README

LLVM Chapter 10 reductions

No strong scalar x86_64 bug-report candidate was found in this pass.

What was checked:

• sdiv by small positive constants
• udiv by small positive constants
• srem/urem by constants
• quotient/remainder pairs by the same constant
• several even divisors
• several odd divisors
• negative signed divisors

Observed result:

• Current-head LLVM consistently lowered these to shifts, biased shifts, and magic-multiply sequences.
• No cases in this sweep fell back to div/idiv.
• No obvious missed combine stood out as clearly suboptimal on x86_64.