Add nccl env intercept example code (flagos-ai#269)

Author: yzhang35

flagcx/adaptor/tuner/Makefile

@@ -0,0 +1,31 @@
+# Compiler and flags for C++
+CXX = g++
+CXXFLAGS = -fPIC -Wall -Wextra -O2 -std=c++11
+LDFLAGS = -shared -pthread
+
+# Target shared library name
+TARGET = libnccl_param.so
+
+# Source files
+SOURCES = nccl_param.cc
+
+# Object files (automatically generated from source files)
+OBJECTS = $(SOURCES:.cc=.o)
+
+# Default target
+all: $(TARGET)
+
+# Build the shared library
+$(TARGET): $(OBJECTS)
+	$(CXX) $(LDFLAGS) -o $@ $(OBJECTS)
+
+# Compile C++ source files to object files
+%.o: %.cc
+	$(CXX) $(CXXFLAGS) -c $< -o $@
+
+# Clean up build files
+clean:
+	rm -f $(OBJECTS) $(TARGET)
+
+# Phony targets
+.PHONY: all clean

flagcx/adaptor/tuner/nccl_param.cc

@@ -0,0 +1,33 @@
+// This is a example of intercept NCCL environment getter functions.
+#include <stdio.h>
+#include <cstring>
+#include <string>
+#include <cstdint>
+#include <pthread.h>
+#include <stdlib.h>  // for setenv, getenv
+
+static void ncclLoadParam(char const* env, int64_t deftVal, int64_t* value) {
+  static pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
+  pthread_mutex_lock(&mutex);
+  const char* str = getenv(env);
+  *value = deftVal;
+  if (str && strlen(str) > 0) {
+    try {
+      *value = std::stoll(str);
+    } catch (const std::exception& e) {
+      *value = deftVal;
+    }
+  }
+  pthread_mutex_unlock(&mutex);
+}
+
+#define NCCL_PARAM(name, env, deftVal)		 \
+  int64_t ncclParam##name() {			 \
+    int64_t value = INT64_MIN;			 \
+    ncclLoadParam("NCCL_" env, deftVal, &value); \
+    return value;				 \
+  }
+
+// In order to intercept a NCCL environment getter function, add an additional line of NCCL_PRARM here.
+// Below is an example of intercepting NCCL_P2P_NVL_CHUNKSIZE env.
+NCCL_PARAM(P2pNvlChunkSize, "P2P_NVL_CHUNKSIZE", (1 << 19)); /* 512 kB */

flagcx/core/flagcx_tuner.cc

@@ -116,11 +116,11 @@ struct flagcxTunerContext {
 static struct flagcxEnvConfig config1 = {
     "defaultConfig1",
     1,
-    {FLAGCX_ENV_TYPE_COLL, "NCCL_P2P_NVL_CHUNKSIZE", "1024", "524288"}};
+    {FLAGCX_ENV_TYPE_CREATION, "NCCL_P2P_NVL_CHUNKSIZE", "1024", "524288"}};
 static struct flagcxEnvConfig config2 = {
     "defaultConfig2",
     1,
-    {FLAGCX_ENV_TYPE_COLL, "NCCL_P2P_NVL_CHUNKSIZE", "524288", "524288"}};
+    {FLAGCX_ENV_TYPE_CREATION, "NCCL_P2P_NVL_CHUNKSIZE", "524288", "524288"}};
 
 bool operator<(const struct flagcxCommTag &lhs,
                const struct flagcxCommTag &rhs) {