diff --git a/egs/wsj/s5/utils/data/get_uniform_subsegments.py b/egs/wsj/s5/utils/data/get_uniform_subsegments.py
index a963f35fa70..dfaa55b18f7 100755
--- a/egs/wsj/s5/utils/data/get_uniform_subsegments.py
+++ b/egs/wsj/s5/utils/data/get_uniform_subsegments.py
@@ -13,11 +13,22 @@
 def get_args():
     parser = argparse.ArgumentParser(
         description=textwrap.dedent("""
-        Creates a subsegments file from an input segments file
-        that has the format
+        Creates a subsegments file from an input segments file.
+
+        The output format is
+
         <subsegment-id> <utterance-id> <start-time> <end-time>,
-        where the timing are relative to the start-time of the
-        <utterance-id> in the input segments file.
+
+        where the timings are relative to the start-time of the
+         <utterance-id> in the input segments file.
+        Reminder: the format of the input segments file is:
+
+         <utterance-id> <recording-id> <start-time> <end-time>
+
+        where the recording-id corresponds to a wav file (or a channel of
+        a wav file) from wav.scp.  Note: you can use
+        utils/data/get_segments_for_data.sh to generate a 'default'
+        segments file for your data if one doesn't already exist.
 
         e.g.: get_uniform_subsegments.py data/dev/segments > \\
                 data/dev_uniform_segments/sub_segments
diff --git a/src/nnet3bin/Makefile b/src/nnet3bin/Makefile
index 67d15d3c38a..544109df832 100644
--- a/src/nnet3bin/Makefile
+++ b/src/nnet3bin/Makefile
@@ -19,7 +19,8 @@ BINFILES = nnet3-init nnet3-info nnet3-get-egs nnet3-copy-egs nnet3-subset-egs \
    nnet3-discriminative-subset-egs nnet3-get-egs-simple \
    nnet3-discriminative-compute-from-egs nnet3-latgen-faster-looped \
    nnet3-egs-augment-image nnet3-xvector-get-egs nnet3-xvector-compute \
-   nnet3-latgen-grammar nnet3-compute-batch nnet3-latgen-faster-batch
+   nnet3-latgen-grammar nnet3-compute-batch nnet3-latgen-faster-batch \
+   cuda-gpu-available
 
 OBJFILES =
 
@@ -33,6 +34,6 @@ ADDLIBS = ../nnet3/kaldi-nnet3.a ../chain/kaldi-chain.a \
           ../lat/kaldi-lat.a ../fstext/kaldi-fstext.a ../hmm/kaldi-hmm.a \
           ../transform/kaldi-transform.a ../gmm/kaldi-gmm.a \
           ../tree/kaldi-tree.a ../util/kaldi-util.a ../matrix/kaldi-matrix.a \
-          ../base/kaldi-base.a 
+          ../base/kaldi-base.a
 
 include ../makefiles/default_rules.mk
diff --git a/src/nnetbin/cuda-gpu-available.cc b/src/nnet3bin/cuda-gpu-available.cc
similarity index 100%
rename from src/nnetbin/cuda-gpu-available.cc
rename to src/nnet3bin/cuda-gpu-available.cc
diff --git a/src/nnetbin/Makefile b/src/nnetbin/Makefile
index 86d59ae503e..5e934ff523a 100644
--- a/src/nnetbin/Makefile
+++ b/src/nnetbin/Makefile
@@ -1,5 +1,9 @@
 
+# cuda-gpu-available was moved to nnet3bin/; remove it in case an outdated
+# version persists.
 all:
+	-rm -f cuda-gpu-available
+
 EXTRA_CXXFLAGS = -Wno-sign-compare
 include ../kaldi.mk
 
@@ -15,7 +19,7 @@ BINFILES = nnet-train-frmshuff \
         nnet-forward nnet-copy nnet-info nnet-concat \
         transf-to-nnet cmvn-to-nnet nnet-initialize \
 	feat-to-post paste-post train-transitions \
-	cuda-gpu-available nnet-set-learnrate
+	nnet-set-learnrate
 
 OBJFILES =
 
@@ -25,6 +29,6 @@ TESTFILES =
 
 ADDLIBS = ../nnet/kaldi-nnet.a ../cudamatrix/kaldi-cudamatrix.a \
           ../lat/kaldi-lat.a ../hmm/kaldi-hmm.a ../tree/kaldi-tree.a \
-          ../util/kaldi-util.a ../matrix/kaldi-matrix.a ../base/kaldi-base.a 
+          ../util/kaldi-util.a ../matrix/kaldi-matrix.a ../base/kaldi-base.a
 
 include ../makefiles/default_rules.mk
diff --git a/tools/Makefile b/tools/Makefile
index e690df3da88..b3a9d88a8ac 100644
--- a/tools/Makefile
+++ b/tools/Makefile
@@ -138,13 +138,9 @@ cub:
 	rm -f cub
 	ln -s cub-$(CUB_VERSION) cub
 
-# OpenBLAS is not compiled by default. Run 'make -j openblas' in this directory to build.
+# OpenBLAS is not compiled by default; you can run `make openblas` to build it,
+# but you should probably just call extras/install_openblas.sh; this is
+# here for legacy reasons.
 .PHONY: openblas
 openblas:
-	@-rm -rf OpenBLAS xianyi-OpenBLAS-*
-	wget -t3 -nv -O- $$( \
-            wget -qO- 'https://api.github.com/repos/xianyi/OpenBLAS/releases/tags/v$(OPENBLAS_VERSION)' | \
-            python -c 'import sys,json;print(json.load(sys.stdin)["tarball_url"])') | \
-	  tar xzf -
-	mv xianyi-OpenBLAS-* OpenBLAS
-	$(MAKE) PREFIX=$$(pwd)/OpenBLAS/install USE_THREAD=0 -C OpenBLAS all install
+	extras/install_openblas.sh
diff --git a/tools/extras/install_openblas.sh b/tools/extras/install_openblas.sh
index 90afe8e9de4..224a9eade69 100755
--- a/tools/extras/install_openblas.sh
+++ b/tools/extras/install_openblas.sh
@@ -1,5 +1,25 @@
 #!/bin/bash
 
-# OpenBLAS is downloaded and built by tools/Makefile, but not automatically by
-# its default 'all' target.
-make -j openblas
+
+OPENBLAS_VERSION=0.3.5
+
+set -e
+
+if ! command -v gfortran 2>/dev/null; then
+  echo "$0: gfortran is not installed.  Please install it, e.g. by:"
+  echo " apt-get install gfortran"
+  echo "(if on Debian or Ubuntu), or:"
+  echo " yum install fortran"
+  echo "(if on RedHat/CentOS).  On a Mac, if brew is installed, it's:"
+  echo " brew install gfortran"
+  exit 1
+fi
+
+
+rm -rf xianyi-OpenBLAS-* OpenBLAS
+
+wget -t3 -nv -O- $(wget -qO- "https://api.github.com/repos/xianyi/OpenBLAS/releases/tags/v${OPENBLAS_VERSION}" | python -c 'import sys,json;print(json.load(sys.stdin)["tarball_url"])') | tar xzf -
+
+mv xianyi-OpenBLAS-* OpenBLAS
+
+make PREFIX=$(pwd)/OpenBLAS/install USE_THREAD=0 -C OpenBLAS all install