feat: support CCL Broadcast

examples/ccl/broadcast.cc

@@ -0,0 +1,272 @@
+/**
+ * InfiniCCL Example: Thread-per-GPU Single-Node Broadcast
+ *
+ * This example spawns one CPU thread per GPU and performs native CCL
+ * broadcasts without an MPI launcher.
+ */
+
+#include <unistd.h>
+
+#include <algorithm>
+#include <array>
+#include <atomic>
+#include <charconv>
+#include <cmath>
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <limits>
+#include <string>
+#include <string_view>
+#include <thread>
+#include <vector>
+
+#include "backend_manifest.h"
+#include "infiniccl.h"
+#include "utils.h"
+
+using namespace infini::ccl;
+
+struct ThreadArgs {
+  int rank;
+  int size;
+  int root;
+  infinicclUniqueId id;
+  size_t num_elements;
+  int warmup_iter;
+  int profile_iter;
+  std::atomic_bool *all_correct;
+};
+
+template <typename T>
+bool ParseIntegerOption(const char *argument, T *value) {
+  if (!argument || !value) {
+    return false;
+  }
+
+  const std::string_view text(argument);
+  if (text.empty()) {
+    return false;
+  }
+
+  T parsed{};
+  const auto [end, error] =
+      std::from_chars(text.data(), text.data() + text.size(), parsed);
+  if (error != std::errc{} || end != text.data() + text.size()) {
+    return false;
+  }
+
+  *value = parsed;
+  return true;
+}
+
+void PrintBroadcastMetrics(size_t num_elements, double elapsed_ms) {
+  constexpr double kBytesPerMiB = 1024.0 * 1024.0;
+  constexpr double kBytesPerGB = 1.0e9;
+  const double data_bytes = static_cast<double>(num_elements) * sizeof(float);
+  const auto original_flags = std::cout.flags();
+  const auto original_precision = std::cout.precision();
+
+  std::cout << std::fixed;
+  std::cout << "Data size:      " << std::setprecision(2)
+            << data_bytes / kBytesPerMiB << " MiB" << std::endl;
+  std::cout << "Time:           " << std::setprecision(3) << elapsed_ms << " ms"
+            << std::endl;
+  if (elapsed_ms > 0.0 && std::isfinite(elapsed_ms)) {
+    const double algorithm_bandwidth =
+        data_bytes / kBytesPerGB / (elapsed_ms / 1000.0);
+    const double bus_bandwidth = algorithm_bandwidth;
+    std::cout << "Throughput:     " << std::setprecision(2) << bus_bandwidth
+              << " GB/s (Bus BW)" << std::endl;
+    std::cout << "Alg Bandwidth:  " << algorithm_bandwidth << " GB/s"
+              << std::endl;
+  } else {
+    std::cout << "Throughput:     N/A (Bus BW)" << std::endl;
+    std::cout << "Alg Bandwidth:  N/A" << std::endl;
+  }
+
+  std::cout.flags(original_flags);
+  std::cout.precision(original_precision);
+}
+
+void WorkerThread(ThreadArgs args) {
+  constexpr Device::Type kDevType =
+      ListGetBest<DevicePriority>(EnabledDevices{});
+  using Rt = Runtime<kDevType>;
+
+  CHECK_RT(Rt, Rt::SetDevice(args.rank));
+
+  infinicclComm_t comm = nullptr;
+  CHECK_INFINI(infinicclCommInitRank(&comm, args.size, args.id, args.rank));
+
+  constexpr float kRootValue = 42.0f;
+  constexpr float kSentinelValue = -1.0f;
+  const size_t total_bytes = args.num_elements * sizeof(float);
+
+  std::vector<float> h_send(args.num_elements, kSentinelValue);
+  std::vector<float> h_recv(args.num_elements, kSentinelValue);
+
+  float *d_send = nullptr;
+  float *d_recv = nullptr;
+  CHECK_RT(Rt, Rt::Malloc(reinterpret_cast<void **>(&d_send), total_bytes));
+  CHECK_RT(Rt, Rt::Malloc(reinterpret_cast<void **>(&d_recv), total_bytes));
+
+  auto ResetBuffers = [&]() {
+    std::fill(h_send.begin(), h_send.end(),
+              args.rank == args.root ? kRootValue : kSentinelValue);
+    std::fill(h_recv.begin(), h_recv.end(), kSentinelValue);
+    CHECK_RT(Rt, Rt::Memcpy(d_send, h_send.data(), total_bytes,
+                            Rt::MemcpyHostToDevice));
+    CHECK_RT(Rt, Rt::Memcpy(d_recv, h_recv.data(), total_bytes,
+                            Rt::MemcpyHostToDevice));
+    CHECK_RT(Rt, Rt::StreamSynchronize(nullptr));
+  };
+
+  auto RunScenario = [&](const std::string &name, float *verify_buff,
+                         auto &&collective_call) {
+    for (int i = 0; i < args.warmup_iter; ++i) {
+      CHECK_INFINI(collective_call());
+    }
+    CHECK_RT(Rt, Rt::StreamSynchronize(nullptr));
+
+    Timer timer;
+    for (int i = 0; i < args.profile_iter; ++i) {
+      CHECK_INFINI(collective_call());
+    }
+    CHECK_RT(Rt, Rt::StreamSynchronize(nullptr));
+    const double elapsed =
+        timer.ElapsedMs() / static_cast<double>(args.profile_iter);
+
+    CHECK_RT(Rt, Rt::Memcpy(h_recv.data(), verify_buff, total_bytes,
+                            Rt::MemcpyDeviceToHost));
+
+    const bool correct = Validator::ValidateResult(
+        h_recv.data(), args.num_elements, kRootValue, args.rank, true, name);
+    if (!correct) {
+      args.all_correct->store(false, std::memory_order_relaxed);
+    }
+
+    if (args.rank == 0) {
+      PrintBroadcastMetrics(args.num_elements, elapsed);
+    }
+  };
+
+  ResetBuffers();
+  RunScenario("Out-of-Place Broadcast", d_recv, [&]() {
+    return infinicclBroadcast(d_send, d_recv, args.num_elements,
+                              infinicclFloat32, args.root, comm, nullptr);
+  });
+
+  ResetBuffers();
+  float *d_in_place = args.rank == args.root ? d_send : d_recv;
+  RunScenario("In-Place Broadcast", d_in_place, [&]() {
+    return infinicclBroadcast(d_in_place, d_in_place, args.num_elements,
+                              infinicclFloat32, args.root, comm, nullptr);
+  });
+
+  ResetBuffers();
+  d_in_place = args.rank == args.root ? d_send : d_recv;
+  RunScenario("Legacy In-Place Bcast", d_in_place, [&]() {
+    return infinicclBcast(d_in_place, args.num_elements, infinicclFloat32,
+                          args.root, comm, nullptr);
+  });
+
+  CHECK_RT(Rt, Rt::Free(d_send));
+  CHECK_RT(Rt, Rt::Free(d_recv));
+  CHECK_INFINI(infinicclCommDestroy(comm));
+}
+
+int main(int argc, char **argv) {
+  int num_gpus = 8;
+  int warmup_iters = 1;
+  int profile_iters = 20;
+  size_t num_elements = 1 << 25;
+
+  int opt;
+  while ((opt = getopt(argc, argv, "g:w:p:n:h")) != -1) {
+    switch (opt) {
+      case 'g':
+        if (!ParseIntegerOption(optarg, &num_gpus)) {
+          std::cerr << "Invalid value for `-g`." << std::endl;
+          return EXIT_FAILURE;
+        }
+        break;
+      case 'w':
+        if (!ParseIntegerOption(optarg, &warmup_iters)) {
+          std::cerr << "Invalid value for `-w`." << std::endl;
+          return EXIT_FAILURE;
+        }
+        break;
+      case 'p':
+        if (!ParseIntegerOption(optarg, &profile_iters)) {
+          std::cerr << "Invalid value for `-p`." << std::endl;
+          return EXIT_FAILURE;
+        }
+        break;
+      case 'n':
+        if (!ParseIntegerOption(optarg, &num_elements)) {
+          std::cerr << "Invalid value for `-n`." << std::endl;
+          return EXIT_FAILURE;
+        }
+        break;
+      case 'h':
+        std::cout << "Usage: " << argv[0] << " [options]\n"
+                  << "Options:\n"
+                  << "  -g <num_gpus>        Number of GPUs (default: 8)\n"
+                  << "  -w <warmup_iters>    Warmup iterations (default: 1)\n"
+                  << "  -p <profile_iters>   Profile iterations (default: 20)\n"
+                  << "  -n <num_elements>    Number of elements (default: "
+                  << (1 << 25) << ")\n";
+        return EXIT_SUCCESS;
+      default:
+        std::cerr << "Invalid argument. Use `-h` for help." << std::endl;
+        return EXIT_FAILURE;
+    }
+  }
+
+  if (num_gpus <= 0 || warmup_iters < 0 || profile_iters <= 0 ||
+      num_elements == 0 ||
+      num_elements > std::numeric_limits<size_t>::max() / sizeof(float)) {
+    std::cerr << "Invalid execution parameter." << std::endl;
+    return EXIT_FAILURE;
+  }
+
+  std::array<char, 256> hostname{};
+  if (gethostname(hostname.data(), hostname.size()) != 0) {
+    std::cerr << "Failed to query the host name." << std::endl;
+    return EXIT_FAILURE;
+  }
+  hostname.back() = '\0';
+
+  const int root = num_gpus > 1 ? num_gpus - 1 : 0;
+  std::cout << "[Main Process] Host: " << hostname.data()
+            << " | Target GPUs: " << num_gpus << " | Root: " << root
+            << std::endl;
+
+  infinicclUniqueId shared_id;
+  CHECK_INFINI(infinicclGetUniqueId(&shared_id));
+
+  std::atomic_bool all_correct{true};
+  std::vector<std::thread> threads;
+  threads.reserve(num_gpus);
+
+  for (int rank = 0; rank < num_gpus; ++rank) {
+    ThreadArgs args{rank,         num_gpus,     root,          shared_id,
+                    num_elements, warmup_iters, profile_iters, &all_correct};
+    threads.emplace_back(WorkerThread, args);
+  }
+
+  for (auto &thread : threads) {
+    if (thread.joinable()) {
+      thread.join();
+    }
+  }
+
+  if (!all_correct.load(std::memory_order_relaxed)) {
+    std::cerr << "Broadcast validation failed." << std::endl;
+    return EXIT_FAILURE;
+  }
+
+  std::cout << "[Main Process] All broadcast scenarios passed." << std::endl;
+  return EXIT_SUCCESS;
+}