feat(lite): add example script and some small change for lar

GitOrigin-RevId: a28ed2f27a

lite/load_and_run/examples/README.md

@@ -0,0 +1,15 @@
+# load and run example scripts
+
+this is some typical examples for using of load and run, you can use
+
+```bash
+./example.sh 
+```
+
+to run those examples, more details is in the script
+
+the usage of script can be see by run this scirpt with:
+
+```bash
+./example.sh -h
+```

lite/load_and_run/examples/example.sh

@@ -0,0 +1,437 @@
+#!/bin/bash
+OLD_IFS="$IFS"
+IFS=$'\n'
+TYPICAL_MODEL_DATD="model_source/resnet50_b1_float32_without_data.mge --input \"data:input_data/resnet50_input.npy\""
+DEVICE_DESC=""
+WORK_DIR_PATH="."
+RUN_ARM_DEVICE="false"
+RUN_TARGET="diff_model"
+ONLY_PREPARE_MODEL="false"
+MODEL_PREAPRED="false"
+ONLY_BUILD="false"
+LAR_BUILT="false"
+CLEAN_ALL="false"
+
+RUN_TARGETS=("diff_model")
+RUN_TARGETS+=("diff_device")
+RUN_TARGETS+=("fast_run")
+RUN_TARGETS+=("io")
+RUN_TARGETS+=("layout")
+RUN_TARGETS+=("optimize")
+RUN_TARGETS+=("plugin")
+RUN_TARGETS+=("all")
+
+
+function usage() {
+    echo "$0 args1 args2 .."
+    echo "available args detail:"
+    echo "-p : prepare example model "
+    echo "-b : build load_and_run for x86/armv7/arm64 cpu and CUDA"
+    echo "-t : set the ssh arm device "
+    echo "-w : set the arm device workspace dir"
+    echo "-c : clean all"
+    echo "-a : run all test"
+    echo "-e : set the running target for test (details use \"-e\" to see)"
+    echo "-h : show usage"
+    exit -1
+}
+
+
+while getopts "pbcahe:w:t:" arg
+do
+    case $arg in
+        t)
+            DEVICE_DESC=$OPTARG
+            RUN_ARM_DEVICE="true"
+            echo "config arm device DEVICE_DESC to ${DEVICE_DESC}"
+            ;;
+        w)
+            WORK_DIR_PATH=$OPTARG
+            echo "config arm device WORK_DIR_PATH to ${WORK_DIR_PATH}"
+            ;;
+        e)
+            tmp_target=null
+            for target in ${RUN_TARGETS[@]}; do
+                if [ "$target" = "$OPTARG" ]; then
+                    echo "CONFIG BUILD RUN_TARGET to : $OPTARG"
+                    tmp_target=$OPTARG
+                    RUN_TARGET=$OPTARG
+                    break
+                fi
+            done
+            if [ "$tmp_target" = "null" ]; then
+                echo "ERR args for target (-e)"
+                echo "available target usage :"
+                for target in ${RUN_TARGETS[@]}; do
+                    echo "  -e $target"
+                done
+                exit -1
+            fi
+            ;;
+        h)
+            echo "show usage"
+            usage
+            ;;
+        a)
+            echo "config RUN_TARGET=all"
+            RUN_TARGET="all"
+            ;;
+        c)
+            echo "clean all directory generated by script"
+            CLEAN_ALL="true"
+
+            ;;
+        b)
+            echo "run build"
+            ONLY_BUILD="true"
+            ;;
+        p)
+            echo "prepare model and input"
+            ONLY_PREPARE_MODEL="true"
+            ;;
+        ?)
+            echo "unkonw argument"
+            usage
+            ;;
+    esac
+done
+
+
+function prepare_model_and_data(){
+    rm -rf model_source && mkdir model_source
+    # dump mgb model
+    python3 script/resnet50_mgb.py -o model_source/resnet50.pkl
+    ../dump_with_testcase.py model_source/resnet50.pkl -o model_source/resnet50_with_data.mgb -d "#rand(0, 255)" --no-assert
+    
+    # prepare simple add model
+    python3 script/add_demo.py --dir model_source
+    python3 script/conv_demo.py --dir model_source
+    
+    #generate trt model
+    script/gen_trt_model.sh
+
+    #prepare mge model
+    python3 script/resnet50_mge.py --dir model_source
+    python3 script/resnet50_mge.py  --dir model_source -d int8
+    python3 script/resnet50_mge.py  --dir model_source --inputs "#rand(0,255)"
+
+    #make input_data
+    rm -rf input_data && mkdir input_data
+    python3 script/mge_input_data.py
+
+    rm -rf tmpdir && mkdir tmpdir
+}
+
+function build_lar(){
+    # build cpu and cuda version 
+     ../../../scripts/cmake-build/host_build.sh -r -t -e load_and_run
+    #WARNING:config the cuda environment before compile
+    ../../../scripts/cmake-build/host_build.sh  -c -t -e load_and_run
+    
+    # # build arm version
+
+    ../../../scripts/cmake-build/cross_build_android_arm_inference.sh -r -a arm64-v8a -e load_and_run
+    ../../../scripts/cmake-build/cross_build_android_arm_inference.sh -r -a armeabi-v7a -e load_and_run
+    
+
+    # link or for test
+    ln -s ../../../build_dir/host/MGE_WITH_CUDA_OFF/MGE_INFERENCE_ONLY_OFF/Release//build/lite/load_and_run/load_and_run lar_cpu
+    ln -s ../../../build_dir/host/MGE_WITH_CUDA_ON/MGE_INFERENCE_ONLY_OFF/Release//build/lite/load_and_run/load_and_run lar_cuda
+    cp ../../../build_dir/android/arm64-v8a/Release/build/lite/load_and_run/load_and_run ./lar_arm64
+    cp ../../../build_dir/android/armeabi-v7a/Release/build/lite/load_and_run/load_and_run ./lar_armv7
+
+    
+    
+}   
+
+function set_arm_device_and_upload(){
+    DEVICE_DESC="${1}"
+    WORK_DIR_PATH="${2}"
+    RUN_ARM_DEVICE="${3}"
+    cmd="rsync -aP -zz  ./lar_arm64 ./lar_armv7  model_source/resnet50_b1_float32_without_data.mge input_data/resnet50_input.npy  $DEVICE_DESC:$WORK_DIR_PATH/"
+    echo $cmd
+    bash -c "$cmd"
+}
+
+function test_different_model(){
+    CmdArray=("./lar_cpu model_source/resnet50_with_data.mgb")
+    CmdArray+=("./lar_cpu model_source/resnet50_b1_float32_with_data.mge")
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD")
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --lite")
+    for cmd in ${CmdArray[@]}; do
+        echo "$cmd"
+        bash -c "$cmd"
+    done
+}
+
+function test_different_device(){
+    #dispatch时，计算任务会加入一个工作队列，由队列统一管理执行 均值 131.278 ms 标准差 15.197ms m_asyc_exec异步执行
+    CmdArray=("./lar_cpu $TYPICAL_MODEL_DATD --cpu")
+    #dispatch时，计算任务直接执行 均值 131.875 ms 标准差 7.758ms m_asyc_exec同步执行
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --cpu-default")
+    #多线程运行 1～8平均运行时间（ms）：129.611, 84.266, 76.963, 55.212, 69.283, 77.338, 58.386, 64.585
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --multithread 4")
+    #主线程锁核，其他任务在线程池中的线程上运行 132.614, 83.095, 69.792, 54.452, 48.890, 48.206, 46.386, 53.908
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --multithread-default 4")
+    #cpu多线程绑核(x86上绑核影响不大)
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --multithread 2 --multi-thread-core-ids 1,5")
+
+    #xpu 设置为cpu上运行 132.740 ms  comp_node:cpu
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --cpu")
+    #xpu 设置为cuda上运行 6.495 ms comp_node:gpu
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --cuda")
+    
+    for cmd in ${CmdArray[@]}; do
+        echo $cmd
+         bash -c "$cmd"
+    done
+
+function test_fast_run(){
+
+    CmdArray=("./lar_cuda $TYPICAL_MODEL_DATD --fast-run")
+
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --full-run")
+    #fast run 搜参  
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --fast-run --fast-run-algo-policy tmpdir/algo_cache_file")
+
+     #fast run 带参执行 
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --fast-run-algo-policy tmpdir/algo_cache_file")
+
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --fast-run  --reproducible")
+
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --fast-run  --fast-run-shared-batch-size 1")
+
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --fast-run --binary-equal-between-batch")
+
+    for cmd in ${CmdArray[@]}; do
+        echo $cmd
+         bash -c "$cmd"
+    done
+}
+
+function test_io(){
+    rm -rf tmpdir/bin_io_info tmpdir/bin_out_info tmpdir/bin_out_info_cuda tmpdir/io_info.txt
+    mkdir tmpdir/bin_io_info tmpdir/bin_out_info tmpdir/bin_io_info_cuda
+
+    CmdArray=("./lar_cpu $TYPICAL_MODEL_DATD")
+
+    CmdArray+=("./lar_cpu model_source/add_demo_f32_without_data.mge --input \"input_data/add_demo_input.json\"") 
+    #! the model must support input with nhwc shape
+    CmdArray+=("./lar_cpu model_source/resnet50_b1_int8_without_data.mge --input \"data:input_data/cat.ppm\"") 
+
+    CmdArray+=("./lar_cpu model_source/add_demo_f32_without_data.mge --input \"data:[2,3,4]\"")
+
+    CmdArray+=("./lar_cpu model_source/add_demo_f32_without_data.mge --input \"data:[2,3,4]\" --io-dump  tmpdir/io_info.txt --iter 1 --warmup-iter 0")
+    
+    CmdArray+=("./lar_cpu model_source/add_demo_f32_without_data.mge --input \"data:[2,3,4]\" --io-dump-stdout --iter 1 --warmup-iter 0")
+    
+    
+    
+    CmdArray+=("./lar_cpu model_source/add_demo_f32_without_data.mge --input \"data:[2,3,4]\" --io-dump-stderr --iter 1 --warmup-iter 0")
+
+    #different data in the given directory the name is the var id which is the same with txt-dump information
+    CmdArray+=("./lar_cpu model_source/add_demo_f32_without_data.mge --input \"data:[2,3,4]\" --bin-io-dump tmpdir/bin_io_info --iter 1 --warmup-iter 0")
+
+    CmdArray+=("./lar_cuda model_source/add_demo_f32_without_data.mge --input \"data:[2,3,4]\" --bin-io-dump tmpdir/bin_io_info_cuda --iter 1 --warmup-iter 0")
+    
+   
+    CmdArray+=("./lar_cpu model_source/add_demo_f32_without_data.mge --input \"data:[2,3,4]\" --bin-out-dump tmpdir/bin_out_info --iter 1 --warmup-iter 0")
+
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --copy-to-host")
+
+    for cmd in ${CmdArray[@]}; do
+        echo $cmd
+         bash -c "$cmd"
+    done
+
+    #compare the binary io information
+    python3 ../../../imperative/python/megengine/tools/compare_binary_iodump.py tmpdir/bin_io_info tmpdir/bin_io_info_cuda
+
+}
+
+function test_layout_related(){
+    # very little speed up
+    CmdArray=("./lar_cuda $TYPICAL_MODEL_DATD --enable-nchw4")
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --enable-chwn4")
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --enable-nchw32")
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --enable-nchw64")
+
+    #speed up 
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --enable-nchw88") 
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --cuda --layout-transform cuda")
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --cuda --layout-transform cuda --layout-transform-dump model_source/resnet50_b1_float32_without_data_with_global_layout_trans_cuda.mge")
+    CmdArray+=("./lar_cuda model_source/resnet50_b1_float32_without_data_with_global_layout_trans_cuda.mge")
+
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --cpu --layout-transform cpu")
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --cpu --layout-transform cpu --layout-transform-dump model_source/resnet50_b1_float32_without_data_with_global_layout_trans_cpu.mge")
+    CmdArray+=("./lar_cpu model_source/resnet50_b1_float32_without_data_with_global_layout_trans_cpu.mge")
+    
+    for cmd in ${CmdArray[@]}; do
+        echo $cmd
+         bash -c "$cmd"
+    done
+
+if [ ${RUN_ARM_DEVICE} == "true" ] ; then 
+    #speed up
+    CmdArray=("./lar_arm64 resnet50_b1_float32_without_data.mge --input \"data:resnet50_input.npy\" --cpu --enable-nchw44")
+
+    #speed up
+    CmdArray+=("./lar_arm64 resnet50_b1_float32_without_data.mge --input \"data:resnet50_input.npy\" --cpu --enable-nchw44-dot")
+
+    for cmd in ${CmdArray[@]}; do
+        echo $cmd
+        ssh -t $DEVICE_DESC "unset LD_PRELOAD && cd $WORK_DIR_PATH && LD_LIBRARY_PATH=./ $cmd" 
+    done
+else
+    echo "SET arm device ON : $RUN_ARM_DEVICE"
+fi
+
+}
+
+function test_optimize(){
+    CmdArray=("./lar_cpu $TYPICAL_MODEL_DATD --enable-fuse-preprocess")
+    #warm up speed up
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --enable-fuse-conv-bias-nonlinearity")
+    
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --enable-fuse-conv-bias-with-z")
+    
+    CmdArray+=("./lar_cuda model_source/trt_conv_demo_with_data.mgb  --tensorrt")
+    
+    CmdArray+=("./lar_cuda model_source/trt_conv_demo_with_data.mgb  --tensorrt --tensorrt-cache tmpdir/TRT_cache")
+    
+    CmdArray+=("./lar_cuda model_source/trt_conv_demo_with_data.mgb  --tensorrt-cache tmpdir/TRT_cache")
+    
+
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --no-sanity-check --record-comp-seq2")
+
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --disable_mem_opt")
+
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --workspace_limit 10000")
+
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --fake-first")
+
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD  --enable_jit ") 
+
+    for cmd in ${CmdArray[@]}; do
+        echo $cmd
+         bash -c "$cmd"
+    done
+
+}
+
+function test_plugin(){
+    
+    rm -rf tmpdir/staticMemInfoDir tmpdir/staticMemInfoDirLogs
+    mkdir tmpdir/staticMemInfoDir
+
+    CmdArray=("./lar_cpu model_source/add_demo_f32_without_data.mge --input \"data:[2,3,4]\" --check-dispatch")
+
+    CmdArray+=("./lar_cpu model_source/add_demo_f32_without_data.mge --input \"data:[2,3,4]\" --check-var-value 5:0")
+
+    CmdArray+=("./lar_cpu model_source/add_demo_f32_without_data.mge --input \"data:[2,3,4]\" --range 2")
+
+    CmdArray+=("./lar_cpu model_source/add_demo_f32_without_data.mge --input \"data:[2,3,4]\" --profile tmpdir/opr_profile.json
+                
+                ")
+    CmdArray+=("./lar_cuda model_source/add_demo_f32_without_data.mge --input \"data:[2,3,4]\" --profile-host tmpdir/opr_profile_host.json")
+
+
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --model-info")
+
+    CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --verbose")
+
+    CmdArray+=("./lar_cpu model_source/resnet50_with_data.mgb --disable-assert-throw")
+
+    # wait gdb attach to given PID
+    # CmdArray+=("./lar_cpu $TYPICAL_MODEL_DATD --wait-gdb")
+
+    CmdArray+=("./lar_cuda $TYPICAL_MODEL_DATD --get-static-mem-info tmpdir/staticMemInfoDir")
+
+    for cmd in ${CmdArray[@]}; do
+        echo $cmd
+         bash -c "$cmd"
+    done
+
+    # view the graph with given url (usally: http://localhost:6006/)
+    # mkdir tmpdir/staticMemInfoDirLogs &&  python3 ../../../imperative/python/megengine/tools/graph_info_analyze.py -i tmpdir/staticMemInfoDir -o tmpdir/staticMemInfoDirLogs
+    # pip3 install tensorboard && tensorboard --logdir tmpdir/staticMemInfoDirLogs
+}
+
+function clean(){
+    rm -rf tmpdir model_source input_data lar_cpu lar_cuda lar_arm64 lar_armv7
+
+}
+
+function main(){
+
+if [ ${CLEAN_ALL} == "true" ] ; then
+    clean
+    exit 0
+fi
+
+if [ ${ONLY_PREPARE_MODEL} == "true" ] ; then 
+    prepare_model_and_data
+    MODEL_PREAPRED="true"
+    exit 0
+fi
+
+if [ ${ONLY_BUILD} == "true" ] ; then 
+    build_lar
+    LAR_BUILT="true"
+    exit 0
+fi
+
+if [ ${RUN_ARM_DEVICE} == "true" ] ; then
+    set_arm_device_and_upload $DEVICE_DESC "$WORK_DIR_PATH" "true"
+fi
+
+if [ ${MODEL_PREAPRED} != "true" ] ; then
+    CHECK_MODEL=$(find . -name add_demo_input.json)
+    if [ ${CHECK_MODEL} == "" ] ; then
+        prepare_model_and_data
+        MODEL_PREAPRED="true"
+    fi
+fi
+
+if [ ${LAR_BUILT} != "true" ] ; then
+    CHECK_LAR=$(find . -name lar_armv7)
+    if [ ${CHECK_LAR} == "" ] ; then
+        build_lar
+        LAR_BUILT="true"
+    fi
+fi
+
+if [ ${RUN_TARGET} == "diff_model" -o ${RUN_TARGET} == "all" ] ; then
+    test_different_model
+fi
+
+if [ ${RUN_TARGET} == "diff_device" -o ${RUN_TARGET} == "all" ] ; then
+   test_different_device
+fi
+
+if [ ${RUN_TARGET} == "fast_run" -o ${RUN_TARGET} == "all" ] ; then
+    test_fast_run
+fi
+
+if [ ${RUN_TARGET} == "io" -o ${RUN_TARGET} == "all" ] ; then
+    test_io
+fi
+
+if [ ${RUN_TARGET} == "layout" -o ${RUN_TARGET} == "all" ] ; then
+    test_layout_related
+fi
+
+if [ ${RUN_TARGET} == "optimize" -o ${RUN_TARGET} == "all" ] ; then
+    test_optimize
+fi
+
+if [ ${RUN_TARGET} == "plugin" -o ${RUN_TARGET} == "all" ] ; then
+    test_plugin
+fi
+
+
+
+}
+
+main
+IFS=$OLD_IFS

lite/load_and_run/examples/script/add_demo.py

@@ -0,0 +1,57 @@
+#!/usr/bin/env python3
+import argparse
+import math
+
+import megengine.functional as F
+import megengine.hub as hub
+import megengine.module as M
+import numpy as np
+from megengine import jit, tensor, Parameter
+
+class Simple(M.Module):
+    def __init__(self):
+        super().__init__()
+        self.a = Parameter([0,1,2], dtype=np.float32)
+
+    def forward(self, x):
+        x = x + self.a
+        return x
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="dump mge model for add_demo",
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter,
+    )
+    parser.add_argument(
+        "--inputs",
+        help="set the inputs data to get a model with testcase",
+        default="",
+        type=str,
+    )
+
+    parser.add_argument(
+        "--dir",
+        help="set the dir where the model to dump",
+        default=".",
+        type=str,
+    )
+
+    args = parser.parse_args()
+    net = Simple()
+    net.eval()
+
+    @jit.trace(symbolic=True, capture_as_const=True)
+    def fun(data):
+        return net(data)
+    data = tensor([3,4,5])
+    fun(data)
+    if args.inputs == "":
+        fun.dump(
+            args.dir + "/add_demo_f32_without_data.mge", arg_names=["data"],
+            no_assert=True,    
+        )
+    else:
+        fun.dump(
+            args.dir + "/add_demo_f32_with_data.mge", arg_names=["data"],
+            input_data=[args.inputs], no_assert=True,
+        )

lite/load_and_run/examples/script/conv_demo.py

@@ -0,0 +1,42 @@
+#!/usr/bin/env python3
+import argparse
+import math
+
+import megengine.functional as F
+import megengine.module as M
+import numpy as np
+from megengine import jit, tensor
+class ConvNet(M.Module):
+     def __init__(self):
+
+         super().__init__()
+         self.conv1 = M.Conv2d(in_channels=3, out_channels=1, kernel_size=3, bias=False)
+
+     def forward(self, input):
+         x = self.conv1(input)
+         return x
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="dump mge model for add_demo",
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter,
+    )
+
+    parser.add_argument(
+        "--dir",
+        help="set the dir where the model to dump",
+        default=".",
+        type=str,
+    )
+
+    args = parser.parse_args()
+    net = ConvNet()
+    net.eval()
+
+    @jit.trace(symbolic=True, capture_as_const=True)
+    def fun(data):
+        return net(data)
+    inp = tensor(np.arange(0, 96).astype("float32").reshape(2, 3, 4, 4))
+    out = fun(inp)
+    fun.dump(args.dir + "/conv_demo_f32_without_data.mge", arg_names=["data"],
+        no_assert=True)

lite/load_and_run/examples/script/dump_trt.py

@@ -0,0 +1,37 @@
+#!/usr/bin/env python3
+from megskull.network import RawNetworkBuilder
+import megskull.opr.all as O
+from megskull.opr.external import TensorRTRuntimeOpr
+from meghair.utils.io import dump
+import argparse
+
+def str2tuple(x):
+    x = x.split(',')
+    x = [int(a) for a in x]
+    x = tuple(x)
+    return x
+
+def make_network(model, isize):
+    data = [O.DataProvider('input{}'.format(i), shape=isizes[i])
+            for i in range(len(isizes))]
+    f = open(model, 'rb')
+    engine = f.read()
+
+    opr = TensorRTRuntimeOpr(data, engine, 1)
+
+    net = RawNetworkBuilder(inputs=[data], outputs=opr.outputs)
+
+    return net
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(dest = 'model')
+    parser.add_argument(dest = 'output')
+    parser.add_argument('--isize', help='input sizes. '
+            'e.g. for models with two (1,3,224,224) inputs, '
+            'the option --isize="1,3,224,224;1,3,224,224" should be used')
+    
+    args = parser.parse_args()
+    isizes = [str2tuple(x) for x in args.isize.split(';')]
+    net = make_network(args.model, isizes)
+    dump(net, args.output)

lite/load_and_run/examples/script/gen_trt_model.sh

@@ -0,0 +1,33 @@
+#!/bin/bash
+
+CUR_DIR="$( cd "$(dirname $0)" >/dev/null 2>&1 ; pwd -P )"
+
+# find correct trtexec version, only works for internal ci and brainpp env setups
+CUDA_VERSION=$(nvcc --version | grep -o "[0-9].\.[0-9]*" | head -n 1)
+SEARCH_PATH=$(echo `which nvcc | xargs dirname`/../../)
+TRT_CANDIDATE=$(find `cd $SEARCH_PATH; pwd` -name "trtexec" | grep "bin/trtexec" | grep $CUDA_VERSION)
+TRT_CANDIDATE=${TRT_CANDIDATE%$'\n'*}
+TRT_LIB_PATH=$(readlink -f "`dirname $TRT_CANDIDATE`/../lib")
+MODELS_PATH=$(readlink -f "${CUR_DIR}/../model_source")
+
+# generate mge model
+rm -rf $MODELS_PATH/conv_demo_f32_without_data.mge
+python3 ${CUR_DIR}/conv_demo.py --dir $MODELS_PATH
+
+# generate caffe model with mge convert
+# INSTALL mgeconvert:
+# python3 -m pip install git+https://github.com/MegEngine/mgeconvert.git --user --install-option="--targets=caffe"
+rm -rf $MODELS_PATH/conv_demo.prototxt $MODELS_PATH/conv_demo.caffemodel
+convert mge_to_caffe -i $MODELS_PATH/conv_demo_f32_without_data.mge -c $MODELS_PATH/conv_demo.prototxt -b $MODELS_PATH/conv_demo.caffemodel
+
+# generate trt model
+rm -rf $MODELS_PATH/conv_demo.trt
+echo "WARNING: config cudnn and cublas path before run trtexec"
+export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$TRT_LIB_PATH 
+echo $LD_LIBRARY_PATH
+$TRT_CANDIDATE --deploy="$MODELS_PATH/conv_demo.prototxt" --model="$MODELS_PATH/conv_demo.caffemodel" --output="conv1_Convolution" --batch=1 --saveEngine="$MODELS_PATH/conv_demo.trt"
+
+# redump trt model into mgb model
+rm -rf $MODELS_PATH/trt_conv_demo.pkl $MODELS_PATH/trt_conv_demo_with_data.mgb
+python3 $CUR_DIR/dump_trt.py $MODELS_PATH/conv_demo.trt $MODELS_PATH/trt_conv_demo.pkl --isize="1,3,4,4"
+$CUR_DIR/../../dump_with_testcase.py $MODELS_PATH/trt_conv_demo.pkl -o $MODELS_PATH/trt_conv_demo_with_data.mgb -d "#rand(0, 255)" --no-assert

lite/load_and_run/examples/script/mge_input_data.py

@@ -0,0 +1,33 @@
+#!/usr/bin/env python3
+import numpy as np
+import cv2
+import megengine.data.transform as T
+import megengine.functional as F
+import json
+import urllib
+url, filename = ("https://data.megengine.org.cn/images/cat.jpg", "input_data/cat.jpg")
+try: urllib.URLopener().retrieve(url, filename)
+except: urllib.request.urlretrieve(url, filename)
+
+# numpy data
+data=np.random.rand(1,3,224,224)
+np.save("input_data/resnet50_input_uint8.npy",data.astype(np.uint8))
+np.save("input_data/resnet50_input.npy",data.astype(np.float32))
+
+#ppm data
+image = cv2.imread("input_data/cat.jpg")
+transform = T.Compose([
+    T.Resize(256),
+    T.CenterCrop(224),
+])
+processed_img = transform.apply(image)
+cv2.imwrite("input_data/cat.ppm",processed_img)
+
+#json 
+data_obj = {
+    "shape": [1,3],
+    "type": "int32",
+    "raw": [2,3,4]
+}
+with open("input_data/add_demo_input.json", "w") as f:
+     json.dump({"data":data_obj},f)

lite/load_and_run/examples/script/resnet50_mgb.py

@@ -0,0 +1,153 @@
+#!/usr/bin/env python3
+
+import megbrain as mgb
+from megskull.graph import FpropEnv
+import megskull as mgsk
+from megskull.opr.compatible.caffepool import CaffePooling2D
+from megskull.opr.arith import ReLU
+from megskull.opr.all import (
+    DataProvider, Conv2D, Pooling2D, FullyConnected,
+    Softmax, Dropout, BatchNormalization, CrossEntropyLoss,
+    ElementwiseAffine, WarpPerspective, WarpPerspectiveWeightProducer,
+    WeightDecay, ParamProvider, ConvBiasActivation, ElemwiseMultiType)
+from megskull.network import RawNetworkBuilder
+from megskull.utils.debug import CallbackInjector
+import megskull.opr.helper.param_init as pinit
+from megskull.opr.helper.elemwise_trans import Identity
+from megskull.opr.netsrc import DataProvider
+from megskull.opr.cnn import Conv2D, Pooling2D, FullyConnected, Softmax, Conv2DImplHelper
+from megskull.opr.loss import CrossEntropyLoss
+from megskull.opr.regularizer import Dropout, BatchNormalization
+
+from megskull.opr.arith import Add, ReLU
+from megskull.opr.netsrc import ConstProvider
+from megskull.network import RawNetworkBuilder
+import numpy as np
+from megskull.network import RawNetworkBuilder, NetworkVisitor
+from megskull.graph import iter_dep_opr
+from megskull.utils.misc import get_2dshape
+import functools
+import re
+import fnmatch
+import argparse
+import sys
+
+def create_bn_relu_float(conv_name, f_in, ksize, stride, pad, num_outputs,
+                         has_relu, args):
+    f = Conv2D(conv_name, f_in, kernel_shape=ksize, stride=stride,
+               padding=pad, output_nr_channel=num_outputs,
+               nonlinearity=mgsk.opr.helper.elemwise_trans.Identity())
+    if has_relu:
+        f = ReLU(f)
+    return f
+
+
+def get_num_inputs(feature, format):
+    if format == 'NCHW':
+        return feature.partial_shape[1]
+    else:
+        assert format == 'NCHW4'
+        return feature.partial_shape[1] * 4
+
+
+def create_bn_relu(prefix, f_in, ksize, stride, pad, num_outputs,
+                   has_relu, conv_name_fun, args):
+    if conv_name_fun:
+        conv_name = conv_name_fun(prefix)
+    else:
+        conv_name = prefix
+    return create_bn_relu_float(conv_name, f_in, ksize, stride, pad,
+                                    num_outputs, has_relu, args)
+
+
+
+
+def create_bottleneck(prefix, f_in, stride, num_outputs1, num_outputs2, args,
+                      has_proj=False):
+    proj = f_in
+    if has_proj:
+        proj = create_bn_relu(prefix, f_in, ksize=1, stride=stride, pad=0,
+                              num_outputs=num_outputs2, has_relu=False,
+                              conv_name_fun=lambda p: "interstellar{}_branch1".format(
+                                  p), args=args)
+
+    f = create_bn_relu(prefix, f_in, ksize=1, stride=1, pad=0,
+                       num_outputs=num_outputs1, has_relu=True,
+                       conv_name_fun=lambda p: "interstellar{}_branch2a".format(
+                           p), args=args)
+
+    f = create_bn_relu(prefix, f, ksize=3, stride=stride, pad=1,
+                       num_outputs=num_outputs1, has_relu=True,
+                       conv_name_fun=lambda p: "interstellar{}_branch2b".format(
+                           p), args=args)
+
+    f = create_bn_relu(prefix, f, ksize=1, stride=1, pad=0,
+                       num_outputs=num_outputs2, has_relu=False,
+                       conv_name_fun=lambda p: "interstellar{}_branch2c".format(
+                           p), args=args)
+
+    f = ReLU(f + proj)
+
+    return f
+
+
+def get(args):
+    img_size = 224
+    num_inputs = 3
+    data = DataProvider('data', shape=(args.batch_size, num_inputs,
+                                           img_size, img_size))
+
+    inp = data
+    f = create_bn_relu("conv1", inp, ksize=7, stride=2, pad=3, num_outputs=64,
+                       has_relu=True, conv_name_fun=None,
+                       args=args)
+    f = Pooling2D("pool1", f, window=3, stride=2, padding=1, mode="MAX",
+                  format=args.format)
+
+    pre = [2, 3, 4, 5]
+    stages = [3, 4, 6, 3]
+    mid_outputs = [64, 128, 256, 512]
+    enable_stride = [False, True, True, True]
+
+    for p, s, o, es in zip(pre, stages, mid_outputs, enable_stride):
+        for i in range(s):
+            has_proj = False if i > 0 else True
+            stride = 1 if not es or i > 0 else 2
+            prefix = "{}{}".format(p, chr(ord("a") + i))
+            f = create_bottleneck(prefix, f, stride, o, o * 4, args, has_proj)
+            print("{}\t{}".format(prefix, f.partial_shape))
+
+    f = Pooling2D("pool5", f, window=7, stride=7, padding=0, mode="AVERAGE",
+                  format=args.format)
+
+    f = FullyConnected("fc1000", f, output_dim=1000,
+                       nonlinearity=mgsk.opr.helper.elemwise_trans.Identity())
+
+    f = Softmax("cls_softmax", f)
+    f.init_weights()
+
+    net = RawNetworkBuilder(inputs=[data], outputs=[f])
+
+    return net
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(
+        description='dump pkl model for resnet50',
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter)
+    parser.add_argument('-b', '--batch-size',
+                        help='batch size of the model', default=1)
+    parser.add_argument('-f', '--format', choices=['NCHW', 'NCHW4'],
+                        help='format of conv',
+                        default='NCHW')
+    parser.add_argument('-o', '--output',
+                        help='output pkl path', required=True)
+    args = parser.parse_args()
+    if args.format != 'NCHW':
+        print('Only suppprt NCHW for float model')
+        parser.print_help()
+        sys.exit(1)
+
+    from meghair.utils import io
+    io.dump(get(args), args.output)
+

lite/load_and_run/examples/script/resnet50_mge.py

@@ -0,0 +1,351 @@
+#!/usr/bin/env python3
+import argparse
+import math
+
+import megengine.functional as F
+import megengine.hub as hub
+import megengine.module as M
+import numpy as np
+from megengine import jit, tensor
+
+
+class BasicBlock(M.Module):
+    expansion = 1
+
+    def __init__(
+        self,
+        in_channels,
+        channels,
+        stride=1,
+        groups=1,
+        base_width=64,
+        dilation=1,
+        norm=M.BatchNorm2d,
+    ):
+        super().__init__()
+        if groups != 1 or base_width != 64:
+            raise ValueError("BasicBlock only supports groups=1 and base_width=64")
+        if dilation > 1:
+            raise NotImplementedError("Dilation > 1 not supported in BasicBlock")
+        self.conv1 = M.Conv2d(
+            in_channels, channels, 3, stride, padding=dilation, bias=False
+        )
+        self.bn1 = norm(channels)
+        self.conv2 = M.Conv2d(channels, channels, 3, 1, padding=1, bias=False)
+        self.bn2 = norm(channels)
+        self.downsample = (
+            M.Identity()
+            if in_channels == channels and stride == 1
+            else M.Sequential(
+                M.Conv2d(in_channels, channels, 1, stride, bias=False), norm(channels),
+            )
+        )
+
+    def forward(self, x):
+        identity = x
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = F.relu(x)
+        x = self.conv2(x)
+        x = self.bn2(x)
+        identity = self.downsample(identity)
+        x += identity
+        x = F.relu(x)
+        return x
+
+
+class Bottleneck(M.Module):
+    expansion = 4
+
+    def __init__(
+        self,
+        in_channels,
+        channels,
+        stride=1,
+        groups=1,
+        base_width=64,
+        dilation=1,
+        norm=M.BatchNorm2d,
+    ):
+        super().__init__()
+        width = int(channels * (base_width / 64.0)) * groups
+        self.conv1 = M.Conv2d(in_channels, width, 1, 1, bias=False)
+        self.bn1 = norm(width)
+        self.conv2 = M.Conv2d(
+            width,
+            width,
+            3,
+            stride,
+            padding=dilation,
+            groups=groups,
+            dilation=dilation,
+            bias=False,
+        )
+        self.bn2 = norm(width)
+        self.conv3 = M.Conv2d(width, channels * self.expansion, 1, 1, bias=False)
+        self.bn3 = norm(channels * self.expansion)
+        self.downsample = (
+            M.Identity()
+            if in_channels == channels * self.expansion and stride == 1
+            else M.Sequential(
+                M.Conv2d(in_channels, channels * self.expansion, 1, stride, bias=False),
+                norm(channels * self.expansion),
+            )
+        )
+
+    def forward(self, x):
+        identity = x
+
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = F.relu(x)
+
+        x = self.conv2(x)
+        x = self.bn2(x)
+        x = F.relu(x)
+
+        x = self.conv3(x)
+        x = self.bn3(x)
+
+        identity = self.downsample(identity)
+
+        x += identity
+        x = F.relu(x)
+
+        return x
+
+
+class ResNet(M.Module):
+    def __init__(
+        self,
+        block,
+        layers,
+        num_classes=1000,
+        zero_init_residual=False,
+        groups=1,
+        width_per_group=64,
+        replace_stride_with_dilation=None,
+        norm=M.BatchNorm2d,
+    ):
+        super().__init__()
+        self.in_channels = 64
+        self.dilation = 1
+        if replace_stride_with_dilation is None:
+            # each element in the tuple indicates if we should replace
+            # the 2x2 stride with a dilated convolution instead
+            replace_stride_with_dilation = [False, False, False]
+        if len(replace_stride_with_dilation) != 3:
+            raise ValueError(
+                "replace_stride_with_dilation should be None "
+                "or a 3-element tuple, got {}".format(replace_stride_with_dilation)
+            )
+        self.groups = groups
+        self.base_width = width_per_group
+        self.conv1 = M.Conv2d(
+            3, self.in_channels, kernel_size=7, stride=2, padding=3, bias=False
+        )
+        self.bn1 = norm(self.in_channels)
+        self.maxpool = M.MaxPool2d(kernel_size=3, stride=2, padding=0)
+        self.layer1 = self._make_layer(block, 64, layers[0], norm=norm)
+        self.layer2 = self._make_layer(
+            block,
+            128,
+            layers[1],
+            stride=2,
+            dilate=replace_stride_with_dilation[0],
+            norm=norm,
+        )
+        self.layer3 = self._make_layer(
+            block,
+            256,
+            layers[2],
+            stride=2,
+            dilate=replace_stride_with_dilation[1],
+            norm=norm,
+        )
+        self.layer4 = self._make_layer(
+            block,
+            512,
+            layers[3],
+            stride=2,
+            dilate=replace_stride_with_dilation[2],
+            norm=norm,
+        )
+        self.fc = M.Linear(512 * block.expansion, num_classes)
+
+        for m in self.modules():
+            if isinstance(m, M.Conv2d):
+                M.init.msra_normal_(m.weight, mode="fan_out", nonlinearity="relu")
+                if m.bias is not None:
+                    fan_in, _ = M.init.calculate_fan_in_and_fan_out(m.weight)
+                    bound = 1 / math.sqrt(fan_in)
+                    M.init.uniform_(m.bias, -bound, bound)
+            elif isinstance(m, M.BatchNorm2d):
+                M.init.ones_(m.weight)
+                M.init.zeros_(m.bias)
+            elif isinstance(m, M.Linear):
+                M.init.msra_uniform_(m.weight, a=math.sqrt(5))
+                if m.bias is not None:
+                    fan_in, _ = M.init.calculate_fan_in_and_fan_out(m.weight)
+                    bound = 1 / math.sqrt(fan_in)
+                    M.init.uniform_(m.bias, -bound, bound)
+
+        # Zero-initialize the last BN in each residual branch,
+        # so that the residual branch starts with zeros, and each residual block
+        # behaves like an identity. According to https://arxiv.org/abs/1706.02677
+        # This improves the model by 0.2~0.3%.
+        if zero_init_residual:
+            for m in self.modules():
+                if isinstance(m, Bottleneck):
+                    M.init.zeros_(m.bn3.weight)
+                elif isinstance(m, BasicBlock):
+                    M.init.zeros_(m.bn2.weight)
+
+    def _make_layer(
+        self, block, channels, blocks, stride=1, dilate=False, norm=M.BatchNorm2d
+    ):
+        previous_dilation = self.dilation
+        if dilate:
+            self.dilation *= stride
+            stride = 1
+
+        layers = []
+        layers.append(
+            block(
+                self.in_channels,
+                channels,
+                stride,
+                groups=self.groups,
+                base_width=self.base_width,
+                dilation=previous_dilation,
+                norm=norm,
+            )
+        )
+        self.in_channels = channels * block.expansion
+        for _ in range(1, blocks):
+            layers.append(
+                block(
+                    self.in_channels,
+                    channels,
+                    groups=self.groups,
+                    base_width=self.base_width,
+                    dilation=self.dilation,
+                    norm=norm,
+                )
+            )
+
+        return M.Sequential(*layers)
+
+    def extract_features(self, x):
+        outputs = {}
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = F.relu(x)
+        x = self.maxpool(x)
+        outputs["stem"] = x
+
+        x = self.layer1(x)
+        outputs["res2"] = x
+        x = self.layer2(x)
+        outputs["res3"] = x
+        x = self.layer3(x)
+        outputs["res4"] = x
+        x = self.layer4(x)
+        outputs["res5"] = x
+        return outputs
+
+    def forward(self, x):
+        x = F.reshape(x, (1,3,224,224))
+        x = self.extract_features(x)["res5"]
+
+        x = F.avg_pool2d(x, 7)
+        x = F.flatten(x, 1)
+        x = self.fc(x)
+
+        return x
+
+
+
+@hub.pretrained(
+    "https://data.megengine.org.cn/models/weights/resnet50_fbaug_76254_4e14b7d1.pkl"
+)
+def resnet50(**kwargs):
+    r"""ResNet-50 model from
+    `"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_
+    """
+    return ResNet(Bottleneck, [3, 4, 6, 3], **kwargs)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="dump mge model for resnet50",
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter,
+    )
+    parser.add_argument(
+        "-b", "--batch-size", help="batch size of the model", default=1, type=int
+    )
+    parser.add_argument(
+        "-d",
+        "--dtype",
+        help="the dtype of the model,which includes float32 and uint8",
+        default="float32",
+        type=str,
+    )
+    parser.add_argument(
+        "--inputs",
+        help="set the inputs data to get a model with testcase",
+        default="",
+        type=str,
+    )
+
+    parser.add_argument(
+        "--dir",
+        help="set the dir where the model to dump",
+        default=".",
+        type=str,
+    )
+
+    parser.add_argument(
+        "--enable-nchw4",
+        help="enable-nchw4 for NVIDIA CUDNN",
+        action='store_true'
+    )
+
+    parser.add_argument(
+        "--enable-chwn4",
+        help="enable-chwn4 for NVIDIA CUDNN",
+        action='store_true'
+    )
+
+    args = parser.parse_args()
+    net = resnet50()
+    net.eval()
+
+    @jit.trace(symbolic=True, capture_as_const=True)
+    def fun(data):
+        return net(data)
+    
+    if args.dtype == "float32" or args.dtype == "uint8":
+        # dump float32
+        data_type=np.float32
+        if args.dtype == "uint8":
+            data_type =np.uint8
+        data = tensor(
+            (np.random.random([args.batch_size, 3, 224, 224])*255).astype(data_type)
+        )
+        fun(data)
+        if args.inputs == "":
+            fun.dump(
+                
+                args.dir + "/resnet50_b" + str(args.batch_size) + "_"+ args.dtype +"_without_data.mge", arg_names=["data"],
+                no_assert=True, enable_nchw4=args.enable_nchw4, enable_chwn4=args.enable_chwn4,
+                
+            )
+        else:
+            fun.dump(
+                args.dir + "/resnet50_b" + str(args.batch_size) + "_"+ args.dtype +"_with_data.mge", arg_names=["data"],
+                input_data=[args.inputs], no_assert=True, enable_nchw4=args.enable_nchw4,
+            )
+
+    else:
+        raise TypeError("dtype should be float32")

lite/load_and_run/src/options/device_options.cpp

@@ -148,7 +148,9 @@ XPUDeviceOption::XPUDeviceOption() {
     }
 
     if (!FLAGS_multi_thread_core_ids.empty()) {
-        mgb_assert(enable_multithread, "core ids should be set after --multithread");
+        mgb_assert(
+                enable_multithread || enable_multithread_default,
+                "core ids should be set after --multithread or --multithread-default");
         std::stringstream id_stream(FLAGS_multi_thread_core_ids);
         std::string id;
         size_t thread_cnt = 0;

lite/load_and_run/src/options/extern_c_opr_options.cpp

@@ -126,7 +126,9 @@ void COprLibOption::init_extern_param(std::shared_ptr<ModelBase> model_ptr) {
 
 void COprLibOption::load_lib() {
     auto handle = dlopen(lib_path.c_str(), RTLD_LAZY);
-    mgb_assert(handle, "failed to open c opr lib %s: %s", lib_path.c_str(), dlerror());
+    mgb_assert(
+            handle, "failed to open c opr lib %s:\n errmsg: %s", lib_path.c_str(),
+            dlerror());
 
     const char* entry = MGB_C_OPR_INIT_FUNC_STR;
     auto func = dlsym(handle, entry);

lite/load_and_run/src/options/io_options.cpp

@@ -286,8 +286,13 @@ DEFINE_string(
 DEFINE_string(io_dump, "", "set the io dump file path in text format");
 DEFINE_bool(io_dump_stdout, false, "dump io opr to stdout in text format");
 DEFINE_bool(io_dump_stderr, false, "dump io opr to stderr in text format");
-DEFINE_string(bin_io_dump, "", "set the io dump file path in binary format");
-DEFINE_string(bin_out_dump, "", "set the out dump file path in binary format");
+DEFINE_string(
+        bin_io_dump, "",
+        "set the io dump directory path where variable in binary format located");
+DEFINE_string(
+        bin_out_dump, "",
+        "set the out dump directory path where output variable in binary format "
+        "located");
 DEFINE_bool(copy_to_host, false, "copy device data to host");
 
 REGIST_OPTION_CREATOR(input, lar::InputOption::create_option);

lite/load_and_run/src/options/optimize_options.cpp

@@ -77,12 +77,6 @@ void WeightPreprocessOption::config_model_internel<ModelLite>(
         if (weight_preprocess) {
             LITE_WARN("enable weight-preprocess optimization");
             model->get_config().options.weight_preprocess = true;
-            //! FIXME: algo searcher enable weight preprocess for opencl(
-            //! implement below has some problem);
-            // #if MGB_OPENCL
-            //             megdnn::opencl::algo_searcher::AlgoSearcherBase::
-            //                     enable_weight_preprocess();
-            // #endif
         }
     }
 }
@@ -95,11 +89,6 @@ void WeightPreprocessOption::config_model_internel<ModelMdl>(
         if (weight_preprocess) {
             mgb_log_warn("enable weight-preprocess optimization");
             graph_option.graph_opt.enable_weight_preprocess();
-            //! FIXME: this implemment is not right
-            // #if MGB_OPENCL
-            //             megdnn::opencl::algo_searcher::AlgoSearcherBase::
-            //                     enable_weight_preprocess();
-            // #endif
         }
     }
 }
@@ -570,12 +559,6 @@ DEFINE_bool(
         enable_jit, false,
         " Execute supported operators with JIT(now only support NVRTC). "
         "Can only be used on Nvidia GPUs");
-#if MGB_ENABLE_ANDROID_NN
-DEFINE_bool(
-        android_nn, false,
-        "Execute supported operators with Android NN. Can only be used "
-        "with --cpu.");
-#endif
 #if MGB_ENABLE_TENSOR_RT
 DEFINE_bool(
         tensorrt, false,

lite/load_and_run/src/options/plugin_options.cpp

@@ -8,10 +8,10 @@
  */
 
 #include "plugin_options.h"
+#include <map>
 #include "misc.h"
 #include "models/model_lite.h"
 #include "models/model_mdl.h"
-
 ///////////////////// Plugin options///////////////////////////
 namespace lar {
 
@@ -153,7 +153,12 @@ void DebugOption::format_and_print(
     auto table = mgb::TextTable(tablename);
     auto&& network = model->get_lite_network();
     table.padding(1);
-    table.align(mgb::TextTable::Align::Mid).add("type").add("name").add("shape").eor();
+    table.align(mgb::TextTable::Align::Mid)
+            .add("type")
+            .add("name")
+            .add("shape")
+            .add("dtype")
+            .eor();
 
     auto to_string = [&](lite::Layout& layout) {
         std::string shape("{");
@@ -165,6 +170,19 @@ void DebugOption::format_and_print(
         shape.append("}");
         return shape;
     };
+    auto get_dtype = [&](lite::Layout& layout) {
+        std::map<LiteDataType, std::string> type_map = {
+                {LiteDataType::LITE_FLOAT, "float32"},
+                {LiteDataType::LITE_HALF, "float16"},
+                {LiteDataType::LITE_INT64, "int64"},
+                {LiteDataType::LITE_INT, "int32"},
+                {LiteDataType::LITE_UINT, "uint32"},
+                {LiteDataType::LITE_INT16, "int16"},
+                {LiteDataType::LITE_UINT16, "uint16"},
+                {LiteDataType::LITE_INT8, "int8"},
+                {LiteDataType::LITE_UINT8, "uint8"}};
+        return type_map[layout.data_type];
+    };
 
     auto input_name = network->get_all_input_name();
     for (auto& i : input_name) {
@@ -173,6 +191,7 @@ void DebugOption::format_and_print(
                 .add("INPUT")
                 .add(i)
                 .add(to_string(layout))
+                .add(get_dtype(layout))
                 .eor();
     }
 
@@ -183,6 +202,7 @@ void DebugOption::format_and_print(
                 .add("OUTPUT")
                 .add(i)
                 .add(to_string(layout))
+                .add(get_dtype(layout))
                 .eor();
     }
 
@@ -196,13 +216,28 @@ void DebugOption::format_and_print(
         const std::string& tablename, std::shared_ptr<ModelMdl> model) {
     auto table = mgb::TextTable(tablename);
     table.padding(1);
-    table.align(mgb::TextTable::Align::Mid).add("type").add("name").add("shape").eor();
-
+    table.align(mgb::TextTable::Align::Mid)
+            .add("type")
+            .add("name")
+            .add("shape")
+            .add("dtype")
+            .eor();
+    auto get_dtype = [&](megdnn::DType data_type) {
+        std::map<megdnn::DTypeEnum, std::string> type_map = {
+                {mgb::dtype::Float32().enumv(), "float32"},
+                {mgb::dtype::Int32().enumv(), "int32"},
+                {mgb::dtype::Int16().enumv(), "int16"},
+                {mgb::dtype::Uint16().enumv(), "uint16"},
+                {mgb::dtype::Int8().enumv(), "int8"},
+                {mgb::dtype::Uint8().enumv(), "uint8"}};
+        return type_map[data_type.enumv()];
+    };
     for (auto&& i : model->get_mdl_load_result().tensor_map) {
         table.align(mgb::TextTable::Align::Mid)
                 .add("INPUT")
                 .add(i.first)
                 .add(i.second->shape().to_string())
+                .add(get_dtype(i.second->dtype()))
                 .eor();
     }
 
@@ -211,6 +246,7 @@ void DebugOption::format_and_print(
                 .add("OUTPUT")
                 .add(i.node()->name())
                 .add(i.shape().to_string())
+                .add(get_dtype(i.dtype()))
                 .eor();
     }
 
@@ -358,18 +394,22 @@ DEFINE_double(
 
 DEFINE_bool(
         check_dispatch, false,
-        "check whether an operator call dispatch on cpu comp nodes");
+        "check whether an operator call dispatch on cpu comp nodes This is used to "
+        "find potential bugs in MegDNN");
 
 DEFINE_string(
         check_var_value, "",
         "--check-var-value [interval]|[interval:init_idx], Enable "
-        "VarValueChecker plugin. Refer to its doc for more details");
+        "VarValueChecker plugin. check values of all vars in a graph from given var "
+        "ID(init_idx) with step interval");
 #if MGB_ENABLE_JSON
 DEFINE_string(
         profile, "",
         "Write profiling result to given file. The output file is in "
         "JSON format");
-DEFINE_string(profile_host, "", "focus on host time profiling For some backends");
+DEFINE_string(
+        profile_host, "",
+        "focus on host time profiling For some backends(such as openCL)");
 #endif
 
 ///////////////////// Debug gflags///////////////////////////

lite/load_and_run/src/strategys/strategy_normal.cpp

@@ -94,15 +94,16 @@ void NormalStrategy::run_subline() {
             m_runtime_param.stage = RunStage::AFTER_RUNNING_WAIT;
             stage_config_model();
             auto cur = timer.get_msecs();
-            printf("iter %lu/%lu: %.3fms (exec=%.3fms)\n", i, run_num, cur, exec_time);
+            printf("iter %lu/%lu: %.3f ms (exec=%.3f ms)\n", i, run_num, cur,
+                   exec_time);
             time_sum += cur;
             time_sqrsum += cur * cur;
             fflush(stdout);
             min_time = std::min(min_time, cur);
             max_time = std::max(max_time, cur);
         }
-        printf("\n=== finished test #%u: time=%.3fms avg_time=%.3fms "
-               "sexec=%.3fms min=%.3fms max=%.3fms\n\n",
+        printf("\n=== finished test #%u: time=%.3f ms avg_time=%.3f ms "
+               "sexec=%.3f ms min=%.3f ms max=%.3f ms\n\n",
                idx, time_sum, time_sum / run_num,
                std::sqrt(
                        (time_sqrsum * run_num - time_sum * time_sum) /
@@ -121,7 +122,7 @@ void NormalStrategy::run_subline() {
         m_runtime_param.stage = RunStage::MODEL_RUNNING;
         stage_config_model();
 
-        if (!idx) {
+        if (idx == 0) {
             warm_up();
         }
         tot_time += run_iter(idx);