From 33c771a3a5ce623f0aad2e00a89068e6204b8b14 Mon Sep 17 00:00:00 2001
From: Jidong Xiao <xiaoj8@rpi.edu>
Date: Mon, 24 Feb 2025 22:17:23 -0500
Subject: [PATCH] adding explaination of volatile

---
 lectures/optimization/gprof/README.md | 83 ++++++++++++++++++++++-----
 1 file changed, 70 insertions(+), 13 deletions(-)

diff --git a/lectures/optimization/gprof/README.md b/lectures/optimization/gprof/README.md
index 5c6a1e3..067b225 100644
--- a/lectures/optimization/gprof/README.md
+++ b/lectures/optimization/gprof/README.md
@@ -11,27 +11,33 @@ $ sudo apt-get install binutils
 ## Compiling a C++ Program for Profiling
 To use `gprof`, compile your program with the `-pg` flag:
 ```sh
-$ g++ -pg -o my_program my_program.cpp
+$ g++ -pg -o test test.cpp
 ```
 
 ## Example C++ Program
-Create a file `my_program.cpp` with the following code:
+Create a file `test.cpp` with the following code:
 ```cpp
 #include <iostream>
-#include <chrono>
-#include <thread>
 
-void slowFunction() {
-    std::this_thread::sleep_for(std::chrono::seconds(1));
+void heavyComputation() {
+    volatile long long sum = 0;
+    for (long long i = 0; i < 500000000; ++i) {
+        sum += i;  // Simple but expensive loop
+    }
 }
 
-void fastFunction() {
-    for (volatile int i = 0; i < 1000000; ++i);
+void lightComputation() {
+    volatile int sum = 0;
+    for (int i = 0; i < 100000; ++i) {
+        sum += i;  // Lighter loop
+    }
 }
 
 int main() {
-    for (int i = 0; i < 5; ++i) slowFunction();
-    for (int i = 100; i < 200; ++i) fastFunction();
+    heavyComputation();  // Call heavy function once
+    for (int i = 0; i < 1000; ++i) {
+        lightComputation();  // Call light function many times
+    }
     return 0;
 }
 ```
@@ -39,15 +45,15 @@ int main() {
 ## Running and Profiling the Program
 1. Compile the program:
    ```sh
-   $ g++ -pg -o my_program my_program.cpp
+   $ g++ -pg -o test test.cpp
    ```
 2. Execute the program to generate `gmon.out`:
    ```sh
-   $ ./my_program
+   $ ./test
    ```
 3. Analyze the profiling data:
    ```sh
-   $ gprof my_program gmon.out > profile.txt
+   $ gprof test gmon.out > profile.txt
    $ cat profile.txt
    ```
 
@@ -62,3 +68,54 @@ int main() {
 
 ## Conclusion
 `gprof` is a powerful tool for detecting performance bottlenecks in C++ programs. By identifying expensive functions, developers can make targeted optimizations.
+
+# Why Use `volatile` in the above program?
+
+## Compiler May Remove the Loops
+When compiling a program, the compiler applies optimizations to make the code run faster. One such optimization is **dead code elimination**, where the compiler removes code that does **not affect the program's observable behavior**.
+
+For example, consider this function:
+
+```cpp
+void heavyComputation() {
+    long long sum = 0;
+    for (long long i = 0; i < 500000000; ++i) {
+        sum += i;
+    }
+}
+```
+
+- The compiler notices that `sum` is **never used outside the function**.
+- Since the result is discarded, the compiler **may completely remove the loop**.
+- This means `heavyComputation()` might do **nothing** at runtime, which ruins our profiling experiment.
+
+---
+
+## How Does `volatile` Help?
+Declaring a variable as `volatile` tells the compiler:
+
+"This variable might change in ways you cannot predict, so do not optimize it away."
+
+For example:
+
+```cpp
+void heavyComputation() {
+    volatile long long sum = 0;  // Mark sum as volatile
+    for (long long i = 0; i < 500000000; ++i) {
+        sum += i;
+    }
+}
+```
+
+- Now, **even if `sum` is never used**, the compiler **must** perform the loop.
+- The `volatile` keyword prevents the compiler from assuming that `sum` is unimportant.
+- This ensures that the loop actually runs during profiling.
+
+---
+
+## **Does `volatile` Affect Performance?**
+Yes, but only slightly.
+- **Without `volatile`**, the compiler can optimize the loop aggressively.
+- **With `volatile`**, every read and write to `sum` is guaranteed to happen exactly as written, preventing some optimizations.
+
+However, this small cost is **worth it for benchmarking**, because it ensures that the loops are not removed.