From: SourceForge.net <no...@so...> - 2003-08-02 01:20:36
|
Bugs item #781753, was opened at 2003-08-01 17:13 Message generated for change (Comment added) made by rtoy You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=104933&aid=781753&group_id=4933 Category: None Group: None Status: Open Resolution: None Priority: 5 Submitted By: Stavros Macrakis (macrakis) Assigned to: Raymond Toy (rtoy) Summary: bfloat->float fails for very large and very small Initial Comment: sm: 2.0b0^-1024.00001b0 => 5.5626...B-309 (OK) float(sm) => 0.0 NO! true for all smaller sm's but floats can represent numbers down to 4.94e-324 float(8.989b307) => non-number but floats can represent numbers up to 1.797693134862316E+308 A much more minor problem: though 2.8e-324 correctly reads in as 4.94e-324 (rounded), 2.7e-324 reads in as 0.0. This is a GCL problem. These may be fixed in more recent GCLs, but should still be checked in Maxima. Maxima 5.9.0 GCL 2.5.0 mingw Windows 2000 Athlon ---------------------------------------------------------------------- >Comment By: Raymond Toy (rtoy) Date: 2003-08-01 21:20 Message: Logged In: YES user_id=28849 The float(sm) issue appears to be a GCL problem. CMUCL doesn't do that. (Perhaps GCL defaults to truncate-to-zero?) The overflow issue is caused by trying to compute 0.5 * 2^1024, and 2^1024 doesn't fit in a double-float. A fix using scale-float instead solves this problem. CMUCL doesn't have problems with 2.7e-324 becoming 0.0. ---------------------------------------------------------------------- Comment By: Raymond Toy (rtoy) Date: 2003-08-01 18:30 Message: Logged In: YES user_id=28849 The float(sm) issue appears to be a GCL problem. CMUCL doesn't do that. (Perhaps GCL defaults to truncate-to-zero?) The overflow issue is caused by trying to compute 0.5 * 2^1024, and 2^1024 doesn't fit in a double-float. A fix using scale-float instead solves this problem. CMUCL doesn't have problems with 2.7e-324 becoming 0.0. ---------------------------------------------------------------------- You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=104933&aid=781753&group_id=4933 |