Chained vs. independent additions on a 4-wide superscalar core

Both snippets perform the same 5 additions on a 4-wide superscalar out-of-order core where each add has a 1-cycle latency but the core can issue up to 4 independent ops per cycle. Version A chains every add through the previous result. Version B computes three additions independently, then combines them. Which finishes sooner, and why, despite identical instruction counts?

main.cpp
// Version A: serial dependency chain
int x = ((((a + b) + c) + d) + e) + f;

// Version B: independent partial sums combined at the end
int p = a + b;
int q = c + d;
int r = e + f;
int x2 = p + q + r;

23 of 47 in Computer Architecture