3 ms·

Funny to think that once x86 was the platform that had too few general-purpose registers, so people sacrificed the frame pointer register in their highly-optimi

• varispeed 6y ago
  I still remember programming on paper mapping U and V pipelines to get as much done in parallel...
• jandrese 6y ago
  I still remember benchmarking the various optimization options in GCC and the only one that consistently and significantly improved performance on real code was -fomit-frame-pointer.
  • sroussey 6y ago
    My custom builds of MySQL and PHP back in the day did exactly that.
  • praptak 6y ago
    It saves one push per function call, which probably helps more than freeing up the register.
    • jandrese 6y ago
      The improvement was dramatic. IIRC 20-25% faster code pretty much every time. Nothing else in the optimization set even came close.
      • mr_toad 6y ago
        Back in the days before on-chip caches register memory was vastly faster than main memory.
• nwmcsween 6y ago
  It still is limited, this isn't general purpose registers which is 16, I semi regularly see register spills on x86_64 while looking at compiler output
  • klodolph 6y ago
    You’ll see register spills no matter how many registers, right? I definitely see register spills on architectures with 32 registers.
• jlokier 6y ago
  Even sacrificed the stack pointer in their very highly optimised assembly routines :-)