/usr/include/dieharder/diehard_sums.h is in libdieharder-dev 3.31.1-7build1.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 | /*
* diehard_sums test header.
*/
/*
* function prototype
*/
int diehard_sums(Test **test,int irun);
static Dtest diehard_sums_dtest __attribute__((unused)) = {
"Diehard Sums Test",
"diehard_sums",
"\
#==================================================================\n\
# Diehard Sums Test\n\
# Integers are floated to get a sequence U(1),U(2),... of uni- \n\
# form [0,1) variables. Then overlapping sums, \n\
# S(1)=U(1)+...+U(100), S2=U(2)+...+U(101),... are formed. \n\
# The S's are virtually normal with a certain covariance mat- \n\
# rix. A linear transformation of the S's converts them to a \n\
# sequence of independent standard normals, which are converted \n\
# to uniform variables for a KSTEST. The p-values from ten \n\
# KSTESTs are given still another KSTEST. \n\
#\n\
# Comments\n\
#\n\
# At this point I think there is rock solid evidence that this test\n\
# is completely useless in every sense of the word. It is broken,\n\
# and it is so broken that there is no point in trying to fix it.\n\
# The problem is that the transformation above is not linear, and\n\
# doesn't work. Don't use it.\n\
#\n\
# For what it is worth, rgb_lagged_sums with ntuple 0 tests for\n\
# exactly the same thing, but scalably and reliably without the\n\
# complication of overlapping samples and covariance. Use it\n\
# instead.\n\
#==================================================================\n",
100,
100,
1,
diehard_sums,
0
};
|