| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
| |
rather than "../config.h" because the dependency generation script
checks for this as a special case and treats it relative to the
build directory.
* dep.mk: Regenerated.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
(comp_clamp): Bugfix: avoid shifting left into sign bit. Function
renamed to comp_trunc.
(logtrunc, ash): New functions.
* eval.c (eval_init): Registered logtrunc and ash intrinsics.
* lib.h (logtrunc, ash): Declared.
* mpi-patches/add-bitops (s_highest_bit_mp): Forward declaration for
added.
(mp_clamp_comp): Bugfix in handling remainder bits. Function
renamed to mp_trunc_comp.
(mp_trunc, mp_shift): New functions.
|
| |
|
|
|
|
|
|
| |
allow arbitrarily wide complements (which causes the code to access
beyond the a argument's digits array). A similar fix is applied in
the new mp_clamp_comp function. Incorrect initializations of mp_int
fixed in all the logic functions: mp_init was used instead of
mp_init_size.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
normalized, otherwise we end up with fixnum-range bignums.
(comp_clamp): New function.
(logcomp): Changed to two argument form. If second argument
is present (not nil) then call comp_clamp.
* eval.c (eval_init): Change registration of logcomp to allow
optional argument.
* lib.h (logcomp): Declaration updated.
* mpi-patches/add-bitops: New mp_clamp_comp function implemented.
|
| |
|
|
|
|
|
|
|
|
| |
* arith.c (logcomp): New function.
* eval.c (eval_init): logcomp registered as intrinsic.
* lib.h (logcomp) declared.
* mpi-patches/add-bitops: Fixed bugs in mp_xor. Implemented mp_comp.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
negative integers behave as an "infinite bit two's complement".
* arith.c (logand, logor, logxor): New functions.
* eval.c (eval_init): New intrinsic functions logand, logior, logxor.
* lib.h (logand, logor, logxor): Declared.
* mpi-patches/series: New patch, add-bitops.
* mpi-patches/add-bitops: New file.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
to work around TXR build problems by running make inside
the MPI directory, which is wrong.
Also fixing an incorrect use of ctype.h functions in
MPI and one case of a superfluous warning.
* mpi-patches/series: new patches added.
* mpi-patches/disable-make: New file.
* mpi-patches/fix-ctype-warnings: New file.
|
| |
|
|
|
|
|
|
| |
code on platforms where mp_digit is smaller than long.
This was left shifting a mp_digit by MP_DIGIT_BIT.
It was found by compiling on NetBSD with a gcc 4.1.3
which has a broken__int128_t, forcing us to use 32 bit
mp_digit with a 64 bit mp_word.
|
| |
|
|
|
|
|
|
| |
(minus): Floating point support.
* mpi-patches/mpi-to-double (mp_to_double): Re-apply lost
bugfix: index incremented instead of decremented.
Didn't refresh patch last time, then did a make distclean.
|
| |
|
|
|
|
|
|
|
|
|
| |
* eval.c (eval_init): flo-int registered as intrinsic.
* lib.h (flo_int): Declared.
* mpi-patches/series: Added mpi-to-double to patch stack.
(mp_to_double): New MPI function.
* mpi-patches/mpi-to-double: New file.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
variable on platforms where the MP digit is smaller than a long integer.
(Not anything TXR is known to run on). Changed algorithm to take the
first and last digit and add them together, rather than just taking the
last digit. The last digit will be zeros for numbers that contain 2 as a
factor with a large enough multiplicity.
* mpi-patches/add-mpi-toradix-with-case: Refreshed.
* mpi-patches/bit-search-optimizations: Likewise.
* mpi-patches/faster-square-root: Likewise.
* mpi-patches/fix-bad-shifts: Likewise.
* mpi-patches/fix-mult-bug: Likewise.
|
| |
|
|
|
| |
patch, resulting in an insufficient bit field width for representing
the allocation size of the MPI integer on 32 bit platforms.
|
| |
|
|
|
|
| |
It will take days to completely wipe the egg off my face.
I forgot to fix this code for unsigned integers before
pasting it into MPI.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
(abso, isqrt): New functions.
(isqrt_fixnum): New static function.
* eval.c (eval_init): Registered abs, sqrt and numberp instrinsics.
* lib.c (numberp): New function.
* lib.h (numberp, abso, isqrt): Declared.
* mpi-patches/series: New patch added.
* mpi-patches/faster-square-root: New patch added.
* txr.1: Documentation stubs for new functions.
|
| |
|
|
|
|
|
|
| |
so that obj_t stays four pointers wide.
* mpi-patches/series: New patch added.
* mpi-patches/shrink-mpi-int: New file.
|
| |
|
|
|
|
|
|
|
| |
storage class specifier.
* mpi-patches/fix-mult-bug (s_mp_sqr): More braindamage found in MPI.
This function performs additions and multiplication mp_digit,
expecting a mp_word precision result without casting. This function
is needed for exponentiation.
|
| |
|
|
|
|
|
|
|
| |
the highest bit, replacing them with an adapation of the
bit searching function used in arith.c.
* mpi-patches/series: Patch added.
* mpi-patches/bit-search-optimizations: New file.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
s_mp_mul_d and added to patch. This one caused malloc corruption and
crashes, because the incorrect arithmetic causes the function
to think that the multiplication will not be needing another
digit, but then there is a carry out which does spill into
a new digit.
* mpi-patches/series: Arg! Somehow the patch fix-bad-shift
went missing from the series file, even though the patch
itself is in the GIT repository.
|
| |
|
|
|
|
|
|
|
|
|
| |
code like 1 << n, where n exceeds the width of the type int.
* arith.c (trunc): New function, reimplementation of removed
trunc from lib.c.
* lib.c (trunc): Removed.
* mpi-patches/fix-bad-shifts: New file.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* arith.c: Regenerated.
* arith.txr (CNUM_BIT): New constant.
(bignum, bignum_dbl_ipt): New static functions.
(@{add-fname}): Use bignum function.
(mul): New functions, rewrite of mul from lib.c.
* lib.c (mul): Function removed.
* mpi-patches/add-mp-set-intptr (mp_set_intptr): Revised patch.
Local variable v should be int_ptr_t not unsigned long.
Also, the mp_set interface doesn't set the sign; it's an unsigned
interface. We must do that ourselves.
* mpi-patches/fix-mult-bug: The main multiplication function is
also broken in the same way, requiring the cast.
* mpi-patches/mpi-set-double-intptr: Fixed use of wrong type for
local variable v.
|
| |
|
|
|
|
| |
* mpi-patches/mpi-set-double-intptr: New file.
* mpi-patches/series (mpi-set-double-intptr): Patch added.
|
| |
|
|
|
|
| |
* mpi-patches/series (mpi-set-mpi-word): New patch.
* mpi-patches/mpi-set-mpi-word: New file.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
low-order bit material from the bignum to fill an unsigned long.
We don't need to walk the entire bignum. If the low order
digit of the bignum is at least as large as an unsigned long,
we just take that as the hash, otherwise we take enough of the
digits to fill an unsigned long. For negative numbers, we just
invert the bits of the hash.
* mpi-patches/add-mpi-toradix-with-case: Refreshed.
* mpi-patches/fix-mult-bug: Refreshed.
|
| |
|
|
|
|
|
|
|
|
| |
* configure (longlong, ulonglong, superlong, usuperlong): Initialize
these variables so that if the detection tests fail, the script
does not access unbound variables.
Avoid adding junk like .bss.* into config.h.
* mpi-patches/config-types: Fixed wrong use of nonexistent
SIZEOF_LONG_T.
|
| |
|
|
|
| |
* mpi-patches/config-types: Added missing definitions
of MP_DIGIT_SIZE in two cases.
|
|
|
Bignums, based on Michael Fromberger's MPI library, are integrated
into the input syntax, stream output, equality testing, the garbage
collector, and hashing.
The plus operation handles transitions between fixnums and bignums.
Other operations are still fixnum only.
* Makefile (CFLAGS): Add mpi directory to include file search.
(OBJS): Include new arith.o module and all of MPI_OBJS.
(MPI_OBJS, MPI_OBJS_BASE): New variables.
* configure (mpi_version, have_quilt, have_patch): New variables.
Script detects whether patch and quilt are available. Unpacks
mpi library, applies patches. Detects 128 bit integer type.
Records more information in config.h about the sizes of types.
* dep.mk: Updated.
* depend.txr: Make work with paths that have directory components.
* eval.c (eval_init): Rename of nump to fixnump.
* gc.c (finalize, mark_obj): Handle BGNUM case.
* hash.c: (hash_c_str): Changed to return unsigned long
instead of long.
(equal_hash): Handle BGNUM case.
(eql_hash): Handle bignums with equal-hash, but other
objects as eq.
* lib.c (num_s): Variable renamed to fixnum_s.
(bignum_s): New symbol variable.
(code2type): Follow rename of num_s. Handle BGNUM case.
(typeof): Follow rename of num_s.
(eql): Handle bignums using equal, and other types using eq.
(equal): Handle BGNUM case.
(chk_calloc): New function.
(c_num): Wording change in error message: is not a fixnum.
(nump): Renamed to fixnump.
(bignump): New function.
(plus): Function removed, reimplemented in arith.c.
(int_str): Handle integers which are too large for wcstol
using bignum conversion. Base 0 is no longer passed to
wcstol but converted to 10 because the special semantics
for 0 would be inconsistent for bignums.
(obj_init): Follow rename of num_s. Initialize bignum_s.
|
