Remaining control flow gaps vs standard Forth

[tetsuo-cpp]

Summary

WarpForth uses a compile-time control-flow stack in the translator, emitting cf.br/cf.cond_br directly. This matches standard Forth's cf-stack semantics and supports interleaved control flow patterns (multi-WHILE, WHILE+UNTIL, etc.).

Remaining Gaps

High impact for GPU kernels

Feature	Standard Forth	Notes
LEAVE / UNLOOP	Early exit from DO...LOOP	Very common in search/scan patterns. Requires adding a forward branch target to LoopContext and branching to the exit block.
EXIT	Early return from a word	Common in utility words with multiple exit conditions. Requires emitting func.return mid-body.

Medium impact

Feature	Standard Forth	Notes
+LOOP	Arbitrary loop step	Useful for strided iteration. The current LOOP always increments by 1; +LOOP would pop the step from the stack. Straightforward to add.

Low/no impact for GPU kernels

Feature	Why
BEGIN...AGAIN	Infinite loops hang the GPU; trivial workaround BEGIN 0 UNTIL
?DO...LOOP	WarpForth's DO already behaves like ?DO (zero-trip safe via arith.cmpi slt check)
RECURSE	Recursion is dangerous on GPUs (tiny call stacks, thread divergence)
CS-PICK / CS-ROLL	Exotic metaprogramming, never used in kernel code

Related

• refactor: lower control flow to cf dialect instead of scf #20 — refactor: lower control flow to cf dialect instead of scf
• Add validation for mismatched control flow words #21 — Add validation for mismatched control flow words

[tetsuo-cpp]

Created sub-issues for each actionable item:

High impact

• feat: implement LEAVE and UNLOOP words #23 — feat: implement LEAVE and UNLOOP words
• feat: implement EXIT word #24 — feat: implement EXIT word

Medium impact

• feat: implement +LOOP word #25 — feat: implement +LOOP word