save workspace

2 files changed, 20 insertions, 13 deletions

diff --git a/kaldi_decode/src/Makefile b/kaldi_decode/src/Makefile
index 118420c..0897798 100644
--- a/kaldi_decode/src/Makefile
+++ b/kaldi_decode/src/Makefile
@@ -1,11 +1,11 @@
 # Change KDIR to `kaldi-trunk' path (Kaldi must be compiled with --share)
-KDIR := /slfs6/users/ymz09/kaldi/
-NERVDIR := /slfs6/users/ymz09/nerv-project/nerv/
+KDIR := /speechlab/tools/KALDI/kaldi-master/
+NERVDIR := /speechlab/users/mfy43/nerv/
 CUDADIR := /usr/local/cuda/
 
 nnet-forward:
 	g++ -msse -msse2 -Wall -I $(KDIR)/src/ -pthread -DKALDI_DOUBLEPRECISION=0 -DHAVE_POSIX_MEMALIGN -Wno-sign-compare -Wno-unused-local-typedefs -Winit-self -DHAVE_EXECINFO_H=1 -rdynamic -DHAVE_CXXABI_H -DHAVE_ATLAS -I $(KDIR)/tools/ATLAS/include -I $(KDIR)/tools/openfst/include -Wno-sign-compare -g  -fPIC -DHAVE_CUDA -I $(CUDADIR)/include  -DKALDI_NO_EXPF  -I $(NERVDIR)/install//include/luajit-2.0/ -I $(NERVDIR)/install/include/ -DLUA_USE_APICHECK -c -o nnet-forward.o nnet-forward.cc
-	g++ -rdynamic -Wl,-rpath=$(KDIR)/tools/openfst/lib -L$(CUDADIR)/lib64 -Wl,-rpath,$(CUDADIR)/lib64 -Wl,-rpath=$(KDIR)/src/lib -L.  -L$(KDIR)/src/nnet/   -L$(KDIR)/src/cudamatrix/   -L$(KDIR)/src/lat/   -L$(KDIR)/src/hmm/   -L$(KDIR)/src/tree/   -L$(KDIR)/src/matrix/   -L$(KDIR)/src/util/   -L$(KDIR)/src/base/ nnet-forward.o $(KDIR)/src/nnet//libkaldi-nnet.so $(KDIR)/src/cudamatrix//libkaldi-cudamatrix.so $(KDIR)/src/lat//libkaldi-lat.so $(KDIR)/src/hmm//libkaldi-hmm.so $(KDIR)/src/tree//libkaldi-tree.so $(KDIR)/src/matrix//libkaldi-matrix.so $(KDIR)/src/util//libkaldi-util.so $(KDIR)/src/base//libkaldi-base.so   -L$(KDIR)/tools/openfst/lib -lfst /usr/lib/liblapack.so /usr/lib/libcblas.so /usr/lib/libatlas.so /usr/lib/libf77blas.so -lm -lpthread -ldl -lcublas -lcudart   -lkaldi-nnet   -lkaldi-cudamatrix   -lkaldi-lat   -lkaldi-hmm   -lkaldi-tree   -lkaldi-matrix   -lkaldi-util   -lkaldi-base -lstdc++ -L$(NERVDIR)/install/lib -lnervcore -lluaT -rdynamic -Wl,-rpath=$(KDIR)//tools/openfst/lib -L$(DUDADIR)/lib64 -Wl,-rpath,$(CUDADIR)/lib64 -Wl,-rpath=$(KDIR)//src/lib -lfst -lm -lpthread -ldl -lcublas -lcudart -L $(NERVDIR)/luajit-2.0/src/ -lluajit -o nnet-forward
+	g++ -rdynamic -Wl,-rpath=$(KDIR)/tools/openfst/lib -L$(CUDADIR)/lib64 -Wl,-rpath,$(CUDADIR)/lib64 -Wl,-rpath=$(KDIR)/src/lib -L.  -L$(KDIR)/src/nnet/   -L$(KDIR)/src/cudamatrix/   -L$(KDIR)/src/lat/   -L$(KDIR)/src/hmm/   -L$(KDIR)/src/tree/   -L$(KDIR)/src/matrix/   -L$(KDIR)/src/util/   -L$(KDIR)/src/base/ nnet-forward.o $(KDIR)/src/nnet//libkaldi-nnet.so $(KDIR)/src/cudamatrix//libkaldi-cudamatrix.so $(KDIR)/src/lat//libkaldi-lat.so $(KDIR)/src/hmm//libkaldi-hmm.so $(KDIR)/src/tree//libkaldi-tree.so $(KDIR)/src/matrix//libkaldi-matrix.so $(KDIR)/src/util//libkaldi-util.so $(KDIR)/src/base//libkaldi-base.so   -L$(KDIR)/tools/openfst/lib -lfst -lm -lpthread -ldl   -lkaldi-nnet   -lkaldi-cudamatrix   -lkaldi-lat   -lkaldi-hmm   -lkaldi-tree   -lkaldi-matrix   -lkaldi-util   -lkaldi-base -lstdc++ -L$(NERVDIR)/install/lib -Wl,-rpath=$(NERVDIR)/install/lib -lnervcore -lluaT -rdynamic -Wl,-rpath=$(KDIR)//tools/openfst/lib -L$(DUDADIR)/lib64 -Wl,-rpath,$(CUDADIR)/lib64 -Wl,-rpath=$(KDIR)//src/lib -lfst -lm -lpthread -ldl -L $(NERVDIR)/luajit-2.0/src/ -lluajit -o nnet-forward -L/home/intel/mkl/lib/intel64/ -Wl,-rpath=/home/intel/mkl/lib/intel64/ -lmkl_rt
 
 clean:
 	-rm nnet-forward.o nnet-forward
diff --git a/kaldi_decode/src/nerv4decode.lua b/kaldi_decode/src/nerv4decode.lua
index b2ff344..898b5a8 100644
--- a/kaldi_decode/src/nerv4decode.lua
+++ b/kaldi_decode/src/nerv4decode.lua
@@ -1,7 +1,19 @@
-package.path="/home/slhome/ymz09/.luarocks/share/lua/5.1/?.lua;/home/slhome/ymz09/.luarocks/share/lua/5.1/?/init.lua;/slfs6/users/ymz09/nerv-project/nerv/install/share/lua/5.1/?.lua;/slfs6/users/ymz09/nerv-project/nerv/install/share/lua/5.1/?/init.lua;"..package.path; 
-package.cpath="/home/slhome/ymz09/.luarocks/lib/lua/5.1/?.so;/slfs6/users/ymz09/nerv-project/nerv/install/lib/lua/5.1/?.so;"..package.cpath;
+print = function(...) io.write(table.concat({...}, "\t")) end
+io.output('/dev/null')
+-- path and cpath are correctly set by `path.sh`
 local k,l,_=pcall(require,"luarocks.loader") _=k and l.add_context("nerv","scm-1")
 require 'nerv'
+nerv.printf("*** NERV: A Lua-based toolkit for high-performance deep learning (alpha) ***\n")
+nerv.info("automatically initialize a default MContext...")
+nerv.MMatrix._default_context = nerv.MContext()
+nerv.info("the default MContext is ok")
+-- only for backward compatibilty, will be removed in the future
+local function _add_profile_method(cls)
+    local c = cls._default_context
+    cls.print_profile = function () c:print_profile() end
+    cls.clear_profile = function () c:clear_profile() end
+end
+_add_profile_method(nerv.MMatrix)
 
 function build_trainer(ifname, feature)
     local param_repo = nerv.ParamRepo()
@@ -34,9 +46,7 @@ function build_trainer(ifname, feature)
             end
             local transformed
             if e.global_transf then
-                local batch = gconf.cumat_type(data[id]:nrow(), data[id]:ncol())
-                batch:copy_fromh(data[id])
-                transformed = nerv.speech_utils.global_transf(batch,
+                transformed = nerv.speech_utils.global_transf(data[id],
                 global_transf,
                 gconf.frm_ext or 0, 0,
                 gconf)
@@ -45,7 +55,7 @@ function build_trainer(ifname, feature)
             end
             table.insert(input, transformed)
         end
-        local output = {nerv.CuMatrixFloat(input[1]:nrow(), network.dim_out[1])}
+        local output = {nerv.MMatrixFloat(input[1]:nrow(), network.dim_out[1])}
         network:batch_resize(input[1]:nrow())
         network:propagate(input, output)
         
@@ -65,12 +75,9 @@ function build_trainer(ifname, feature)
 end
 
 function init(config, feature)
-    local tmp = io.write
-    io.write = function(...)
-    end
     dofile(config)
+    gconf.use_cpu = true -- use CPU to decode
     trainer = build_trainer(gconf.decode_param, feature)
-    io.write = tmp
 end
 
 function feed()