you should compile with TTMATH_MULTITHREADS
and use TTMATH_MULTITHREADS_HELPER macro somewhere in your *.cpp file
added file: ttmaththreads.h (look there for more info)
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@183 e52654a7-88a9-db11-a3e9-0013d4bc506e
* added: gamma() function to the parser
* added: Big::IsInteger() method
returns true if the value is integer
* added: CGamma<ValueType> class
is used with Gamma() and Factorial() in multithreaded environment
* changed: Factorial() is using the Gamma() function now
* removed: Parser<>::SetFactorialMax() method
the factorial() is such a fast now that we don't need the method longer
* removed: ErrorCode::err_too_big_factorial
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@178 e52654a7-88a9-db11-a3e9-0013d4bc506e
wide characters are used when macro TTMATH_USE_WCHAR is defined
this macro is defined automatically when there is macro UNICODE or _UNICODE defined
some types have been changed
char -> tt_char
std::string -> tt_string
std::ostringstream -> tt_ostringstream
std::ostream -> tt_ostream
std::istream -> tt_istream
normally tt_char is equal char but when you are using wide characters then tt_char will be wchar_t (and so on)
(all typedef's are in ttmathtypes.h)
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@177 e52654a7-88a9-db11-a3e9-0013d4bc506e
fixed: some methods should set 'err_improper_argument' when the argument is a NaN object
(was: err_overflow)
changed: version of the library: 0.8.5 now
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@163 e52654a7-88a9-db11-a3e9-0013d4bc506e
the remainder from a division '%' was changed with a bitwise And operation '&'
((i % 5) == 0) was changed to: ((i & 3) == 0) - it means ((i % 4) == 0)
now the test if performed after 4 iterations (early were after 5 iterations)
we can do that when the divisor is a power of 2
changed: optimization in Factorial()
we're testing WasStopSignal() only after a few iterations
it's faster now about 4 times on GCC 4.3.3 (when stop object is provided to the factorial)
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@161 e52654a7-88a9-db11-a3e9-0013d4bc506e
we can copy 'old_value = *this' only when 'testing' is true
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@160 e52654a7-88a9-db11-a3e9-0013d4bc506e
changed: some cosmetic changes to get rid of warnings generated by
MS Visual 2008 when warning level is 4
changed: names of labels in asm code: p to ttmath_loop, end to ttmath_end
p2 to ttmath_loop2
Robert Muir reported that there was a confict with boost::end
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@156 e52654a7-88a9-db11-a3e9-0013d4bc506e
bool Big::IsNan() - returns true if the NaN flag is set
void Big::SetNan() - sets the NaN flag
The NaN flag is set by default after creating an object:
Big<1, 2> a; // NaN is set (it means the object has not a valid number)
std::cout << a; // cout gives "NaN"
a = 123; // now NaN is not set
std::cout << a; // cout gives "123"
The NaN is set if there was a carry during calculations
a.Mul(very_big_value); // a will have a NaN set
The NaN is set if an argument is NaN too
b.SetNan();
a.Add(b); // a will have NaN because b has NaN too
If you try to do something on a NaN object, the result is a NaN too
a.SetNan();
a.Add(2); // a is still a NaN
The NaN is set if you use incorrect arguments
a.Ln(-10); // a will have the NaN flag
The only way to clear the NaN flag is to assign a correct value or other correct object
supposing 'a' has NaN flag, to remove the flag you can either:
a = 10;
a.FromInt(30);
a.SetOne();
a.FromBig(other_object_without_nan);
etc.
changed: renamed macro CONSTANTSGENERATOR to TTMATH_CONSTANTSGENERATOR
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@152 e52654a7-88a9-db11-a3e9-0013d4bc506e
void UInt::Mul3Big(const UInt<value_size> & ss2, UInt<value_size*2> & result)
a new multiplication algorithm: Karatsuba multiplication,
on a vector UInt<100> with all items different from zero this algorithm is faster
about 3 times than Mul2Big(), and on a vector UInt<1000> with all items different from
zero this algorithm is faster more than 5 times than Mul2Big()
(measured on 32bit platform with GCC 4.3.3 with -O3 and -DTTMATH_RELEASE)
added: uint MulFastest(const UInt<value_size> & ss2)
void MulFastestBig(const UInt<value_size> & ss2, UInt<value_size*2> & result)
those methods are trying to select the fastest multiplication algorithm
changed: uint Mul(const UInt<value_size> & ss2, uint algorithm = 100)
void MulBig(const UInt<value_size> & ss2, UInt<value_size*2> & result, uint algorithm = 100)
those methods by default use MulFastest() and MulFastestBig()
changed: changed a little Mul2Big() to cooperate with Mul3Big()
changed: names of methods in macros TTMATH_LOG()
added: uint AddVector(const uint * ss1, const uint * ss2, uint ss1_size, uint ss2_size, uint * result)
uint SubVector(const uint * ss1, const uint * ss2, uint ss1_size, uint ss2_size, uint * result)
three forms: asm x86, asm x86_64, no-asm
those methods are used by the Karatsuba multiplication algorithm
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@148 e52654a7-88a9-db11-a3e9-0013d4bc506e
and the result was sometimes very big (even greater than x)
fixed: global function Mod(x) didn't set an ErrorCode object
fixed: global function Round() didn't test a carry
now it sets ErrorCode object
changed: function Sin(x) to Sin(x, ErrorCode * err=0)
when x was very big the function returns zero
now it sets ErrorCode object to err_overflow
and the result is undefined
the same is to Cos() function
changed: PrepareSin(x) is using Big::Mod() now when reducing 2PI period
should be a little accurate especially on a very big 'x'
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@143 e52654a7-88a9-db11-a3e9-0013d4bc506e
'typedef unsigned long long int ulint' has been put inside '#ifdef TTMATH_NOASM'
in order to not confuse a compiler while compiling with strict iso c++
(errors about not defining 'long long' in the c++ standard)
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@137 e52654a7-88a9-db11-a3e9-0013d4bc506e
changed: UInt::SubInt() in no-asm code is a little faster now
changed: small cosmetic changes in commentaries
deleted: some debug #ifdef's from UInt::Div() (in no-asm code)
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@136 e52654a7-88a9-db11-a3e9-0013d4bc506e
sometimes when using -O2 or -O3 GCC doesn't set correctly
the stack pointer (actually the stack is used for other things)
and you can't use instructions like push/pop in assembler code.
All the asm code in x86_64 have been rewritten, now instructions
push/pop are not used, other thing which have access to stack
(like "m" (mask) constraints in Rcl2 and Rcr2) have also gone away,
now the library works well with -O2 and -O3 and the asm code
is a little faster
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@127 e52654a7-88a9-db11-a3e9-0013d4bc506e
all the methods which are using assembler code have been
rewritten to no-asm forms, now we have:
1. asm for x86 file: ttmathuint_x86.h
2. asm for x86_64 file: ttmathuint_x86_64.h
3. no asm file: ttmathuint_noasm.h
(it's used when macro TTMATH_NOASM is defined)
The third form can be used on x86 and x86_64 as well and
on other platforms with a little effort.
(Temporarily I left there some '#ifdef's for debugging.)
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@126 e52654a7-88a9-db11-a3e9-0013d4bc506e
used for debugging purposes by macro TTMATH_LOG(msg)
(it is used in nearly all methods in UInt class)
added: macro TTMATH_DEBUG_LOG: when defined then TTMATH_LOG()
put some debug information (to std::cout)
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@125 e52654a7-88a9-db11-a3e9-0013d4bc506e
(radix) -- operators '#' and '&', e.g.:
'#sin(1)' was equal '0' -- there was a zero from '#' and then
it was multipied by 'sin(1)'
the parser didn't check whether Big::FromString() has actually
read a proper value -- the method Big::FromString() didn't have
something to report such a situation
fixed: Big::FromString() when the base is 10, the method reads the scientific
part only if such a part it correctly supplied, e.g:
'1234e10', '1234e+10', '1234e-5'
previous '1234e' was treated as: '1234e0' (now parsing stops on 'e' and
the 'e' can be parsed by other parsers, e.g. the mathematical
parser -- now in the parser would be: '1234e' = '1234 * e' = '3354,3597...' )
added: to Int::FromString(): parameter 'const char ** after_source = 0'
if exists it's pointing at the end of the parsed string
added: to UInt::FromString(), Int::FromString(), Big::FromString():
parameter 'bool * value_read = 0' - (if exists) tells
whether something has actually been read (at least one digit)
added: the parser checks itself for the operators for changing the base
(operators '#' and '&')
changed: in the parser: the form with operators '#' and '&' is as follows:
[-|+][#|&]numeric_value
previous was: [-|+][#|&][-|+]numeric_value
removed: Big::FromString() this method doesn't longer recognize operators
for changing the base ('#' and '&')
changed: in the parser: the short form of multiplication has the same
priority as the normal multiplication, e.g.:
'2x^3' = 2 * (x^3)
previous the priority was greater than powering priority
previous: '2x^3' = (2*x) ^ 3
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@113 e52654a7-88a9-db11-a3e9-0013d4bc506e
to make division first and then mutliplication -- the result is more
accurate especially when x is: 90,180,270 or 360
added: global template functions in ttmath.h:
ValueType GradToRad(const ValueType & x, ErrorCode * err = 0)
ValueType RadToGrad(const ValueType & x, ErrorCode * err = 0)
ValueType DegToGrad(const ValueType & x, ErrorCode * err = 0)
ValueType DegToGrad(const ValueType & d, const ValueType & m,
const ValueType & s, ErrorCode * err = 0)
ValueType GradToDeg(const ValueType & x, ErrorCode * err = 0)
added: Parser::SetDegRadGrad(int angle) - 0 deg, 1 rad (default), 2 grad
this affects following functions (in the parser only): sin, cos, tan, cot,
asin, acos, atan, acot
added: functions to the parser: gradtorad(grad), radtograd(rad), degtograd(deg),
degtograd(d,m,s), gradtodeg(grad)
removed: Big::Ln() and Big::Log() some old info was removed
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@110 e52654a7-88a9-db11-a3e9-0013d4bc506e
it's using PowInt() only when pow.exponent is in range (-man*TTMATH_BITS_PER_UINT; 0]
previously the powering 'hung' on an input like this: "(1+ 1e-10000) ^ 10e100000000"
(the was 10e100000000 iterations in PowInt())
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@108 e52654a7-88a9-db11-a3e9-0013d4bc506e
didn't correctly set the exponent (when the mantisses had different size -
when 'man' was different from 'another_man')
this had impact on operator= too
sample:
Big<2,3> a = 100;
Big<3,5> b;
b = a; // b had a wrong value
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@106 e52654a7-88a9-db11-a3e9-0013d4bc506e
changed: the size of built-in variables (constants) in ttmathbig.h
now they consist of 256 32bit words
macro TTMATH_BUILTIN_VARIABLES_SIZE is equal: 256u on a 32bit platform and 128ul on a 64bit platform
added: macro CONSTANTSGENERATOR which is useful for generating constants
(it's used by 'gen' program in 'constgen' directory)
after declaring this macro the methods: ExpSurrounding0() and LnSurrounding1() will be public visible
and the inner loop can do more iterations than TTMATH_ARITHMETIC_MAX_LOOP
changed: in methods: ExpSurrounding0() and LnSurrounding1() - the way of comparising with the last word
(now we're comparing not with the last state but with a state from a few words back)
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@105 e52654a7-88a9-db11-a3e9-0013d4bc506e
fixed: Big::Add incorrectly rounded 'this' when both exponents were equal
it caused that sometimes when adding a zero the result has changed
this had impact among other things on FromString() method
"0,8" had different binary representation from "0,80"
changed: renamed: Big::PowUInt(UInt<pow_size> pow) -> Big::Pow(UInt<pow_size> pow)
it returns 2 when there is: 0^0
changed: renamed: Big::PowInt(Int<pow_size> pow) -> Big::Pow(Int<pow_size> pow)
it returns 2 when there is: 0^0 or 0^(-something)
changed: renamed: Big::PowBUInt() -> PowUInt(), Big::PowBInt() -> Big::PowInt()
they return 2 when the arguments are incorrect (like above)
changed: UInt::SetBitInWord(uint & value, uint bit) is taking the first argument by a reference now,
the specific bit is set in the 'value' and the method returns the last state of the bit (zero or one)
added: UInt::GetBit(uint bit_index) - returning the state of the specific bit
changed: UInt::SetBit(uint bit_index) - it's using TTMATH_ASSERT now
changed: Big::Mod2() - it's using mantissa.GetBit() now
added: Big::operator=(double) and Big::Big(double)
added: TTMATH_ASSERT somewhere in ttmathuint64.h
added: UInt::Pow(UInt<value_size> pow) and Int::Pow(Int<value_size> pow)
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@104 e52654a7-88a9-db11-a3e9-0013d4bc506e
UInt::FindLeadingBitInWord, UInt::SetBitInWord
fixed: UInt::Div() didn't return a correct result when the divisor was equal 1
there was an error in UInt::DivInt() - when the divisor was 1 it returned
zero and the carry was set
this error was from the beginning of the TTMath library
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@103 e52654a7-88a9-db11-a3e9-0013d4bc506e
fixed: UInt::FindLeadingBitInWord(x) didn't correctly return result -1 when the x was 0
the mistake made by Thomas Braby, it was not in the 0.8.2 release
changed: tests: we do not test for lahf/sahf instructions now (in 64bit code)
we don't use these instructions
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@97 e52654a7-88a9-db11-a3e9-0013d4bc506e
the proper carry, (when 'bits' was greater than or equal to 'value_size')
this had impact on Rcl() and Rcr(), they also returned the wrong carry
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@85 e52654a7-88a9-db11-a3e9-0013d4bc506e
they are much faster than UInt::Rcl2(1,c) and Rcr2(1,c)
changed: UInt::Rcl() and UInt::Rcr()
we don't longer make the things with moving a half in the left and a half in the right
we simply use Rcl2_one() and Rcl2() /Rcr2_one() and Rcr2()/
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@84 e52654a7-88a9-db11-a3e9-0013d4bc506e
(much faster now)
changed: asm code in: UInt::Rcl2, UInt::Rcr2 (32 and 64bit)
previous versions of Rcl2 and Rcr2 had O(n2) complexity,
now they have O(n) and are much faster
changed: now we do not use LAHF and SAHF instructions (both in 32 and 64 bit code)
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@83 e52654a7-88a9-db11-a3e9-0013d4bc506e
32 and 64 bits, much faster now
added: tests for UInt::AddInt
fixed: tests: test_lahf() returned incorrect value for 32bit platform
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@82 e52654a7-88a9-db11-a3e9-0013d4bc506e
there were used 'lahf' and 'sahf' instructions
probably they are supported in a new version of GCC
with '-msahf' option but GCC 4.2.1 has no such option
at the moment I'm using opcodes:
lahf -> 9f
sahf -> 9e
Warning! these instructions are not on all 64bit processors
from: http://gcc.gnu.org/onlinedocs/gcc/i386-and-x86_002d64-Options.html
"Early Intel CPUs with Intel 64 lacked LAHF and SAHF instructions supported
by AMD64 until introduction of Pentium 4 G1 step in December 2005."
changed: UInt::Add on 64bit platform
changed: UInt::Add on 32bit platform (a little)
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@80 e52654a7-88a9-db11-a3e9-0013d4bc506e
which is pointing at the end of the parsing string
added: initial import of tests: testing UInt::Add method
(not finished)
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@79 e52654a7-88a9-db11-a3e9-0013d4bc506e
the changes which were made in revision 60 and 63
(I forgot to change the info there)
git-svn-id: svn://ttmath.org/publicrep/ttmath/trunk@67 e52654a7-88a9-db11-a3e9-0013d4bc506e
Tomasz Sowa