Tree optimization: Avoid negating the most negative signed value during loop analysis
GCC's tree-ssa-loop-niter pass now avoids incorrect negation of the most negative signed integer, preventing potential compiler misbehavior with loop iteration…
GCC’s number_of_iterations_until_wrap function, part of the tree-ssa-loop-niter optimization pass, has been improved to prevent the problematic negation of the most negative signed integer value. This issue, identified by PR125602, could lead to incorrect INTEGER_CST results during loop analysis. The fix restructures the negation and type conversion logic to avoid this overflow, ensuring accurate calculation of loop bounds and preventing potential miscompilations.
In Details
The tree-ssa-loop-niter pass in GCC is responsible for analyzing loops to determine their iteration counts, which is crucial for various loop optimizations. This commit addresses a specific edge case in number_of_iterations_until_wrap related to PR tree-optimization/125602. The problem arises when attempting to negate the MIN_INT value (the most negative signed integer) during iteration count analysis, which results in an overflow and incorrect INTEGER_CST representation. The fix ensures that the negation and conversion to niter_type are reordered and handled with an assumption ch…
For Context
When a compiler like GCC optimizes your code, it tries to figure out how many times a loop will run. This is called 'loop iteration analysis' and helps with many performance improvements. There's a special problem with signed whole numbers (integers): if you try to flip the sign of the most negative possible number (like -2,147,483,648 for a 32-bit integer), it doesn't have a positive equivalent, and the operation can overflow or wrap around, producing an incorrect result. This commit fixes a bug in GCC's loop analysis where it was trying to do exactly this. By changing how the compiler calculates loop bounds when dealing with negative step values, it avoids this mathematical trap, ensuring your loops are correctly analyzed and optimized, preventing crashes or unexpected behavior in your programs.