bug in PostProc, calculating ELF with parallel processing

When using PP to calculate the ELF of a Si crystal in the diamond phase, the output depends on the number of processors used.

With 1 processor, the output is smooth (images created with XCrysDen, isosurface with elf=0.1):

1

but with 32 processors the output is "noisy":

32

The pp.x input file is the same in both cases:

9013102.vc-relax.PP_in

The input data is also the same in both cases, calculated with PWscf using wf_collect=.true. :

9013102.vc-relax.pw_in

Quantum Espresso 6.2.1, compiled with gfortran 4.4.7, Intel MKL Libraries v 2018.1.163 and OpenMPI 3.0.0.