Benchmarks: Build Pipeline - Support rocm in third_party/makefile (#140)

**Description**
Support rocm in third_party/makefile.

**Major Revision**
- Split rocm and cuda target in makefile
- Add target in dockerfile

dockerfile/cuda11.1.1.dockerfile

@@ -93,7 +93,7 @@ ENV PATH="${PATH}" \
 WORKDIR ${SB_HOME}
 
 ADD third_party third_party
-RUN make -j -C third_party
+RUN make -j -C third_party cuda
 
 ADD . .
 RUN python3 -m pip install .[nvidia,torch] && \

third_party/Makefile

@@ -6,10 +6,13 @@
 SB_MICRO_PATH ?= "/usr/local"
 MPI_HOME_PATH ?= "/usr/local/mpi"
 
-.PHONY: all cutlass bandwidthTest nccl_tests perftest fio
+.PHONY: all cuda rocm common cuda_cutlass cuda_bandwidthTest cuda_nccl_tests cuda_perftest rocm_perftest fio
 
 # Build all targets.
-all: cutlass bandwidthTest nccl_tests perftest fio
+all: cuda rocm
+cuda: common cuda_cutlass cuda_bandwidthTest cuda_nccl_tests cuda_perftest
+rocm: common rocm_perftest
+common: fio
 
 # Create $(SB_MICRO_PATH)/bin and $(SB_MICRO_PATH)/lib, no error if existing, make parent directories as needed.
 sb_micro_path:
@@ -17,7 +20,7 @@ sb_micro_path:
 	mkdir -p $(SB_MICRO_PATH)/lib
 
 # Build cutlass.
-cutlass:
+cuda_cutlass:
 ifneq (,$(wildcard cutlass/CMakeLists.txt))
 	cmake -DCMAKE_INSTALL_BINDIR=$(SB_MICRO_PATH)/bin -DCMAKE_INSTALL_LIBDIR=$(SB_MICRO_PATH)/lib -DCMAKE_BUILD_TYPE=Release \
 		-DCUTLASS_NVCC_ARCHS='70;80' -DCUTLASS_ENABLE_EXAMPLES=OFF -DCUTLASS_ENABLE_TESTS=OFF -S ./cutlass -B ./cutlass/build
@@ -28,7 +31,7 @@ endif
 # cuda-samples is released together with CUDA, they have the exact same version. Like v10.0, v11.1 and so on.
 # The version we use is the released tag of cuda-samples which is consistent with the cuda version in the environment or docker.
 # The Makefile of bandwidthTest does not have 'install' target, so need to copy bin to $(SB_MICRO_PATH)/bin/ and create $(SB_MICRO_PATH)/bin/ if not existing.
-bandwidthTest: sb_micro_path
+cuda_bandwidthTest: sb_micro_path
 	if [ -d cuda-samples ]; then rm -rf cuda-samples; fi
 	git clone -b v$(shell nvcc --version | grep 'release' | awk '{print $$6}' | cut -c2- | cut -d '.' -f1-2) https://github.com/NVIDIA/cuda-samples.git ./cuda-samples
 	cd ./cuda-samples/Samples/bandwidthTest && make clean && make TARGET_ARCH=x86_64 SMS="70 75 80 86"
@@ -36,7 +39,7 @@ bandwidthTest: sb_micro_path
 
 # Build nccl-tests.
 # The version we use is commit 44df0bf from master branch, since it didn't update release tag for long time.
-nccl_tests: sb_micro_path
+cuda_nccl_tests: sb_micro_path
 ifneq (,$(wildcard nccl-tests/Makefile))
 	cd ./nccl-tests && make MPI=1 MPI_HOME=$(MPI_HOME_PATH) -j
 	cp -v ./nccl-tests/build/* $(SB_MICRO_PATH)/bin/
@@ -44,10 +47,14 @@ endif
 
 # Build perftest.
 # The version we use is the tag v4.5-0.2.
-perftest:
+cuda_perftest:
 ifneq (,$(wildcard perftest/autogen.sh))
 	cd perftest && ./autogen.sh && ./configure CUDA_H_PATH=/usr/local/cuda/include/cuda.h --prefix=$(SB_MICRO_PATH) && make -j && make install
 endif
+rocm_perftest:
+ifneq (,$(wildcard perftest/autogen.sh))
+	cd perftest && ./autogen.sh && ./configure --enable-rocm --with-rocm=/opt/rocm --prefix=$(SB_MICRO_PATH) && make -j && make install
+endif
 
 # Build FIO from commit 0313e9 (fio-3.27 tag).
 fio: