[GPU-Plugin] Multi-GPU gpu_id bug fixes for grow_gpu_hist and grow_gpu methods, and additional documentation for the gpu plugin. (#2463)

Makefile

@@ -96,7 +96,7 @@ endif
 CFLAGS += $(OPENMP_FLAGS)
 
 # for using GPUs
-GPU_COMPUTE_VER ?= 50 52 60 61
+GPU_COMPUTE_VER ?= 35 50 52 60 61
 NVCC = nvcc
 INCLUDES = -Iinclude -I$(DMLC_CORE)/include -I$(RABIT)/include
 INCLUDES += -I$(CUB_PATH)
@@ -106,14 +106,13 @@ NVCC_FLAGS = --std=c++11 $(CODE) $(INCLUDES) -lineinfo --expt-extended-lambda
 NVCC_FLAGS += -Xcompiler=$(OPENMP_FLAGS) -Xcompiler=-fPIC
 ifeq ($(PLUGIN_UPDATER_GPU),ON)
   CUDA_ROOT = $(shell dirname $(shell dirname $(shell which $(NVCC))))
-  INCLUDES += -I$(CUDA_ROOT)/include
-  LDFLAGS += -L$(CUDA_ROOT)/lib64 -lcudart
+  INCLUDES += -I$(CUDA_ROOT)/include -Inccl/src/
+  LDFLAGS += -L$(CUDA_ROOT)/lib64 -lcudart  -lcudadevrt -Lnccl/build/lib/ -lnccl_static -lm -ldl -lrt
 endif
 
 # specify tensor path
 .PHONY: clean all lint clean_all doxygen rcpplint pypack Rpack Rbuild Rcheck java pylint
 
-
 all: lib/libxgboost.a $(XGBOOST_DYLIB) xgboost
 
 $(DMLC_CORE)/libdmlc.a: $(wildcard $(DMLC_CORE)/src/*.cc $(DMLC_CORE)/src/*/*.cc)
@@ -143,7 +142,7 @@ build/%.o: src/%.cc
 	$(CXX) -c $(CFLAGS) $< -o $@
 
 # order of this rule matters wrt %.cc rule below!
-build_plugin/%.o: plugin/%.cu
+build_plugin/%.o: plugin/%.cu build_nccl
 	@mkdir -p $(@D)
 	$(NVCC) -c $(NVCC_FLAGS) $< -o $@
 
@@ -152,6 +151,11 @@ build_plugin/%.o: plugin/%.cc
 	$(CXX) $(CFLAGS) -MM -MT build_plugin/$*.o $< >build_plugin/$*.d
 	$(CXX) -c $(CFLAGS) $< -o $@
 
+build_nccl:
+	@mkdir -p build/include
+	cd build/include ; ln -sf ../../nccl/src/nccl.h .
+	cd nccl ; make -j ; cd ..
+
 # The should be equivalent to $(ALL_OBJ)  except for build/cli_main.o
 amalgamation/xgboost-all0.o: amalgamation/xgboost-all0.cc
 	$(CXX) -c $(CFLAGS) $< -o $@
@@ -173,6 +177,7 @@ jvm-packages/lib/libxgboost4j.so: jvm-packages/xgboost4j/src/native/xgboost4j.cp
 	@mkdir -p $(@D)
 	$(CXX) $(CFLAGS) $(JAVAINCFLAGS) -shared -o $@ $(filter %.cpp %.o %.a, $^) $(LDFLAGS)
 
+
 xgboost: $(CLI_OBJ) $(ALL_DEP)
 	$(CXX) $(CFLAGS) -o $@  $(filter %.o %.a, $^)  $(LDFLAGS)