gcc45 @4.5.3_4 gfortran gives wrong results for log and exp functions on OSX lion

[SiggyF]

The gfortran compiler generates programs that give incorrect results on OSX lion. I have tested the following program with gfortran-mp-4.5 and gfortran-mp-4.7.

program test_exp
  write(*,*) exp(1.0), log(2.7182818284590451)
end program test_exp

This should result in the following output: 2.71... 1.0...

However on OSX Lion with gfortran it results in: 2.7699862 1.3489696

I have tested it with:

gcc45 @4.5.3_4

and

gcc47 @4.7.0_3

See also ​http://stackoverflow.com/questions/10673701/can-i-call-the-fortran-log-function-with-a-number for several related test cases.

[SiggyF]

test program to check for problem with log/exp

[SiggyF]

I rebuild the gfortran47 compiler using port -s gcc47, but it also gives the wrong results.

[tenomoto (Takeshi Enomoto)]

Cc Me!

[SiggyF]

It seems to be related to a broken mpfr.

​http://llvm.org/bugs/show_bug.cgi?id=11111

When I do:

port build mpfr

and then run

make check 

in the build folder I get errors like this:

[tversion] MPFR tuning parameters from src/x86_64/core2/mparam.h
PASS: tversion
PASS: tinternals
PASS: tinits
PASS: tisqrt
PASS: tsgn
PASS: tcheck
PASS: tisnan
PASS: texceptions
PASS: tset_exp
PASS: tset
PASS: mpf_compat
PASS: mpfr_compat
Error for mpfr_ui_pow(a, b, a) for b=1764214285 
a=1.10010010000111111011010101001e1
expected1.00011111011011011011010011100e92, got 1.10000100011010001011111101100e87
FAIL: reuse
PASS: tabs
results differ for x=1.11001100001000101110010110101100001010000000000e-27 prec=55 rnd_mode=MPFR_RNDN
got      1.100100011101111000010011110101010001100111011011001010
expected 1.100100011101111000010011111100011011011011001111000100
approx   0.11001000111011110000100111111000110110110110011110000111111001011E1
FAIL: tacos
Error: mpfr_acosh (1)
FAIL: tacosh
PASS: tadd
PASS: tadd1sp
PASS: tadd_d
PASS: tadd_ui
results differ for x=1.111111000111011100011011111101110100101101000001011110101000010110001000111000011010101001100000100111011100111100010111000000110000010111010010001111110101001101001101101101111110101110100000110e-1
u=1.100000111110010000111001000100100000011110010110000111001111000101010100001011010111100011011011011001100100100100000111111100101101110001000010110111100111001111101001010101111010111110010011110e-2 prec=222 rnd_mode=MPFR_RNDU
got      1.01001100010110111001000011110110001011111100000110100110111010010000010100101111111001111001110101010001011100010110000001100000101100011111011110110111111111000100000011000000010111110010010110010100101111000001011101100e-1
expected 1.01001100010110111001000011110110001011111100000110100110111010010000010100101111111001111001110101010001011100010110000001100000101100011111011110110111111111000100000011000000010111110010010110010101110011110101010110001e-1
approx   0.1010011000101101110010000111101100010111111000001101001101110100100000101001011111110011110011101010100010111000101100000011000001011000111110111101101111111110001000000110000000101111100100101100101011100111101010101100001110001000E0
FAIL: tagm

This appears to be related to #33361.

[SiggyF]

I managed to solve this by changing the compiler in gmp from clang to macports-gcc-4.5. After that the test cases in mpfr all passed and also my fortran test.

[tenomoto (Takeshi Enomoto)]

Has the update of gmp in r94119 fixed the problem? I didn't encounter the problem (maybe because I 'm using Xcode 4.3), I can't tell. BTW, the changes in r94119 is to remove the use of clang when llvm-gcc-4.2 is specified. The default is clang on Lion so I'm not sure if the changes fixed the gfortran problem.

[ryandesign (Ryan Carsten Schmidt)]

r94119 should have fixed it. Assuming it did, since there was no response to the above question.

[SiggyF]

Thanks that fixed it. The output of my test program is now:

   2.7182817      0.99999994