Fix bug: static link to MKL 11.3.2 failed.

Add dimension check in online-nnet3 decoding code, so we get more mea…

Add missing dependencies to Makefiles

$ ./configure --mkl-root=/opt/intel/mkl --static-math=yes
...
Configuring MKL library directory: Found: /opt/intel/mkl/lib/intel64
MKL configured with threading: sequential, libs:  -Wl,--start-group /opt/intel/mkl/lib/intel64/libmkl_intel_lp64.a /opt/intel/mkl/lib/intel64/libmkl_core.a /opt/intel/mkl/lib/intel64/libmkl_sequential.a -Wl,--end-group
MKL include directory configured as: /opt/intel/mkl/include
Configuring MKL threading as sequential
MKL threading libraries configured as   -lpthread -lm
Using Intel MKL as the linear algebra library.
/opt/intel/mkl/lib/intel64/libmkl_core.a(mkl_memory_patched.o): In function `mkl_serv_set_memory_limit':
mkl_memory.c:(.text+0x49c): undefined reference to `dlsym'
mkl_memory.c:(.text+0x4b2): undefined reference to `dlsym'
mkl_memory.c:(.text+0x4c8): undefined reference to `dlsym'
/opt/intel/mkl/lib/intel64/libmkl_core.a(mkl_memory_patched.o): In function `mkl_serv_allocate':
mkl_memory.c:(.text+0x1251): undefined reference to `dlsym'
mkl_memory.c:(.text+0x1267): undefined reference to `dlsym'
...

added utils/combine_ali_dirs.sh (fixes #553).

some cosmetic changes: add comments to RNNLM rescoring utilities to r…

Speed up CuMatrix<Real>::Transpose() and transposed copy from matrix

smbr: Fixed minor bug in generating diagnostics egs

2 CUDA kernels for TraceMatMat with/without transpose for all matrix size.

New:
LOG (TestCuMatrixTraceMatMat():cu-matrix-speed-test.cc:458) For CuMatrix::TraceMatMat<float>, for dim = 1024, speed was 10.1076 gigaflops.
LOG (TestCuMatrixTraceMatMat():cu-matrix-speed-test.cc:458) For CuMatrix::TraceMatMat<float> [transposed], for dim = 1024, speed was 11.8711 gigaflops.
LOG (TestCuMatrixTraceMatMat():cu-matrix-speed-test.cc:458) For CuMatrix::TraceMatMat<double>, for dim = 1024, speed was 7.10019 gigaflops.
LOG (TestCuMatrixTraceMatMat():cu-matrix-speed-test.cc:458) For CuMatrix::TraceMatMat<double> [transposed], for dim = 1024, speed was 7.81977 gigaflops.

Old:
LOG (TestCuMatrixTraceMatMat():cu-matrix-speed-test.cc:458) For CuMatrix::TraceMatMat<float>, for dim = 1024, speed was 4.57783 gigaflops.
LOG (TestCuMatrixTraceMatMat():cu-matrix-speed-test.cc:458) For CuMatrix::TraceMatMat<float> [transposed], for dim = 1024, speed was 7.96795 gigaflops.
LOG (TestCuMatrixTraceMatMat():cu-matrix-speed-test.cc:458) For CuMatrix::TraceMatMat<double>, for dim = 1024, speed was 3.61182 gigaflops.
LOG (TestCuMatrixTraceMatMat():cu-matrix-speed-test.cc:458) For CuMatrix::TraceMatMat<double> [transposed], for dim = 1024, speed was 6.39571 gigaflops.

add new results for Multi-splice version of online recipe of Librispeech, including those on test set.

LOG (TestCuMatrixTransposeCross():cu-matrix-speed-test.cc:91) For CuMatrix::TransposeCross<float>, for dim = 1024, speed was 14.0498 gigaflops.
LOG (TestCuMatrixTransposeS():cu-matrix-speed-test.cc:72) For CuMatrix::TransposeS<float>, for dim = 1024, speed was 16.845 gigaflops.
LOG (TestCuMatrixTransposeNS():cu-matrix-speed-test.cc:56) For CuMatrix::TransposeNS<float>, for dim = 1024, speed was 14.2464 gigaflops.
LOG (TestCuMatrixTransposeCross():cu-matrix-speed-test.cc:91) For CuMatrix::TransposeCross<double>, for dim = 1024, speed was 10.4523 gigaflops.
LOG (TestCuMatrixTransposeS():cu-matrix-speed-test.cc:72) For CuMatrix::TransposeS<double>, for dim = 1024, speed was 9.65529 gigaflops.
LOG (TestCuMatrixTransposeNS():cu-matrix-speed-test.cc:56) For CuMatrix::TransposeNS<double>, for dim = 1024, speed was 8.52148 gigaflops.

Add barrier for correct timing.

Original performance:
LOG (TestCuMatrixTransposeCross():cu-matrix-speed-test.cc:91) For CuMatrix::TransposeCross<float>, for dim = 1024, speed was 4.26727 gigaflops.
LOG (TestCuMatrixTransposeS():cu-matrix-speed-test.cc:72) For CuMatrix::TransposeS<float>, for dim = 1024, speed was 5.97203 gigaflops.
LOG (TestCuMatrixTransposeNS():cu-matrix-speed-test.cc:56) For CuMatrix::TransposeNS<float>, for dim = 1024, speed was 3.0816 gigaflops.
LOG (TestCuMatrixTransposeCross():cu-matrix-speed-test.cc:91) For CuMatrix::TransposeCross<double>, for dim = 1024, speed was 3.95059 gigaflops.
LOG (TestCuMatrixTransposeS():cu-matrix-speed-test.cc:72) For CuMatrix::TransposeS<double>, for dim = 1024, speed was 4.36189 gigaflops.
LOG (TestCuMatrixTransposeNS():cu-matrix-speed-test.cc:56) For CuMatrix::TransposeNS<double>, for dim = 1024, speed was 2.39275 gigaflops.

base/kaldi_error : the error messages are no longer printed 2x

A new CUDA kernel for CuMatrixBase<Real>::FindRowMaxId;

- the binary can be replaced (so we could eventually append posteriors, features, etc.)

align-equal-compiled.cc: correct the usage description

Old:
LOG (TestCuFindRowMaxId():cu-matrix-speed-test.cc:264) For CuMatrix::FindRowMaxId<float>, for dim = 1024, speed was 3.99218 gigaflops.
LOG (TestCuFindRowMaxId():cu-matrix-speed-test.cc:264) For CuMatrix::FindRowMaxId<double>, for dim = 1024, speed was 3.46283 gigaflops.

New:
LOG (TestCuFindRowMaxId():cu-matrix-speed-test.cc:264) For CuMatrix::FindRowMaxId<float>, for dim = 1024, speed was 66.2965 gigaflops.
LOG (TestCuFindRowMaxId():cu-matrix-speed-test.cc:264) For CuMatrix::FindRowMaxId<double>, for dim = 1024, speed was 58.442 gigaflops.

Update nnet1-to-raw-nnet.cc

Update perturb_data_dir_volume.sh

Add seed for random number generator in utils/data/perturb_data_dir.sh

Now GetParams requires vector allocated already. Changes in nnet\nnet-various.h at Apr 21, 2016

improve speed of split_data.sh; includes change to filter_scps.pl (thanks to remi francis for noticing the issue)