update

mllib-dal/src/main/native/GPU.cpp

@@ -25,6 +25,7 @@ std::vector<sycl::device> get_gpus() {
 }
 
 static int getLocalRank(ccl::communicator &comm, int size, int rank) {
+    const int MPI_MAX_PROCESSOR_NAME = 128;
     /* Obtain local rank among nodes sharing the same host name */
     char zero = static_cast<char>(0);
     std::vector<char> name(MPI_MAX_PROCESSOR_NAME + 1, zero);
@@ -112,45 +113,3 @@ sycl::queue getQueue(const ComputeDevice device) {
     }
     }
 }
-
-preview::spmd::communicator<preview::spmd::device_memory_access::usm>
-createDalCommunicator(const jint executorNum, const jint rank,
-                      const ccl::string ccl_ip_port) {
-    auto gpus = get_gpus();
-
-    auto t1 = std::chrono::high_resolution_clock::now();
-
-    ccl::init();
-
-    auto t2 = std::chrono::high_resolution_clock::now();
-    auto duration =
-        (float)std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1)
-            .count();
-
-    logger::println(logger::INFO, "OneCCL singleton init took %f secs",
-                    duration / 1000);
-
-    t1 = std::chrono::high_resolution_clock::now();
-
-    auto kvs_attr = ccl::create_kvs_attr();
-
-    kvs_attr.set<ccl::kvs_attr_id::ip_port>(ccl_ip_port);
-
-    ccl::shared_ptr_class<ccl::kvs> kvs = ccl::create_main_kvs(kvs_attr);
-
-    t2 = std::chrono::high_resolution_clock::now();
-    duration =
-        (float)std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1)
-            .count();
-    logger::println(logger::INFO, "OneCCL (native): create kvs took %f secs",
-                    duration / 1000);
-    sycl::queue queue{gpus[0]};
-    t1 = std::chrono::high_resolution_clock::now();
-    auto comm = preview::spmd::make_communicator<preview::spmd::backend::ccl>(
-        queue, executorNum, rank, kvs);
-    t2 = std::chrono::high_resolution_clock::now();
-    duration =
-        (float)std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1)
-            .count();
-    return comm;
-}

mllib-dal/src/main/native/GPU.h

@@ -6,8 +6,6 @@
 #include <jni.h>
 #include <oneapi/ccl.hpp>
 
-#include "Communicator.hpp"
-
 sycl::queue getAssignedGPU(const ComputeDevice device, jint *gpu_indices);
 
 sycl::queue getQueue(const ComputeDevice device);

mllib-dal/src/main/native/Profile.hpp

@@ -11,8 +11,7 @@ class Profiler {
 
     void startProfile(std::string s = "") {
         action = s;
-        logger::println(logger::INFO, "%s (native): start %s", subject.c_str(),
-                        action.c_str());
+	logger::println(logger::INFO, "%s (native): start %s", subject.c_str(), action.c_str());
         startTime = std::chrono::high_resolution_clock::now();
     }
 
@@ -21,14 +20,11 @@ class Profiler {
         auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(
                             end_time - startTime)
                             .count();
-        logger::println(logger::INFO, "%s (native): start %s took %f secs",
-                        subject.c_str(), action.c_str(),
-                        (float)duration / 1000);
+	logger::println(logger::INFO, "%s (native): start %s took %f secs", subject.c_str(), action.c_str(), (float)duration / 1000);
     }
 
     void println(std::string msg) {
-        logger::println(logger::INFO, "%s (native): %s", subject.c_str(),
-                        msg.c_str());
+	logger::println(logger::INFO, "%s (native): %s", subject.c_str(), msg.c_str());
     }
 
   private: