Benchmarks: micro benchmarks - add nvbandwidth build

.gitmodules

@@ -30,3 +30,6 @@
 [submodule "third_party/Megatron/Megatron-DeepSpeed"]
 	path = third_party/Megatron/Megatron-DeepSpeed
 	url = https://github.com/microsoft/Megatron-DeepSpeed.git
+[submodule "third_party/nvbandwidth"]
+	path = third_party/nvbandwidth
+	url = https://github.com/NVIDIA/nvbandwidth.git

third_party/Makefile

@@ -16,12 +16,12 @@ ROCM_VER ?= $(shell hipconfig -R | grep -oP '\d+\.\d+\.\d+' || echo "0.0.0")
 
 NUM_MAKE_JOBS ?= $(shell nproc --ignore=2)
 
-.PHONY: all cuda_with_msccl cuda rocm common cuda_cutlass cuda_bandwidthTest cuda_nccl_tests cuda_perftest cuda_msccl rocm_perftest fio rocm_rccl_tests rocm_rocblas rocm_bandwidthTest gpcnet cuda_gpuburn cpu_stream cpu_hpl directx_amf_encoding_latency directx_amd rocm_hipblaslt megatron_lm megatron_deepspeed apex_rocm
+.PHONY: all cuda_with_msccl cuda rocm common cuda_cutlass cuda_bandwidthTest cuda_nccl_tests cuda_perftest cuda_msccl rocm_perftest fio rocm_rccl_tests rocm_rocblas rocm_bandwidthTest gpcnet cuda_gpuburn cpu_stream cpu_hpl directx_amf_encoding_latency directx_amd rocm_hipblaslt megatron_lm megatron_deepspeed apex_rocm nvbandwidth
 
 # Build targets.
 all: cuda rocm
 cuda_with_msccl: cuda cuda_msccl
-cuda: common cuda_cutlass cuda_bandwidthTest cuda_nccl_tests cuda_perftest gpcnet cuda_gpuburn megatron_lm megatron_deepspeed
+cuda: common cuda_cutlass cuda_bandwidthTest cuda_nccl_tests cuda_perftest gpcnet cuda_gpuburn megatron_lm megatron_deepspeed nvbandwidth
 rocm: common rocm_perftest rocm_rccl_tests rocm_rocblas rocm_bandwidthTest rocm_hipblaslt megatron_deepspeed apex_rocm
 cpu: common cpu_perftest cpu_stream
 common: fio
@@ -239,3 +239,16 @@ ifneq (,$(wildcard msccl/tests/msccl-tests-nccl/Makefile))
 	mkdir -p $(SB_MICRO_PATH)/bin/msccl-tests-nccl && \
 	cp -r -v ./msccl/tests/msccl-tests-nccl/build/* $(SB_MICRO_PATH)/bin/msccl-tests-nccl/
 endif
+
+# Build nvbandwidth.
+REQUIRED_CMAKE_VERSION := 3.20.0
+CMAKE_VERSION := $(shell cmake --version 2>/dev/null | head -n 1 | sed 's/[^0-9.]*\([0-9.]*\).*/\1/')
+nvbandwidth: sb_micro_path
+	@if [ "$(CMAKE_VERSION)" = "" ]; then \
+        echo "Skip nvbandwidth for CMake is not installed."; \
+    elif [ "$(shell printf '%s\n' "$(REQUIRED_CMAKE_VERSION)" "$(CMAKE_VERSION)" | sort -V | head -n 1)" = "$(REQUIRED_CMAKE_VERSION)" ]; then \
+        cd ./nvbandwidth && cmake . && make && cd ..; \
+        cp -v ./nvbandwidth/nvbandwidth $(SB_MICRO_PATH)/bin; \
+    else \
+        echo "Skip nvbandwidth for CMake version $(CMAKE_VERSION) is too old. Required version is $(REQUIRED_CMAKE_VERSION) or higher."; \
+    fi