Update the xgboost to latest version

Intel has optimized the xgboost in latest version. Performance has
obvious acceleration in benchmark. In svm training with Yahoo LTR data,
latest version gets 7.8X acceleration in host, and 2X acceleration in occlum.

Signed-off-by: yuanwu <yuan.wu@intel.com>

demos/xgboost/download_and_build_xgboost.sh

@@ -9,11 +9,14 @@ apt-get update
 apt-get install -y python3-pip python3-setuptools
 pip3 install kubernetes
 
+#install the cmake
+./install_cmake.sh
+
 # Download and build XGBoost
 rm -rf xgboost_src && mkdir xgboost_src
 pushd xgboost_src
 git clone https://github.com/dmlc/xgboost .
-git checkout 6d5b34d82486cd1d0480c548f5d1953834659bd6
+git checkout 9e955fb9b06cac32a06c92c4715f749d9d87e932
 git submodule init
 git submodule update
 git apply ../patch/xgboost-01.diff

demos/xgboost/install_cmake.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+set -e
+cd ~
+wget https://github.com/Kitware/CMake/releases/download/v3.15.5/cmake-3.15.5.tar.gz && tar xf cmake-3.15.5.tar.gz
+cd cmake-3.15.5
+./bootstrap
+make -j4
+sudo make install

demos/xgboost/patch/dmlc-core-01.diff

@@ -1,18 +1,18 @@
 diff --git a/tracker/dmlc_tracker/local.py b/tracker/dmlc_tracker/local.py
-index dff7c17..fca0f9d 100644
+index 6e4af12..09df15e 100644
 --- a/tracker/dmlc_tracker/local.py
 +++ b/tracker/dmlc_tracker/local.py
-@@ -24,6 +24,13 @@ def exec_cmd(cmd, role, taskid, pass_env):
+@@ -26,6 +26,13 @@ def exec_cmd(cmd, num_attempt, role, taskid, pass_env):
- 
+     num_retry = env.get('DMLC_NUM_ATTEMPT', num_attempt)
-     num_retry = env.get('DMLC_NUM_ATTEMPT', 0)
+     num_trial = 0
  
 +    cmd_str = ''
 +    for k, v in env.items():
 +        if str(k)[0:4] == 'DMLC':
 +            strenv = str(k) + '=' + str(v);
 +            cmd_str = cmd_str + ' ' + strenv
-+    cmd = cmd + cmd_str
++    cmdline = cmdline + ' ' + cmd_str
 +
+     logging.debug('num of retry %d',num_retry)
+ 
      while True:
-         if os.name == 'nt':
-             ret = subprocess.call(cmd, shell=True, env=env)

demos/xgboost/patch/rabit-01.diff

@@ -1,8 +1,8 @@
-diff --git a/src/allreduce_base.cc b/src/allreduce_base.cc
+diff --git a/rabit/src/allreduce_base.cc b/rabit/src/allreduce_base.cc
-index 143db6e..a6daf20 100644
+index d1959eaa..68cd377a 100644
---- a/src/allreduce_base.cc
+--- a/rabit/src/allreduce_base.cc
-+++ b/src/allreduce_base.cc
++++ b/rabit/src/allreduce_base.cc
-@@ -486,12 +486,13 @@ AllreduceBase::TryAllreduceTree(void *sendrecvbuf_,
+@@ -551,12 +551,13 @@ AllreduceBase::TryAllreduceTree(void *sendrecvbuf_,
      // select must return
      watcher.Poll();
      // exception handling
@@ -18,7 +18,7 @@ index 143db6e..a6daf20 100644
      // read data from childs
      for (int i = 0; i < nlink; ++i) {
        if (i != parent_index && watcher.CheckRead(links[i].sock)) {
-@@ -641,12 +642,13 @@ AllreduceBase::TryBroadcast(void *sendrecvbuf_, size_t total_size, int root) {
+@@ -731,12 +732,13 @@ AllreduceBase::TryBroadcast(void *sendrecvbuf_, size_t total_size, int root) {
      // select
      watcher.Poll();
      // exception handling

demos/xgboost/patch/xgboost-01.diff

@@ -1,10 +1,10 @@
 diff --git a/src/cli_main.cc b/src/cli_main.cc
-index faa93ad..a201615 100644
+index 5c602f37..d18a77c2 100644
 --- a/src/cli_main.cc
 +++ b/src/cli_main.cc
-@@ -339,6 +339,16 @@ int CLIRunTask(int argc, char *argv[]) {
+@@ -444,6 +444,16 @@ class CLI {
-     printf("Usage: <config>\n");
+       this->PrintHelp();
-     return 0;
+       exit(1);
      }
 +
 +    // FIXME:workaroud for local distributed machine learning on Occlum
@@ -16,6 +16,6 @@ index faa93ad..a201615 100644
 +      }
 +    }
 +
-   rabit::Init(argc, argv);
+     for (int i = 0; i < argc; ++i) {
- 
+       std::string str {argv[i]};
-   std::vector<std::pair<std::string, std::string> > cfg;
+       if (str == "-h" || str == "--help") {