Smoothness Parameter Ignored in Test Script

[ben-e-whitney]

The test script src/mgard_test.c does not use the s parameter. Two illustrations:

1. The output does not vary with s. Try running the following script. It is possible, but very unlikely, that this is due to the data.

#!/bin/bash

S_VALUES="-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 inf"

for s in $S_VALUES
do
	./mgard_test data/zort_111_160_15.{dat,mgard.$s} 111 160 15 1e-3 $s
done

for s in $S_VALUES
do
	diff --report-identical-files data/zort_111_160_15.mgard.{0.0,$s}
done

2. As far as I can tell, s is used this way:

   1. main in mgard_test.c passes s to mgard_compress.

   2. mgard_compress passes s to mgard::refactor_qz.

   3. mgard::refactor_qz doesn't use s! To see this, compile with -Wunused-parameter:

      $ g++ -std=c++11 -c  -Wall -Wfatal-errors -Wunused-parameter -Iinclude  -Iblosc/include  -O3 -fPIC   -c src/mgard.cpp -o obj/mgard.o 2>&1 | grep --after-context 1 "unused parameter ‘s’"
      src/mgard.cpp:45:94: warning: unused parameter ‘s’ [-Wunused-parameter]
       refactor_qz (int nrow, int ncol, int nfib, const double *u, int &outsize, double tol, double s)
      --
      src/mgard.cpp:110:97: warning: unused parameter ‘s’ [-Wunused-parameter]
      	 double* recompose_udq(int nrow, int ncol, int nfib, unsigned char *data, int data_len, double s)
      

Additionally, the script improperly compares the tolerance tol (which, if I understand correctly, is on the relative error in the s norm) with the relative error in the L^∞ norm. Hopefully as a result, it reports a failure with this dataset (run after applying the commits in #8).

$ ./mgard_test data.{bin,mgard} 64 64 64 0.1 0
compress_memory_blosc: 1048584
compressor = zstd ;csize = 1946
compress_memory_blosc done: 137 milliseconds

In size:     2097152  Out size:       1946  Compression ratio:       1077 
Rel. L^2 error tolerance: 1.00000E-01 
Rel. L^infty error: 1.61896E-01 
FAILURE: Error tolerance NOT met!

[tugluk]

Currently MGARD defaults to L^∞ compression. This will be changed in the "release" due in 3 weeks. I will check the unmet tolerance for the given data.

Thanks for the informative bug report.

[tugluk]

Unmet tolerance issue is corrected, the compression was being performed using absolute tolerances. Many thanks.