From: mlevedahl AT erols DOT com (mark levedahl)
Newsgroups: comp.os.msdos.djgpp
Subject: Re: Round off errors
Date: Fri, 20 Feb 1998 20:38:49 GMT
Organization: Erol's Internet Services
Lines: 67
Message-ID: <34efe8fb.923747@news.erols.com>
References: <01bd3e0c$200881c0$LocalHost AT alfredoc>
NNTP-Posting-Host: 207-172-112-238.s238.tnt4.ann.erols.com
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Precedence: bulk

On Fri, 20 Feb 1998 08:30:41 -0600, "Alfredo J. Cole" <ajc AT acyc DOT com>
wrote:

Works fine using double rather than float. float will give 6 to 7
digits of accuracy, so your example fails for known reasons. double
offers about 15 significant figures. Unless you're involved in
tracking the Federal budget and/or National Debt, this should be more
than adequate, even across a large number of calculations. Still, BCD
would be safer..

>Compiling the following code with DJGPP gives the erroneous result
>(5551.20). The same code compiled with Borland's bcd class gives the
>correct result (5551.22). I am developing an accounting system with
>DJGPP and it requires that the results be exact. Is there a similar
>class or option under DJGPP that will eliminate these round off
>errors?
>
>Thank you.
>
>------------------------------
>/*  Example of round-off error */
>/*  When this program is executed with IEEE binary floating
>    point arithmetic, the answer is incorrect due to accumulated
>    round-off error.  When BCD is used, the correct answer
>    is obtained.
>    */
>
>#include <stdio.h>
>
>float data[100] = {
>       4997.01, 921.05, -1921.06, 4621.03, -605.20,
>      -3105.12, 4665.01, -271.23, -212.23, -1548.18,
>      -2351.12, -2787.12, 1474.19, 1817.07, -2299.12,
>       2648.13, -3787.12, 3948.13, -3401.12, 170.14,
>       4030.14, -334.22, -704.20, -2233.12, 3369.13,
>      -4218.24, 2214.13, -4707.24, 2369.13, 4816.01,
>      -4009.12, 332.15, 4501.01, -1349.18, -438.22,
>      -4996.24, 4020.13, -2623.12, 2541.13, -608.20,
>       3929.14, -3300.12, -4314.24, 839.16, 2173.14,
>       1243.07, -2291.12, 693.16, 4061.14, -2625.12,
>      -4048.12, 2162.13, 686.05, 2133.13, -2149.12,
>       3884.13, -199.24, -3611.12, 3282.13, -95.24,
>      -4614.24, 1875.07, 2203.13, -1371.18, -1616.06,
>       1210.07, -3014.12, 4197.01, -4141.24, 3255.13,
>      -3203.12, 547.16, 2713.14, -1039.18, 982.05,
>      -1924.06, 3419.13, -2488.12, 4334.03, -4469.24,
>       4088.14, -3840.12, 585.05, -1932.06, 4496.01,
>       -918.20, -445.22, -1856.18, 2007.07, -2631.12,
>        351.03, 774.05, 84.01, -1089.06, 4278.01,
>      -4860.24, 1165.07, 4482.03, -8.23, -3432.12 };
>
>int main () {
>   /* add 100 numbers and print the result */
>   /* In this example, the array is initialized to contain
>      the data.  In real programs the data might be read
>      from a file or the keyboard using scanf.
>      */
>   float answer = 0.0;
>   int i;
>   for (i = 0; i < 100; ++i) answer += data[i];
>   printf("The sum is %10.2f\n",answer);
>   printf("The correct answer is 5551.22\n");
>   return 0;
>   }
>
>------------------------------