angr.analyses.decompiler.peephole_optimizations.narrow_fp_ops

class angr.analyses.decompiler.peephole_optimizations.narrow_fp_ops.NarrowFPOperations

Bases: PeepholeOptimizationExprBase

Narrow floating-point operations when all FP operands are VEX FP promotions from the same source width.

VEX emulates x87 with F64, so float operations appear as:

AddF64(Conv(32F->64F, a), Conv(32F->64F, b))

Since both operands are 32-bit floats promoted for emulation, the operation is semantically 32-bit. Narrow to:

Conv(32F->64F, AddF32(a, b))

The conversion keeps the width the parent expression was built with. A parent whose other operand is also a promotion is narrowed in turn.

NAME = 'Narrow FP operations with promoted float operands'
expr_classes = (<class 'angr.ailment.expression.BinaryOp'>,)
optimize(expr, **kwargs)
Parameters:

expr (BinaryOp)