You can subscribe to this list here.
2000 
_{Jan}

_{Feb}

_{Mar}

_{Apr}

_{May}

_{Jun}

_{Jul}

_{Aug}

_{Sep}
(371) 
_{Oct}
(167) 
_{Nov}
(412) 
_{Dec}
(208) 

2001 
_{Jan}
(378) 
_{Feb}
(302) 
_{Mar}
(269) 
_{Apr}
(296) 
_{May}
(306) 
_{Jun}
(381) 
_{Jul}
(346) 
_{Aug}
(315) 
_{Sep}
(195) 
_{Oct}
(216) 
_{Nov}
(280) 
_{Dec}
(227) 
2002 
_{Jan}
(309) 
_{Feb}
(333) 
_{Mar}
(328) 
_{Apr}
(407) 
_{May}
(517) 
_{Jun}
(519) 
_{Jul}
(400) 
_{Aug}
(580) 
_{Sep}
(1273) 
_{Oct}
(984) 
_{Nov}
(683) 
_{Dec}
(538) 
2003 
_{Jan}
(578) 
_{Feb}
(454) 
_{Mar}
(312) 
_{Apr}
(366) 
_{May}
(505) 
_{Jun}
(431) 
_{Jul}
(415) 
_{Aug}
(374) 
_{Sep}
(470) 
_{Oct}
(578) 
_{Nov}
(372) 
_{Dec}
(309) 
2004 
_{Jan}
(308) 
_{Feb}
(247) 
_{Mar}
(372) 
_{Apr}
(413) 
_{May}
(333) 
_{Jun}
(323) 
_{Jul}
(269) 
_{Aug}
(239) 
_{Sep}
(469) 
_{Oct}
(383) 
_{Nov}
(400) 
_{Dec}
(332) 
2005 
_{Jan}
(411) 
_{Feb}
(363) 
_{Mar}
(346) 
_{Apr}
(316) 
_{May}
(275) 
_{Jun}
(248) 
_{Jul}
(396) 
_{Aug}
(396) 
_{Sep}
(279) 
_{Oct}
(340) 
_{Nov}
(319) 
_{Dec}
(218) 
2006 
_{Jan}
(317) 
_{Feb}
(263) 
_{Mar}
(304) 
_{Apr}
(296) 
_{May}
(209) 
_{Jun}
(349) 
_{Jul}
(246) 
_{Aug}
(198) 
_{Sep}
(174) 
_{Oct}
(138) 
_{Nov}
(201) 
_{Dec}
(270) 
2007 
_{Jan}
(223) 
_{Feb}
(182) 
_{Mar}
(350) 
_{Apr}
(350) 
_{May}
(259) 
_{Jun}
(221) 
_{Jul}
(299) 
_{Aug}
(465) 
_{Sep}
(356) 
_{Oct}
(265) 
_{Nov}
(417) 
_{Dec}
(225) 
2008 
_{Jan}
(421) 
_{Feb}
(327) 
_{Mar}
(219) 
_{Apr}
(389) 
_{May}
(375) 
_{Jun}
(262) 
_{Jul}
(215) 
_{Aug}
(289) 
_{Sep}
(257) 
_{Oct}
(383) 
_{Nov}
(237) 
_{Dec}
(209) 
2009 
_{Jan}
(232) 
_{Feb}
(327) 
_{Mar}
(306) 
_{Apr}
(251) 
_{May}
(146) 
_{Jun}
(247) 
_{Jul}
(302) 
_{Aug}
(252) 
_{Sep}
(263) 
_{Oct}
(376) 
_{Nov}
(270) 
_{Dec}
(244) 
2010 
_{Jan}
(225) 
_{Feb}
(184) 
_{Mar}
(300) 
_{Apr}
(290) 
_{May}
(275) 
_{Jun}
(535) 
_{Jul}
(192) 
_{Aug}
(237) 
_{Sep}
(304) 
_{Oct}
(142) 
_{Nov}
(384) 
_{Dec}
(186) 
2011 
_{Jan}
(305) 
_{Feb}
(337) 
_{Mar}
(331) 
_{Apr}
(318) 
_{May}
(306) 
_{Jun}
(299) 
_{Jul}
(205) 
_{Aug}
(271) 
_{Sep}
(232) 
_{Oct}
(179) 
_{Nov}
(252) 
_{Dec}
(216) 
2012 
_{Jan}
(195) 
_{Feb}
(268) 
_{Mar}
(142) 
_{Apr}
(226) 
_{May}
(203) 
_{Jun}
(132) 
_{Jul}
(211) 
_{Aug}
(429) 
_{Sep}
(289) 
_{Oct}
(291) 
_{Nov}
(182) 
_{Dec}
(188) 
2013 
_{Jan}
(205) 
_{Feb}
(259) 
_{Mar}
(224) 
_{Apr}
(125) 
_{May}
(295) 
_{Jun}
(181) 
_{Jul}
(209) 
_{Aug}
(167) 
_{Sep}
(330) 
_{Oct}
(212) 
_{Nov}
(95) 
_{Dec}
(114) 
2014 
_{Jan}
(40) 
_{Feb}
(63) 
_{Mar}
(62) 
_{Apr}
(65) 
_{May}
(82) 
_{Jun}
(105) 
_{Jul}
(56) 
_{Aug}
(175) 
_{Sep}
(79) 
_{Oct}
(49) 
_{Nov}
(51) 
_{Dec}
(47) 
2015 
_{Jan}
(26) 
_{Feb}
(69) 
_{Mar}
(82) 
_{Apr}
(55) 
_{May}
(35) 
_{Jun}
(57) 
_{Jul}
(54) 
_{Aug}
(56) 
_{Sep}
(25) 
_{Oct}
(21) 
_{Nov}
(8) 
_{Dec}
(27) 
2016 
_{Jan}
(49) 
_{Feb}
(44) 
_{Mar}
(132) 
_{Apr}
(39) 
_{May}
(39) 
_{Jun}
(49) 
_{Jul}
(70) 
_{Aug}
(43) 
_{Sep}
(69) 
_{Oct}
(79) 
_{Nov}
(65) 
_{Dec}
(32) 
2017 
_{Jan}
(99) 
_{Feb}
(88) 
_{Mar}
(42) 
_{Apr}
(47) 
_{May}
(56) 
_{Jun}

_{Jul}
(79) 
_{Aug}
(9) 
_{Sep}
(29) 
_{Oct}
(4) 
_{Nov}

_{Dec}

S  M  T  W  T  F  S 






1
(9) 
2

3
(3) 
4
(4) 
5
(19) 
6
(16) 
7
(5) 
8
(12) 
9
(7) 
10
(4) 
11

12
(10) 
13
(4) 
14
(12) 
15
(5) 
16
(8) 
17
(1) 
18
(7) 
19
(8) 
20
(2) 
21
(17) 
22
(1) 
23
(18) 
24
(4) 
25
(16) 
26
(11) 
27
(62) 
28
(20) 
29
(13) 
30
(20) 
From: K. Frank <kfrank29.c@gm...>  20110430 22:26:41

Hi TSalm! On Sat, Apr 30, 2011 at 5:55 PM, TSalm <TSalm@...> wrote: > Hi, > > Is there away to catch an exception and display his stacktrace ? I believe that the answer is no, if I understand your question. Ordinary, builtin c++ exceptions do not take a snapshot of the stack or build a stacktrace as they unwind the stack. This is by design, as c++ doesn't want to impose on you the cost of building and maintaining this stacktrace. The stack gets unwound between where you throw the exception and where you catch it. This can be inconvenient when debugging, because if your code throws an exception deep down in some nested function calls, seeing the stacktrace (at the time the exception was thrown) can help make the bug apparent. (I think java exceptions do give you the stacktrace, and this can be very helpful.) When I'm doing development and debugging, I generally put in assets. These cause the program to core dump (or send control back to the debugger) right at the point where the error is detected (i.e., the assert fails), and then you can look at the state of the stack when the error occurred (unless your bug has corrupted the stack...). > TIA, > TSalm Good luck. K. Frank 
From: Kai Tietz <ktietz70@go...>  20110430 22:14:14

2011/4/30 Kai Tietz <ktietz70@...>: > 2011/4/30 Bob Delaney <delaneyrm@...>: >> >> On Apr 30, 2011, at 10:15 AM, K. Frank wrote: >> >>> I get: >>> >>> C:\>gcc o pow_test pow_test.c >>> >>> C:\>pow_test >>> 2.20000000000000017764^3.10000000000000008882 = 11.52153412678571875460 >> >> >> Victor Shoup's NTL and my own fp, to 22 digit accuracy, give: >> >> 2.20000000000000017764^3.10000000000000008882 = 11.52153412678571783832 >> >> so your result is accurate to only 16 digits. Your >> >> 2.20000000000000000004^3.09999999999999999991 = 11.52153412678571414718 >> >> should be: >> >> 2.20000000000000000004^3.09999999999999999991 = 11.52153412678571414732 >> >> so your result is accurate to about 20 digits. >> >> Bob > > Well, what is to be expected as we used here in test application an > %.20F. By specifying here a %.40F its calculated result is > 11.5215341267857141471786519559827866032720 and it is has binary > identity for what gmp calculates for 80bit IEEE 754 floating point. > Which means for our purposes in gcc, it has right accuracy. > > Kai > To be more precise here IEEE 754 floats for Intel/Amd architecture x86/x86_64 has following definitions: float (32bit): digits for mantissa:24 digit:6, min exp:125 (37 10th) max exp: 128 (38 10th) Epsilon:1.19209e007 double (64bit): digits for mantissa:53 digit:15, min exp:1021 (307 10th) max exp: 1024 (308 10th) Epsilon:2.22045e016 long double (80bit): digits for mantissa:64 digit:18, min exp:16381 (4931 10th) max exp: 16384 (4932 10th) Epsilon:1.0842e019 Which means that indeed the accuracy of 20 digits after decimal point are pretty precise here. Regards, Kai 
From: K. Frank <kfrank29.c@gm...>  20110430 22:13:15

Hello Bob and Kai! On Sat, Apr 30, 2011 at 5:41 PM, Kai Tietz <ktietz70@...> wrote: > 2011/4/30 Bob Delaney <delaneyrm@...>: >> >> On Apr 30, 2011, at 10:15 AM, K. Frank wrote: >> >>> I get: >>> >>> C:\>gcc o pow_test pow_test.c >>> >>> C:\>pow_test >>> 2.20000000000000017764^3.10000000000000008882 = 11.52153412678571875460 >> >> >> Victor Shoup's NTL and my own fp, to 22 digit accuracy, give: >> >> 2.20000000000000017764^3.10000000000000008882 = 11.52153412678571783832 >> >> so your result is accurate to only 16 digits. Your This is exactly what I would expect (short of looking at the actual binary representation). An IEEE double is 64 bits, with a 52bit mantissa, which, with the "implicit" bit, gives 53 bits of precision. This gives just under 16 decimal digits of precision, which is what you see. >> >> 2.20000000000000000004^3.09999999999999999991 = 11.52153412678571414718 >> >> should be: >> >> 2.20000000000000000004^3.09999999999999999991 = 11.52153412678571414732 >> >> so your result is accurate to about 20 digits. As I understand it, gcc defines long double as the x87 80bit floating point number. This has a 64bit mantissa, and I believe that it does not use an implicit bit, so this gives slightly more than 19 decimal digits of precision, again consistent with your result. >> >> Bob > > Well, what is to be expected as we used here in test application an > %.20F. By specifying here a %.40F its calculated result is > 11.5215341267857141471786519559827866032720 and it is has binary > identity for what gmp calculates for 80bit IEEE 754 floating point. > Which means for our purposes in gcc, it has right accuracy. (Kai's comment seems to confirm that gcc's long double is, indeed, the 80bit floatingpoint number.) > Kai Thanks to all for the insight. K. Frank 
From: TSalm <TSalm@fr...>  20110430 21:55:08

Hi, Is there away to catch an exception and display his stacktrace ? TIA, TSalm ____________ Analyse antivirus effectuée par G Data AntiVirus 
From: Kai Tietz <ktietz70@go...>  20110430 21:42:03

2011/4/30 Bob Delaney <delaneyrm@...>: > > On Apr 30, 2011, at 10:15 AM, K. Frank wrote: > >> I get: >> >> C:\>gcc o pow_test pow_test.c >> >> C:\>pow_test >> 2.20000000000000017764^3.10000000000000008882 = 11.52153412678571875460 > > > Victor Shoup's NTL and my own fp, to 22 digit accuracy, give: > > 2.20000000000000017764^3.10000000000000008882 = 11.52153412678571783832 > > so your result is accurate to only 16 digits. Your > > 2.20000000000000000004^3.09999999999999999991 = 11.52153412678571414718 > > should be: > > 2.20000000000000000004^3.09999999999999999991 = 11.52153412678571414732 > > so your result is accurate to about 20 digits. > > Bob Well, what is to be expected as we used here in test application an %.20F. By specifying here a %.40F its calculated result is 11.5215341267857141471786519559827866032720 and it is has binary identity for what gmp calculates for 80bit IEEE 754 floating point. Which means for our purposes in gcc, it has right accuracy. Kai 
From: Bob Delaney <delaneyrm@ea...>  20110430 21:21:26

On Apr 30, 2011, at 10:15 AM, K. Frank wrote: > I get: > > C:\>gcc o pow_test pow_test.c > > C:\>pow_test > 2.20000000000000017764^3.10000000000000008882 = 11.52153412678571875460 Victor Shoup's NTL and my own fp, to 22 digit accuracy, give: 2.20000000000000017764^3.10000000000000008882 = 11.52153412678571783832 so your result is accurate to only 16 digits. Your 2.20000000000000000004^3.09999999999999999991 = 11.52153412678571414718 should be: 2.20000000000000000004^3.09999999999999999991 = 11.52153412678571414732 so your result is accurate to about 20 digits. Bob 
From: Lars Johannesen <lj@ha...>  20110430 16:57:44

Try look at boost /lars On Sat, Apr 30, 2011 at 12:26 PM, Ocean <Ocean@...> wrote: > >> Original Message >> From: Greg Chicares [mailto:gchicares@...] >> Sent: Saturday, April 30, 2011 1:09 AM >> To: mingwusers@... >> Subject: Re: [Mingwusers] A beginner's questions about RegEx support >> >> On 20110430 02:04Z, Ocean wrote: >> [...using C++...] >> > 1. Does MinGW support RegEx "out of the box", or do I need to install a >> > RegEx library seperately? >> >> Yes: a regex facility was added to the C++ standard library in TR1. >> To use it, #include <regex>. >> > > > Thanks! From what I can find out, the standard library in TR1 for > RegEx is somewhat limited. How can I use a different RegEx libary (such as > PCRE)? > > > >  > WhatsUp Gold  Download Free Network Management Software > The most intuitive, comprehensive, and costeffective network > management toolset available today. Delivers lowest initial > acquisition cost and overall TCO of any competing solution. > http://p.sf.net/sfu/whatsupgoldsd > _______________________________________________ > MinGWusers mailing list > MinGWusers@... > > This list observes the Etiquette found at > http://www.mingw.org/Mailing_Lists. > We ask that you be polite and do the same. Disregard for the list etiquette may cause your account to be moderated. > > _______________________________________________ > You may change your MinGW Account Options or unsubscribe at: > https://lists.sourceforge.net/lists/listinfo/mingwusers > Also: mailto:mingwusersrequest@...?subject=unsubscribe > 
From: Ocean <Ocean@co...>  20110430 16:26:29

> Original Message > From: Greg Chicares [mailto:gchicares@...] > Sent: Saturday, April 30, 2011 1:09 AM > To: mingwusers@... > Subject: Re: [Mingwusers] A beginner's questions about RegEx support > > On 20110430 02:04Z, Ocean wrote: > [...using C++...] > > 1. Does MinGW support RegEx "out of the box", or do I need to install a > > RegEx library seperately? > > Yes: a regex facility was added to the C++ standard library in TR1. > To use it, #include <regex>. > Thanks! From what I can find out, the standard library in TR1 for RegEx is somewhat limited. How can I use a different RegEx libary (such as PCRE)? 
From: Kai Tietz <ktietz70@go...>  20110430 15:48:06

2011/4/30 Kai Tietz <ktietz70@...>: > 2011/4/30 K. Frank <kfrank29.c@...>: >> Hello Kai! >> >> On Sat, Apr 30, 2011 at 5:51 AM, Kai Tietz <ktietz70@...> wrote: >>> 2011/4/30 Kai Tietz <ktietz70@...>: >>>> 2011/4/30 Greg Chicares <gchicares@...>: >>>>> On 20110429 19:55Z, K. Frank wrote: >>>>>> >>>>>> By the way, could someone confirm that mingw does use msvcrt for >>>>>> sqrt and pow? >>>>> >>>>> http://cygwin.com/cgibin/cvsweb.cgi/src/winsup/mingw/mingwex/math/?cvsroot=src >>>>> >>>>> Looks to me like sqrt(double) calls into msvcrt. But pow() is a >>>>> different story. Danny implemented it in libmingwex because of >>>>> some qualityofimplementation problem with msvcrt; then people >>>>> complained that the libmingwex version was slower, and I can't >>>>> remember where it wound up, but it's all in the archives. >>>>> >>>>>> sqrt (x) and pow (x, 0.5) ought to give the same result (even if not required >>>>>> to by IEEE754). >>>>> ... >>>>> Not so many years ago (perhaps when msvcrt was written), it was >>>>> thought that correctlyrounded transcendental functions weren't >>>>> feasible in practice, so library authors had a looser attitude >>>>> toward exactness. If you have an ideal implementation of sqrt() >>>>> but a pow() that's only approximate, should you specialcase >>>>> the power of exactly 0.5? If you do, then pow() and sqrt() will >>>>> be consistent, but then the powers >>>>> x^(0.5*(1epsilon)) >>>>> x^(0.5 ) >>>>> x^(0.5*(1+epsilon)) >>>>> might not be monotone. Some would call that a bad tradeoff. >>>> >>>> Well, I spent some efforts into log, exp, and pow implementation. Also >>>> some other basemath functions I improved in mingww64's math library >>>> in two terms. a) Make those functions behave as ISOC99 specification >>>> tells in cornercases. and b) Improve accuracy in terms of gcc's >>>> internal used gmp. As we noticed here result differences. >>>> >>>> The following example illustrates the issues pretty well: >>>> >>>> #include <stdio.h> >>>> #include <math.h> >>>> >>>> double abc[3][2] = { { 2.2, 3.1 }, { 2.0, 3.0 }, { 4.0, 0.5 } }; >>>> long double ab[3][2] = { { 2.2L, 3.1L }, { 2.0L, 3.0L }, { 4.0L, 0.5L } }; >>>> int main() >>>> { >>>> double r[3]; >>>> long double rl[3]; >>>> int i; >>>> for (i = 0; i < 3; i++) >>>> __mingw_printf ("%.20F^%.20F = %.20F\n", abc[i][0], abc[i][1], >>>> (r[i] = pow (abc[i][0], abc[i][1]))); >>>> __mingw_printf ("%.20F %.20F %.20F\n", pow (2.2, 3.1), pow (2.0, >>>> 3.0), pow (4.0, 0.5)); >>>> r[0] = pow (2.2, 3.1); >>>> r[1] = pow (2.0, 3.0); >>>> r[2] = pow (4.0, 0.5); >>>> __mingw_printf ("%.20F %.20F %.20F\n", r[0], r[1], r[2]); >>>> >>>> for (i = 0; i < 3; i++) >>>> __mingw_printf ("%.20LF^%.20LF = %.20LF\n", ab[i][0], ab[i][1], >>>> (rl[i] = powl (ab[i][0], ab[i][1]))); >>>> __mingw_printf ("%.20LF %.20LF %.20LF\n", >>>> powl (2.2L, 3.1L), powl (2.0L, 3.0L) , powl >>>> (4.0L, 0.5L)); >>>> rl[0] = powl (2.2L, 3.1L); >>>> rl[1] = powl (2.0L, 3.0L); >>>> rl[2] = powl (4.0L, 0.5L); >>>> __mingw_printf ("%.20LF %.20LF %.20LF\n", rl[0], rl[1], rl[2]); >>>> return 1; >>>> } >> >> Do I understand your point correctly that you are comparing gcc's >> compiletime and runtime values for pow? >> >> For the record, when I run this on a mingw64 build: >> >> gcc (GCC) 4.5.2 20101002 (prerelease) [svn/rev.164902  mingww64/oz] >> >> I get: >> >> C:\>gcc o pow_test pow_test.c >> >> C:\>pow_test >> 2.20000000000000017764^3.10000000000000008882 = 11.52153412678571875460 >> 2.00000000000000000000^3.00000000000000000000 = 8.00000000000000000000 >> 4.00000000000000000000^0.50000000000000000000 = 2.00000000000000000000 >> 11.52153412678571697825 8.00000000000000000000 2.00000000000000000000 >> 0.00000000000000177636 0.00000000000000000000 0.00000000000000000000 >> 2.20000000000000000004^3.09999999999999999991 = 11.52153412678571414718 >> 2.00000000000000000000^3.00000000000000000000 = 8.00000000000000000000 >> 4.00000000000000000000^0.50000000000000000000 = 2.00000000000000000000 >> 11.52153412678571414718 8.00000000000000000000 2.00000000000000000000 >> 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 >> >> As I understand it, this means I am seeing a small difference between >> compile and runtime evaluation only for powl (2.2, 3.1), i.e., the >> longdouble case. >> >>>> The interesting issue is here that gcc uses for constant >>>> mathcalculations gmplibrary to get result. For more complex >>>> variants, it uses crt's math. By this it is important for a gcc based >>>> runtime to be in the IEEE 754 floating point compatible to gmp's >>>> results. >> >> Yes, that's a good point. I hadn't considered the issue of compiletime >> evaluation, and, ideally, it should agree with runtime evaluation, adding >> yet another constraint to our wish list. >> >> Furthermore, if we take the idea of a crosscompiler seriously, we >> would also like the compiletime evaluations of a crosscompiler and >> native compiler to agree. >> >> Hmm... >> >> So, we would like sqrt (x) and pow (x, 0.5) to agree. We would like >> compiletime and runtime evaluations to agree. We would like >> crosscompilers and native compilers to agree. >> >> Taken at face value, this would seem to preclude using the platform >> library and platform hardware to compute things like sqrt and pow >> (unless we know they comply with some agreedupon standard, >> such as IEEE754). >> >> Just to make the reasoning explicit, if I perform my runtime evaluation >> on platform A, e.g., using platform A's FPU, then a B>A crosscompiler >> running on platform B won't have platform A's FPU available to perform >> consistent compiletime evaluations. >> >> So you either give up some of the abovewishedfor consistency, you >> forgo the potential performance benefits of using the platform's library >> and / or hardware, or you get everybody to use a common standard. >> >>>> ... >>>> I admit that this might be not that obvious to users, why results here >>>> are different, but for a gcc based toolchain we need to play nice with >>>> gcc's internal assumptions. >>>> >>>> Regards, >>>> Kai >>> >>> On closer look for long double, we can have rouding issues for long >>> double precission. I see that powl (0.01L, 0.5L) we have a rounding >>> difference to sqrtl (0.01L). Interestingly not for float or double >>> types. So we we might need to specialcase in pow the case for y == >>> 0.5 to solve this. >> >> Be, however, cognizant of Greg's concern. Forcing pow to agree with >> sqrt just for y = 0.5 has the potential to make pow, in a sense, internally >> inconsistent, It's a tricky tradeoff. (Ideally, both pow and sqrt would >> be "exact" and therefore agree with one another automatically.) >> >> By the way, I also see a problem with runtime powl for powl (2.2L, 0.5L). >> It disagrees slightly from sqrtl (2.2L) and from the compiletime value >> for powl (2.2L, 0.5L). >> >> I modified your test program to compare sqrt with pow (x, 0.5) (and sqrtl >> with powl). For the record: >> >> C:\>gcc version >> gcc (GCC) 4.5.2 20101002 (prerelease) [svn/rev.164902  mingww64/oz] >> >> C:\>gcc o sqrt_pow sqrt_pow.c >> >> C:\>sqrt_pow >> 1.04880884817015163080, 1.04880884817015163080, 0.00000000000000000000 >> 1.48323969741913264109, 1.48323969741913264109, 0.00000000000000000000 >> 1.81659021245849494619, 1.81659021245849494619, 0.00000000000000000000 >> 1.04880884817015163080 1.04880884817015163080 >> 1.48323969741913264109 1.48323969741913264109 >> 1.81659021245849494619 1.81659021245849494619 >> 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 >> 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 >> 1.04880884817015154699, 1.04880884817015154699, 0.00000000000000000000 >> 1.48323969741913258980, 1.48323969741913258970, 0.00000000000000000011 >> 1.81659021245849499921, 1.81659021245849499921, 0.00000000000000000000 >> 1.04880884817015154699 1.04880884817015154699 >> 1.48323969741913258980 1.48323969741913258980 >> 1.81659021245849499921 1.81659021245849499921 >> 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 >> 0.00000000000000000000 0.00000000000000000011 0.00000000000000000000 >> >> where sqrt_pow.c is: >> >> #include <stdio.h> >> #include <math.h> >> >> double abc[3] = { 1.1, 2.2, 3.3 }; >> long double abcl[3] = { 1.1L, 2.2L, 3.3L }; >> int main() >> { >> double r1[3], r2[3], r3[3]; >> long double rl1[3], rl2[3], rl3[3]; >> int i; >> for (i = 0; i < 3; i++) >> __mingw_printf ("%.20F, %.20F, %.20F\n", >> (r1[i] = sqrt (abc[i])), >> (r2[i] = pow (abc[i], 0.5)), >> (r3[i] = pow (abc[i], 0.5)  sqrt (abc[i]))); >> __mingw_printf ("%.20F %.20F\n", sqrt (1.1), pow (1.1, 0.5)); >> __mingw_printf ("%.20F %.20F\n", sqrt (2.2), pow (2.2, 0.5)); >> __mingw_printf ("%.20F %.20F\n", sqrt (3.3), pow (3.3, 0.5)); >> r1[0] = sqrt (1.1); >> r1[1] = sqrt (2.2); >> r1[2] = sqrt (3.3); >> __mingw_printf ("%.20F %.20F %.20F\n", r1[0], r1[1], r1[2]); >> r2[0] = pow (1.1, 0.5); >> r2[1] = pow (2.2, 0.5); >> r2[2] = pow (3.3, 0.5); >> __mingw_printf ("%.20F %.20F %.20F\n", r2[0], r2[1], r2[2]); >> >> for (i = 0; i < 3; i++) >> __mingw_printf ("%.20LF, %.20LF, %.20LF\n", >> (rl1[i] = sqrtl (abcl[i])), >> (rl2[i] = powl (abcl[i], 0.5L)), >> (rl3[i] = powl (abcl[i], 0.5L)  sqrtl (abcl[i]))); >> __mingw_printf ("%.20LF %.20LF\n", sqrtl (1.1L), powl (1.1L, 0.5L)); >> __mingw_printf ("%.20LF %.20LF\n", sqrtl (2.2L), powl (2.2L, 0.5L)); >> __mingw_printf ("%.20LF %.20LF\n", sqrtl (3.3L), powl (3.3L, 0.5L)); >> rl1[0] = sqrtl (1.1L); >> rl1[1] = sqrtl (2.2L); >> rl1[2] = sqrtl (3.3L); >> __mingw_printf ("%.20LF %.20LF %.20LF\n", rl1[0], rl1[1], rl1[2]); >> rl2[0] = powl (1.1L, 0.5L); >> rl2[1] = powl (2.2L, 0.5L); >> rl2[2] = powl (3.3L, 0.5L); >> __mingw_printf ("%.20LF %.20LF %.20LF\n", rl2[0], rl2[1], rl2[2]); >> >> return 1; >> } >> >>> >>> Kai >> >> Thanks, Kai, for your insight and explanations. >> >> >> K. Frank > > I add a special case to mingww64's pow(fl) function about y=0.5 and > get now for your adjusted testcase the following results, which are > looking much better IMHO. > > $ ./tst.exe > 1.04880884817015163080, 1.04880884817015163080, 0.00000000000000004142 > 1.48323969741913264109, 1.48323969741913264109, 0.00000000000000000857 > 1.81659021245849494619, 1.81659021245849494619, 0.00000000000000000412 > 1.04880884817015163080 1.04880884817015163080 > 1.48323969741913264109 1.48323969741913264109 > 1.81659021245849494619 1.81659021245849494619 > 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 > 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 > 1.04880884817015154699, 1.04880884817015154699, 0.00000000000000000000 > 1.48323969741913258980, 1.48323969741913258980, 0.00000000000000000000 > 1.81659021245849499921, 1.81659021245849499921, 0.00000000000000000000 > 1.04880884817015154699 1.04880884817015154699 > 1.48323969741913258980 1.48323969741913258980 > 1.81659021245849499921 1.81659021245849499921 > 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 > 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 > > I used for this a 32bit variant and 64bit variant, which made no > differences in output. > The issue here is that 1.0 branch still uses the inaccurate variant > from Danny, but nowaday's trunk uses for pow, sqrt, log, and exp > functions new implementations, which are treating rounding and > shortpath calculation in an ISOC variant (as gcc expects internally > such a math). Also complexmath part is completely rewritten on > mingww64's trunk version. > > IMHO it might be a good idea to replace step by step the classical > runtime functions, so that they are not using here msvcrt in first > place. Of course the importsymbols of msvcrt (for compatibility with > VC generated objects need to remain in .def exports, but the actual > functions should come directly from mingwex. > PS: The differences shown in results for 32bit aren't caused by inaccuarcy of used math directly. Instead it is caused by the fact that 32bit use FPU registers (which are 80bit always). By this the inline difference without using store, have for double small differences. By first storing values and then printing those stored values you see also for 32bit a delta of zero. Interesting this behavior is on 64bit x86_64 different, as it uses always precise accuracy to store results in registers. By this you have really just a 64bit result for double type, and not indirecty 80bit (as shown in first case). I altered your testcase a bit in following lines ... for (i = 0; i < 3; i++) { r1[i] = sqrt (abc[i]); r2[i] = pow (abc[i], 0.5); r3[i] = pow (abc[i], 0.5)  sqrt (abc[i]); __mingw_printf ("%.20F, %.20F, %.20F\n", r1[i], r2[i], r3[i]); } ... and get now $ ./tst.exe 1.04880884817015163080, 1.04880884817015163080, 0.00000000000000000000 1.48323969741913264109, 1.48323969741913264109, 0.00000000000000000000 1.81659021245849494619, 1.81659021245849494619, 0.00000000000000000000 1.04880884817015163080 1.04880884817015163080 1.48323969741913264109 1.48323969741913264109 1.81659021245849494619 1.81659021245849494619 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 1.04880884817015154699, 1.04880884817015154699, 0.00000000000000000000 1.48323969741913258980, 1.48323969741913258980, 0.00000000000000000000 1.81659021245849499921, 1.81659021245849499921, 0.00000000000000000000 1.04880884817015154699 1.04880884817015154699 1.48323969741913258980 1.48323969741913258980 1.81659021245849499921 1.81659021245849499921 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 as result. it is an interesting fact that this happens. Regards, Kai 
From: Kai Tietz <ktietz70@go...>  20110430 15:34:00

2011/4/30 K. Frank <kfrank29.c@...>: > Hello Kai! > > On Sat, Apr 30, 2011 at 5:51 AM, Kai Tietz <ktietz70@...> wrote: >> 2011/4/30 Kai Tietz <ktietz70@...>: >>> 2011/4/30 Greg Chicares <gchicares@...>: >>>> On 20110429 19:55Z, K. Frank wrote: >>>>> >>>>> By the way, could someone confirm that mingw does use msvcrt for >>>>> sqrt and pow? >>>> >>>> http://cygwin.com/cgibin/cvsweb.cgi/src/winsup/mingw/mingwex/math/?cvsroot=src >>>> >>>> Looks to me like sqrt(double) calls into msvcrt. But pow() is a >>>> different story. Danny implemented it in libmingwex because of >>>> some qualityofimplementation problem with msvcrt; then people >>>> complained that the libmingwex version was slower, and I can't >>>> remember where it wound up, but it's all in the archives. >>>> >>>>> sqrt (x) and pow (x, 0.5) ought to give the same result (even if not required >>>>> to by IEEE754). >>>> ... >>>> Not so many years ago (perhaps when msvcrt was written), it was >>>> thought that correctlyrounded transcendental functions weren't >>>> feasible in practice, so library authors had a looser attitude >>>> toward exactness. If you have an ideal implementation of sqrt() >>>> but a pow() that's only approximate, should you specialcase >>>> the power of exactly 0.5? If you do, then pow() and sqrt() will >>>> be consistent, but then the powers >>>> x^(0.5*(1epsilon)) >>>> x^(0.5 ) >>>> x^(0.5*(1+epsilon)) >>>> might not be monotone. Some would call that a bad tradeoff. >>> >>> Well, I spent some efforts into log, exp, and pow implementation. Also >>> some other basemath functions I improved in mingww64's math library >>> in two terms. a) Make those functions behave as ISOC99 specification >>> tells in cornercases. and b) Improve accuracy in terms of gcc's >>> internal used gmp. As we noticed here result differences. >>> >>> The following example illustrates the issues pretty well: >>> >>> #include <stdio.h> >>> #include <math.h> >>> >>> double abc[3][2] = { { 2.2, 3.1 }, { 2.0, 3.0 }, { 4.0, 0.5 } }; >>> long double ab[3][2] = { { 2.2L, 3.1L }, { 2.0L, 3.0L }, { 4.0L, 0.5L } }; >>> int main() >>> { >>> double r[3]; >>> long double rl[3]; >>> int i; >>> for (i = 0; i < 3; i++) >>> __mingw_printf ("%.20F^%.20F = %.20F\n", abc[i][0], abc[i][1], >>> (r[i] = pow (abc[i][0], abc[i][1]))); >>> __mingw_printf ("%.20F %.20F %.20F\n", pow (2.2, 3.1), pow (2.0, >>> 3.0), pow (4.0, 0.5)); >>> r[0] = pow (2.2, 3.1); >>> r[1] = pow (2.0, 3.0); >>> r[2] = pow (4.0, 0.5); >>> __mingw_printf ("%.20F %.20F %.20F\n", r[0], r[1], r[2]); >>> >>> for (i = 0; i < 3; i++) >>> __mingw_printf ("%.20LF^%.20LF = %.20LF\n", ab[i][0], ab[i][1], >>> (rl[i] = powl (ab[i][0], ab[i][1]))); >>> __mingw_printf ("%.20LF %.20LF %.20LF\n", >>> powl (2.2L, 3.1L), powl (2.0L, 3.0L) , powl >>> (4.0L, 0.5L)); >>> rl[0] = powl (2.2L, 3.1L); >>> rl[1] = powl (2.0L, 3.0L); >>> rl[2] = powl (4.0L, 0.5L); >>> __mingw_printf ("%.20LF %.20LF %.20LF\n", rl[0], rl[1], rl[2]); >>> return 1; >>> } > > Do I understand your point correctly that you are comparing gcc's > compiletime and runtime values for pow? > > For the record, when I run this on a mingw64 build: > > gcc (GCC) 4.5.2 20101002 (prerelease) [svn/rev.164902  mingww64/oz] > > I get: > > C:\>gcc o pow_test pow_test.c > > C:\>pow_test > 2.20000000000000017764^3.10000000000000008882 = 11.52153412678571875460 > 2.00000000000000000000^3.00000000000000000000 = 8.00000000000000000000 > 4.00000000000000000000^0.50000000000000000000 = 2.00000000000000000000 > 11.52153412678571697825 8.00000000000000000000 2.00000000000000000000 > 0.00000000000000177636 0.00000000000000000000 0.00000000000000000000 > 2.20000000000000000004^3.09999999999999999991 = 11.52153412678571414718 > 2.00000000000000000000^3.00000000000000000000 = 8.00000000000000000000 > 4.00000000000000000000^0.50000000000000000000 = 2.00000000000000000000 > 11.52153412678571414718 8.00000000000000000000 2.00000000000000000000 > 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 > > As I understand it, this means I am seeing a small difference between > compile and runtime evaluation only for powl (2.2, 3.1), i.e., the > longdouble case. > >>> The interesting issue is here that gcc uses for constant >>> mathcalculations gmplibrary to get result. For more complex >>> variants, it uses crt's math. By this it is important for a gcc based >>> runtime to be in the IEEE 754 floating point compatible to gmp's >>> results. > > Yes, that's a good point. I hadn't considered the issue of compiletime > evaluation, and, ideally, it should agree with runtime evaluation, adding > yet another constraint to our wish list. > > Furthermore, if we take the idea of a crosscompiler seriously, we > would also like the compiletime evaluations of a crosscompiler and > native compiler to agree. > > Hmm... > > So, we would like sqrt (x) and pow (x, 0.5) to agree. We would like > compiletime and runtime evaluations to agree. We would like > crosscompilers and native compilers to agree. > > Taken at face value, this would seem to preclude using the platform > library and platform hardware to compute things like sqrt and pow > (unless we know they comply with some agreedupon standard, > such as IEEE754). > > Just to make the reasoning explicit, if I perform my runtime evaluation > on platform A, e.g., using platform A's FPU, then a B>A crosscompiler > running on platform B won't have platform A's FPU available to perform > consistent compiletime evaluations. > > So you either give up some of the abovewishedfor consistency, you > forgo the potential performance benefits of using the platform's library > and / or hardware, or you get everybody to use a common standard. > >>> ... >>> I admit that this might be not that obvious to users, why results here >>> are different, but for a gcc based toolchain we need to play nice with >>> gcc's internal assumptions. >>> >>> Regards, >>> Kai >> >> On closer look for long double, we can have rouding issues for long >> double precission. I see that powl (0.01L, 0.5L) we have a rounding >> difference to sqrtl (0.01L). Interestingly not for float or double >> types. So we we might need to specialcase in pow the case for y == >> 0.5 to solve this. > > Be, however, cognizant of Greg's concern. Forcing pow to agree with > sqrt just for y = 0.5 has the potential to make pow, in a sense, internally > inconsistent, It's a tricky tradeoff. (Ideally, both pow and sqrt would > be "exact" and therefore agree with one another automatically.) > > By the way, I also see a problem with runtime powl for powl (2.2L, 0.5L). > It disagrees slightly from sqrtl (2.2L) and from the compiletime value > for powl (2.2L, 0.5L). > > I modified your test program to compare sqrt with pow (x, 0.5) (and sqrtl > with powl). For the record: > > C:\>gcc version > gcc (GCC) 4.5.2 20101002 (prerelease) [svn/rev.164902  mingww64/oz] > > C:\>gcc o sqrt_pow sqrt_pow.c > > C:\>sqrt_pow > 1.04880884817015163080, 1.04880884817015163080, 0.00000000000000000000 > 1.48323969741913264109, 1.48323969741913264109, 0.00000000000000000000 > 1.81659021245849494619, 1.81659021245849494619, 0.00000000000000000000 > 1.04880884817015163080 1.04880884817015163080 > 1.48323969741913264109 1.48323969741913264109 > 1.81659021245849494619 1.81659021245849494619 > 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 > 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 > 1.04880884817015154699, 1.04880884817015154699, 0.00000000000000000000 > 1.48323969741913258980, 1.48323969741913258970, 0.00000000000000000011 > 1.81659021245849499921, 1.81659021245849499921, 0.00000000000000000000 > 1.04880884817015154699 1.04880884817015154699 > 1.48323969741913258980 1.48323969741913258980 > 1.81659021245849499921 1.81659021245849499921 > 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 > 0.00000000000000000000 0.00000000000000000011 0.00000000000000000000 > > where sqrt_pow.c is: > > #include <stdio.h> > #include <math.h> > > double abc[3] = { 1.1, 2.2, 3.3 }; > long double abcl[3] = { 1.1L, 2.2L, 3.3L }; > int main() > { > double r1[3], r2[3], r3[3]; > long double rl1[3], rl2[3], rl3[3]; > int i; > for (i = 0; i < 3; i++) > __mingw_printf ("%.20F, %.20F, %.20F\n", > (r1[i] = sqrt (abc[i])), > (r2[i] = pow (abc[i], 0.5)), > (r3[i] = pow (abc[i], 0.5)  sqrt (abc[i]))); > __mingw_printf ("%.20F %.20F\n", sqrt (1.1), pow (1.1, 0.5)); > __mingw_printf ("%.20F %.20F\n", sqrt (2.2), pow (2.2, 0.5)); > __mingw_printf ("%.20F %.20F\n", sqrt (3.3), pow (3.3, 0.5)); > r1[0] = sqrt (1.1); > r1[1] = sqrt (2.2); > r1[2] = sqrt (3.3); > __mingw_printf ("%.20F %.20F %.20F\n", r1[0], r1[1], r1[2]); > r2[0] = pow (1.1, 0.5); > r2[1] = pow (2.2, 0.5); > r2[2] = pow (3.3, 0.5); > __mingw_printf ("%.20F %.20F %.20F\n", r2[0], r2[1], r2[2]); > > for (i = 0; i < 3; i++) > __mingw_printf ("%.20LF, %.20LF, %.20LF\n", > (rl1[i] = sqrtl (abcl[i])), > (rl2[i] = powl (abcl[i], 0.5L)), > (rl3[i] = powl (abcl[i], 0.5L)  sqrtl (abcl[i]))); > __mingw_printf ("%.20LF %.20LF\n", sqrtl (1.1L), powl (1.1L, 0.5L)); > __mingw_printf ("%.20LF %.20LF\n", sqrtl (2.2L), powl (2.2L, 0.5L)); > __mingw_printf ("%.20LF %.20LF\n", sqrtl (3.3L), powl (3.3L, 0.5L)); > rl1[0] = sqrtl (1.1L); > rl1[1] = sqrtl (2.2L); > rl1[2] = sqrtl (3.3L); > __mingw_printf ("%.20LF %.20LF %.20LF\n", rl1[0], rl1[1], rl1[2]); > rl2[0] = powl (1.1L, 0.5L); > rl2[1] = powl (2.2L, 0.5L); > rl2[2] = powl (3.3L, 0.5L); > __mingw_printf ("%.20LF %.20LF %.20LF\n", rl2[0], rl2[1], rl2[2]); > > return 1; > } > >> >> Kai > > Thanks, Kai, for your insight and explanations. > > > K. Frank I add a special case to mingww64's pow(fl) function about y=0.5 and get now for your adjusted testcase the following results, which are looking much better IMHO. $ ./tst.exe 1.04880884817015163080, 1.04880884817015163080, 0.00000000000000004142 1.48323969741913264109, 1.48323969741913264109, 0.00000000000000000857 1.81659021245849494619, 1.81659021245849494619, 0.00000000000000000412 1.04880884817015163080 1.04880884817015163080 1.48323969741913264109 1.48323969741913264109 1.81659021245849494619 1.81659021245849494619 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 1.04880884817015154699, 1.04880884817015154699, 0.00000000000000000000 1.48323969741913258980, 1.48323969741913258980, 0.00000000000000000000 1.81659021245849499921, 1.81659021245849499921, 0.00000000000000000000 1.04880884817015154699 1.04880884817015154699 1.48323969741913258980 1.48323969741913258980 1.81659021245849499921 1.81659021245849499921 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 I used for this a 32bit variant and 64bit variant, which made no differences in output. The issue here is that 1.0 branch still uses the inaccurate variant from Danny, but nowaday's trunk uses for pow, sqrt, log, and exp functions new implementations, which are treating rounding and shortpath calculation in an ISOC variant (as gcc expects internally such a math). Also complexmath part is completely rewritten on mingww64's trunk version. IMHO it might be a good idea to replace step by step the classical runtime functions, so that they are not using here msvcrt in first place. Of course the importsymbols of msvcrt (for compatibility with VC generated objects need to remain in .def exports, but the actual functions should come directly from mingwex. 
From: K. Frank <kfrank29.c@gm...>  20110430 15:15:19

Hello Kai! On Sat, Apr 30, 2011 at 5:51 AM, Kai Tietz <ktietz70@...> wrote: > 2011/4/30 Kai Tietz <ktietz70@...>: >> 2011/4/30 Greg Chicares <gchicares@...>: >>> On 20110429 19:55Z, K. Frank wrote: >>>> >>>> By the way, could someone confirm that mingw does use msvcrt for >>>> sqrt and pow? >>> >>> http://cygwin.com/cgibin/cvsweb.cgi/src/winsup/mingw/mingwex/math/?cvsroot=src >>> >>> Looks to me like sqrt(double) calls into msvcrt. But pow() is a >>> different story. Danny implemented it in libmingwex because of >>> some qualityofimplementation problem with msvcrt; then people >>> complained that the libmingwex version was slower, and I can't >>> remember where it wound up, but it's all in the archives. >>> >>>> sqrt (x) and pow (x, 0.5) ought to give the same result (even if not required >>>> to by IEEE754). >>> ... >>> Not so many years ago (perhaps when msvcrt was written), it was >>> thought that correctlyrounded transcendental functions weren't >>> feasible in practice, so library authors had a looser attitude >>> toward exactness. If you have an ideal implementation of sqrt() >>> but a pow() that's only approximate, should you specialcase >>> the power of exactly 0.5? If you do, then pow() and sqrt() will >>> be consistent, but then the powers >>> x^(0.5*(1epsilon)) >>> x^(0.5 ) >>> x^(0.5*(1+epsilon)) >>> might not be monotone. Some would call that a bad tradeoff. >> >> Well, I spent some efforts into log, exp, and pow implementation. Also >> some other basemath functions I improved in mingww64's math library >> in two terms. a) Make those functions behave as ISOC99 specification >> tells in cornercases. and b) Improve accuracy in terms of gcc's >> internal used gmp. As we noticed here result differences. >> >> The following example illustrates the issues pretty well: >> >> #include <stdio.h> >> #include <math.h> >> >> double abc[3][2] = { { 2.2, 3.1 }, { 2.0, 3.0 }, { 4.0, 0.5 } }; >> long double ab[3][2] = { { 2.2L, 3.1L }, { 2.0L, 3.0L }, { 4.0L, 0.5L } }; >> int main() >> { >> double r[3]; >> long double rl[3]; >> int i; >> for (i = 0; i < 3; i++) >> __mingw_printf ("%.20F^%.20F = %.20F\n", abc[i][0], abc[i][1], >> (r[i] = pow (abc[i][0], abc[i][1]))); >> __mingw_printf ("%.20F %.20F %.20F\n", pow (2.2, 3.1), pow (2.0, >> 3.0), pow (4.0, 0.5)); >> r[0] = pow (2.2, 3.1); >> r[1] = pow (2.0, 3.0); >> r[2] = pow (4.0, 0.5); >> __mingw_printf ("%.20F %.20F %.20F\n", r[0], r[1], r[2]); >> >> for (i = 0; i < 3; i++) >> __mingw_printf ("%.20LF^%.20LF = %.20LF\n", ab[i][0], ab[i][1], >> (rl[i] = powl (ab[i][0], ab[i][1]))); >> __mingw_printf ("%.20LF %.20LF %.20LF\n", >> powl (2.2L, 3.1L), powl (2.0L, 3.0L) , powl >> (4.0L, 0.5L)); >> rl[0] = powl (2.2L, 3.1L); >> rl[1] = powl (2.0L, 3.0L); >> rl[2] = powl (4.0L, 0.5L); >> __mingw_printf ("%.20LF %.20LF %.20LF\n", rl[0], rl[1], rl[2]); >> return 1; >> } Do I understand your point correctly that you are comparing gcc's compiletime and runtime values for pow? For the record, when I run this on a mingw64 build: gcc (GCC) 4.5.2 20101002 (prerelease) [svn/rev.164902  mingww64/oz] I get: C:\>gcc o pow_test pow_test.c C:\>pow_test 2.20000000000000017764^3.10000000000000008882 = 11.52153412678571875460 2.00000000000000000000^3.00000000000000000000 = 8.00000000000000000000 4.00000000000000000000^0.50000000000000000000 = 2.00000000000000000000 11.52153412678571697825 8.00000000000000000000 2.00000000000000000000 0.00000000000000177636 0.00000000000000000000 0.00000000000000000000 2.20000000000000000004^3.09999999999999999991 = 11.52153412678571414718 2.00000000000000000000^3.00000000000000000000 = 8.00000000000000000000 4.00000000000000000000^0.50000000000000000000 = 2.00000000000000000000 11.52153412678571414718 8.00000000000000000000 2.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 As I understand it, this means I am seeing a small difference between compile and runtime evaluation only for powl (2.2, 3.1), i.e., the longdouble case. >> The interesting issue is here that gcc uses for constant >> mathcalculations gmplibrary to get result. For more complex >> variants, it uses crt's math. By this it is important for a gcc based >> runtime to be in the IEEE 754 floating point compatible to gmp's >> results. Yes, that's a good point. I hadn't considered the issue of compiletime evaluation, and, ideally, it should agree with runtime evaluation, adding yet another constraint to our wish list. Furthermore, if we take the idea of a crosscompiler seriously, we would also like the compiletime evaluations of a crosscompiler and native compiler to agree. Hmm... So, we would like sqrt (x) and pow (x, 0.5) to agree. We would like compiletime and runtime evaluations to agree. We would like crosscompilers and native compilers to agree. Taken at face value, this would seem to preclude using the platform library and platform hardware to compute things like sqrt and pow (unless we know they comply with some agreedupon standard, such as IEEE754). Just to make the reasoning explicit, if I perform my runtime evaluation on platform A, e.g., using platform A's FPU, then a B>A crosscompiler running on platform B won't have platform A's FPU available to perform consistent compiletime evaluations. So you either give up some of the abovewishedfor consistency, you forgo the potential performance benefits of using the platform's library and / or hardware, or you get everybody to use a common standard. >> ... >> I admit that this might be not that obvious to users, why results here >> are different, but for a gcc based toolchain we need to play nice with >> gcc's internal assumptions. >> >> Regards, >> Kai > > On closer look for long double, we can have rouding issues for long > double precission. I see that powl (0.01L, 0.5L) we have a rounding > difference to sqrtl (0.01L). Interestingly not for float or double > types. So we we might need to specialcase in pow the case for y == > 0.5 to solve this. Be, however, cognizant of Greg's concern. Forcing pow to agree with sqrt just for y = 0.5 has the potential to make pow, in a sense, internally inconsistent, It's a tricky tradeoff. (Ideally, both pow and sqrt would be "exact" and therefore agree with one another automatically.) By the way, I also see a problem with runtime powl for powl (2.2L, 0.5L). It disagrees slightly from sqrtl (2.2L) and from the compiletime value for powl (2.2L, 0.5L). I modified your test program to compare sqrt with pow (x, 0.5) (and sqrtl with powl). For the record: C:\>gcc version gcc (GCC) 4.5.2 20101002 (prerelease) [svn/rev.164902  mingww64/oz] C:\>gcc o sqrt_pow sqrt_pow.c C:\>sqrt_pow 1.04880884817015163080, 1.04880884817015163080, 0.00000000000000000000 1.48323969741913264109, 1.48323969741913264109, 0.00000000000000000000 1.81659021245849494619, 1.81659021245849494619, 0.00000000000000000000 1.04880884817015163080 1.04880884817015163080 1.48323969741913264109 1.48323969741913264109 1.81659021245849494619 1.81659021245849494619 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 1.04880884817015154699, 1.04880884817015154699, 0.00000000000000000000 1.48323969741913258980, 1.48323969741913258970, 0.00000000000000000011 1.81659021245849499921, 1.81659021245849499921, 0.00000000000000000000 1.04880884817015154699 1.04880884817015154699 1.48323969741913258980 1.48323969741913258980 1.81659021245849499921 1.81659021245849499921 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000000 0.00000000000000000011 0.00000000000000000000 where sqrt_pow.c is: #include <stdio.h> #include <math.h> double abc[3] = { 1.1, 2.2, 3.3 }; long double abcl[3] = { 1.1L, 2.2L, 3.3L }; int main() { double r1[3], r2[3], r3[3]; long double rl1[3], rl2[3], rl3[3]; int i; for (i = 0; i < 3; i++) __mingw_printf ("%.20F, %.20F, %.20F\n", (r1[i] = sqrt (abc[i])), (r2[i] = pow (abc[i], 0.5)), (r3[i] = pow (abc[i], 0.5)  sqrt (abc[i]))); __mingw_printf ("%.20F %.20F\n", sqrt (1.1), pow (1.1, 0.5)); __mingw_printf ("%.20F %.20F\n", sqrt (2.2), pow (2.2, 0.5)); __mingw_printf ("%.20F %.20F\n", sqrt (3.3), pow (3.3, 0.5)); r1[0] = sqrt (1.1); r1[1] = sqrt (2.2); r1[2] = sqrt (3.3); __mingw_printf ("%.20F %.20F %.20F\n", r1[0], r1[1], r1[2]); r2[0] = pow (1.1, 0.5); r2[1] = pow (2.2, 0.5); r2[2] = pow (3.3, 0.5); __mingw_printf ("%.20F %.20F %.20F\n", r2[0], r2[1], r2[2]); for (i = 0; i < 3; i++) __mingw_printf ("%.20LF, %.20LF, %.20LF\n", (rl1[i] = sqrtl (abcl[i])), (rl2[i] = powl (abcl[i], 0.5L)), (rl3[i] = powl (abcl[i], 0.5L)  sqrtl (abcl[i]))); __mingw_printf ("%.20LF %.20LF\n", sqrtl (1.1L), powl (1.1L, 0.5L)); __mingw_printf ("%.20LF %.20LF\n", sqrtl (2.2L), powl (2.2L, 0.5L)); __mingw_printf ("%.20LF %.20LF\n", sqrtl (3.3L), powl (3.3L, 0.5L)); rl1[0] = sqrtl (1.1L); rl1[1] = sqrtl (2.2L); rl1[2] = sqrtl (3.3L); __mingw_printf ("%.20LF %.20LF %.20LF\n", rl1[0], rl1[1], rl1[2]); rl2[0] = powl (1.1L, 0.5L); rl2[1] = powl (2.2L, 0.5L); rl2[2] = powl (3.3L, 0.5L); __mingw_printf ("%.20LF %.20LF %.20LF\n", rl2[0], rl2[1], rl2[2]); return 1; } > > Kai Thanks, Kai, for your insight and explanations. K. Frank 
From: RSPsoftware <arabcoder@ya...>  20110430 14:16:17

in the function int __stdcallnormalize_money_csharp( char*data2, char*_out_,int val) if 'val' isnot used Wall do´nt detect it 
From: James K Beard <beardjamesk@ve...>  20110430 13:36:33

My comments don't seem to be making it into the thread this week, possibly because of a switch between my replyto email address and another. So, if it gets lost, perhaps nightstrike can intervene, or echo the information. Perhaps not. The sqrt() function can be implemented with excellent speed to within a bit of full mantissa accuracy using iterations of Newton's method, as is classically done in libraries. Not so with ln() or exp(), which are classically done by extracting the exponent and using the reducedrange mantissa in a modified Chebychev approximation. Arithmetic coprocessors do sqrt(), ln(), and exp() internally with 80bit floating point, then store in 32, 64, or 80 bit floating point format. So, the coprocessors can get full mantissa accuracy with 32bit and 64bit sqrt() and pow(x,0.5). So, the problem is nonexistent for Intel/PPC and most other common workstation processors. Yet we are looking at software transcendental function libraries. James K Beard Original Message From: Greg Chicares [mailto:gchicares@...] Sent: Friday, April 29, 2011 7:38 PM To: MinGW Users List Cc: mingw64 Subject: Re: [Mingww64public] [Mingwusers] Math library discrepancies that surprised me. On 20110429 19:55Z, K. Frank wrote: > > By the way, could someone confirm that mingw does use msvcrt for > sqrt and pow? http://cygwin.com/cgibin/cvsweb.cgi/src/winsup/mingw/mingwex/math/?cvsroot= src Looks to me like sqrt(double) calls into msvcrt. But pow() is a different story. Danny implemented it in libmingwex because of some qualityofimplementation problem with msvcrt; then people complained that the libmingwex version was slower, and I can't remember where it wound up, but it's all in the archives. > sqrt (x) and pow (x, 0.5) ought to give the same result (even if not required > to by IEEE754). If they should give exactly the same result always, then the first thing that needs to be fixed is any standard that allows them to differ. The language standards are so much looser than the numerical standard, even now, that I'd rather see C and C++ catch up to IEEE7541985 before asking them to go beyond. > Browsing some gcc lists, I did see some comments that suggest that > gcc (on linux) does try to transform pow (x, 0.5) to sqrt (x). This would > make sqrt and pow consistent (whether or not technically correct). Not so many years ago (perhaps when msvcrt was written), it was thought that correctlyrounded transcendental functions weren't feasible in practice, so library authors had a looser attitude toward exactness. If you have an ideal implementation of sqrt() but a pow() that's only approximate, should you specialcase the power of exactly 0.5? If you do, then pow() and sqrt() will be consistent, but then the powers x^(0.5*(1epsilon)) x^(0.5 ) x^(0.5*(1+epsilon)) might not be monotone. Some would call that a bad tradeoff.   WhatsUp Gold  Download Free Network Management Software The most intuitive, comprehensive, and costeffective network management toolset available today. Delivers lowest initial acquisition cost and overall TCO of any competing solution. http://p.sf.net/sfu/whatsupgoldsd _______________________________________________ Mingww64public mailing list Mingww64public@... https://lists.sourceforge.net/lists/listinfo/mingww64public 
From: JonY <jon_y@us...>  20110430 11:27:59

On 4/30/2011 09:11, Dongsheng Song wrote: > On Sat, Apr 30, 2011 at 05:16, Jim Bell <Jim@...> wrote: > >> So to purge msvcrt.dll from a nontrivial project, I need to rebuild >> libstdc++6.dll and libgcc_s_dw21.dll with msvcr100 instead of msvcrt, >> right? >> >> If I were to be foolish enough to attempt this, where would I start? >> Download gcc and work through its docs & dependencies? Or are there >> MinGW instructions/hints? (I couldn't find any.) >> >> > I think the most simple way is: > 1. cp mingww64crt/lib32/msvcr100.def mingww64crt/lib32/msvcrt.def > 2. cp mingww64crt/lib64/msvcr100.def mingww64crt/lib64/msvcrt.def > > Then do a full mingww64 build, that's all. > > Or you can modify gcc source, change the default library from 'lmsvcrt' to > 'lmsvcr100', > Then do a full mingww64 build, but I can not assume it can works for all > case. > >  > Dongsheng > Hello, I need to stress that the OP is using the original mingw headers and crt, so those don't exist there. Also, mingww64 doesn't use dw2, so read the docs carefully if you want to use mingww64. Dongsheng's point still stands, you could just build the original mingw crt and w32api normally, but switch libmsvcrt.a with libmsvcr100.a. After that, you can rebuild GCC. 
From: Kai Tietz <ktietz70@go...>  20110430 09:51:40

2011/4/30 Kai Tietz <ktietz70@...>: > 2011/4/30 Greg Chicares <gchicares@...>: >> On 20110429 19:55Z, K. Frank wrote: >>> >>> By the way, could someone confirm that mingw does use msvcrt for >>> sqrt and pow? >> >> http://cygwin.com/cgibin/cvsweb.cgi/src/winsup/mingw/mingwex/math/?cvsroot=src >> >> Looks to me like sqrt(double) calls into msvcrt. But pow() is a >> different story. Danny implemented it in libmingwex because of >> some qualityofimplementation problem with msvcrt; then people >> complained that the libmingwex version was slower, and I can't >> remember where it wound up, but it's all in the archives. >> >>> sqrt (x) and pow (x, 0.5) ought to give the same result (even if not required >>> to by IEEE754). >> >> If they should give exactly the same result always, then the >> first thing that needs to be fixed is any standard that allows >> them to differ. The language standards are so much looser than >> the numerical standard, even now, that I'd rather see C and C++ >> catch up to IEEE7541985 before asking them to go beyond. >> >>> Browsing some gcc lists, I did see some comments that suggest that >>> gcc (on linux) does try to transform pow (x, 0.5) to sqrt (x). This would >>> make sqrt and pow consistent (whether or not technically correct). >> >> Not so many years ago (perhaps when msvcrt was written), it was >> thought that correctlyrounded transcendental functions weren't >> feasible in practice, so library authors had a looser attitude >> toward exactness. If you have an ideal implementation of sqrt() >> but a pow() that's only approximate, should you specialcase >> the power of exactly 0.5? If you do, then pow() and sqrt() will >> be consistent, but then the powers >> x^(0.5*(1epsilon)) >> x^(0.5 ) >> x^(0.5*(1+epsilon)) >> might not be monotone. Some would call that a bad tradeoff. > > Well, I spent some efforts into log, exp, and pow implementation. Also > some other basemath functions I improved in mingww64's math library > in two terms. a) Make those functions behave as ISOC99 specification > tells in cornercases. and b) Improve accuracy in terms of gcc's > internal used gmp. As we noticed here result differences. > > The following example illustrates the issues pretty well: > > #include <stdio.h> > #include <math.h> > > double abc[3][2] = { { 2.2, 3.1 }, { 2.0, 3.0 }, { 4.0, 0.5 } }; > long double ab[3][2] = { { 2.2L, 3.1L }, { 2.0L, 3.0L }, { 4.0L, 0.5L } }; > int main() > { > double r[3]; > long double rl[3]; > int i; > for (i = 0; i < 3; i++) > __mingw_printf ("%.20F^%.20F = %.20F\n", abc[i][0], abc[i][1], > (r[i] = pow (abc[i][0], abc[i][1]))); > __mingw_printf ("%.20F %.20F %.20F\n", pow (2.2, 3.1), pow (2.0, > 3.0), pow (4.0, 0.5)); > r[0] = pow (2.2, 3.1); > r[1] = pow (2.0, 3.0); > r[2] = pow (4.0, 0.5); > __mingw_printf ("%.20F %.20F %.20F\n", r[0], r[1], r[2]); > > for (i = 0; i < 3; i++) > __mingw_printf ("%.20LF^%.20LF = %.20LF\n", ab[i][0], ab[i][1], > (rl[i] = powl (ab[i][0], ab[i][1]))); > __mingw_printf ("%.20LF %.20LF %.20LF\n", > powl (2.2L, 3.1L), powl (2.0L, 3.0L) , powl > (4.0L, 0.5L)); > rl[0] = powl (2.2L, 3.1L); > rl[1] = powl (2.0L, 3.0L); > rl[2] = powl (4.0L, 0.5L); > __mingw_printf ("%.20LF %.20LF %.20LF\n", rl[0], rl[1], rl[2]); > return 1; > } > > The interesting issue is here that gcc uses for constant > mathcalculations gmplibrary to get result. For more complex > variants, it uses crt's math. By this it is important for a gcc based > runtime to be in the IEEE 754 floating point compatible to gmp's > results. > > Btw new pow, log, and exp variant are slightly faster then variant in > msvcrt, but this is more or less just a sideeffect. > > I admit that this might be not that obvious to users, why results here > are different, but for a gcc based toolchain we need to play nice with > gcc's internal assumptions. > > Regards, > Kai > On closer look for long double, we can have rouding issues for long double precission. I see that powl (0.01L, 0.5L) we have a rounding difference to sqrtl (0.01L). Interestingly not for float or double types. So we we might need to specialcase in pow the case for y == 0.5 to solve this. Kai 
From: Kai Tietz <ktietz70@go...>  20110430 08:42:16

2011/4/30 Greg Chicares <gchicares@...>: > On 20110429 19:55Z, K. Frank wrote: >> >> By the way, could someone confirm that mingw does use msvcrt for >> sqrt and pow? > > http://cygwin.com/cgibin/cvsweb.cgi/src/winsup/mingw/mingwex/math/?cvsroot=src > > Looks to me like sqrt(double) calls into msvcrt. But pow() is a > different story. Danny implemented it in libmingwex because of > some qualityofimplementation problem with msvcrt; then people > complained that the libmingwex version was slower, and I can't > remember where it wound up, but it's all in the archives. > >> sqrt (x) and pow (x, 0.5) ought to give the same result (even if not required >> to by IEEE754). > > If they should give exactly the same result always, then the > first thing that needs to be fixed is any standard that allows > them to differ. The language standards are so much looser than > the numerical standard, even now, that I'd rather see C and C++ > catch up to IEEE7541985 before asking them to go beyond. > >> Browsing some gcc lists, I did see some comments that suggest that >> gcc (on linux) does try to transform pow (x, 0.5) to sqrt (x). This would >> make sqrt and pow consistent (whether or not technically correct). > > Not so many years ago (perhaps when msvcrt was written), it was > thought that correctlyrounded transcendental functions weren't > feasible in practice, so library authors had a looser attitude > toward exactness. If you have an ideal implementation of sqrt() > but a pow() that's only approximate, should you specialcase > the power of exactly 0.5? If you do, then pow() and sqrt() will > be consistent, but then the powers > x^(0.5*(1epsilon)) > x^(0.5 ) > x^(0.5*(1+epsilon)) > might not be monotone. Some would call that a bad tradeoff. Well, I spent some efforts into log, exp, and pow implementation. Also some other basemath functions I improved in mingww64's math library in two terms. a) Make those functions behave as ISOC99 specification tells in cornercases. and b) Improve accuracy in terms of gcc's internal used gmp. As we noticed here result differences. The following example illustrates the issues pretty well: #include <stdio.h> #include <math.h> double abc[3][2] = { { 2.2, 3.1 }, { 2.0, 3.0 }, { 4.0, 0.5 } }; long double ab[3][2] = { { 2.2L, 3.1L }, { 2.0L, 3.0L }, { 4.0L, 0.5L } }; int main() { double r[3]; long double rl[3]; int i; for (i = 0; i < 3; i++) __mingw_printf ("%.20F^%.20F = %.20F\n", abc[i][0], abc[i][1], (r[i] = pow (abc[i][0], abc[i][1]))); __mingw_printf ("%.20F %.20F %.20F\n", pow (2.2, 3.1), pow (2.0, 3.0), pow (4.0, 0.5)); r[0] = pow (2.2, 3.1); r[1] = pow (2.0, 3.0); r[2] = pow (4.0, 0.5); __mingw_printf ("%.20F %.20F %.20F\n", r[0], r[1], r[2]); for (i = 0; i < 3; i++) __mingw_printf ("%.20LF^%.20LF = %.20LF\n", ab[i][0], ab[i][1], (rl[i] = powl (ab[i][0], ab[i][1]))); __mingw_printf ("%.20LF %.20LF %.20LF\n", powl (2.2L, 3.1L), powl (2.0L, 3.0L) , powl (4.0L, 0.5L)); rl[0] = powl (2.2L, 3.1L); rl[1] = powl (2.0L, 3.0L); rl[2] = powl (4.0L, 0.5L); __mingw_printf ("%.20LF %.20LF %.20LF\n", rl[0], rl[1], rl[2]); return 1; } The interesting issue is here that gcc uses for constant mathcalculations gmplibrary to get result. For more complex variants, it uses crt's math. By this it is important for a gcc based runtime to be in the IEEE 754 floating point compatible to gmp's results. Btw new pow, log, and exp variant are slightly faster then variant in msvcrt, but this is more or less just a sideeffect. I admit that this might be not that obvious to users, why results here are different, but for a gcc based toolchain we need to play nice with gcc's internal assumptions. Regards, Kai 
From: Greg Chicares <gchicares@sb...>  20110430 05:08:47

On 20110430 02:04Z, Ocean wrote: [...using C++...] > 1. Does MinGW support RegEx "out of the box", or do I need to install a > RegEx library seperately? Yes: a regex facility was added to the C++ standard library in TR1. To use it, #include <regex>. 
From: Ocean <Ocean@co...>  20110430 02:18:10

I am a beginner to C++ in general, as well as MinGW. I would like to use RegEx in my code, but I'm unfamiliar with how. To clarify, I know how to use RegEx, but not in the context of C++ coding. Doing a good bit of searching, I still don't have a clear set of answers, so please forgive me if these questions have been asked before, and if it makes any difference, I'm using Win 7 x64. 1. Does MinGW support RegEx "out of the box", or do I need to install a RegEx library seperately? 2. If it does support RegEx natively, what do I need to do to be able to access the RegEx functionality within a source file? And where can I look up reference information about the specific functions, their syntax, the RegEx flavor, etc...? 3. How would I install a RegEx library if MinGW doesn't support it by default, or if I wanted to use another engine? Thanks in advance to any and all who are kind enough to help! :) 
From: Dongsheng Song <dongsheng.song@gm...>  20110430 01:11:31

On Sat, Apr 30, 2011 at 05:16, Jim Bell <Jim@...> wrote: > So to purge msvcrt.dll from a nontrivial project, I need to rebuild > libstdc++6.dll and libgcc_s_dw21.dll with msvcr100 instead of msvcrt, > right? > > If I were to be foolish enough to attempt this, where would I start? > Download gcc and work through its docs & dependencies? Or are there > MinGW instructions/hints? (I couldn't find any.) > > I think the most simple way is: 1. cp mingww64crt/lib32/msvcr100.def mingww64crt/lib32/msvcrt.def 2. cp mingww64crt/lib64/msvcr100.def mingww64crt/lib64/msvcrt.def Then do a full mingww64 build, that's all. Or you can modify gcc source, change the default library from 'lmsvcrt' to 'lmsvcr100', Then do a full mingww64 build, but I can not assume it can works for all case.  Dongsheng 
From: K. Frank <kfrank29.c@gm...>  20110430 00:51:08

Hi Greg! On Fri, Apr 29, 2011 at 7:38 PM, Greg Chicares <gchicares@...> wrote: > On 20110429 19:55Z, K. Frank wrote: >> >> By the way, could someone confirm that mingw does use msvcrt for >> sqrt and pow? > > http://cygwin.com/cgibin/cvsweb.cgi/src/winsup/mingw/mingwex/math/?cvsroot=src > > Looks to me like sqrt(double) calls into msvcrt. But pow() is a > different story. Danny implemented it in libmingwex because of > some qualityofimplementation problem with msvcrt; then people > complained that the libmingwex version was slower, and I can't > remember where it wound up, but it's all in the archives. Thanks for that  good to know. From the source link you gave: http://cygwin.com/cgibin/cvsweb.cgi/src/winsup/mingw/mingwex/math/pow.c?annotate=1.1.10.1&cvsroot=src the code for pow (the regular double pow (double, double)) has these comments about accuracy. 28: * ACCURACY: 29: * 30: * Relative error: 31: * arithmetic domain # trials peak rms 32: * IEEE 26,26 30000 4.2e16 7.7e17 33: * DEC 26,26 60000 4.8e17 9.1e18 34: * 1/26 < x < 26, with log(x) uniformly distributed. 35: * 26 < y < 26, y uniformly distributed. 36: * IEEE 0,8700 30000 1.5e14 2.1e15 37: * 0.99 < x < 1.01, 0 < y < 8700, uniformly distributed. This commentary suggest that pow is not "exact." This would reasonably explain Arthur seeing pow not match sqrt. But it still leaves the puzzle of why the different mingw versions don't behave the same. One minor puzzle: I don't see anything in the pow code that would let the compiler / library take advantage of hardware support for pow, which I think has been available in the x86 architecture for quite some time now. > >> sqrt (x) and pow (x, 0.5) ought to give the same result (even if not required >> to by IEEE754). > > If they should give exactly the same result always, then the > first thing that needs to be fixed is any standard that allows > them to differ. Well, it's a question of engineering compromise whether they should always agree. In an ideal world, it would be nice if pow were "exact" (in the sense of closet floatingpoint number). Then if pow were "exact," pow and sqrt should agree. But there is expense in getting pow to be "exact," so it's arguable that the language / IEEE standards should give implementers some flexibility. (And, as you point out below, if pow isn't exact, then forcing pow and sqrt to agree could actually cause other problems.) > The language standards are so much looser than > the numerical standard, even now, that I'd rather see C and C++ > catch up to IEEE7541985 before asking them to go beyond. > >> Browsing some gcc lists, I did see some comments that suggest that >> gcc (on linux) does try to transform pow (x, 0.5) to sqrt (x). This would >> make sqrt and pow consistent (whether or not technically correct). > > Not so many years ago (perhaps when msvcrt was written), it was > thought that correctlyrounded transcendental functions weren't > feasible in practice, so library authors had a looser attitude > toward exactness. If you have an ideal implementation of sqrt() > but a pow() that's only approximate, should you specialcase > the power of exactly 0.5? If you do, then pow() and sqrt() will > be consistent, but then the powers > x^(0.5*(1epsilon)) > x^(0.5 ) > x^(0.5*(1+epsilon)) > might not be monotone. Some would call that a bad tradeoff. I actually agree with your concern. The transformation of pow (x, 0.5) to sqrt (x) that it sounded like gcc / linux was doing sounds like a bad idea if it's being done this simply and naively. Any speculation on why Arthur gets different results with different mingw's? Thanks for your perspective. K. Frank 
From: Greg Chicares <gchicares@sb...>  20110429 23:38:10

On 20110429 19:55Z, K. Frank wrote: > > By the way, could someone confirm that mingw does use msvcrt for > sqrt and pow? http://cygwin.com/cgibin/cvsweb.cgi/src/winsup/mingw/mingwex/math/?cvsroot=src Looks to me like sqrt(double) calls into msvcrt. But pow() is a different story. Danny implemented it in libmingwex because of some qualityofimplementation problem with msvcrt; then people complained that the libmingwex version was slower, and I can't remember where it wound up, but it's all in the archives. > sqrt (x) and pow (x, 0.5) ought to give the same result (even if not required > to by IEEE754). If they should give exactly the same result always, then the first thing that needs to be fixed is any standard that allows them to differ. The language standards are so much looser than the numerical standard, even now, that I'd rather see C and C++ catch up to IEEE7541985 before asking them to go beyond. > Browsing some gcc lists, I did see some comments that suggest that > gcc (on linux) does try to transform pow (x, 0.5) to sqrt (x). This would > make sqrt and pow consistent (whether or not technically correct). Not so many years ago (perhaps when msvcrt was written), it was thought that correctlyrounded transcendental functions weren't feasible in practice, so library authors had a looser attitude toward exactness. If you have an ideal implementation of sqrt() but a pow() that's only approximate, should you specialcase the power of exactly 0.5? If you do, then pow() and sqrt() will be consistent, but then the powers x^(0.5*(1epsilon)) x^(0.5 ) x^(0.5*(1+epsilon)) might not be monotone. Some would call that a bad tradeoff. 
From: Jim Bell <Jim@JCBell.com>  20110429 21:16:50

So to purge msvcrt.dll from a nontrivial project, I need to rebuild libstdc++6.dll and libgcc_s_dw21.dll with msvcr100 instead of msvcrt, right? If I were to be foolish enough to attempt this, where would I start? Download gcc and work through its docs & dependencies? Or are there MinGW instructions/hints? (I couldn't find any.) Thanks in Advance, Jim 
From: K. Frank <kfrank29.c@gm...>  20110429 19:56:00

Hello Arthur! (I've take the liberty of copying this to the mingw list, as well.) On Fri, Apr 29, 2011 at 9:54 AM, Arthur Norman wrote: > The following short program compares the value of pow(x, 0.5) to that > returned by sqrt(x) for a range of values of x, and prints out any first > case where the two do not agree. Well, this is unfortunate. Ideally they should, and as a qualityofimplementation issue, I think we should be able to expect them to agree. > With the mingw family of compilers I had > expected the Microsoft C library to be in use. > > What I appear to see is > linux (ubuntu 11.04 64bit tested) no output > Mac (snow leopard) no output > gcc (cygwin) no output > gcc3 mnocygwin ) > i686w64mingw32gcc ) all different from each other > x86_64w64mingw32gcc ) but all find discrepancies. Very interesting. > It is a totally fair cop if the Microsoft C library gives different > results in the last bit for floating point elementary functions to gnu > libraries, Well, from the mingw perspective, if the microsoft runtime library is flawed in this way, then it's a legitimate cop for mingw not to fix it. By the way, could someone confirm that mingw does use msvcrt for sqrt and pow? > but is there an obvious reason why the three mingw varients > all give different answers. If I understand your point correctly, if the problem is with msvcrt, then the three mingw variants should all give the same flawed results. > This showed up in regression testing something > much larger across platforms. > > Arthur > ... As I understand it, IEEE754 compliance requires sqrt to be calculated "correctly", i.e., to return the floatingpoint number closest to the exact mathematical value. (The c++ standard, I believe, does not, however, require IEEE754 compliance.) I think, however, the IEEE754 does not require all mathematical functions, and, in particular, pow, to be "correct," in the above sense sense. It would be interesting to print out in binary (or hex) the results for a few of the values that lead to discrepancies with both mingw and a linuxbased gcc, and see whether the problem is in sqrt (for which IEEE754 specifies the result) or in pow (for which I think IEEE754 permits some leeway). In my opinion: mingw / msvcrt should be IEEE754 compliant. Of course, if msvcrt isn't, then it's not mingw's fault. sqrt (x) and pow (x, 0.5) ought to give the same result (even if not required to by IEEE754). Browsing some gcc lists, I did see some comments that suggest that gcc (on linux) does try to transform pow (x, 0.5) to sqrt (x). This would make sqrt and pow consistent (whether or not technically correct). Maybe gcc / linux pow and msvcrt pow are both not quite exact, but this is hidden in your test, because gcc / linux replaces pow (x, 0.5) with sqrt (x), and msvcrt doesn't. (Just speculation.) It is odd, however, that the various mingw's don't agree, and this speaks to a possible bug in mingw. Thanks. K. Frank > ... > ==================== > #include <stdio.h> > #include <math.h> > > int main(int argc, char *argv[]) > { > double a, b, c; > int i; > for (i=0, a=1.0; i<10000000; i++) > { a *= 1.00000001; > b = sqrt(a); c = pow(a, 0.5); > if (b == c) continue; > printf("%.17g %.17g %.17g\n", a, b, c); > return 0; > } > return 0; > } > ==================== > > acn1@... temp > $ gcc expt.c o expt > acn1@... temp > $ ./expt > acn1@... temp > $ gcc3 mnocygwin expt.c o expt > acn1@... temp > $ ./expt > 1.0241210164238141 1.011988644414459 1.0119886444144588 > acn1@... temp > $ i686w64mingw32gcc expt.c o expt > acn1@... temp > $ ./expt > 1.0004603659307831 1.000230156479389 1.0002301564793892 > acn1@... temp > $ x86_64w64mingw32gcc expt.c o expt > acn1@... temp > $ ./expt > 1.0000179101601867 1.0000089550399969 1.0000089550399971 > acn1@... temp > ============================================================== 
From: TSalm <TSalm@fr...>  20110429 17:57:41

Le 29/04/2011 15:12, LRN a écrit : > On 29.04.2011 16:11, LRN wrote: >> On 29.04.2011 14:56, TSalm wrote: >>> Le 29/04/2011 13:28, Greg Chicares a écrit : >>>> On 20110429 10:49Z, TSalm wrote: >>>>> Le 29/04/2011 10:44, Gisle Vanem a écrit : >>>> [...] >>>>>> I think it's simpler (as the >>>>>> comment says) to just define "D_WIN32_WINNT=0x501" on the >>>>>> gcc commandline. >>>> [...] >>>>> In fact, I've added this in a "Configure.h" file , and include it >>>>> before >>>>> every #include block : >>>>> #define WINVER 0x0501 >>>>> #define _WIN32_WINNT 0x0501 >>>>> But G++ doesn't find GetProcessImageFileNameA : >>>>> D:\dev\C++\WinList\WinList\src\main.cpp:98: error: >>>>> `GetProcessImageFileNameA' was not declared in this scope >>>>> I don't understand where I'm wrong... >>>> Try specifying "D_WIN32_WINNT=0x501" on the gcc command line instead: >>>> that's guaranteed to be parsed before any header is included. If that >>>> works, then "Configure.h" wasn't working the way you expected. >>>> >>>> When the preprocessor doesn't behave as hoped, the gcc commandline >>>> option 'E' is helpful for determining the cause. >>> No, it's not resolv the problem too :( >>> [...] >>> >> Whoawhoa, are you sure you're using MinGWgcc and not Msysgcc? > I mean, mingww32api, not msysw32api > Yes, it's in fact an old version of MinGW...(gcc3.4.5). I've just installing the latest version, and now, the problem disappear. Thanks to everybody for your help. ____________ Analyse antivirus effectuée par G Data AntiVirus 
From: Earnie <earnie@us...>  20110429 14:46:48

TSalm wrote: > Le 29/04/2011 13:28, Greg Chicares a écrit : >> On 20110429 10:49Z, TSalm wrote: >>> Le 29/04/2011 10:44, Gisle Vanem a écrit : >> [...] >>>> I think it's simpler (as the >>>> comment says) to just define "D_WIN32_WINNT=0x501" on the >>>> gcc commandline. >> [...] >>> In fact, I've added this in a "Configure.h" file , and include it before >>> every #include block : >>> #define WINVER 0x0501 >>> #define _WIN32_WINNT 0x0501 >>> But G++ doesn't find GetProcessImageFileNameA : >>> D:\dev\C++\WinList\WinList\src\main.cpp:98: error: >>> `GetProcessImageFileNameA' was not declared in this scope >>> I don't understand where I'm wrong... >> Try specifying "D_WIN32_WINNT=0x501" on the gcc command line instead: >> that's guaranteed to be parsed before any header is included. If that >> works, then "Configure.h" wasn't working the way you expected. >> >> When the preprocessor doesn't behave as hoped, the gcc commandline >> option 'E' is helpful for determining the cause. > No, it's not resolv the problem too :( > mingw32g++.exe Wall fexceptions D_WIN32_WINNT=0x501 DWINVER=0x0501 > O2 Isrc I"C:\Program Files\MinGW\include" I..\..\fltk1.3.xr6829 > IC:\MinGW\include IC:\MinGW\include\c++ IC:\MinGW\include\c++\3.4.2 > IC:\MinGW\include\c++\3.4.2\mingw32 > IC:\MinGW\lib\gcc\mingw32\3.4.2\include IC:\MinGW\include\sys c > D:\dev\C++\WinList\WinList\src\main.cpp o obj\Release\src\main.o > Does the following work for you? mingw32g++ Wall fexceptions D_WIN32_WINNT=0x0501 DWINVER=0x0501 Isrc I..\..\fltk1.3.xr6829 c d:\dev\C++\WinList\WinList\src\main.cpp o obj\Release\src\main.o I've removed the specified include directories that relate to MinGW itself. You should not need to include those, if you do, then something is wrong with your environment.  Earnie  http://www.formykids.com 