µDewy Language Specification

µDewy is archived and no longer developed. This is a snapshot of its documentation. → Dewy

A.2.1 Codegen Notes

The x86_64 backend still follows udewy's logical value-stack model, but it does not map every save_value() directly to a machine push.

• The current visible expression result stays in rax.
• A small prefix of the logical saved-value stack is cached in callee-saved registers before falling back to spill slots on the real stack.
• When a call has more than 6 arguments, the extra arguments are written into an outbound stack-argument area and the first 6 are placed in rdi, rsi, rdx, rcx, r8, and r9.
• Call lowering also keeps the machine stack aligned to the ABI-required 16-byte boundary.
• Busy locals live in rbx/r15 (across calls) or r8/r9/r11 (between calls) through the shared local register allocation described in Backend Architecture; ordinary builds only.
• The visible value may be pending instead of in rax: a local plus a constant offset that was named but not yet loaded, a constant, a load not yet performed, or the flags of the last comparison. save_value moves a pending local, constant or load straight into its cache register; a comparison uses a pending local, constant or load as the cmpq operand; a load or store through a pending local uses it as the base register (movq 24(%rbx), %r12, movq %r12, 24(%rbx)); a constant stores directly (movq $0, %r15); and if, loop and short-circuit tests branch on the pending flags (jne, jge, ...) instead of materializing -1/0; not on a pending comparison inverts it. Every other operation materializes the value first. A loop test such as i <? n is therefore movq, cmpq, jge, and a tag test __load_i64__(cell + 8) =? 140 is cmpq $140, 8(%rbx), jne. While a local is still in memory, a site may hold two instructions in one code line (the slot load, then the operation through rax); allocation rewrites it to the one-instruction register form.