests/tfmma.c] Test (1/2)x + (1/2)x = x near underflow. (r10419) Vincent Lefèvre
12:44:20 Source Code scm commit: [tests/tfmma.c] Replaced precision 64 by GMP_NUMB_BITS (to make sure that the add1sp and sub1sp conditions are satisfied with the current src code, these tests may rely on the fact that there are no trailing bits, i.e. that the precision is a multiple of GMP_NUMB_BITS). (r10418) Vincent Lefèvre
12:26:21 Source Code scm commit: added reference for Lambert W function (r10417) Paul Zimmermann
09:21:00 Source Code scm commit: [src/{add,sub}.c] Consider UBF numbers as special cases so that mpfr_sub1sp and mpfr_add1sp, which do not support UBF, are never called on UBF numbers. This should also (very slightly) speed up the normal cases. (r10416) Vincent Lefèvre
09:05:52 Source Code scm commit: [src/mpfr-impl.h] UBF support: added MPFR_IS_SINGULAR_OR_UBF and MPFR_ARE_SINGULAR_OR_UBF macros. (r10415) Vincent Lefèvre
08:21:44 Source Code scm commit: declare prec in unsigned long (r10414) Paul Zimmermann
07:24:43 Source Code scm commit: RNDF: specified the fits functions (r10413) Paul Zimmermann
06:10:39 Source Code scm commit: RNDF: fixed tfactorial (r10412) Paul Zimmermann
2016-06-02
23:46:02 Source Code scm commit: C++ compatibility (corrected). (r10411) Vincent Lefèvre
16:20:26 Source Code scm commit: RNDF: fixed texp2 (r10410) Paul Zimmermann
16:16:48 Source Code scm commit: avoid mpfr_printf in tests (r10409) Paul Zimmermann
16:15:31 Source Code scm commit: C++ compatibility (r10408) Paul Zimmermann
16:12:28 Source Code scm commit: suggestion from Vincent (r10407) Paul Zimmermann
16:06:45 Source Code scm commit: RNDF: fixed texp10 and tgeneric (r10406) Paul Zimmermann
16:03:44 Source Code scm commit: RNDF: fixed mpfr_sum and tsum (r10405) Paul Zimmermann
15:50:24 Source Code scm commit: RNDF: fixed texp (r10404) Paul Zimmermann
15:36:44 Source Code scm commit: [tests/tfmma.c] Added test cases where the precision of the result is twice the precision of each input, which can currently involve add1sp.c and sub1sp.c code. (r10403) Vincent Lefèvre
14:59:10 Source Code scm commit: RNDF: partially fixed texp (r10402) Paul Zimmermann
14:55:53 Source Code scm commit: RNDF: fixed tcot (r10401) Paul Zimmermann
14:49:09 Source Code scm commit: RNDF: fixed tcos (r10400) Paul Zimmermann
14:46:04 Source Code scm commit: RNDF: fixed tconst_pi (r10399) Paul Zimmermann
14:41:40 Source Code scm commit: RNDF: fixed li2 and tconst_log2 (r10398) Paul Zimmermann
14:37:48 Source Code scm commit: Merged the latest changes from the trunk. (r10397) Vincent Lefèvre
14:34:35 Source Code scm commit: [src/exceptions.c] Fixed the MPFR_LOG_FUNC in mpfr_underflow and mpfr_overflow (x is not an input in these functions). (r10396) Vincent Lefèvre
14:26:51 Source Code scm commit: [src/sub1.c] In r10347, logging was introduced instead of #ifdef DEBUG and printf. Removed 2 MPFR_LOG_VAR, as the variable is not normalized yet. Re-added #ifdef DEBUG for the MPFR_LOG_MSG's since this may be too low level and not much useful except in case of bugs. (r10395) Vincent Lefèvre
14:20:24 Source Code scm commit: RNDF: fixed issue with Bessel functions for large arguments (r10394) Paul Zimmermann