Update RCCL-tests in examples folder (#261)

- Create a local copy for ROCm/rccl-tests for our examples.
- Update argument parsing to no longer use getopt_long.
- Workaround for Dyninst instrumentation.

---------

Co-authored-by: David Galiffi <David.Galiffi@amd.com>

[ROCm/rocprofiler-systems commit: 4e5029221b]

projects/rocprofiler-systems/examples/rccl/rccl-tests/src/gather.cpp

@@ -0,0 +1,143 @@
+/*************************************************************************
+ * Copyright (c) 2016-2022, NVIDIA CORPORATION. All rights reserved.
+ * Modifications Copyright (c) 2019-2022 Advanced Micro Devices, Inc. All rights reserved.
+ *
+ * See LICENSE.txt for license information
+ ************************************************************************/
+
+#include "common.h"
+#include "cuda_runtime.h"
+
+void
+GatherGetCollByteCount(size_t* sendcount, size_t* recvcount, size_t* paramcount,
+                       size_t* sendInplaceOffset, size_t* recvInplaceOffset, size_t count,
+                       size_t eltSize, int nranks)
+{
+    *sendcount         = (count / nranks) & -(16 / eltSize);
+    *recvcount         = (*sendcount) * nranks;
+    *sendInplaceOffset = *sendcount;
+    *recvInplaceOffset = 0;
+    *paramcount        = *sendcount;
+}
+
+testResult_t
+GatherInitData(struct threadArgs* args, ncclDataType_t type, ncclRedOp_t op, int root,
+               int rep, int in_place)
+{
+    size_t sendcount = args->sendBytes / wordSize(type);
+    size_t recvcount = args->expectedBytes / wordSize(type);
+    int    nranks    = args->nProcs * args->nThreads * args->nGpus;
+
+    for(int i = 0; i < args->nGpus; i++)
+    {
+        CUDACHECK(cudaSetDevice(args->gpus[i]));
+        int rank = ((args->proc * args->nThreads + args->thread) * args->nGpus + i);
+        CUDACHECK(cudaMemset(args->recvbuffs[i], 0, args->expectedBytes));
+        void* data = in_place ? ((char*) args->recvbuffs[i]) + rank * args->sendBytes
+                              : args->sendbuffs[i];
+        TESTCHECK(InitData(data, sendcount, rank * sendcount, type, ncclSum, rep, 1, 0));
+        CUDACHECK(cudaMemcpy(args->expected[i], args->recvbuffs[i], args->expectedBytes,
+                             cudaMemcpyDefault));
+        if(rank == root)
+        {
+            TESTCHECK(InitData(args->expected[i], nranks * sendcount, 0, type, ncclSum,
+                               rep, 1, 0));
+        }
+        CUDACHECK(cudaDeviceSynchronize());
+    }
+    return testSuccess;
+}
+
+void
+GatherGetBw(size_t count, int typesize, double sec, double* algBw, double* busBw,
+            int nranks)
+{
+    double baseBw = (double) (count * nranks * typesize) / 1.0E9 / sec;
+
+    *algBw        = baseBw;
+    double factor = ((double) (nranks - 1)) / ((double) (nranks));
+    *busBw        = baseBw * factor;
+}
+
+testResult_t
+GatherRunColl(void* sendbuff, void* recvbuff, size_t count, ncclDataType_t type,
+              ncclRedOp_t op, int root, ncclComm_t comm, cudaStream_t stream)
+{
+    int nRanks;
+    NCCLCHECK(ncclCommCount(comm, &nRanks));
+    int rank;
+    NCCLCHECK(ncclCommUserRank(comm, &rank));
+    size_t rankOffset = count * wordSize(type);
+    if(count == 0) return testSuccess;
+
+    NCCLCHECK(ncclGroupStart());
+    NCCLCHECK(ncclSend(sendbuff, count, type, root, comm, stream));
+    if(rank == root)
+    {
+        for(int r = 0; r < nRanks; r++)
+        {
+            NCCLCHECK(ncclRecv(((char*) recvbuff) + r * rankOffset, count, type, r, comm,
+                               stream));
+        }
+    }
+    NCCLCHECK(ncclGroupEnd());
+
+    return testSuccess;
+}
+
+struct testColl gatherTest = { "Gather", GatherGetCollByteCount, GatherInitData,
+                               GatherGetBw, GatherRunColl };
+
+void
+GatherGetBuffSize(size_t* sendcount, size_t* recvcount, size_t count, int nranks)
+{
+    size_t paramcount, sendInplaceOffset, recvInplaceOffset;
+    GatherGetCollByteCount(sendcount, recvcount, &paramcount, &sendInplaceOffset,
+                           &recvInplaceOffset, count, /*eltSize=*/1, nranks);
+}
+
+testResult_t
+GatherRunTest(struct threadArgs* args, int root, ncclDataType_t type,
+              const char* typeName, ncclRedOp_t op, const char* opName)
+{
+    args->collTest = &gatherTest;
+    ncclDataType_t* run_types;
+    const char**    run_typenames;
+    int             type_count;
+    int             begin_root, end_root;
+
+    if((int) type != -1)
+    {
+        type_count    = 1;
+        run_types     = &type;
+        run_typenames = &typeName;
+    }
+    else
+    {
+        type_count    = test_typenum;
+        run_types     = test_types;
+        run_typenames = test_typenames;
+    }
+
+    if(root != -1)
+    {
+        begin_root = end_root = root;
+    }
+    else
+    {
+        begin_root = 0;
+        end_root   = args->nProcs * args->nThreads * args->nGpus - 1;
+    }
+
+    for(int i = 0; i < type_count; i++)
+    {
+        for(int j = begin_root; j <= end_root; j++)
+        {
+            TESTCHECK(TimeTest(args, run_types[i], run_typenames[i], (ncclRedOp_t) 0,
+                               "none", j));
+        }
+    }
+    return testSuccess;
+}
+
+struct testEngine ncclTestEngine = { GatherGetBuffSize, GatherRunTest };