This accepts a 10-byte binary encoded IEEE 754 80-bit real and unpacks it into a IEEE 64-bit real (Perl NV).
Testing is limited to Intel P4 (Does anyone else do 80-bit FP math?) and Perl 5.8.6/7 under WinXP.
The test data below shows a selected range of values and compares the output from the generating program's vprintf equivalent, and Perl's printf on the converted results. The values are selected to highlight the range and accuracy limitations of both formats at all four! ends of the scale.
It shows correct handling of underflow and overflow by conversion to [+|-]1.#INF as appropriate.
I believe it will handle #NaN (& #QNaN) correctly, but this has not been tested, as I haven't worked out how to generate these in the writer.
A version that doesn't go through ASCII encoded binary has so far eluded me due to the limitations of 32-bit integers for the bit twiddling.
sub IEEE80toIEEE64 { my( $discard, $mantissa, $hidden, $exponent, $sign ) = unpack 'a11 a52 a1 a15 a1', unpack 'b80', $_[ 0 ]; $exponent = unpack( 'v', pack 'b15', $exponent ) - 16383 + 1023; $exponent = 32767, $mantissa = '0' x 52 if $exponent < 0 or $exponent > 2047; $exponent = unpack 'b11', pack 'v', $exponent; return unpack 'd', pack 'b64', $mantissa . $exponent . $sign; }
realtest x.bin C:\dmd\test>\test\80bit.pl -1.0935598595647423806e-4950 -1.#INF 1.0935598595647423806e-4950 1.#INF -1.2345678901234567887e-4932 -1.#INF 1.2345678901234567887e-4932 1.#INF -1.2345678901234567887e-309 -1.#INF 1.2345678901234567887e-309 1.#INF -1.2345678901234567912e-206 -1.2345678901234567e-206 1.2345678901234567912e-206 1.2345678901234567e-206 -1.2345678901234567937e-103 -1.2345678901234568e-103 1.2345678901234567937e-103 1.2345678901234568e-103 -1.234567890123456796 -1.2345678901234567 1.234567890123456796 1.2345678901234567 -1.2345678901234567983e+103 -1.2345678901234567e+103 1.2345678901234567983e+103 1.2345678901234567e+103 -1.2345678901234568006e+206 -1.2345678901234568e+206 1.2345678901234568006e+206 1.2345678901234568e+206 -1.2345678901234568031e+309 -1.#INF 1.2345678901234568031e+309 1.#INF -1.2345678901234567889e+4930 -1.#INF 1.2345678901234567889e+4930 1.#INF -1.2345678901234567892e+4931 -1.#INF 1.2345678901234567892e+4931 1.#INF 0.005000 0.0049999999999999992 -0.005000 -0.0049999999999999992 0.050000 0.049999999999999996 -0.050000 -0.049999999999999996 0.500000 0.5 -0.500000 -0.5 5.000000 5 -5.000000 -5 50.000000 50 -50.000000 -50
|
|---|