GPU Threads vs CPU Threads: Warps and Wavefronts

Write a C program that simulates warp execution and demonstrates divergence penalties.

Requirements:

• Define WARP_SIZE as 32
• Create an array of 32 thread values (use indices 0..31)
• Simulate an if/else branch where even threads do one operation and odd threads do another
• Show the two serialized execution phases: first even threads (odd masked), then odd threads (even masked)
• Calculate and print the divergence penalty: a warp with full divergence takes 2x the instructions
• Simulate a memory load where threads access consecutive integers (coalesced) vs scattered integers (uncoalesced)
• Print the number of memory transactions for each pattern (coalesced = 1 transaction, uncoalesced = up to 32)
• Include a summary explaining why GPU threads are not like CPU threads