Merge branch 'dev-v0.6.0' into poa-gfa

.gitmodules

@@ -33,3 +33,6 @@
 [submodule "3rdparty/kseqpp"]
 	path = 3rdparty/kseqpp
 	url = https://github.com/cartoonist/kseqpp.git
+[submodule "3rdparty/htslib"]
+	path = 3rdparty/htslib
+	url = https://github.com/samtools/htslib.git

CMakeLists.txt

@@ -23,8 +23,8 @@ string(STRIP ${GW_VERSION} GW_VERSION)
 project(${GW_PROJECT_NAME})
 
 # Process options.
-option(gw_enable_tests "Build GenomeWorks unit tests" OFF)
-option(gw_enable_benchmarks "Build GenomeWorks benchmarks" OFF)
+option(gw_enable_tests "Build GenomeWorks unit tests" ON)
+option(gw_enable_benchmarks "Build GenomeWorks benchmarks" ON)
 option(gw_build_shared "Build GenomeWorks libraries as shared objects" OFF)
 option(gw_device_synchronize_kernels "Run cudaDeviceSynchronize() in GW_CU_CHECK_ERR calls" OFF)
 option(gw_optimize_for_native_cpu "Build with march=native" ON)
@@ -40,7 +40,9 @@ option(gw_enable_cudapoa_nw_print "Enable verbose prints within cudapoa NW kerne
 option(gw_profiling "Compile a binary for profiling with NVTX markers." OFF)
 option(gw_enable_caching_allocator "Enable caching allocator." ON)
 option(gw_generate_docs "Generate Doxygen documentation" ON)
-option(gw_cuda_gen_all_arch "ON: Generate optimized CUDA code for all architectures | OFF: for detected architectures only" ON)
+option(gw_cuda_gen_all_arch "ON: Generate optimized CUDA code for all architectures | OFF: for detected architectures only" OFF)
+# Optionally build htslib for SAM/BAM support. Requires autoconf to be installed
+option(gw_build_htslib "Build 3rdparty htslib that allows output in SAM/BAM format" ON)
 
 # Must be included before others for options value validation
 include(cmake/Utils.cmake)
@@ -83,6 +85,7 @@ add_subdirectory(common/io)
 add_subdirectory(cudapoa)
 add_subdirectory(cudamapper)
 add_subdirectory(cudaaligner)
+add_subdirectory(cudaextender)
 
 # Add documentation generation.
 validate_boolean(gw_generate_docs)

README.md

@@ -6,81 +6,17 @@ GenomeWorks is a GPU-accelerated library for biological sequence analysis. This
 For more detailed API documentation please refer to the [documentation](#enable-doc-generation).
 
 * Modules
-    * [cudamapper](#cudamapper) - CUDA-accelerated sequence to sequence mapping
-    * [cudapoa](#cudapoa) - CUDA-accelerated partial order alignment
-    * [cudaaligner](#cudaaligner) - CUDA-accelerated pairwise sequence alignment
+    * [cudamapper](cudamapper/README.md) - CUDA-accelerated sequence to sequence mapping
+    * [cudapoa](cudapoa/README.md) - CUDA-accelerated partial order alignment
+    * [cudaaligner](cudaaligner/README.md) - CUDA-accelerated pairwise sequence alignment
+    * [cudaextender](cudaextender/README.md) - CUDA-accelerated seed extension
 * Setup GenomeWorks
     * [Clone GenomeWorks](#clone-genomeworks)
     * [System Requirements](#system-requirements)
     * [GenomeWorks Installation](#genomeworks-setup)
 * [Python API](#genomeworks-python-api)
 * [Development Support](#development-support)
 
-### cudamapper
-
-The `cudamapper` package provides minimizer-based GPU-accelerated approximate mapping.
-
-#### Tool - *cudamapper*
-
-`cudamapper` is an end-to-end command line to for sequence to sequence mapping. `cudamapper` outputs
-mappings in the PAF format and is currently optimised for all-vs-all long read (ONT, Pacific Biosciences) sequences.
-
-To run all-vs all overlaps use the following command:
-
-`cudamapper in.fasta in.fasta`
-
-A query fasta can be mapped to a reference as follows:
-
-`cudamapper query.fasta target.fasta`
-
-To access more information about running cudamapper, run `cudamapper --help`.
-
-#### Library - *libcudamapper.so*
-
-* `Indexer` module to generate an index of minimizers from a list of sequences.
-* `Matcher` module to find locations of matching pairs of minimizers between sequences using minimizer indices.
-* `Overlapper` module to generate overlaps from sequence of minimizer matches generated by matcher.
-
-#### Sample - *sample_cudamapper*
-
-A prototypical binary highlighting the usage of `libcudamapper.so` APIs (indexer, matcher and overlapper) and
-techniques to tie them into an application.
-
-### cudapoa
-
-The `cudapoa` package provides a GPU-accelerated implementation of the [Partial Order Alignment](https://simpsonlab.github.io/2015/05/01/understanding-poa/)
-algorithm. It is heavily influenced by [SPOA](https://github.com/rvaser/spoa) and in many cases can be considered a GPU-accelerated replacement. Features include:
-
-#### Tool - *cudapoa*
-
-A command line tool for generating consensus and MSA from a list of `fasta`/`fastq` files. The tool
-is built on top of `libcudapoa.so` and showcases optimization strategies for writing high performance
-applications with `libcudapoa.so`.
-
-#### Library - *libcudapoa.so*
-
-* Generation of consensus sequences
-* Generation of multi-sequence alignments (MSAs)
-* Custom adaptive band implementation of POA
-* Support for long and short read sequences
-
-#### Sample - *sample_cudapoa*
-
-A prototypical binary to showcase the use of `libcudapoa.so` APIs.
-
-### cudaaligner
-
-The `cudaaligner` package provides GPU-accelerated global alignment. Features include:
-
-#### Library - *libcudaaligner.so*
-
-* Short and long read support
-* Banded implementation with configurable band width for flexible performance and accuracy trade-off
-
-#### Sample - *sample_cudaaligner*
-
-A prototypical binary to showcase the use of `libcudaaligner.so` APIs.
-
 ## Clone GenomeWorks 
 
 ### Latest released version
@@ -104,9 +40,12 @@ Minimum requirements -
 
 1. Ubuntu 16.04 or Ubuntu 18.04
 2. CUDA 9.0+ (official instructions for installing CUDA are available [here](https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html))
-3. gcc/g++ 5.4.0+
-4. Python 3.6.7+
-5. CMake (>= 3.0)
+3. GPU generation Pascal and later (compute capability >= 6.0)
+4. gcc/g++ 5.4.0+ / 7.x.x
+5. Python 3.6.7+
+6. CMake (>= 3.10.2)
+7. autoconf (required to output SAM/BAM files)
+8. automake (required to output SAM/BAM files)
 
 ## GenomeWorks Setup
 
@@ -124,6 +63,8 @@ NOTE : The `gw_cuda_gen_all_arch=OFF` option pre-generates optimized code only f
 For building a binary that pre-generates opimized code for all common GPU architectures, please remove the option
 or set it to `ON`.
 
+NOTE : (OPTIONAL) To enable outputting overlaps in SAM/BAM format, pass the `gw_build_htslib=ON` option.
+
 ### Package generation
 Package generation puts the libraries, headers and binaries built by the `make` command above
 into a `.deb`/`.rpm` for portability and easy installation. The package generation itself doesn't

VERSION

@@ -1 +1,2 @@
 0.6.0
+

ci/checks/check_copyright.py

@@ -145,7 +145,7 @@ def copyright_present(f):
     f - Path to file
     """
     with io.open(f, "r", encoding="utf-8") as fh:
-        return re.search('Copyright 20[0-9]+-20[0-9]+', fh.read())
+        return re.search('Copyright (20[0-9][0-9]-)?20[0-9][0-9] NVIDIA CORPORATION', fh.read())
 
 
 def parse_args():

ci/common/build-test-sdk.sh

@@ -31,6 +31,8 @@ cmake .. "${CMAKE_COMMON_VARIABLES[@]}" \
     -Dgw_enable_tests=ON \
     -Dgw_enable_benchmarks=ON \
     -Dgw_build_shared=ON \
+    -Dgw_cuda_gen_all_arch=ON \
+    -Dgw_build_htslib=ON \
     -DCMAKE_INSTALL_PREFIX="${LOCAL_BUILD_DIR}/install" \
     -GNinja
 

ci/common/prep-init-env.sh

@@ -53,8 +53,19 @@ python --version
 logger "Conda install GenomeWorks custom packages - clang-format"
 conda install --override-channels -c sarcasm clang-format
 
-logger "Conda install GenomeWorks custom packages - doxygen ninja cmake"
-conda install --override-channels -c conda-forge doxygen ninja cmake">=3.10.2"
+logger "Conda install GenomeWorks custom packages - doxygen ninja"
+conda install --override-channels -c conda-forge doxygen ninja
+
+# Building cudamapper, using CUDA_SELECT_NVCC_ARCH_FLAGS() with the 'common' argument, generates
+# the "-gencode;arch=compute_72,code=sm_72" arch flag which is incompatible with CUDA 9.0 and causes nvcc command to fail
+# Using cmake with a version older than 3.18 does not create 'compute_72' arch.
+if [ "$(echo ${CUDA_VERSION} | cut -d"." -f1-2)" = "9.0" ]; then
+  logger "Conda install cmake for CUDA 9.0"
+  conda install --override-channels -c conda-forge cmake"=3.17"
+else
+  logger "Conda install cmake"
+  conda install --override-channels -c conda-forge cmake">=3.10.2"
+fi
 
 logger "Update LD_LIBRARY_PATH"
 export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib

ci/cpu/prebuild.sh

@@ -22,4 +22,4 @@
 # Note we still _BUILD_ for GPU, we just don't (can't) test on it
 export BUILD_FOR_GPU=1
 export TEST_ON_GPU=0
-export CONDA_ENV_NAME="gdf"
+export CONDA_ENV_NAME="parabricks"

ci/gpu/prebuild.sh

@@ -21,4 +21,4 @@
 
 export BUILD_FOR_GPU=1
 export TEST_ON_GPU=1
-export CONDA_ENV_NAME="gdf"
+export CONDA_ENV_NAME="parabricks"

cmake/3rdparty.cmake

@@ -50,3 +50,10 @@ set_property(TARGET cub APPEND PROPERTY INTERFACE_INCLUDE_DIRECTORIES "${CUB_DIR
 
 set(KSEQPP_DIR ${PROJECT_SOURCE_DIR}/3rdparty/kseqpp/src CACHE STRING
     "Path to kseqpp repo")
+
+if (gw_build_htslib)
+    include(cmake/BuildHTSLib.cmake)
+    build_htslib_source()
+else()
+    message(STATUS "Not building htslib, overlap output to SAM & BAM unavailable. Enable with -Dgw_build_htslib=ON")
+endif()

cmake/BuildHTSLib.cmake

@@ -0,0 +1,42 @@
+#
+# Copyright 2019-2020 NVIDIA CORPORATION.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+function(build_htslib_source)
+    message(STATUS "Building htslib")
+    set(HTSLIB_DIR ${PROJECT_SOURCE_DIR}/3rdparty/htslib/ CACHE STRING
+        "Path to htslib repo")
+    set(MAKE_COMMAND make)
+    set(HTSLIB_INSTALL ${MAKE_COMMAND} install prefix=${CMAKE_BINARY_DIR}/3rdparty/htslib)
+    set(htslib_PREFIX ${CMAKE_BINARY_DIR}/3rdparty/htslib)
+    include(ExternalProject)
+    ExternalProject_Add(htslib_project
+            PREFIX ${htslib_PREFIX}
+            SOURCE_DIR ${PROJECT_SOURCE_DIR}/3rdparty/htslib
+            BUILD_IN_SOURCE 1
+            CONFIGURE_COMMAND autoheader && autoconf && ${HTSLIB_DIR}configure --disable-bz2 --disable-lzma --disable-libcurl --disable-s3 --disable-gcs
+            BUILD_COMMAND "${HTSLIB_INSTALL}"
+            INSTALL_COMMAND ""
+            BUILD_BYPRODUCTS ${CMAKE_BINARY_DIR}/3rdparty/htslib/lib/libhts.a
+            LOG_CONFIGURE 0
+            LOG_BUILD 0
+            LOG_TEST 0
+            LOG_INSTALL 0
+            )
+
+    include_directories(${CMAKE_BINARY_DIR}/3rdparty/htslib/include/htslib)
+    add_library(htslib STATIC IMPORTED)
+    set_property(TARGET htslib APPEND PROPERTY IMPORTED_LOCATION ${CMAKE_BINARY_DIR}/3rdparty/htslib/lib/libhts.a)
+endfunction()

common/base/CMakeLists.txt

@@ -26,7 +26,9 @@ message(STATUS "nvcc flags for ${MODULE_NAME}: ${CUDA_NVCC_FLAGS}")
 get_property(gw_library_type GLOBAL PROPERTY gw_library_type)
 add_library(${MODULE_NAME} ${gw_library_type}
         src/cudautils.cpp
-        src/logging.cpp)
+        src/logging.cpp
+        src/graph.cpp
+        )
 target_link_libraries(${MODULE_NAME} PUBLIC spdlog ${CUDA_LIBRARIES})
 
 if (gw_profiling)

common/base/include/claraparabricks/genomeworks/logging/logging.hpp

@@ -95,6 +95,10 @@
 // Due to a header file incompatibility with nvcc in CUDA 9.0
 // logging through the logger class in GW is disabled for any .cu files.
 #pragma message("Logging disabled for CUDA Toolkit < 9.2")
+#elif __GNUC__ >= 9
+// Due to a ISO C++ standard incompatibility the spdlog fails to pass
+// pedantic requirements.
+#pragma message("Logging disabled for GCC >= 9")
 #else
 #include <spdlog/spdlog.h>
 #endif
@@ -137,6 +141,8 @@ LoggingStatus SetHeader(bool logTime, bool logLocation);
 /// parameters as per https://github.com/gabime/spdlog/blob/v1.x/README.md
 #ifdef GW_CUDA_BEFORE_9_2
 #define GW_LOG_DEBUG(...)
+#elif __GNUC__ >= 9
+#define GW_LOG_DEBUG(...)
 #else
 #define GW_LOG_DEBUG(...) SPDLOG_DEBUG(__VA_ARGS__)
 #endif
@@ -148,6 +154,8 @@ LoggingStatus SetHeader(bool logTime, bool logLocation);
 /// parameters as per https://github.com/gabime/spdlog/blob/v1.x/README.md
 #ifdef GW_CUDA_BEFORE_9_2
 #define GW_LOG_INFO(...)
+#elif __GNUC__ >= 9
+#define GW_LOG_INFO(...)
 #else
 #define GW_LOG_INFO(...) SPDLOG_INFO(__VA_ARGS__)
 #endif
@@ -159,6 +167,8 @@ LoggingStatus SetHeader(bool logTime, bool logLocation);
 /// parameters as per https://github.com/gabime/spdlog/blob/v1.x/README.md
 #ifdef GW_CUDA_BEFORE_9_2
 #define GW_LOG_WARN(...)
+#elif __GNUC__ >= 9
+#define GW_LOG_WARN(...)
 #else
 #define GW_LOG_WARN(...) SPDLOG_WARN(__VA_ARGS__)
 #endif
@@ -170,6 +180,8 @@ LoggingStatus SetHeader(bool logTime, bool logLocation);
 /// parameters as per https://github.com/gabime/spdlog/blob/v1.x/README.md
 #ifdef GW_CUDA_BEFORE_9_2
 #define GW_LOG_ERROR(...)
+#elif __GNUC__ >= 9
+#define GW_LOG_ERROR(...)
 #else
 #define GW_LOG_ERROR(...) SPDLOG_ERROR(__VA_ARGS__)
 #endif
@@ -181,6 +193,8 @@ LoggingStatus SetHeader(bool logTime, bool logLocation);
 /// parameters as per https://github.com/gabime/spdlog/blob/v1.x/README.md
 #ifdef GW_CUDA_BEFORE_9_2
 #define GW_LOG_CRITICAL(...)
+#elif __GNUC__ >= 9
+#define GW_LOG_CRITICAL(...)
 #else
 #define GW_LOG_CRITICAL(...) SPDLOG_CRITICAL(__VA_ARGS__)
 #endif

common/base/include/claraparabricks/genomeworks/types.hpp

@@ -21,6 +21,7 @@
 #if __cplusplus >= 201703
 #include <optional>
 #include <string_view>
+#include <cstddef>
 #else
 #include <experimental/optional>
 #include <experimental/string_view>
@@ -45,21 +46,23 @@ using position_in_read_t = std::uint32_t;
 using number_of_basepairs_t = position_in_read_t;
 
 // TODO: Once minimal supported GCC version is moved to GCC 7.1
-// or higher, thegw_optional_t and gw_string_view_t aliases
-// can be removed and std::optional and std::string_view can
+// or higher, gw_optional_t, gw_string_view_t and gw_byte_t aliases
+// can be removed and std::optional, std::string_view and std::byte can
 // be used directly instead
 #if __cplusplus >= 201703
 template <typename T>
 using gw_optional_t               = std::optional<T>;
 using gw_nullopt_t                = std::nullopt_t;
 constexpr gw_nullopt_t gw_nullopt = std::nullopt;
 using gw_string_view_t            = std::string_view;
+using gw_byte_t                   = std::byte;
 #else
 template <typename T>
 using gw_optional_t               = std::experimental::optional<T>;
 using gw_nullopt_t                = std::experimental::nullopt_t;
 constexpr gw_nullopt_t gw_nullopt = std::experimental::nullopt;
 using gw_string_view_t            = std::experimental::string_view;
+using gw_byte_t                   = unsigned char;
 #endif
 
 } // namespace genomeworks