Fix to the reorder_addlibs.sh script (was not handling library names with numbers in them correctly) and the corresponding fix to Makefiles.

bench result:
CuMatrix::DiffGroupPnorm<float>,    16   0.019   0.009  2.11x
CuMatrix::DiffGroupPnorm<float>,    32   0.074   0.036  2.06x
CuMatrix::DiffGroupPnorm<float>,    64   0.297   0.142  2.10x
CuMatrix::DiffGroupPnorm<float>,   128   1.142   0.520  2.20x
CuMatrix::DiffGroupPnorm<float>,   256   3.442   1.553  2.22x
CuMatrix::DiffGroupPnorm<float>,   512   6.856   2.943  2.33x
CuMatrix::DiffGroupPnorm<float>,  1024  11.653   3.915  2.98x
CuMatrix::DiffGroupPnorm<float>,  2048  13.812   4.263  3.24x
CuMatrix::DiffGroupPnorm<float>,  4096  14.431   4.381  3.29x
CuMatrix::DiffGroupPnorm<double>,    16   0.019   0.009  2.17x
CuMatrix::DiffGroupPnorm<double>,    32   0.073   0.033  2.20x
CuMatrix::DiffGroupPnorm<double>,    64   0.296   0.133  2.22x
CuMatrix::DiffGroupPnorm<double>,   128   1.068   0.457  2.34x
CuMatrix::DiffGroupPnorm<double>,   256   2.999   1.159  2.59x
CuMatrix::DiffGroupPnorm<double>,   512   4.921   1.705  2.89x
CuMatrix::DiffGroupPnorm<double>,  1024   6.932   1.993  3.48x
CuMatrix::DiffGroupPnorm<double>,  2048   7.499   2.087  3.59x
CuMatrix::DiffGroupPnorm<double>,  4096   7.684   2.104  3.65x

fix bug

unit test for diff group pnorm

easy test for now

back to full test

fix p=inf for MatrixBase::GrouPnormDeriv

standard inf

del TODO

fix bug

Add script to automatically put the Kaldi libraries we link with in the right order; use it to modify the Makefiles.  Minor top-level Makefile fix.

Some minor changes: replace '*1.0e-03' with '/1000.0f' for more consistent rounding; minor documentation and cosmetic changes

Shrink in-value of ClipGradientComponent toward some smaller value when clipping proportion exceeds some threshold (#803)

 (also minor bug fix in profiling in cu-vector.cc)

New: For CuMatrix::Softmax<float>, for dim = 16, speed was 0.0153621 gigaflops.
Old: For CuMatrix::Softmax<float>, for dim = 16, speed was 0.0138999 gigaflops.
New: For CuMatrix::Softmax<float>, for dim = 32, speed was 0.0614275 gigaflops.
Old: For CuMatrix::Softmax<float>, for dim = 32, speed was 0.0507328 gigaflops.
New: For CuMatrix::Softmax<float>, for dim = 64, speed was 0.235765 gigaflops.
Old: For CuMatrix::Softmax<float>, for dim = 64, speed was 0.203548 gigaflops.
New: For CuMatrix::Softmax<float>, for dim = 128, speed was 0.729239 gigaflops.
Old: For CuMatrix::Softmax<float>, for dim = 128, speed was 0.725481 gigaflops.
New: For CuMatrix::Softmax<float>, for dim = 256, speed was 2.30126 gigaflops.
Old: For CuMatrix::Softmax<float>, for dim = 256, speed was 1.71863 gigaflops.
New: For CuMatrix::Softmax<float>, for dim = 512, speed was 5.0565 gigaflops.
Old: For CuMatrix::Softmax<float>, for dim = 512, speed was 3.69659 gigaflops.
New: For CuMatrix::Softmax<float>, for dim = 1024, speed was 10.2482 gigaflops.
Old: For CuMatrix::Softmax<float>, for dim = 1024, speed was 6.38335 gigaflops.
New: For CuMatrix::Softmax<double>, for dim = 16, speed was 0.0143354 gigaflops.
Old: For CuMatrix::Softmax<double>, for dim = 16, speed was 0.013143 gigaflops.
New: For CuMatrix::Softmax<double>, for dim = 32, speed was 0.0590478 gigaflops.
Old: For CuMatrix::Softmax<double>, for dim = 32, speed was 0.0495458 gigaflops.
New: For CuMatrix::Softmax<double>, for dim = 64, speed was 0.228611 gigaflops.
Old: For CuMatrix::Softmax<double>, for dim = 64, speed was 0.193465 gigaflops.
New: For CuMatrix::Softmax<double>, for dim = 128, speed was 0.668961 gigaflops.
Old: For CuMatrix::Softmax<double>, for dim = 128, speed was 0.676449 gigaflops.
New: For CuMatrix::Softmax<double>, for dim = 256, speed was 2.1013 gigaflops.
Old: For CuMatrix::Softmax<double>, for dim = 256, speed was 1.51862 gigaflops.
New: For CuMatrix::Softmax<double>, for dim = 512, speed was 4.13055 gigaflops.
Old: For CuMatrix::Softmax<double>, for dim = 512, speed was 3.1547 gigaflops.
New: For CuMatrix::Softmax<double>, for dim = 1024, speed was 6.43429 gigaflops.
Old: For CuMatrix::Softmax<double>, for dim = 1024, speed was 5.02974 gigaflops.

minor changes

Fix issue with 'explicit' constructors that prevented inclusion in stl vector, plus various cosmetic and documentation changes

- fix a bug (unknown feature) which occurs when “—endpoint=true” with
use of fbank feature

resize the result matrix before back-propagation

New: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 16, speed was 0.0165568 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 16, speed was 0.00355242 gigaflops.
New: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 32, speed was 0.0678791 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 32, speed was 0.0145515 gigaflops.
New: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 64, speed was 0.24739 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 64, speed was 0.0583246 gigaflops.
New: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 128, speed was 0.898427 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 128, speed was 0.225076 gigaflops.
New: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 256, speed was 2.89009 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 256, speed was 0.834096 gigaflops.
New: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 512, speed was 6.72164 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 512, speed was 1.92722 gigaflops.
New: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 1024, speed was 10.4916 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<float>, for dim = 1024, speed was 2.78281 gigaflops.
New: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 16, speed was 0.0148584 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 16, speed was 0.00260567 gigaflops.
New: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 32, speed was 0.0586865 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 32, speed was 0.0121077 gigaflops.
New: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 64, speed was 0.22893 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 64, speed was 0.0527767 gigaflops.
New: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 128, speed was 0.763462 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 128, speed was 0.175736 gigaflops.
New: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 256, speed was 2.40457 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 256, speed was 0.58351 gigaflops.
New: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 512, speed was 4.55165 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 512, speed was 1.42464 gigaflops.
New: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 1024, speed was 4.36421 gigaflops.
Old: For CuMatrix::DiffSoftmaxPerRow<double>, for dim = 1024, speed was 1.94971 gigaflops.

Modify error message and documentation to suggest nvidia-smi -c 3 (process exclusive mode) as thread exclusive mode is deprecated now

fix bug in determinize-star-inl.h that affected goodness of hashing function (thanks: EnochLiou); fix problem in lattice-oracle error message found by Nicolas Serrano.

 * Fixed snip_edges=false behaviour to have an expected feature delay of frame_shift/2 instead of being identical to the delay of snip_edges=true. This makes the behaviour of snip_edges=false consistent with other feature extractions tools.
 * Added a unit test that ensure snip_edges=false behaves as expected, by calculating the actual delay between outputs only differing by snip_edge through cross-correlation.
 * Refactored ExtractFrame to treat both edge cases symmetrically
 * Calculating window size / shift by dividing by 1000 instead of multiplying by 0.001 to fix unexpected behaviour with non integer window size / shift.
 * Made sure all changes pass google code validation
 * Changed all feature window functions to use precalculated factor: in my opinion it make the functions easier to compare, and it may be a bit faster.

Note that the feature delay of snip_edges=true (the default) is rather difficult to predict (and almost certainly not what people assume), which almost certainly makes the use of pitch features problematic at the moment.
I would suggest switching to snip_edges=false for new recipes.

Add some diagnostic scripts to analyze initial/final silence lengths and related things; call them from alignment and decoding scripts.

For compatbility with HTK USEPOWER option have added --use-power option (default[=true] is same as current code) to filterbank generation code. To use add --use-power=[false|true] to feature config file.

    stronger unit test

New: For CuMatrix::GroupPnorm<float>, for dim = 16, speed was 0.014416 gigaflops.
Old: For CuMatrix::GroupPnorm<float>, for dim = 16, speed was 0.0138561 gigaflops.
New: For CuMatrix::GroupPnorm<float>, for dim = 32, speed was 0.0616648 gigaflops.
Old: For CuMatrix::GroupPnorm<float>, for dim = 32, speed was 0.0542906 gigaflops.
New: For CuMatrix::GroupPnorm<float>, for dim = 64, speed was 0.241291 gigaflops.
Old: For CuMatrix::GroupPnorm<float>, for dim = 64, speed was 0.213442 gigaflops.
New: For CuMatrix::GroupPnorm<float>, for dim = 128, speed was 0.869675 gigaflops.
Old: For CuMatrix::GroupPnorm<float>, for dim = 128, speed was 0.821949 gigaflops.
New: For CuMatrix::GroupPnorm<float>, for dim = 256, speed was 3.07193 gigaflops.
Old: For CuMatrix::GroupPnorm<float>, for dim = 256, speed was 2.90466 gigaflops.
New: For CuMatrix::GroupPnorm<float>, for dim = 512, speed was 8.8404 gigaflops.
Old: For CuMatrix::GroupPnorm<float>, for dim = 512, speed was 6.48644 gigaflops.
New: For CuMatrix::GroupPnorm<float>, for dim = 1024, speed was 16.7489 gigaflops.
Old: For CuMatrix::GroupPnorm<float>, for dim = 1024, speed was 9.3791 gigaflops.
New: For CuMatrix::GroupPnorm<double>, for dim = 16, speed was 0.0159731 gigaflops.
Old: For CuMatrix::GroupPnorm<double>, for dim = 16, speed was 0.0101083 gigaflops.
New: For CuMatrix::GroupPnorm<double>, for dim = 32, speed was 0.0605624 gigaflops.
Old: For CuMatrix::GroupPnorm<double>, for dim = 32, speed was 0.0393037 gigaflops.
New: For CuMatrix::GroupPnorm<double>, for dim = 64, speed was 0.249944 gigaflops.
Old: For CuMatrix::GroupPnorm<double>, for dim = 64, speed was 0.153672 gigaflops.
New: For CuMatrix::GroupPnorm<double>, for dim = 128, speed was 0.840825 gigaflops.
Old: For CuMatrix::GroupPnorm<double>, for dim = 128, speed was 0.598191 gigaflops.
New: For CuMatrix::GroupPnorm<double>, for dim = 256, speed was 3.13722 gigaflops.
Old: For CuMatrix::GroupPnorm<double>, for dim = 256, speed was 1.78274 gigaflops.
New: For CuMatrix::GroupPnorm<double>, for dim = 512, speed was 6.86864 gigaflops.
Old: For CuMatrix::GroupPnorm<double>, for dim = 512, speed was 2.96384 gigaflops.
New: For CuMatrix::GroupPnorm<double>, for dim = 1024, speed was 12.5614 gigaflops.
Old: For CuMatrix::GroupPnorm<double>, for dim = 1024, speed was 3.79237 gigaflops.

    loop unroll by template

    generalize to group transform reduce

    _transform_reduce for vec, mat-col and group

    fix min bug

    fix bug

    fix template param bug

Good performance on large group sizes (>10).
New: For CuMatrix::GroupMax<float>, for dim = 16, speed was 0.0190836 gigaflops.
Old: For CuMatrix::GroupMax<float>, for dim = 16, speed was 0.0193129 gigaflops.
New: For CuMatrix::GroupMax<float>, for dim = 32, speed was 0.0791846 gigaflops.
Old: For CuMatrix::GroupMax<float>, for dim = 32, speed was 0.0768508 gigaflops.
New: For CuMatrix::GroupMax<float>, for dim = 64, speed was 0.311131 gigaflops.
Old: For CuMatrix::GroupMax<float>, for dim = 64, speed was 0.299519 gigaflops.
New: For CuMatrix::GroupMax<float>, for dim = 128, speed was 1.13589 gigaflops.
Old: For CuMatrix::GroupMax<float>, for dim = 128, speed was 1.14847 gigaflops.
New: For CuMatrix::GroupMax<float>, for dim = 256, speed was 4.22264 gigaflops.
Old: For CuMatrix::GroupMax<float>, for dim = 256, speed was 3.92072 gigaflops.
New: For CuMatrix::GroupMax<float>, for dim = 512, speed was 12.2629 gigaflops.
Old: For CuMatrix::GroupMax<float>, for dim = 512, speed was 10.0812 gigaflops.
New: For CuMatrix::GroupMax<float>, for dim = 1024, speed was 21.6979 gigaflops.
Old: For CuMatrix::GroupMax<float>, for dim = 1024, speed was 16.5123 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 16, speed was 0.0188551 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 16, speed was 0.0163827 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 32, speed was 0.0701613 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 32, speed was 0.0620238 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 64, speed was 0.271106 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 64, speed was 0.215268 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 128, speed was 0.931745 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 128, speed was 0.723582 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 256, speed was 3.53189 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 256, speed was 1.9751 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 512, speed was 9.95109 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 512, speed was 3.91183 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 1024, speed was 17.2099 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<float>, for dim = 1024, speed was 4.92671 gigaflops.
New: For CuMatrix::GroupMax<double>, for dim = 16, speed was 0.0199497 gigaflops.
Old: For CuMatrix::GroupMax<double>, for dim = 16, speed was 0.0148693 gigaflops.
New: For CuMatrix::GroupMax<double>, for dim = 32, speed was 0.079538 gigaflops.
Old: For CuMatrix::GroupMax<double>, for dim = 32, speed was 0.0718237 gigaflops.
New: For CuMatrix::GroupMax<double>, for dim = 64, speed was 0.314509 gigaflops.
Old: For CuMatrix::GroupMax<double>, for dim = 64, speed was 0.237838 gigaflops.
New: For CuMatrix::GroupMax<double>, for dim = 128, speed was 1.08104 gigaflops.
Old: For CuMatrix::GroupMax<double>, for dim = 128, speed was 0.788395 gigaflops.
New: For CuMatrix::GroupMax<double>, for dim = 256, speed was 3.7741 gigaflops.
Old: For CuMatrix::GroupMax<double>, for dim = 256, speed was 2.87856 gigaflops.
New: For CuMatrix::GroupMax<double>, for dim = 512, speed was 8.65988 gigaflops.
Old: For CuMatrix::GroupMax<double>, for dim = 512, speed was 5.87111 gigaflops.
New: For CuMatrix::GroupMax<double>, for dim = 1024, speed was 14.0373 gigaflops.
Old: For CuMatrix::GroupMax<double>, for dim = 1024, speed was 8.88655 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 16, speed was 0.0174585 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 16, speed was 0.0136057 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 32, speed was 0.0694617 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 32, speed was 0.0500527 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 64, speed was 0.265809 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 64, speed was 0.177945 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 128, speed was 0.973417 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 128, speed was 0.588654 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 256, speed was 3.43166 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 256, speed was 1.57864 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 512, speed was 8.26032 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 512, speed was 3.14173 gigaflops.
New: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 1024, speed was 12.1338 gigaflops.
Old: For CuMatrix::GroupMax (all group sizes)<double>, for dim = 1024, speed was 3.05406 gigaflops.

fix typo; rename and comment

fix space