#!/bin/sh
# curl http://192.168.1.34/nvidia.cuda13.sh | sh
# yum install -y make gcc kernel-devel-$(uname -r) kernel-headers-$(uname -r)
# sudo sh cuda_*_linux.run
export CUDA_HOME=/usr/local/cuda
export PATH=$PATH:$CUDA_HOME/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CUDA_HOME/lib64
###init
GPU_COUNT=$(nvidia-smi --list-gpus | wc -l)
dir=$(pwd)
logname=nvidia$(date +%Y%m%d_%H%M%S).out
### Install cuda-samples-12.8
#
# unzip cuda-samples-12.8.zip
# cd cuda-samples-12.8
# mkdir build && cd build
# cmake ..
# make -j$(nproc)
#
### Install nccl    https://developer.nvidia.com/nccl/nccl-download
# sudo dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel9/x86_64/cuda-rhel9.repo
# sudo dnf install libnccl-2.28.9-1+cuda13.0 libnccl-devel-2.28.9-1+cuda13.0 libnccl-static-2.28.9-1+cuda13.0
# make CUDA_HOME=/usr/local/cuda -j$(nproc)
# 
### 设置软链接
cd $dir
#ln -s cuda-samples-12.8/build/Samples/ $dir/Samples
ln -s nccl-tests-2.16.4/build nccl

hostnamectl | tee $dir/$logname
### 硬件信息查询
#lspci
echo -e "###lspci -d 10de:" | tee -a $dir/$logname
lspci -d 10de: | tee -a $dir/$logname
sleep 5
#
#nvidia-smi
echo -e "\n###nvidia-smi\n" | tee -a $dir/$logname
nvidia-smi | tee -a $dir/$logname
sleep 5
#
#topo
echo -e "\n###nvidia-smi topo -m\n" | tee -a $dir/$logname
nvidia-smi topo -m | tee -a $dir/$logname
sleep 5
#
#query
echo -e "\n###nvidia-smi -q -u\n" | tee -a $dir/$logname
nvidia-smi -q -u | tee -a $dir/$logname
sleep 5
#
#query
echo -e "\n###nvidia-smi -q -i 0\n" | tee -a $dir/$logname
nvidia-smi -q -i 0 | tee -a $dir/$logname
sleep 5
#
#Configured
echo -e "\n###nvidia-smi --format=csv --query-gpu=gpu_name,serial,gpu_bus_id,vbios_version,memory.total,pcie.link.gen.max,pcie.link.gen.current,pcie.link.width.max,pcie.link.width.current\n"  | tee -a $dir/$logname
nvidia-smi --format=csv --query-gpu=gpu_name,serial,gpu_bus_id,vbios_version,memory.total,pcie.link.gen.max,pcie.link.gen.current,pcie.link.width.max,pcie.link.width.current | tee -a $dir/$logname
sleep 5
#
### GPU测试
#nccl
cd $dir/nccl
NCCL_NTHREADS=$(nproc) NCCL_MIN_NCHANNELS=$GPU_COUNT ./all_reduce_perf -b 1m -e 1g -f 2 -g $GPU_COUNT  | tee -a $dir/$logname
sleep 10
#
#p2pBandwidthLatencyTest
echo -e "\n###p2pBandwidthLatencyTest\n" | tee -a $dir/$logname
cd $dir/Samples/5_Domain_Specific/p2pBandwidthLatencyTest
./p2pBandwidthLatencyTest | tee -a $dir/$logname
sleep 10
#
##bandwidthTest
#echo -e "###bandwidthtest" | tee -a $dir/$logname
#for (( i=0; i<GPU_COUNT; i++ )); do
#    {
#    cd $dir/Samples/1_Utilities/bandwidthTest
#    ./bandwidthTest --device=$i >>  $dir/$logname 
#    }
#done
#./bandwidthTest --device=all >>  $dir/$logname
#cat $dir/$logname | grep 3200000 
#sleep 10
#
#matrixMul
echo -e "\n###matrixMul\n" | tee -a $dir/$logname
cd $dir/Samples/0_Introduction/matrixMul
./matrixMul | tee -a $dir/$logname
sleep 10
#
#alignedTypes
echo -e "\n###alignedTypes\n" | tee -a $dir/$logname
cd $dir/Samples/6_Performance
./alignedTypes/alignedTypes | tee -a $dir/$logname
sleep 5
#
#LargeKernelParameter
echo -e "\n###LargeKernelParameter\n" | tee -a $dir/$logname
cd $dir/Samples/6_Performance
./LargeKernelParameter/LargeKernelParameter | tee -a $dir/$logname
sleep 5
#
#transpose
echo -e "\n###transpose\n" | tee -a $dir/$logname
cd $dir/Samples/6_Performance
./transpose/transpose | tee -a $dir/$logname
sleep 5
#
#UnifiedMemoryPerf
echo -e "\n###UnifiedMemoryPerf\n" | tee -a $dir/$logname
cd $dir/Samples/6_Performance
./UnifiedMemoryPerf/UnifiedMemoryPerf | tee -a $dir/$logname
sleep 5
#
#cudaGraphsPerfScaling
echo -e "\n###cudaGraphsPerfScaling\n" | tee -a $dir/$logname
cd $dir/Samples/6_Performance
./cudaGraphsPerfScaling/cudaGraphsPerfScaling | tee -a $dir/$logname
sleep 5
#
#OpenCL-Benchmark
echo -e "\n###OpenCL-Benchmark\n" | tee -a $dir/$logname
cd $dir
chmod +x OpenCL-Benchmark-Linux
./OpenCL-Benchmark-Linux | tee -a $dir/$logname
sleep 5

#lspci tv
lspci -tv >> $dir/$logname
#lspci vvv
lspci -d 10de: -vvv >> $dir/$logname