Merge pull request #1 from hasindu2008/gpu-varcall-update

cleaned up GPU consensus calling

.gitignore

@@ -17,7 +17,7 @@ include/
 lib/
 share/
 
-hdf5-1.8.14.tar.gz
+hdf5-1.*.tar.gz
 3.2.5.tar.bz2
 eigen/
 local*

.travis.yml

@@ -1,27 +1,84 @@
 # travis.yml for github.com/jts/nanopolish
 
 dist: trusty
+services: docker
 sudo: false
 language: generic
 cache: apt
 git:
     depth: 1
 
+.nanopolish.ci.matrix-definitions:
+    - &cron-only
+      if: type = cron OR env(CI_CRON) = true
+    - &arch
+      before_install:
+          - |
+            sed -i "/^FROM / s/arm64/${ARCH}/" Dockerfile-arm
+          - |
+            docker run --rm --privileged \
+                multiarch/qemu-user-static:register --reset && \
+                docker build --rm -t nanopolish -f Dockerfile-arm .
+      script:
+          - |
+            docker run --rm -t \
+                -e HDF5="${HDF5:-install}" \
+                -e H5_CFLAGS="${H5_CFLAGS}" \
+                -e HDF5_VERSION="1.10.4" \
+                -e H5_INCLUDE="${H5_INCLUDE}" \
+                -e LDFLAGS="${LDFLAGS}" \
+                nanopolish
+
 matrix:
     include:
-    # Set env for both nanoplish and the dependency hdf5.
-    - env:
-        - CC=gcc-4.8
-        - CXX=g++-4.8
-        - AR=gcc-ar-4.8
-        - NM=gcc-nm-4.8
-        - RANLIB=gcc-ranlib-4.8
-    - env:
-        - CC=gcc-8
-        - CXX=g++-8
-        - AR=gcc-ar-8
-        - NM=gcc-nm-8
-        - RANLIB=gcc-ranlib-8
+        # Set env for both nanoplish and the dependency hdf5.
+        - env:
+              - CC=gcc-4.8
+              - CXX=g++-4.8
+              - AR=gcc-ar-4.8
+              - NM=gcc-nm-4.8
+              - RANLIB=gcc-ranlib-4.8
+        - env:
+              - CC=gcc-8
+              - CXX=g++-8
+              - AR=gcc-ar-8
+              - NM=gcc-nm-8
+              - RANLIB=gcc-ranlib-8
+        # aarch64 - ARM 64-bit
+        - name: aarch64
+          sudo: required
+          env:
+              - ARCH=arm64
+          <<: *arch
+          <<: *cron-only
+        - name: aarch64-system-hdf5
+          sudo: required
+          env:
+              - ARCH=arm64
+              - HDF5="system"
+              - H5_INCLUDE="-I/usr/include/hdf5/serial"
+              - LDFLAGS="-L/usr/lib/aarch64-linux-gnu/hdf5/serial"
+          <<: *arch
+        # armv7l - ARM 32-bit
+        - name: armv7l
+          sudo: required
+          env:
+              - ARCH=armhf
+          <<: *arch
+          <<: *cron-only
+        - name: armv7l-system-hdf5
+          sudo: required
+          env:
+              - ARCH=armhf
+              - HDF5="system"
+              - H5_INCLUDE="-I/usr/include/hdf5/serial"
+              - LDFLAGS="-L/usr/lib/arm-linux-gnueabihf/hdf5/serial"
+          <<: *arch
+    allow_failures:
+        # The jobs installing hdf5 from source in docker finishes with error
+        # because of the job exceeded the maximum time limit (50 minutes).
+        - name: aarch64
+        - name: armv7l
 
 # Install and export newer gcc
 before_install:
@@ -38,9 +95,11 @@ before_install:
           sudo apt-get install -qq "${CXX}"
       fi
 
-script:
+before_script:
     # Suppress all compiler warnings for hdf5 Makefile
     # to display the log without downloading the raw log on Travis log page.
     # Travis finishs with error when exceeding the limit of 4 MB of log length.
     - export H5_CFLAGS="-w"
-    - make nanopolish && make test
+
+script:
+    - make && make test

DEVELOPERS.md

@@ -0,0 +1,10 @@
+# Developer Notes
+
+## Updating Bioconda on tagged releases
+The following is a quick step-by-step checklist on updating the bioconda release for nanopolish, to be done after each tagged release, and is a condensed/updated version of [these slides](https://monashbioinformaticsplatform.github.io/bioconda-tutorial/#/) by Andrew Perry.
+1. On Github, fork `https://github.com/bioconda/bioconda-recipes` to `https://github.com/{USER}/bioconda-recipes` and clone the latter repository to a local directory; `cd` into the cloned directory.
+2. Check out a new branch via `git branch nanopolish-bioconda-bump && git checkout nanopolish-bioconda-bump`.
+3. Update the `bioconda-recipes/recipes/nanopolish/meta.yaml` file by editing the version tag and the SHA hash; the SHA256 hash can be obtained by running `sha256sum nanopolish-v{$VERSION}.tar.gz` on the command line (where `{VERSION}` is the new, updated version tag); commit the changes to the `meta.yaml` file via, e.g., `git commit -a -m 'bump nanopolish to version {VERSION}'`.
+4. Push the changes to your forked repo via `git push origin nanopolish-bioconda-bump`; then, make a pull request to merge the updates into the master branch of the upstream `bioconda-recipes` repository.
+5. If all goes well, the automated TravisCI tests on the upstream repository will pass and an owner will merge the changes.
+6. Otherwise, if further edits are requested or if the TravisCI tests fail, make further commits to the local cloned repository and push to the forked repository on Github; the changes should automatically appear in the pull request and will trigger an automated TravisCI check.

Dockerfile-arm

@@ -0,0 +1,18 @@
+FROM multiarch/ubuntu-debootstrap:arm64-bionic
+RUN uname -a
+RUN apt-get update -qq && \
+  apt-get install -yq --no-install-suggests --no-install-recommends \
+  bzip2 \
+  ca-certificates \
+  gcc \
+  g++ \
+  make \
+  software-properties-common
+RUN add-apt-repository -y universe && \
+  apt-get update -qq && \
+  apt-get install -yq libhdf5-dev
+RUN find /usr/include -name "hdf5.h" || true
+RUN find /usr/lib -name "libhdf5.a" || true
+WORKDIR /nanopolish
+COPY . .
+CMD exec bash -c "make && make test"

Makefile

@@ -1,7 +1,7 @@
 #
 
 # Sub directories containing source code, except for the main programs
-SUBDIRS := src src/hmm src/thirdparty src/thirdparty/scrappie src/common src/alignment src/pore_model src/cuda_kernels
+SUBDIRS := src src/hmm src/thirdparty src/thirdparty/scrappie src/common src/alignment src/pore_model
 
 #
 # Set libraries, paths, flags and options
@@ -10,19 +10,20 @@ SUBDIRS := src src/hmm src/thirdparty src/thirdparty/scrappie src/common src/ali
 #Basic flags every build needs
 LIBS = -lz
 CXXFLAGS ?= -g -O3
-CXXFLAGS += -std=c++11 -fopenmp -fsigned-char
-CFLAGS ?= -std=c99 -O3
+CXXFLAGS += -std=c++11 -fopenmp -fsigned-char -D_FILE_OFFSET_BITS=64 #D_FILE_OFFSET_BITS=64 makes nanopolish work in 32 bit systems
+CFLAGS ?= -O3 -std=c99 -fsigned-char -D_FILE_OFFSET_BITS=64 
+LDFLAGS ?=
 CXX ?= g++
 CC ?= gcc
-NVCC = nvcc
-NVCCFLAGS ?= -std=c++11 -I. -I/usr/local/cuda-9.0include -O3 -use_fast_math --default-stream per-thread -restrict
-CURTFLAGS ?= -L/usr/local/cuda-9.0/lib64 -lcudart
 
 # Change the value of HDF5, EIGEN, or HTS below to any value to disable compilation of bundled code
 HDF5 ?= install
 EIGEN ?= install
 HTS ?= install
 
+HDF5_VERSION ?= 1.8.14
+EIGEN_VERSION ?= 3.2.5
+
 # Check operating system, OSX doesn't have -lrt
 UNAME_S := $(shell uname -s)
 ifeq ($(UNAME_S),Linux)
@@ -37,7 +38,7 @@ ifeq ($(HDF5), install)
 else
     # Use system-wide hdf5
     H5_LIB =
-    H5_INCLUDE =
+    H5_INCLUDE ?=
     LIBS += -lhdf5
 endif
 
@@ -69,94 +70,90 @@ EIGEN_INCLUDE = -I./eigen/
 # Include the src subdirectories
 NP_INCLUDE = $(addprefix -I./, $(SUBDIRS))
 
-CUDA_INCLUDE=-I/usr/local/cuda-9.0/include
-
 # Add include flags
-CPPFLAGS += $(H5_INCLUDE) $(HTS_INCLUDE) $(FAST5_INCLUDE) $(NP_INCLUDE) $(EIGEN_INCLUDE) $(CUDA_INCLUDE)
+CPPFLAGS += $(H5_INCLUDE) $(HTS_INCLUDE) $(FAST5_INCLUDE) $(NP_INCLUDE) $(EIGEN_INCLUDE)
 
 # Main programs to build
 PROGRAM = nanopolish
 TEST_PROGRAM = nanopolish_test
 
 .PHONY: all
-all: $(PROGRAM) $(TEST_PROGRAM)
+all: depend $(PROGRAM)
 
 #
 # Build libhts
 #
 htslib/libhts.a:
-	cd htslib && make || exit 255
+	cd htslib && make htslib_default_libs="-lz -lm -lbz2" || exit 255
 
 #
 # If this library is a dependency the user wants HDF5 to be downloaded and built.
 #
 lib/libhdf5.a:
-	if [ ! -e hdf5-1.8.14.tar.gz ]; then \
-		wget https://support.hdfgroup.org/ftp/HDF5/releases/hdf5-1.8/hdf5-1.8.14/src/hdf5-1.8.14.tar.gz; \
+	if [ ! -e hdf5-$(HDF5_VERSION).tar.gz ]; then \
+		version_major_minor=`echo "$(HDF5_VERSION)" | sed -E 's/\.[0-9]+$$//'`; \
+		wget https://support.hdfgroup.org/ftp/HDF5/releases/hdf5-$${version_major_minor}/hdf5-$(HDF5_VERSION)/src/hdf5-$(HDF5_VERSION).tar.gz; \
 	fi
-	tar -xzf hdf5-1.8.14.tar.gz || exit 255
-	cd hdf5-1.8.14 && \
-		./configure --enable-threadsafe --prefix=`pwd`/.. && \
+
+	tar -xzf hdf5-$(HDF5_VERSION).tar.gz || exit 255
+	cd hdf5-$(HDF5_VERSION) && \
+		./configure --enable-threadsafe --disable-hl --libdir=`pwd`/../lib --includedir=`pwd`/../include --prefix=`pwd`/.. && \
 		make && make install
 
 # Download and install eigen if not already downloaded
 eigen/INSTALL:
-	if [ ! -e 3.2.5.tar.bz2 ]; then \
-		wget http://bitbucket.org/eigen/eigen/get/3.2.5.tar.bz2; \
+	if [ ! -e $(EIGEN_VERSION).tar.bz2 ]; then \
+		wget http://bitbucket.org/eigen/eigen/get/$(EIGEN_VERSION).tar.bz2; \
 	fi
-	tar -xjf 3.2.5.tar.bz2 || exit 255
-	mv eigen-eigen-bdd17ee3b1b3 eigen || exit 255
+	tar -xjf $(EIGEN_VERSION).tar.bz2 || exit 255
+	mv eigen-eigen-* eigen || exit 255
 
 #
 # Source files
 #
 
 # Find the source files by searching subdirectories
 CPP_SRC := $(foreach dir, $(SUBDIRS), $(wildcard $(dir)/*.cpp))
-CU_SRC := $(foreach dir, $(SUBDIRS), $(wildcard $(dir)/*.cu))
 C_SRC := $(foreach dir, $(SUBDIRS), $(wildcard $(dir)/*.c))
 EXE_SRC = src/main/nanopolish.cpp src/test/nanopolish_test.cpp
 
 # Automatically generated object names
-CPP_OBJ=$(CPP_SRC:.cpp=.o)
-C_OBJ=$(C_SRC:.c=.o)
-CU_OBJ=$(CU_SRC:.cu=.o)
+CPP_OBJ = $(CPP_SRC:.cpp=.o)
+C_OBJ = $(C_SRC:.c=.o)
 
-.SUFFIXES: .cu
+ifdef cuda
+include cuda.mk
+endif
 
 # Generate dependencies
 .PHONY: depend
 depend: .depend
 
-.depend: $(CPP_SRC) $(C_SRC) $(CU_SRC) $(EXE_SRC) $(H5_LIB) $(EIGEN_CHECK)
+.depend: $(CPP_SRC) $(C_SRC) $(EXE_SRC) $(H5_LIB) $(EIGEN_CHECK)
 	rm -f ./.depend
 	$(CXX) $(CXXFLAGS) $(CPPFLAGS) -MM $(CPP_SRC) $(C_SRC) > ./.depend;
 
-include .depend
-
 # Compile objects
 .cpp.o:
 	$(CXX) -o $@ -c $(CXXFLAGS) $(CPPFLAGS) -fPIC $<
 
 .c.o:
 	$(CC) -o $@ -c $(CFLAGS) $(CPPFLAGS) $(H5_INCLUDE) -fPIC $<
 
-.cu.o:
-	$(NVCC) -o $@ -c $(NVCCFLAGS) $(CPPFLAGS) $<
-
 # Link main executable
-$(PROGRAM): src/main/nanopolish.o $(CU_OBJ) $(CPP_OBJ) $(C_OBJ) $(HTS_LIB) $(H5_LIB) $(EIGEN_CHECK)
-	$(CXX) -o $@ $(CXXFLAGS) $(CPPFLAGS) -fPIC $< $(CPP_OBJ) $(CU_OBJ) $(C_OBJ) $(HTS_LIB) $(H5_LIB) $(LIBS) $(LDFLAGS) $(CURTFLAGS)
+$(PROGRAM): src/main/nanopolish.o $(CPP_OBJ) $(C_OBJ) $(HTS_LIB) $(H5_LIB) $(EIGEN_CHECK)
+	$(CXX) -o $@ $(CXXFLAGS) $(CPPFLAGS) -fPIC $< $(CPP_OBJ) $(C_OBJ) $(HTS_LIB) $(H5_LIB) $(LIBS) $(LDFLAGS)
 
 # Link test executable
-$(TEST_PROGRAM): src/test/nanopolish_test.o $(CPP_OBJ) $(CU_OBJ) $(C_OBJ) $(HTS_LIB) $(H5_LIB)
-	$(CXX) -o $@ $(CXXFLAGS) $(CPPFLAGS) -fPIC $< $(CPP_OBJ) $(CU_OBJ) $(C_OBJ) $(HTS_LIB) $(H5_LIB) $(LIBS) $(LDFLAGS) $(CURTFLAGS)
+$(TEST_PROGRAM): src/test/nanopolish_test.o $(CPP_OBJ) $(C_OBJ) $(HTS_LIB) $(H5_LIB)
+	$(CXX) -o $@ $(CXXFLAGS) $(CPPFLAGS) -fPIC $< $(CPP_OBJ) $(C_OBJ) $(HTS_LIB) $(H5_LIB) $(LIBS) $(LDFLAGS)
 
 .PHONY: test
 test: $(TEST_PROGRAM)
 	./$(TEST_PROGRAM)
 
 .PHONY: clean
 clean:
-	rm -f $(PROGRAM) $(TEST_PROGRAM) $(CPP_OBJ) $(CU_OBJ) $(C_OBJ) src/main/nanopolish.o src/test/nanopolish_test.o
-		src/main/nanopolish.o src/test/nanopolish_test.o
+	rm -f $(PROGRAM) $(TEST_PROGRAM) $(CPP_OBJ) $(C_OBJ) \
+		src/cuda_kernels/GpuAligner.o \
+		src/main/nanopolish.o src/test/nanopolish_test.o

README.md

@@ -6,6 +6,10 @@ Software package for signal-level analysis of Oxford Nanopore sequencing data. N
 
 ## Release notes
 
+* 0.11.1: `nanopolish polya` now supports SQK-RNA-002 kits with automatic backwards-compatibility with SQK-RNA-001
+
+* 0.11.0: support for multi-fast5 files. `nanopolish methyltrain` now subsamples input data, improving speed and memory usage
+
 * 0.10.2: added new program `nanopolish polya` to estimate the length of poly-A tails on direct RNA reads (by @paultsw)
 
 * 0.10.1: `nanopolish variants --consensus` now only outputs a VCF file instead of a fasta sequence. The VCF file describes the changes that need to be made to turn the draft sequence into the polished assembly. A new program, `nanopolish vcf2fasta`, is provided to generate the polished genome (this replaces `nanopolish_merge.py`, see usage instructions below). This change is to avoid issues when merging segments that end on repeat boundaries (reported by Michael Wykes and Chris Wright).
@@ -86,7 +90,7 @@ samtools index reads.sorted.bam
 Now, we use nanopolish to compute the consensus sequence (the genome is polished in 50kb blocks and there will be one output file per block). We'll run this in parallel:
 
 ```
-python nanopolish_makerange.py draft.fa | parallel --results nanopolish.results -P 8 \
+python3 nanopolish_makerange.py draft.fa | parallel --results nanopolish.results -P 8 \
     nanopolish variants --consensus -o polished.{1}.vcf -w {1} -r reads.fa -b reads.sorted.bam -g draft.fa -t 4 --min-candidate-frequency 0.1
 ```
 
@@ -116,10 +120,11 @@ docker run -v /path/to/local/data/data/:/data/ -it :image_id  ./nanopolish event
 
 ## GPU acceleration
 
-The nanopolish consensus improvement algorithm can be performed faster using CUDA-enabled GPU acceleration. This is an experimental feature, to try this feature run with the `--gpu` flag e.g:
+The nanopolish consensus improvement algorithm can be performed faster using CUDA-enabled GPU acceleration. This is an experimental feature, to try this feature run with the `--gpu=1` flag e.g:
 ```
 nanopolish variants --consensus polished_gpu.fa -w "tig00000001:200000-230000" -r reads.fasta -b reads.sorted.bam -g draft.fa --threads=8 --gpu=1
 ```
+Note that this feature requires nanopolish to be compiled with `make cuda=1`. You should have the [CUDA toolkit installed and configured](https://docs.nvidia.com/cuda/cuda-quick-start-guide/). If your CUDA installation is not in the default location, you can provide the path to make as `make cuda=1 NVCC=/path/to/nvidia_c_compiler CUDA_LIB=/path/to/cuda/lib CUDA_INCLUDE=/path/to/cuda/include`.
 
 ## Credits and Thanks
 

cuda.mk

@@ -0,0 +1,27 @@
+#Make file options for CUDA support
+
+NVCC ?= nvcc
+CUDA_ROOT = /usr/local/cuda
+CUDA_LIB ?= $(CUDA_ROOT)/lib64
+CUDA_INCLUDE ?= $(CUDA_ROOT)/include
+CURTFLAGS = -L$(CUDA_LIB) -lcudart_static -lrt 
+NVCCFLAGS ?= -std=c++11 -I. -I$(CUDA_INCLUDE) -O3 -use_fast_math --default-stream per-thread -restrict
+
+CPPFLAGS += -I$(CUDA_INCLUDE)
+CPPFLAGS += -DHAVE_CUDA=1
+
+# Sub directories containing CUDA source code
+SUBDIRS += src/cuda_kernels
+# Find the source files by searching subdirectories
+CU_SRC := $(foreach dir, $(SUBDIRS), $(wildcard $(dir)/*.cu))
+# Automatically generated object names
+CU_OBJ = $(CU_SRC:.cu=.o)
+CPP_OBJ += $(CU_OBJ)
+LDFLAGS += $(CURTFLAGS)
+
+.SUFFIXES: .cu
+
+# Compile objects
+.cu.o:
+	$(NVCC) -o $@ -c $(NVCCFLAGS) $(CPPFLAGS) $<
+