ludic/selfhost/emit_intrin.ludic
Orkuncakilkaya df6955e609 Phase 7g: the compiler builds its IR with interpolation, not +
Answering your readability point directly: the compiler's own string-building —
left uglier by the 7c `+` migration, e.g. `emit_bind(("load i32, ptr " + ip))` —
now reads as interpolation:

  emit_bind(("load i32, ptr " + ip))            -> emit_bind(`load i32, ptr {ip}`)
  emit_bind(("icmp eq i32 " + (kv + (", " + itoa(ak)))))
                                                -> emit_bind(`icmp eq i32 {kv}, {itoa(ak)}`)
  perr(("assign to unknown " + t.s))            -> perr(`assign to unknown {t.s}`)

164 concat chains across selfhost converted by a tool that flattens the `+` tree,
keeps call/index parens (only grouping parens are rewritten), and converts only
**brace-free** literals — LLVM IR structure strings full of `{`/`}` stay as `+`
rather than becoming awkward `{{`/`}}`. No new language surface; interpolation
already desugars to the same concat.

Reseeded (22565 lines); C-free fixpoint holds byte-for-byte (the strongest proof
the reconstruction is exact); goldens identical; 18/18.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-08-28 01:30:37 +03:00

119 lines
5 KiB
Text

# emit_intrin.ludic — the low-level intrinsics the self-host source uses, each
# lowered to the same libc/inline IR the C backend emits. Returns a Val via
# g_intrin_val and sets g_intrin_ok when `name` was an intrinsic.
var g_intrin_ok: bool = false
# is `name` a low-level intrinsic? A pure name check, so it can gate dispatch
# without evaluating arguments (which could clobber shared state).
fn is_intrinsic(name: ptr) -> bool {
if (name == "mem_alloc") or (name == "mem_realloc") { return true }
if (name == "peek8") or (name == "poke8") { return true }
if (name == "file_open") or (name == "file_read") or (name == "file_write") { return true }
if (name == "file_seek") or (name == "file_tell") or (name == "file_close") { return true }
if (name == "os_argc") or (name == "os_arg") or (name == "os_exit") or (name == "file_stderr") { return true }
if (name == "file_stdout") { return true }
if (name == "os_system") or (name == "os_getenv") { return true }
if (name == "peek32") or (name == "poke32") { return true }
return false
}
# emit " <r> = <rest>\n" and return r
fn emit_bind(rest: ptr) -> ptr { let r = nreg(); emit(" "); emit(r); emit(" = "); emit(rest); emit("\n"); return r }
fn arg_code(e: Node, i: int) -> ptr { let v = emit_expr(e.kids[i]); return v.code }
fn emit_intrinsic(name: ptr, e: Node) -> Val {
g_intrin_ok = true
# ptr_null / ptr_is_null are the `null` literal and `x == null` now.
if (name == "mem_alloc") {
let n = arg_code(e, 0)
let w = emit_bind(`zext i32 {n} to i64`)
return val(emit_bind(`call ptr @malloc(i64 {w})`), "ptr")
}
if (name == "mem_realloc") {
let p = arg_code(e, 0); let n = arg_code(e, 1)
let w = emit_bind(`zext i32 {n} to i64`)
return val(emit_bind(`call ptr @realloc(ptr {p}, i64 {w})`), "ptr")
}
if (name == "peek8") {
let p = arg_code(e, 0); let i = arg_code(e, 1)
let a = emit_gep_i8(p, i)
let b = emit_bind(`load i8, ptr {a}`)
return val(emit_bind(`zext i8 {b} to i32`), "int")
}
if (name == "poke8") {
let p = arg_code(e, 0); let i = arg_code(e, 1); let v = arg_code(e, 2)
let a = emit_gep_i8(p, i)
let t = emit_bind(`trunc i32 {v} to i8`)
emit(" store i8 "); emit(t); emit(", ptr "); emit(a); emit("\n")
return val("0", "void")
}
if (name == "file_open") {
let p = arg_code(e, 0); let m = arg_code(e, 1)
return val(emit_bind(`call ptr @fopen(ptr {p}, ptr {m})`), "ptr")
}
if (name == "file_read") or (name == "file_write") {
let f = arg_code(e, 0); let b = arg_code(e, 1); let n = arg_code(e, 2)
let w = emit_bind(`zext i32 {n} to i64`)
var fn2 = "@fread"
if (name == "file_write") { fn2 = "@fwrite" }
let r = emit_bind(`call i64 {fn2}(ptr {b}, i64 1, i64 {w}, ptr {f})`)
return val(emit_bind(`trunc i64 {r} to i32`), "int")
}
if (name == "file_seek") {
let f = arg_code(e, 0); let off = arg_code(e, 1); let wh = arg_code(e, 2)
let o = emit_bind(`sext i32 {off} to i64`)
return val(emit_bind(`call i32 @fseek(ptr {f}, i64 {o}, i32 {wh})`), "int")
}
if (name == "file_tell") {
let f = arg_code(e, 0)
let r = emit_bind(`call i64 @ftell(ptr {f})`)
return val(emit_bind(`trunc i64 {r} to i32`), "int")
}
if (name == "file_close") {
let f = arg_code(e, 0)
emit(" call i32 @fclose(ptr "); emit(f); emit(")\n")
return val("0", "void")
}
# print_str / print_int are the polymorphic `print(x)` builtin now (emit_call).
if (name == "os_argc") { return val(emit_bind("load i32, ptr @L_argc"), "int") }
if (name == "os_arg") {
let i = arg_code(e, 0)
let v = emit_bind("load ptr, ptr @L_argv")
let q = emit_bind(`getelementptr ptr, ptr {v}, i32 {i}`)
return val(emit_bind(`load ptr, ptr {q}`), "str")
}
if (name == "os_exit") {
let n = arg_code(e, 0)
emit(" call void @exit(i32 "); emit(n); emit(")\n")
emit(" unreachable\n")
g_term = true
return val("0", "void")
}
if (name == "file_stderr") { return val(emit_bind("load ptr, ptr @__stderrp"), "ptr") }
if (name == "file_stdout") { return val(emit_bind("load ptr, ptr @__stdoutp"), "ptr") }
if (name == "os_system") {
let c = arg_code(e, 0)
return val(emit_bind(`call i32 @system(ptr {c})`), "int")
}
if (name == "os_getenv") {
let n = arg_code(e, 0)
return val(emit_bind(`call ptr @getenv(ptr {n})`), "str")
}
# bitwise ops are the operators & | ^ << >> ~ now (see emit_bin / p_mul).
if (name == "peek32") {
let p = arg_code(e, 0); let i = arg_code(e, 1)
let g = nreg()
emit(" "); emit(g); emit(" = getelementptr inbounds i32, ptr "); emit(p); emit(", i32 "); emit(i); emit("\n")
return val(emit_bind(`load i32, ptr {g}`), "int")
}
if (name == "poke32") {
let p = arg_code(e, 0); let i = arg_code(e, 1); let v = arg_code(e, 2)
let g = nreg()
emit(" "); emit(g); emit(" = getelementptr inbounds i32, ptr "); emit(p); emit(", i32 "); emit(i); emit("\n")
emit(" store i32 "); emit(v); emit(", ptr "); emit(g); emit("\n")
return val("0", "void")
}
g_intrin_ok = false
return val("0", "void")
}