[tree-ssa] Miscompilation (SIGSEGV) with -ffast-math (!?)

Diego Novillo dnovillo@redhat.com
Mon Feb 23 15:20:00 GMT 2004


On Mon, 2004-02-23 at 08:34, Richard Guenther wrote:

> Any ideas though?
> 
Can't seem to reproduce it:

$ ~/perf/sbox/tree-ssa-branch/local.i686/inst.tobiano/bin/g++ --version
g++ (GCC) 3.5-tree-ssa 20040223 (merged 20040211)

$ ~/perf/sbox/tree-ssa-branch/local.i686/inst.tobiano/bin/g++ -O2 -ffast-math tramp3d-v2.cpp -o tramp3d-v2
tramp3d-v2.cpp:15627: warning: `leafify' attribute directive ignored
tramp3d-v2.cpp:15636: warning: `leafify' attribute directive ignored
tramp3d-v2.cpp:27426: warning: `leafify' attribute directive ignored
tramp3d-v2.cpp:31114: warning: `leafify' attribute directive ignored
tramp3d-v2.cpp:31390: warning: `leafify' attribute directive ignored
tramp3d-v2.cpp:40288: warning: `leafify' attribute directive ignored
tramp3d-v2.cpp:40296: warning: `leafify' attribute directive ignored
tramp3d-v2.cpp:41213: warning: `leafify' attribute directive ignored
tramp3d-v2.cpp:41399: warning: `leafify' attribute directive ignored

$ ./tramp3d-v2 -n 10
Using
  initial state from tramp3d.in
  dump file every 2147483647 to tramp3d.out
  using [1,1,1] block setup for computation on domain [0:63:1,0:63:1,0:63:1]
  solving eeq
  not honouring gravitational potential
  time increments from [0, 1.79769e+308]
  starting at t = 0, i = 1
  rh physical/total domain [0:62:1,0:62:1,0:62:1], [-2:64:1,-2:64:1,-2:64:1]
  v  physical/total domain [1:61:1,1:61:1,1:61:1], [-1:63:1,-1:63:1,-1:63:1]
  pg physical/total domain [0:62:1,0:62:1,0:62:1], [-2:64:1,-2:64:1,-2:64:1]
  T  physical/total domain [0:62:1,0:62:1,0:62:1], [-2:64:1,-2:64:1,-2:64:1]
  flm physical/total domain [1:61:1,1:61:1,1:61:1], [-1:63:1,-1:63:1,-1:63:1]
  flvv physical/total domain [1:62:1,1:62:1,1:62:1], [-1:64:1,-1:64:1,-1:64:1]
  flvcX physical/total domain [0:62:1,1:61:1,1:61:1], [-2:64:1,1:61:1,1:61:1]
  flvcY physical/total domain [1:61:1,0:62:1,1:61:1], [1:61:1,-2:64:1,1:61:1]
  flvcZ physical/total domain [1:61:1,1:61:1,0:62:1], [1:61:1,1:61:1,-2:64:1]
  fle physical/total domain [1:61:1,1:61:1,1:61:1], [-1:63:1,-1:63:1,-1:63:1]
i = 1    t = 0.00407921  dt = 0.00407921 (0.184337s/it)
i = 2    t = 0.00679089  dt = 0.00271169 (6.6041s/it)
i = 3    t = 0.00937923  dt = 0.00258833 (6.62765s/it)
i = 4    t = 0.0120634   dt = 0.0026842 (6.59487s/it)
i = 5    t = 0.0150383   dt = 0.00297491 (6.62053s/it)
i = 6    t = 0.0183952   dt = 0.00335687 (6.61296s/it)
i = 7    t = 0.0217122   dt = 0.00331702 (6.61103s/it)
i = 8    t = 0.0250681   dt = 0.00335591 (6.60463s/it)
i = 9    t = 0.0284961   dt = 0.00342798 (6.61474s/it)
i = 10   t = 0.032036    dt = 0.00353984 (6.59785s/it)
Correctness:
        sum(rh) difference = 0
        sum(vx) = 4.45691e-15
        sum(vy) = -3.01812e-15
        sum(vz) = 1.12288e-16
        sum(T) difference = 2094.52



More information about the Gcc mailing list