Update rocm-rel-6.4 to use TransferBench v1.62.00 (#183)

* updating metadata

* v1.58.00 Fixing DMA copy-on-engine (#152)

* Leo's review

* Update use-transferbench.rst

* Update Doxyfile

* Refining API library

* Update TransferBench.hpp

* Update TransferBench.hpp

* Update TransferBench.hpp

* Bump rocm-docs-core from 1.9.2 to 1.11.0 in /docs/sphinx (#153)

Bumps [rocm-docs-core](https://github.com/ROCm/rocm-docs-core) from 1.9.2 to 1.11.0.
- [Release notes](https://github.com/ROCm/rocm-docs-core/releases)
- [Changelog](https://github.com/ROCm/rocm-docs-core/blob/develop/CHANGELOG.md)
- [Commits](https://github.com/ROCm/rocm-docs-core/compare/v1.9.2...v1.11.0

)

---
updated-dependencies:
- dependency-name: rocm-docs-core
  dependency-type: direct:production
  update-type: version-update:semver-minor
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* Bump rocm-docs-core from 1.11.0 to 1.12.0 in /docs/sphinx (#155)

Bumps [rocm-docs-core](https://github.com/ROCm/rocm-docs-core) from 1.11.0 to 1.12.0.
- [Release notes](https://github.com/ROCm/rocm-docs-core/releases)
- [Changelog](https://github.com/ROCm/rocm-docs-core/blob/develop/CHANGELOG.md)
- [Commits](https://github.com/ROCm/rocm-docs-core/compare/v1.11.0...v1.12.0

)

---
updated-dependencies:
- dependency-name: rocm-docs-core
  dependency-type: direct:production
  update-type: version-update:semver-minor
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* External CI: enable CI triggers

* Apply suggestions from code review
Co-authored-by: Mustafa Abduljabbar <mustafa.abduljabbar@amd.com>

* Update LICENSE.md

* Bump rocm-docs-core from 1.12.0 to 1.13.0 in /docs/sphinx (#160)

Bumps [rocm-docs-core](https://github.com/ROCm/rocm-docs-core) from 1.12.0 to 1.13.0.
- [Release notes](https://github.com/ROCm/rocm-docs-core/releases)
- [Changelog](https://github.com/ROCm/rocm-docs-core/blob/develop/CHANGELOG.md)
- [Commits](https://github.com/ROCm/rocm-docs-core/compare/v1.12.0...v1.13.0

)

---
updated-dependencies:
- dependency-name: rocm-docs-core
  dependency-type: direct:production
  update-type: version-update:semver-minor
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* TransferBench V1.59 (#162)

Adding NIC execution capabilities, various bug fixes introduced by header-only-library refactor
---------
Co-authored-by: Mustafa Abduljabbar <mustafa.abduljabbar@amd.com>

* Adding ability to specify A2A_MODE=numSrcs:numDsts (#164)
Co-authored-by: Mustafa Abduljabbar <mustafa.abduljabbar@amd.com>

* Fixing specific DMA engine transfers, enabling GFX_SINGLE_TEAM=1 by default (#166)

* Bump rocm-docs-core from 1.13.0 to 1.15.0 in /docs/sphinx (#165)

Bumps [rocm-docs-core](https://github.com/ROCm/rocm-docs-core) from 1.13.0 to 1.15.0.
- [Release notes](https://github.com/ROCm/rocm-docs-core/releases)
- [Changelog](https://github.com/ROCm/rocm-docs-core/blob/develop/CHANGELOG.md)
- [Commits](https://github.com/ROCm/rocm-docs-core/compare/v1.13.0...v1.15.0

)

---
updated-dependencies:
- dependency-name: rocm-docs-core
  dependency-type: direct:production
  update-type: version-update:semver-minor
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* Bump rocm-docs-core from 1.15.0 to 1.17.0 in /docs/sphinx (#171)

Bumps [rocm-docs-core](https://github.com/ROCm/rocm-docs-core) from 1.15.0 to 1.17.0.
- [Release notes](https://github.com/ROCm/rocm-docs-core/releases)
- [Changelog](https://github.com/ROCm/rocm-docs-core/blob/develop/CHANGELOG.md)
- [Commits](https://github.com/ROCm/rocm-docs-core/compare/v1.15.0...v1.17.0

)

---
updated-dependencies:
- dependency-name: rocm-docs-core
  dependency-type: direct:production
  update-type: version-update:semver-minor
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* TransferBench v1.61 (#174)
Co-authored-by: Mustafa Abduljabbar <mustafa.abduljabbar@amd.com>

* Bump rocm-docs-core from 1.17.0 to 1.18.1 in /docs/sphinx (#176)

Bumps [rocm-docs-core](https://github.com/ROCm/rocm-docs-core) from 1.17.0 to 1.18.1.
- [Release notes](https://github.com/ROCm/rocm-docs-core/releases)
- [Changelog](https://github.com/ROCm/rocm-docs-core/blob/develop/CHANGELOG.md)
- [Commits](https://github.com/ROCm/rocm-docs-core/compare/v1.17.0...v1.18.1

)

---
updated-dependencies:
- dependency-name: rocm-docs-core
  dependency-type: direct:production
  update-type: version-update:semver-minor
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* Bump rocm-docs-core from 1.18.1 to 1.18.2 in /docs/sphinx (#177)

Bumps [rocm-docs-core](https://github.com/ROCm/rocm-docs-core) from 1.18.1 to 1.18.2.
- [Release notes](https://github.com/ROCm/rocm-docs-core/releases)
- [Changelog](https://github.com/ROCm/rocm-docs-core/blob/develop/CHANGELOG.md)
- [Commits](https://github.com/ROCm/rocm-docs-core/compare/v1.18.1...v1.18.2

)

---
updated-dependencies:
- dependency-name: rocm-docs-core
  dependency-version: 1.18.2
  dependency-type: direct:production
  update-type: version-update:semver-patch
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* TransferBench v1.62.00 (#181)

* Adding non-temporal loads and stores via GFX_TEMPORAL
* Adding additional summary details to a2a preset
* Add SHOW_MIN_ONLY for a2asweep preset
* Adding new P CPU memory type which is indexed by closest GPU

* Bump rocm-docs-core from 1.18.2 to 1.20.0 in /docs/sphinx (#180)

Bumps [rocm-docs-core](https://github.com/ROCm/rocm-docs-core) from 1.18.2 to 1.20.0.
- [Release notes](https://github.com/ROCm/rocm-docs-core/releases)
- [Changelog](https://github.com/ROCm/rocm-docs-core/blob/develop/CHANGELOG.md)
- [Commits](https://github.com/ROCm/rocm-docs-core/compare/v1.18.2...v1.20.0

)

---
updated-dependencies:
- dependency-name: rocm-docs-core
  dependency-version: 1.20.0
  dependency-type: direct:production
  update-type: version-update:semver-minor
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

---------
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: srawat <120587655+SwRaw@users.noreply.github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Daniel Su <danielsu@amd.com>
Co-authored-by: Mustafa Abduljabbar <mustafa.abduljabbar@amd.com>

src/client/Presets/Presets.hpp

@@ -24,6 +24,8 @@ THE SOFTWARE.
 
 // Included after EnvVars and Executors
 #include "AllToAll.hpp"
+#include "AllToAllN.hpp"
+#include "AllToAllSweep.hpp"
 #include "HealthCheck.hpp"
 #include "OneToAll.hpp"
 #include "PeerToPeer.hpp"
@@ -38,14 +40,16 @@ typedef void (*PresetFunc)(EnvVars&          ev,
 
 std::map<std::string, std::pair<PresetFunc, std::string>> presetFuncMap =
 {
-  {"a2a",         {AllToAllPreset,   "Tests parallel transfers between all pairs of GPU devices"}},
-  {"healthcheck", {HealthCheckPreset,"Simple bandwidth health check (MI300X series only)"}},
-  {"one2all",     {OneToAllPreset,   "Test all subsets of parallel transfers from one GPU to all others"}},
-  {"p2p"   ,      {PeerToPeerPreset, "Peer-to-peer device memory bandwidth test"}},
-  {"rsweep",      {SweepPreset,      "Randomly sweep through sets of Transfers"}},
-  {"scaling",     {ScalingPreset,    "Run scaling test from one GPU to other devices"}},
-  {"schmoo",      {SchmooPreset,     "Scaling tests for local/remote read/write/copy"}},
-  {"sweep",       {SweepPreset,      "Ordered sweep through sets of Transfers"}},
+  {"a2a",         {AllToAllPreset,      "Tests parallel transfers between all pairs of GPU devices"}},
+  {"a2a_n",       {AllToAllRdmaPreset,  "Tests parallel transfers between all pairs of GPU devices using Nearest NIC RDMA transfers"}},
+  {"a2asweep",    {AllToAllSweepPreset, "Test GFX-based all-to-all transfers swept across different CU and GFX unroll counts"}},
+  {"healthcheck", {HealthCheckPreset,   "Simple bandwidth health check (MI300X series only)"}},
+  {"one2all",     {OneToAllPreset,      "Test all subsets of parallel transfers from one GPU to all others"}},
+  {"p2p"   ,      {PeerToPeerPreset,    " Peer-to-peer device memory bandwidth test"}},
+  {"rsweep",      {SweepPreset,         "Randomly sweep through sets of Transfers"}},
+  {"scaling",     {ScalingPreset,       "Run scaling test from one GPU to other devices"}},
+  {"schmoo",      {SchmooPreset,        "Scaling tests for local/remote read/write/copy"}},
+  {"sweep",       {SweepPreset,         "Ordered sweep through sets of Transfers"}},
 };
 
 void DisplayPresets()

src/client/Presets/Sweep.hpp

@@ -22,19 +22,21 @@ THE SOFTWARE.
 
 void LogTransfers(FILE *fp, int const testNum, std::vector<Transfer> const& transfers)
 {
-  fprintf(fp, "# Test %d\n", testNum);
-  fprintf(fp, "%d", -1 * (int)transfers.size());
-  for (auto const& transfer : transfers)
-  {
-    fprintf(fp, " (%s->%c%d->%s %d %lu)",
-            MemDevicesToStr(transfer.srcs).c_str(),
-            ExeTypeStr[transfer.exeDevice.exeType], transfer.exeDevice.exeIndex,
-            MemDevicesToStr(transfer.dsts).c_str(),
-            transfer.numSubExecs,
-            transfer.numBytes);
+  if (fp) {
+    fprintf(fp, "# Test %d\n", testNum);
+    fprintf(fp, "%d", -1 * (int)transfers.size());
+    for (auto const& transfer : transfers)
+    {
+      fprintf(fp, " (%s->%c%d->%s %d %lu)",
+              MemDevicesToStr(transfer.srcs).c_str(),
+              ExeTypeStr[transfer.exeDevice.exeType], transfer.exeDevice.exeIndex,
+              MemDevicesToStr(transfer.dsts).c_str(),
+              transfer.numSubExecs,
+              transfer.numBytes);
+    }
+    fprintf(fp, "\n");
+    fflush(fp);
   }
-  fprintf(fp, "\n");
-  fflush(fp);
 }
 
 void SweepPreset(EnvVars&           ev,
@@ -54,6 +56,7 @@ void SweepPreset(EnvVars&           ev,
   int         numGpuSubExecs = EnvVars::GetEnvVar("NUM_GPU_SE"          , 4);
   std::string sweepDst       = EnvVars::GetEnvVar("SWEEP_DST"           , "CG");
   std::string sweepExe       = EnvVars::GetEnvVar("SWEEP_EXE"           , "CDG");
+  std::string sweepFile      = EnvVars::GetEnvVar("SWEEP_FILE"          , "/tmp/lastSweep.cfg");
   int         sweepMax       = EnvVars::GetEnvVar("SWEEP_MAX"           , 24);
   int         sweepMin       = EnvVars::GetEnvVar("SWEEP_MIN"           , 1);
   int         sweepRandBytes = EnvVars::GetEnvVar("SWEEP_RAND_BYTES"    , 0);
@@ -78,6 +81,7 @@ void SweepPreset(EnvVars&           ev,
     ev.Print("NUM_GPU_SE",        numGpuSubExecs,   "Using %d subExecutors/CUs per GPU executed Transfer", numGpuSubExecs);
     ev.Print("SWEEP_DST",         sweepDst.c_str(), "Destination Memory Types to sweep");
     ev.Print("SWEEP_EXE",         sweepExe.c_str(), "Executor Types to sweep");
+    ev.Print("SWEEP_FILE",        sweepFile.c_str(),"File to store the executing sweep configuration");
     ev.Print("SWEEP_MAX",         sweepMax,         "Max simultaneous transfers (0 = no limit)");
     ev.Print("SWEEP_MIN",         sweepMin,         "Min simultaenous transfers");
     ev.Print("SWEEP_RAND_BYTES",  sweepRandBytes,   "Using %s number of bytes per Transfer", (sweepRandBytes ? "random" : "constant"));
@@ -283,10 +287,14 @@ void SweepPreset(EnvVars&           ev,
   std::uniform_int_distribution<int> distribution(sweepMin, maxParallelTransfers);
 
   // Log sweep to configuration file
-  FILE *fp = fopen("lastSweep.cfg", "w");
+  char absPath[1024];
+  auto const res = realpath(sweepFile.c_str(), absPath);
+
+  FILE *fp = fopen(sweepFile.c_str(), "w");
   if (!fp) {
-    printf("[ERROR] Unable to open lastSweep.cfg.  Check permissions\n");
-    exit(1);
+    printf("[WARN] Unable to open %s.  Skipping output of sweep configuration file\n", res ? absPath : sweepFile.c_str());
+  } else {
+    printf("Sweep configuration saved to: %s\n", res ? absPath : sweepFile.c_str());
   }
 
   // Create bitmask of numPossible triplets, of which M will be chosen
@@ -333,7 +341,7 @@ void SweepPreset(EnvVars&           ev,
 
     // Check for test limit
     if (numTestsRun == sweepTestLimit) {
-      printf("Test limit reached\n");
+      printf("Sweep Test limit reached\n");
       break;
     }
 
@@ -341,7 +349,7 @@ void SweepPreset(EnvVars&           ev,
     auto cpuDelta = std::chrono::high_resolution_clock::now() - cpuStart;
     double totalCpuTime = std::chrono::duration_cast<std::chrono::duration<double>>(cpuDelta).count();
     if (sweepTimeLimit && totalCpuTime > sweepTimeLimit) {
-      printf("Time limit exceeded\n");
+      printf("Sweep Time limit exceeded\n");
       break;
     }
 
@@ -357,5 +365,5 @@ void SweepPreset(EnvVars&           ev,
         bitmask[i] = (i < M) ? 1 : 0;
     }
   }
-  fclose(fp);
+  if (fp) fclose(fp);
 }

src/client/Topology.hpp

@@ -38,21 +38,56 @@ static int RemappedCpuIndex(int origIdx)
   return remappingCpu[origIdx];
 }
 
+static void PrintNicToGPUTopo(bool outputToCsv)
+{
+#ifdef NIC_EXEC_ENABLED
+  printf(" NIC | Device Name | Active | PCIe Bus ID  | NUMA | Closest GPU(s) | GID Index | GID Descriptor\n");
+  if(!outputToCsv)
+    printf("-----+-------------+--------+--------------+------+----------------+-----------+-------------------\n");
+
+  int numGpus = TransferBench::GetNumExecutors(EXE_GPU_GFX);
+  auto const& ibvDeviceList = GetIbvDeviceList();
+  for (int i = 0; i < ibvDeviceList.size(); i++) {
+
+    std::string closestGpusStr = "";
+    for (int j = 0; j < numGpus; j++) {
+      if (TransferBench::GetClosestNicToGpu(j) == i) {
+        if (closestGpusStr != "") closestGpusStr += ",";
+        closestGpusStr += std::to_string(j);
+      }
+    }
+
+    printf(" %-3d | %-11s | %-6s | %-12s | %-4d | %-14s | %-9s | %-20s\n",
+           i, ibvDeviceList[i].name.c_str(),
+           ibvDeviceList[i].hasActivePort ? "Yes" : "No",
+           ibvDeviceList[i].busId.c_str(),
+           ibvDeviceList[i].numaNode,
+           closestGpusStr.c_str(),
+           ibvDeviceList[i].isRoce && ibvDeviceList[i].hasActivePort?  std::to_string(ibvDeviceList[i].gidIndex).c_str() : "N/A",
+           ibvDeviceList[i].isRoce && ibvDeviceList[i].hasActivePort?  ibvDeviceList[i].gidDescriptor.c_str() : "N/A"
+          );
+  }
+  printf("\n");
+#endif
+}
+
 void DisplayTopology(bool outputToCsv)
 {
   int numCpus = TransferBench::GetNumExecutors(EXE_CPU);
   int numGpus = TransferBench::GetNumExecutors(EXE_GPU_GFX);
-
+  int numNics = TransferBench::GetNumExecutors(EXE_NIC);
   char sep = (outputToCsv ? ',' : '|');
 
   if (outputToCsv) {
     printf("NumCpus,%d\n", numCpus);
     printf("NumGpus,%d\n", numGpus);
+    printf("NumNics,%d\n", numNics);
   } else {
     printf("\nDetected Topology:\n");
     printf("==================\n");
     printf("  %d configured CPU NUMA node(s) [%d total]\n", numCpus, numa_max_node() + 1);
     printf("  %d GPU device(s)\n", numGpus);
+    printf("  %d Supported NIC device(s)\n", numNics);
   }
 
   // Print out detected CPU topology
@@ -91,8 +126,10 @@ void DisplayTopology(bool outputToCsv)
   }
   printf("\n");
 
-  // Print out detected GPU topology
+  // Print out detected NIC topology
+  PrintNicToGPUTopo(outputToCsv);
 
+  // Print out detected GPU topology
 #if defined(__NVCC__)
   for (int i = 0; i < numGpus; i++) {
     hipDeviceProp_t prop;
@@ -118,12 +155,12 @@ void DisplayTopology(bool outputToCsv)
   printf("        %c", sep);
   for (int j = 0; j < numGpus; j++)
     printf(" GPU %02d %c", j, sep);
-  printf(" PCIe Bus ID  %c #CUs %c NUMA %c #DMA %c #XCC\n", sep, sep, sep, sep);
+  printf(" PCIe Bus ID  %c #CUs %c NUMA %c #DMA %c #XCC %c NIC\n", sep, sep, sep, sep, sep);
 
   if (!outputToCsv) {
     for (int j = 0; j <= numGpus; j++)
       printf("--------+");
-    printf("--------------+------+------+------+------\n");
+    printf("--------------+------+------+------+------+------\n");
   }
 
   // Loop over each GPU device
@@ -149,12 +186,13 @@ void DisplayTopology(bool outputToCsv)
 
     char pciBusId[20];
     HIP_CALL(hipDeviceGetPCIBusId(pciBusId, 20, i));
-    printf(" %11s %c %4d %c %4d %c %4d %c %4d\n",
+    printf(" %-11s %c %-4d %c %-4d %c %-4d %c %-4d %c %-4d\n",
            pciBusId, sep,
            TransferBench::GetNumSubExecutors({EXE_GPU_GFX, i}), sep,
            TransferBench::GetClosestCpuNumaToGpu(i), sep,
            TransferBench::GetNumExecutorSubIndices({EXE_GPU_DMA, i}), sep,
-           TransferBench::GetNumExecutorSubIndices({EXE_GPU_GFX, i}));
+           TransferBench::GetNumExecutorSubIndices({EXE_GPU_GFX, i}), sep,
+           TransferBench::GetClosestNicToGpu(i));
   }
 #endif
 }