The problem is that your perl or C-library is setting (or leaving) the IEEE rounding mode set to RC_CHOP (truncate). Any of the other three modes -- UP, DOWN, OR NEAREST -- would fix this.
The following shows the output on my intel system for teh four modes of IEEE, FP operations:
#! perl -slw use strict; use Inline C => Config => BUILD_NOISY => 1; use Inline C => <<'END_C', NAME => 'ieeroundmode', CLEAN_AFTER_BUILD +=> 0; unsigned int controlFP( unsigned int new, unsigned int mask ) { return _controlfp( new, mask ); } END_C use constant { RC_MASK => 0x00000300, RC_CHOP => 0x00000300, RC_UP => 0x00000200, RC_DOWN => 0x00000100, RC_NEAR => 0x00000000, }; print "Setting mode CHOP; controlFP returned: ", controlFP( RC_CHOP, R +C_MASK ); for( my $n = 0.000005; $n < 0.0001; $n += 0.00001 ) { printf "%.15f %.5f\n", $n, $n; } print "Setting mode UP; controlFP returned: ", controlFP( RC_UP, RC_MA +SK ); for( my $n = 0.000005; $n < 0.0001; $n += 0.00001 ) { printf "%.15f %.5f\n", $n, $n; } print "Setting mode DOWN; controlFP returned: ", controlFP( RC_DOWN, R +C_MASK ); for( my $n = 0.000005; $n < 0.0001; $n += 0.00001 ) { printf "%.15f %.5f\n", $n, $n; } print "Setting mode NEAR; controlFP returned: ", controlFP( RC_NEAR, R +C_MASK ); for( my $n = 0.000005; $n < 0.0001; $n += 0.00001 ) { printf "%.15f %.5f\n", $n, $n; } __END__ C:\test>ieeroundmode.pl Setting mode CHOP; controlFP returned: 525087 0.000005000000000 0.00001 0.000015000000000 0.00002 0.000025000000000 0.00003 0.000035000000000 0.00003 0.000045000000000 0.00004 0.000055000000000 0.00005 0.000065000000000 0.00006 0.000075000000000 0.00007 0.000085000000000 0.00008 0.000095000000000 0.00009 Setting mode UP; controlFP returned: 524831 0.000005000000000 0.00001 0.000015000000000 0.00002 0.000025000000000 0.00003 0.000035000000000 0.00004 0.000045000000000 0.00005 0.000055000000000 0.00006 0.000065000000000 0.00007 0.000075000000000 0.00008 0.000085000000000 0.00009 0.000095000000000 0.00010 Setting mode DOWN; controlFP returned: 524575 0.000005000000000 0.00001 0.000015000000000 0.00002 0.000025000000000 0.00003 0.000035000000000 0.00003 0.000045000000000 0.00004 0.000055000000000 0.00005 0.000065000000000 0.00006 0.000075000000000 0.00007 0.000085000000000 0.00008 0.000095000000000 0.00009 Setting mode NEAR; controlFP returned: 524319 0.000005000000000 0.00001 0.000015000000000 0.00002 0.000025000000000 0.00003 0.000035000000000 0.00004 0.000045000000000 0.00005 0.000055000000000 0.00006 0.000065000000000 0.00007 0.000075000000000 0.00008 0.000085000000000 0.00009 0.000095000000000 0.00010
Maybe you can set the rounding mode yourself in a similar way?
In reply to Re: RFC: Large Floating Point Numbers - Rounding Errors (Rounding mode)
by BrowserUk
in thread RFC: Large Floating Point Numbers - Rounding Errors
by GAVollink
| For: | Use: | ||
| & | & | ||
| < | < | ||
| > | > | ||
| [ | [ | ||
| ] | ] |