Improved debug stride mode

Fix bugs with random stride mode

Add random stride mode for MallocPitch

Reformat CUDA kernel code to strictly follow the 80-char-per-line rule

aspire : added blstm+chain recipe and blstm+xent recipe

         2. added chain recipes for both BLSTM and TDNN models with lesser number of parameters, for comparison
         3. added blstm+xent recipe

Remove unused pooling-related options from example scripts and chain-…

Replacing the default lmwt ranges in scoring scripts [both local and …

Replacing the default lmwt ranges in scoring scripts [both local and global] with more reasonable values [chain systems need values <10, and no system really needs lmwt=20].

WIP: combine_data.sh: combine data with and without segments

Make slurm.pl more resilient to random squeue failures

Fix scripts about srilm in vystadial example

get_segments_for_data.sh: uses get_utt2dur.sh to set
segment end durations, writes to stdout

combine_segments.sh: Handles missing segments files
as with utt2uniq

Fix bug in log-softmax kernel [RE matrices with different strides]

Fix bug regarding newlines in strings, introduced in commit d90d3545

the env.sh is in tool/extras

fix minor error in python script for recent ver. numpy

make sure index to be integer