f77 legendre transform error

[Former user]

Hi,

Thanks for adding the legendre transform to the f77 API.

I'm having trouble getting it to work in your fortran example however. Compiling and running the code gives a segmentation fualt and core dump whenever the new shtns_sh_to_spat_ml routine is called. Any suggestions?

[Nathanaël Schaeffer]

I'm sorry, but it works here. Could you please edit the Makefile and add -g after -O2 on the CFLAGS= line ? (should be line 5) Also add -g on line 91 after gfortran.

Then recompile the fortran example "make SHT_fort_ex" and run it in the debugger

gdb ./SHT_fort_ex

and then type "run". When it crashes, type

backtrace

and send me the output.

Thank you.

[Former user]

I've run the program in the debugger and get the following message

Program received signal SIGSEGV, Segmentation fault.

0x0000000000000000 in ?? ()

(gdb) backtrace

#0 0x0000000000000000 in ?? ()

#1 0x00000000004022a8 in sht_example () at SHT_example.f:105

#2 0x000000000048ee2a in main ()

#3 0x0000003ebe01ed5d in __libc_start_main (main=0x48ee00 <main>, argc=1, ubp_av=0x7fffffffe368, init=<value optimized out>, fini=<value optimized out>, rtld_fini=<value optimized out>, stack_end=0x7fffffffe358) at libc-start.c:226

#4 0x0000000000401619 in _start ()

Hope this helps

[Nathanaël Schaeffer]

Mmm, unfortunately this does not help. Can you also post the output of the program, before the arrays of numbers ?

Also you could try reconfiguring with the --disable-simd option and see if the problem is still there.

[Former user]

The output of the program is as follows:

NLM= 6 [SHTns 2.6.3] built May 29 2015, 11:38:27, id: avx Lmax=5, Mmax*Mres=4, Mres=2, Nlm=12 [1 threads, orthonormalized] => using FFTW : Mmax=2, Nphi=6, Nlat=8 (data layout : phi_inc=1, theta_inc=6, phi_embed=6) => using Gauss nodes + polar optimization threshold = 1.0e-10 + SHT accuracy = 1.44e-15 => SHTns is ready. [SHTns 2.6.3] built May 29 2015, 11:38:27, id: avx Lmax=5, Mmax*Mres=4, Mres=2, Nlm=12 [1 threads, orthonormalized] Gauss grid : Nlat=8, Nphi=6 syn ana vsy van gsp gto v3s v3a std: mem mem mem mem mem mem mem mem ltr: mem mem mem mem mem mem mem mem m: none none none none none none none none

The -disable-simd option makes no difference.

[Nathanaël Schaeffer]

In the file SHT_example.f, could you please put nlat = 32 in line 41 ? I think it should work then... Sorry, too small grids seem to be problematic right now. I will try to fix this soon, but for now, please don't use too small grids. Maybe nlat = 16 is enough.

[Nathanaël Schaeffer]

This bug affects all Legendre transforms on small grids (Nlat < 8, 16 or 32 depending on the system), whether called from Fortran or from C.

Workaround: use large enough grids if you need a pure Legendre transform.

[Former user]

Thanks for your quick reply and workaround.

There seems to be a similar bug concerning nphi, for which the routine doesn't work on my machine when nphi<4.

The shtns_spat_to_sh_ml routine also gives a segmentation fault when using regular grids and DCT acceleration.

Debugging in gdb gives the output

[SHTns 2.6.3] built May 29 2015, 11:38:27, id: avx

Lmax=100, Mmax*Mres=1, Mres=1, Nlm=201 [1 threads, orthonormalized] Regular grid (mtr_dct=1) : Nlat=256, Nphi=4

syn ana vsy van gsp gto v3s v3a

std: dct dct dct dct dct dct s+v dct

ltr: dct dct dct dct dct dct s+v dct

m: fly4 fly2 fly2 fly2 fly2 fly2 s+v fly2

Program received signal SIGSEGV, Segmentation fault.

spat_to_SH_m_fly2_l (shtns=<value optimized out>, im=2117866500, Vr=0x697650, Qlm=0x6fe010, llim=100) at SHT/spat_to_SH_fly.c:467

467 y0[j] *= vread(wg, k+j); y1[j] *= vread(wg, k+j); // weight appears here (must be after the previous accuracy loop).

(gdb) backtrace

#0 spat_to_SH_m_fly2_l (shtns=<value optimized out>, im=2117866500, Vr=0x697650, Qlm=0x6fe010, llim=100) at SHT/spat_to_SH_fly.c:467

#1 0x0000000000401ef3 in sht_example () at SHT_example.f:116

#2 0x000000000048e92a in main ()

#3 0x0000003ebe01ed5d in __libc_start_main (main=0x48e900 <main>, argc=1, ubp_av=0x7fffffffe368, init=<value optimized out>, fini=<value optimized out>, rtld_fini=<value optimized out>, stack_end=0x7fffffffe358) at libc-start.c:226

#4 0x0000000000401559 in _start ()