Static Single Assignment
Subject: Computer and Technology noun /ˈstæt.ɪk ˈsɪŋ.ɡəl əˈsaɪn.mənt/ 1 definition
- compiler-design
- compiler-optimization
- static-analysis
- programming-languages
-
Definition (noun)
Entry #268 · UpdatedStatic Single Assignment (SSA) is a property of a compiler intermediate representation (IR) in which every variable is assigned a value exactly once in the program text and defined before every use. When source code reassigns a variable, converting the IR into SSA form splits it into distinct versioned names—such as
x1andx2—and inserts pseudo-instructions called phi functions at control-flow merge points to select the active version based on which predecessor basic block executed. By making use-definition chains explicit and eliminating value overwrites, SSA dramatically simplifies static analysis and optimizations such as constant propagation, dead-code elimination, and global value numbering. Because physical processors operate on mutable, finite registers rather than unbounded immutable names, SSA acts strictly as an analysis and optimization representation—widely used in compilers such as LLVM, GCC, and V8—and must be lowered back to conventional destructive-assignment instructions during code generation.Common Forms
Minimal SSA: Places phi functions strictly at the iterated dominance frontier of each variable's assignments, producing the smallest structural set of merge nodes dictated by the control-flow graph without checking whether variables remain live.
Pruned SSA: Refines minimal SSA with global liveness analysis to omit phi functions at merge blocks where a variable is dead, eliminating unnecessary pseudo-assignments and shrinking the IR.
Semi-Pruned SSA: Eliminates phi functions for variables confined within a single basic block without running full global liveness analysis, offering a faster trade-off between minimal SSA and pruned SSA.
Block-Argument SSA: Replaces phi instructions with parameterized basic blocks—as used in Swift SIL, MLIR, and Cranelift—so branch instructions explicitly pass SSA values as arguments to successor blocks rather than encoding predecessor-indexed merge nodes.
Example
When Clang compiles a conditional branch that sets
x = a + 1in thethenblock andx = a * 2in theelseblock, LLVM'smem2regpass promotes the mutable stackallocainto distinct SSA virtual registers%x1 = add i32 %a, 1and%x2 = mul i32 %a, 2. At the converging%mergebasic block, the IR inserts%x3 = phi i32 [ %x1, %then ], [ %x2, %else ], allowing subsequent optimization passes to reason directly about immutable value flow.