Matrix factorization

.gitignore

@@ -179,4 +179,4 @@ src/interface_py/build_info.txt
 
 # Temp requirement file generated for Conda
 src/interface_py/requirements_conda.txt
-open_data/
+open_data/*

CMakeLists.txt

@@ -10,7 +10,7 @@ FIND_PACKAGE(BLAS REQUIRED)
 FIND_PACKAGE(SWIG REQUIRED)
 FIND_PACKAGE(PythonLibs REQUIRED) # SWIG
 
-# SWIG behaviour was changed 
+# SWIG behaviour was changed
 if (${CMAKE_VERSION} VERSION_GREATER "3.13")
         cmake_policy(SET CMP0078 OLD)
 endif()
@@ -20,6 +20,7 @@ INCLUDE(${SWIG_USE_FILE})
 #============= OPTIONS & SETTINGS
 OPTION(USE_CUDA  "Build with GPU acceleration" ON)
 OPTION(DEV_BUILD  "Dev build" OFF)
+OPTION(DEV_SYNC  "call cudaDeviceSynchronize after each cuda call" OFF)
 
 # Compiler flags
 SET(CMAKE_CXX_STANDARD 11)
@@ -108,12 +109,16 @@ if(USE_CUDA)
                 MESSAGE(STATUS "Building DEVELOPER compute capability version.")
                 SET(GPU_COMPUTE_VER 61)
                 SET(CMAKE_BUILD_TYPE Debug)
-                # add_compile_definitions(SYNC)
         else()
                 MESSAGE(STATUS "Building RELEASE compute capability version.")
                 SET(GPU_COMPUTE_VER 35;50;52;60;61)
         endif()
 
+        if(DEV_SYNC)
+            MESSAGE(STATUS "Building with device sync.")
+            add_compile_definitions(SYNC)
+        endif()
+
         if(((CUDA_VERSION_MAJOR EQUAL 9) OR (CUDA_VERSION_MAJOR GREATER 9)) AND NOT DEV_BUILD)
                 MESSAGE(STATUS "CUDA GREATER OR EQUAL THAN 9.0 detected, adding Volta compute capability (7.0).")
                 SET(GPU_COMPUTE_VER "${GPU_COMPUTE_VER};70")
@@ -127,7 +132,7 @@ if(USE_CUDA)
         SET(GENCODE_FLAGS "")
         FORMAT_GENCODE_FLAGS("${GPU_COMPUTE_VER}" GENCODE_FLAGS)
         MESSAGE(STATUS "cuda architecture flags: ${GENCODE_FLAGS}")
-        SET(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS};-Xcompiler -fPIC; -std=c++11;--expt-extended-lambda;--expt-relaxed-constexpr;${GENCODE_FLAGS};-lineinfo; -w;")
+        SET(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS};-Xcompiler -fPIC; -g; -std=c++11;--expt-extended-lambda;--expt-relaxed-constexpr;${GENCODE_FLAGS};-lineinfo; -w;")
 
         FILE(GLOB_RECURSE GPU_SOURCES
                 src/*.cu
@@ -144,7 +149,7 @@ if(USE_CUDA)
                 SET(NVTX_LIBRARY nvToolsExt)
         endif()
 
-        find_library(CUBLAS_STATIC_LIB NAMES libcublas_static.a 
+        find_library(CUBLAS_STATIC_LIB NAMES libcublas_static.a
                 PATHS   "${CUDA_TOOLKIT_ROOT_DIR}/lib64" # cuda-9
                         "${CUDA_TOOLKIT_ROOT_DIR}/targets/x86_64-linux/lib/") # cuda-10
         find_library(CULIBOS_STATIC_LIB NAMES libculibos.a
@@ -156,7 +161,7 @@ if(USE_CUDA)
         find_library(CUSOLVER_STATIC_LIB NAMES libcusolver_static.a
                 PATHS   "${CUDA_TOOLKIT_ROOT_DIR}/lib64"
                         "${CUDA_TOOLKIT_ROOT_DIR}/targets/x86_64-linux/lib/")
-        
+
 
         TARGET_LINK_LIBRARIES(gpuh2o4gpu
                 ${CUBLAS_STATIC_LIB}
@@ -176,7 +181,7 @@ if(USE_CUDA)
         else()
                 SWIG_ADD_LIBRARY(ch2o4gpu_gpu LANGUAGE python SOURCES src/swig/ch2o4gpu_gpu.i)
         endif()
-        SWIG_LINK_LIBRARIES(ch2o4gpu_gpu gpuh2o4gpu 
+        SWIG_LINK_LIBRARIES(ch2o4gpu_gpu gpuh2o4gpu
                 ${PYTHON_LIBRARIES}
                 ${CUBLAS_STATIC_LIB}
                 ${CULIBOS_STATIC_LIB}

DEVEL.md

@@ -52,9 +52,9 @@ rm -rf /opt/boost
 mkdir -p /opt/boost
 export CUDA_HOME=/usr/local/cuda/
 apt-get install -y opencl-headers libicu-dev bzip2 libbz2-dev zlib1g-dev python-dev && \
-wget http://sourceforge.net/projects/boost/files/boost/1.61.0/boost_1_61_0.tar.bz2 && \
-tar xjf boost_1_61_0.tar.bz2 && \
-cd boost_1_61_0 && \
+wget https://s3.amazonaws.com/0xdata-public/boost/boost_1_69_0.tar.bz2 && \
+tar xjf boost_1_69_0.tar.bz2 && \
+cd boost_1_69_0 && \
 export PYTHON_PREFIX=`python3-config --prefix` && \
 ./bootstrap.sh --prefix=/opt/boost/ --with-python=python3 && \
 export PYTHON_INCLUDES=`python3-config --includes` && \

Dockerfile-build

@@ -161,7 +161,7 @@ RUN \
     git clone https://github.com/NVIDIA/nccl.git && \
     cd nccl && \
     git checkout tags/v2.4.2-1 && \
-    make src.build;
+    make -j src.build;
 
 
 COPY src/interface_py/requirements_buildonly.txt requirements_buildonly.txt
@@ -187,9 +187,9 @@ ENV LANGUAGE en_US:en
 RUN \
     export CUDA_HOME=/usr/local/cuda/ && \
 	yum install -y opencl-headers icu libicu-devel bzip2 bzip2-devel zlib-devel python-devel && \
-    wget http://sourceforge.net/projects/boost/files/boost/1.61.0/boost_1_61_0.tar.bz2 && \
-    tar xjf boost_1_61_0.tar.bz2 && \
-    cd boost_1_61_0 && \
+    wget https://s3.amazonaws.com/0xdata-public/boost/boost_1_69_0.tar.bz2  && \
+    tar xjf boost_1_69_0.tar.bz2 && \
+    cd boost_1_69_0 && \
     export PYTHONPATH=/opt/h2oai/h2o4gpu/python/ && \
     ./bootstrap.sh --prefix=/opt/boost/ --with-python=python3 && \
     export CPPFLAGS="-I/opt/h2oai/h2o4gpu/python/include/python3.6m/ -fPIC" && \