NeutralCode
diff --git a/‎CMakeModules/FindNCCL.cmake‎
Lines changed: 24 additions & 0 deletions b/‎CMakeModules/FindNCCL.cmake‎
Lines changed: 24 additions & 0 deletions
diff --git a/‎CMakeModules/FindclBLAS.cmake‎
Lines changed: 1 addition & 2 deletions b/‎CMakeModules/FindclBLAS.cmake‎
Lines changed: 1 addition & 2 deletions
diff --git a/‎src/CMakeLists.txt‎
Lines changed: 18 additions & 1 deletion b/‎src/CMakeLists.txt‎
Lines changed: 18 additions & 1 deletion
diff --git a/‎src/gpuarray_buffer_opencl.c‎
Lines changed: 2 additions & 1 deletion b/‎src/gpuarray_buffer_opencl.c‎
Lines changed: 2 additions & 1 deletion
@@ -0,0 +1,24 @@
+# Find the NCCL libraries
+#
+# The following variables are optionally searched for defaults
+#  NCCL_ROOT_DIR:    Base directory where all NCCL components are found
+#
+# The following are set after configuration is done:
+#  NCCL_FOUND
+#  NCCL_INCLUDE_DIR
+#  NCCL_LIBRARY
+
+find_path(NCCL_INCLUDE_DIR NAMES nccl.h
+    PATHS ${NCCL_ROOT_DIR}/include
+    )
+
+find_library(NCCL_LIBRARY NAMES nccl
+    PATHS ${NCCL_ROOT_DIR}/lib ${NCCL_ROOT_DIR}/lib64)
+
+include(FindPackageHandleStandardArgs)
+find_package_handle_standard_args(NCCL DEFAULT_MSG NCCL_INCLUDE_DIR NCCL_LIBRARY)
+
+if(NCCL_FOUND)
+  message(STATUS "Found NCCL (include: ${NCCL_INCLUDE_DIR}, library: ${NCCL_LIBRARY})")
+  mark_as_advanced(NCCL_INCLUDE_DIR NCCL_LIBRARY)
+endif()
@@ -4,7 +4,6 @@
 #  CLBLAS_FOUND - system has clBLAS
 #  CLBLAS_INCLUDE_DIRS - location of clBLAS.h
 #  CLBLAS_LIBRARIES - location of libclBLAS
-#  
 
 IF(CLBLAS_INCLUDE_DIRS)
   # Already in cache, be silent
@@ -13,7 +12,7 @@ ENDIF (CLBLAS_INCLUDE_DIRS)
 
 FIND_PATH(CLBLAS_ROOT_DIR
     NAMES include/clBLAS.h
-    HINTS /usr/local/ $ENV{CLBLAS_ROOT} 
+    HINTS /usr/local/ $ENV{CLBLAS_ROOT}
     DOC "clBLAS root directory.")
 
 FIND_PATH(_CLBLAS_INCLUDE_DIRS
 
@@ -4,6 +4,7 @@ set(CMAKE_C_FLAGS_DEBUG "${CMAKE_C_FLAGS_DEBUG} -DDEBUG")
 
 find_package(CUDA)
 find_package(OpenCL)
+find_package(NCCL)
 
 include_directories("${CMAKE_CURRENT_SOURCE_DIR}")
 
@@ -40,8 +41,10 @@ gpuarray_error.c
 gpuarray_util.c
 gpuarray_buffer.c
 gpuarray_buffer_blas.c
+gpuarray_buffer_collectives.c
 gpuarray_array.c
 gpuarray_array_blas.c
+gpuarray_array_collectives.c
 gpuarray_kernel.c
 gpuarray_extension.c
 gpuarray_elemwise.c
@@ -102,6 +105,14 @@ if (CUDA_FOUND)
   if (CUBLAS_SGEMMEX)
     add_definitions(-DHAVE_CUBLAS_SGEMMEX)
   endif()
+
+  if (NCCL_FOUND)
+    message("Building with NCCL")
+    add_definitions(-DWITH_CUDA_NCCL)
+    set(GPUARRAY_SRC ${GPUARRAY_SRC} gpuarray_collectives_cuda_nccl.c)
+    include_directories("${NCCL_INCLUDE_DIR}")
+    set(CMAKE_REQUIRED_LIBRARIES ${CMAKE_REQUIRED_LIBRARIES} ${NCCL_LIBRARY})
+  endif()
 endif()
 
 if(OPENCL_FOUND)
@@ -136,7 +147,11 @@ add_library(gpuarray-static STATIC ${GPUARRAY_SRC})
 
 if(CUDA_FOUND)
   target_link_libraries(gpuarray ${CUDADRV_LIBRARY} ${CUDA_CUBLAS_LIBRARIES})
-  target_link_libraries(gpuarray-static ${CUDADRV_LIBRARY} ${CUDA_CUBLAS_LIBRARY})
+  target_link_libraries(gpuarray-static ${CUDADRV_LIBRARY} ${CUDA_CUBLAS_LIBRARIES})
+  if (NCCL_FOUND)
+    target_link_libraries(gpuarray ${NCCL_LIBRARY})
+    target_link_libraries(gpuarray-static ${NCCL_LIBRARY})
+  endif()
 endif()
 
 if(OPENCL_FOUND)
@@ -151,8 +166,10 @@ endif()
 SET(headers
   gpuarray/array.h
   gpuarray/blas.h
+  gpuarray/collectives.h
   gpuarray/buffer.h
   gpuarray/buffer_blas.h
+  gpuarray_buffer_collectives.c
   gpuarray/config.h
   gpuarray/elemwise.h
   gpuarray/error.h
 
@@ -1161,7 +1161,8 @@ static int cl_property(gpucontext *c, gpudata *buf, gpukernel *k, int prop_id,
 #endif
 
   case GA_CTX_PROP_COMM_OPS:
-    // Complete in the future if a multi-gpu collectives API for opencl appears
+    // TODO Complete in the future whenif a multi-gpu collectives API for
+    // opencl appears
     *((void **)res) = NULL;
     return GA_DEVSUP_ERROR;