Unbox ^double record field reads (#231)

A record field tagged ^double now reads back as a flonum and feeds the numeric
pass, so hintless arithmetic over those fields lowers to fl-ops — the leaf-numeric
analog of the ^Vec3 nested-field hints. Combined with the whole-program :double
param inference, a vec3-dot over a ^double-fielded record unboxes end to end with
no per-fn hints.

records.ss: a ^double field tag passes through resolution, and the ctor (and a
mutable-field set!) coerce a ^double field to a flonum — JVM primitive-field
parity (jolt returned an exact 1, not 1.0, before), and what makes reading the
field back as :double sound for an fl-op.

types.clj: field-type-from-tag maps "double" -> :double, and a keyword/get lookup
whose result is :double annotates the node :num-read :double. numeric.clj reads
that annotation and classifies the field read as a :double operand, so the
enclosing arithmetic specializes — the read itself keeps its jrec-field-at/jolt-get
emit.

run-fieldnum.ss gate: ctor coercion (int field -> flonum), field-field arithmetic
emitting fl*/fl+, and an untagged field staying generic. make test / shakesmoke
green, selfhost holds, 0 new divergences.

Co-authored-by: Yogthos <yogthos@gmail.com>
This commit is contained in:
Dmitri Sotnikov 2026-06-26 14:43:00 +00:00 committed by GitHub
parent 8ae45057d6
commit 4671e1b67e
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
6 changed files with 292 additions and 180 deletions

View file

@ -4,7 +4,7 @@
# build step. `make test` is the full gate. `make remint` rebuilds the seed after a # build step. `make test` is the full gate. `make remint` rebuilds the seed after a
# source change. # source change.
.PHONY: test ci values corpus unit smoke buildsmoke selfhost sci certify ffi transient infer wp devirt fieldread numwp directlink numeric inline shakesmoke remint .PHONY: test ci values corpus unit smoke buildsmoke selfhost sci certify ffi transient infer wp devirt fieldread numwp fieldnum directlink numeric inline shakesmoke remint
# Full gate (dev machine). Includes the self-host byte-fixpoint, which only holds # Full gate (dev machine). Includes the self-host byte-fixpoint, which only holds
# on the same Chez that minted the seed. # on the same Chez that minted the seed.
@ -15,7 +15,7 @@ test: selfhost ci
# lockfile) — it RUNS correctly on any Chez, but `selfhost` rebuilds it and a # lockfile) — it RUNS correctly on any Chez, but `selfhost` rebuilds it and a
# different Chez version may emit byte-different (gensym/order) output, so the # different Chez version may emit byte-different (gensym/order) output, so the
# byte-fixpoint is a dev-machine check, not a CI one (jolt-8479). # byte-fixpoint is a dev-machine check, not a CI one (jolt-8479).
ci: values corpus unit smoke buildsmoke sci ffi transient infer wp devirt fieldread numwp directlink numeric inline certify ci: values corpus unit smoke buildsmoke sci ffi transient infer wp devirt fieldread numwp fieldnum directlink numeric inline certify
@echo "OK: CI gates passed" @echo "OK: CI gates passed"
# Self-host fixpoint: bootstrap.ss rebuild == checked-in seed. # Self-host fixpoint: bootstrap.ss rebuild == checked-in seed.
@ -86,6 +86,12 @@ fieldread:
numwp: numwp:
@chez --script host/chez/run-numwp.ss @chez --script host/chez/run-numwp.ss
# Double record fields: a ^double-tagged field reads back as a flonum (coerced at
# construction and set!), so hintless arithmetic over those fields unboxes to fl-ops;
# an untagged field stays generic.
fieldnum:
@chez --script host/chez/run-fieldnum.ss
# Direct-linking emission: a closed-world build binds top-level app defs to jv$ # Direct-linking emission: a closed-world build binds top-level app defs to jv$
# Scheme bindings and routes app->app calls/refs to them, skipping var-deref + # Scheme bindings and routes app->app calls/refs to them, skipping var-deref +
# jolt-invoke; ^:dynamic/^:redef and nested defs opt out. # jolt-invoke; ^:dynamic/^:redef and nested defs opt out.

View file

@ -48,10 +48,17 @@
(define chez-record-shapes-tbl (make-hashtable string-hash string=?)) (define chez-record-shapes-tbl (make-hashtable string-hash string=?))
;; method var-key "ns/method" -> (cons proto-name method-name). ;; method var-key "ns/method" -> (cons proto-name method-name).
(define chez-protocol-methods-tbl (make-hashtable string-hash string=?)) (define chez-protocol-methods-tbl (make-hashtable string-hash string=?))
;; type-tag "ns.Name" -> #(bool ...) marking which fields are ^double, so the ctor
;; and set! coerce them to flonums (JVM primitive-field semantics, and what makes
;; reading the field back as :double sound for fl-ops).
(define chez-record-dbl-tbl (make-hashtable string-hash string=?))
(define (chez-double-tag? t) (and (string? t) (string=? t "double")))
(define (register-record-shape! ctor-key field-kws field-tags type-tag) (define (register-record-shape! ctor-key field-kws field-tags type-tag)
(hashtable-set! chez-record-shapes-tbl ctor-key (hashtable-set! chez-record-shapes-tbl ctor-key
(vector field-kws field-tags type-tag))) (vector field-kws field-tags type-tag))
(hashtable-set! chez-record-dbl-tbl type-tag
(list->vector (map chez-double-tag? field-tags))))
;; simple name of a dotted/slashed string: the segment after the last . or /. ;; simple name of a dotted/slashed string: the segment after the last . or /.
(define (chez-shape-simple-name s) (define (chez-shape-simple-name s)
@ -67,6 +74,7 @@
(define (chez-resolve-field-tag tag by-name) (define (chez-resolve-field-tag tag by-name)
(cond ((or (not tag) (jolt-nil-t? tag)) jolt-nil) (cond ((or (not tag) (jolt-nil-t? tag)) jolt-nil)
((string=? tag "num") "num") ((string=? tag "num") "num")
((string=? tag "double") "double") ; a ^double field reads back as a flonum
(else (let ((ck (hashtable-ref by-name (chez-shape-simple-name tag) #f))) (else (let ((ck (hashtable-ref by-name (chez-shape-simple-name tag) #f)))
(if ck ck jolt-nil))))) (if ck ck jolt-nil)))))
@ -163,7 +171,13 @@
(define (jolt-set-field! inst k v) (define (jolt-set-field! inst k v)
(if (jrec? inst) (if (jrec? inst)
(let ((i (jrec-field-index inst k))) (let ((i (jrec-field-index inst k)))
(if i (begin (vector-set! (jrec-vals inst) i v) v) (if i (let* ((flags (hashtable-ref chez-record-dbl-tbl (jrec-tag inst) #f))
;; a ^double field stays a flonum across set!, like the ctor —
;; keeps a later field read sound to unbox.
(v2 (if (and flags (fx< i (vector-length flags)) (vector-ref flags i)
(number? v) (not (flonum? v)))
(exact->inexact v) v)))
(vector-set! (jrec-vals inst) i v2) v2)
(error #f "set! of an unknown field" k))) (error #f "set! of an unknown field" k)))
(error #f "set! of a field on a non-record" inst))) (error #f "set! of a field on a non-record" inst)))
(define (jrec-ext=? ea eb) (define (jrec-ext=? ea eb)
@ -431,6 +445,10 @@
(let* ((tag (string-append (chez-current-ns) "." (symbol-t-name name-sym))) (let* ((tag (string-append (chez-current-ns) "." (symbol-t-name name-sym)))
(kws (seq->list field-kws)) (kws (seq->list field-kws))
(field-tags (if (pair? rest-args) (seq->list (car rest-args)) '())) (field-tags (if (pair? rest-args) (seq->list (car rest-args)) '()))
;; which fields are ^double — coerced to a flonum on construction (JVM
;; primitive-field parity), so reading them back is a genuine flonum.
(dbl-flags (list->vector (map chez-double-tag? field-tags)))
(ndbl (vector-length dbl-flags))
(desc (make-jrdesc tag kws)) (desc (make-jrdesc tag kws))
(nf (length kws)) (nf (length kws))
(ctor (lambda args (ctor (lambda args
@ -439,7 +457,12 @@
(let ((v (make-vector nf jolt-nil))) (let ((v (make-vector nf jolt-nil)))
(let loop ((as args) (i 0)) (let loop ((as args) (i 0))
(if (or (null? as) (= i nf)) (make-jrec desc v jolt-nil) (if (or (null? as) (= i nf)) (make-jrec desc v jolt-nil)
(begin (vector-set! v i (car as)) (loop (cdr as) (+ i 1))))))))) (let ((a (car as)))
(vector-set! v i
(if (and (fx< i ndbl) (vector-ref dbl-flags i)
(number? a) (not (flonum? a)))
(exact->inexact a) a))
(loop (cdr as) (+ i 1)))))))))
;; Register the ctor globally by simple class name (like StringBuilder) so ;; Register the ctor globally by simple class name (like StringBuilder) so
;; (Name. …) interop resolves ns-agnostically: a deftype used across files works ;; (Name. …) interop resolves ns-agnostically: a deftype used across files works
;; even when the runtime current ns is the caller's, not the defining ns ;; even when the runtime current ns is the caller's, not the defining ns

73
host/chez/run-fieldnum.ss Normal file
View file

@ -0,0 +1,73 @@
;; run-fieldnum.ss — ^double record field reads unbox to fl-ops (jolt-evr9 R2).
;;
;; A record field tagged ^double reads back as a flonum (:double in the lattice),
;; so hintless arithmetic over those fields — (* (:x a) (:x b)) — lowers to fl-ops,
;; the same machinery as a ^double param. Two halves pinned here: (1) the ctor
;; coerces a ^double field to a flonum at construction (JVM parity, and what makes
;; the fl-op sound), and (2) field-field arithmetic over a record param (typed by
;; the whole-program fixpoint) emits fl*.
;;
;; chez --script host/chez/run-fieldnum.ss
(import (chezscheme))
(load "host/chez/rt.ss")
(set-chez-ns! "clojure.core")
(load "host/chez/seed/prelude.ss")
(load "host/chez/post-prelude.ss")
(set-chez-ns! "user")
(load "host/chez/host-contract.ss")
(load "host/chez/seed/image.ss")
(load "host/chez/compile-eval.ss")
(define analyze (var-deref "jolt.analyzer" "analyze"))
(define set-record-shapes! (var-deref "jolt.passes.types" "set-record-shapes!"))
(define set-protocol-methods! (var-deref "jolt.passes.types" "set-protocol-methods!"))
(define wp-infer! (var-deref "jolt.passes.types" "wp-infer!"))
(define run-passes (var-deref "jolt.passes" "run-passes"))
(define emit (var-deref "jolt.backend-scheme" "emit"))
(define (anode src) (analyze (make-analyze-ctx "user") (jolt-ce-read src)))
(define (evals src) (jolt-compile-eval (string-append "(do " src ")") "user"))
(define (contains-sub? s sub)
(let ((n (string-length s)) (m (string-length sub)))
(let loop ((i 0))
(cond ((> (+ i m) n) #f)
((string=? (substring s i (+ i m)) sub) #t)
(else (loop (+ i 1)))))))
(define fails 0) (define total 0)
(define (check label actual expected)
(set! total (+ total 1))
(unless (equal? actual expected)
(set! fails (+ fails 1))
(printf " FAIL ~a: got ~s expected ~s\n" label actual expected)))
;; a record with ^double fields; the ctor must coerce an integer arg to a flonum.
(evals "(defrecord V [^double x ^double y])")
(check "ctor coerces ^double field to flonum" (flonum? (evals "(:x (->V 1 2))")) #t)
(check "coerced field value matches" (evals "(:x (->V 1 2))") 1.0)
(check "a flonum arg passes through" (evals "(:y (->V 1.5 2.5))") 2.5)
;; dot is hintless; its caller passes V instances, so the fixpoint types a/b as V
;; records, the ^double fields read :double, and the field-field arithmetic unboxes.
(define dot (anode "(def dot (fn [a b] (+ (* (:x a) (:x b)) (* (:y a) (:y b)))))"))
(define used (anode "(def used (fn [] (dot (->V 1.0 2.0) (->V 3.0 4.0))))"))
(set-record-shapes! (chez-record-shapes-map))
(set-protocol-methods! (jolt-hash-map))
(wp-infer! (jolt-vector dot used))
(set-optimize! #t)
(define dot-emit (emit (run-passes dot (make-analyze-ctx "user"))))
(check "field-field arithmetic unboxes to fl*" (contains-sub? dot-emit "fl*") #t)
(check "field-field arithmetic unboxes to fl+" (contains-sub? dot-emit "fl+") #t)
;; an UNTAGGED field stays generic — no fl-op (the read is :any, not :double).
(evals "(defrecord W [p q])")
(define dotw (anode "(def dotw (fn [a b] (* (:p a) (:p b))))"))
(define usew (anode "(def usew (fn [] (dotw (->W 1.0 2.0) (->W 3.0 4.0))))"))
(set-record-shapes! (chez-record-shapes-map))
(wp-infer! (jolt-vector dotw usew))
(check "untagged field stays generic (no fl*)"
(contains-sub? (emit (run-passes dotw (make-analyze-ctx "user"))) "fl*") #f)
(if (= fails 0)
(begin (printf "fieldnum gate: ~a/~a passed\n" total total) (exit 0))
(begin (printf "fieldnum gate: ~a/~a passed (~a failed)\n" (- total fails) total fails) (exit 1)))

File diff suppressed because one or more lines are too long

View file

@ -156,6 +156,10 @@
node1 (assoc node :args argnodes) node1 (assoc node :args argnodes)
n (count ars)] n (count ars)]
(cond (cond
;; a field read the structural inference proved is a flonum (a ^double record
;; field) is a :double operand — so (* (:x v) (:x v)) unboxes. The read itself
;; isn't lowered here; it keeps its keyword/jrec-field-at emit.
(= :double (get node :num-read)) [:double node1]
;; a call to a var with a declared numeric return (^double/^long) yields that ;; a call to a var with a declared numeric return (^double/^long) yields that
;; kind, so an accumulator over the result types. The call itself isn't an ;; kind, so an accumulator over the result types. The call itself isn't an
;; arithmetic op to lower — its body already coerces the return. ;; arithmetic op to lower — its body already coerces the return.

View file

@ -69,6 +69,7 @@
(cond (cond
(or (nil? tag) (<= depth 0)) :any (or (nil? tag) (<= depth 0)) :any
(= tag "num") :num (= tag "num") :num
(= tag "double") :double ; a ^double field reads back as a flonum
:else (let [e (get shapes tag)] :else (let [e (get shapes tag)]
(if e (record-type-from-entry e depth shapes) :any)))) (if e (record-type-from-entry e depth shapes) :any))))
(defn- record-type-from-entry [rs depth shapes] (defn- record-type-from-entry [rs depth shapes]
@ -224,9 +225,13 @@
mt (ty mr) mt (ty mr)
msub (if (struct-safe? mt) (mark-struct (nd mr) mt) (nd mr)) msub (if (struct-safe? mt) (mark-struct (nd mr) mt) (nd mr))
ft (field-type mt (get fnode :val)) ft (field-type mt (get fnode :val))
dr (when (= n 2) (infer (nth args 1) tenv env))] dr (when (= n 2) (infer (nth args 1) tenv env))
[(if dr (join ft (ty dr)) ft) rt (if dr (join ft (ty dr)) ft)
(assoc node :args (if dr [msub (nd dr)] [msub]))])) node' (assoc node :args (if dr [msub (nd dr)] [msub]))]
;; a flonum field read is a :double operand for the numeric pass (fl-ops); the
;; lookup itself still emits as a keyword/jrec-field-at read, this only feeds
;; its kind up so (* (:x v) (:x v)) over a ^double-fielded record unboxes.
[rt (if (= rt :double) (assoc node' :num-read :double) node')]))
(defn- infer-get-lookup (defn- infer-get-lookup
"(get m :k [default]): the keyword-lookup result type, when the key is a constant "(get m :k [default]): the keyword-lookup result type, when the key is a constant
@ -237,9 +242,10 @@
msub (if (struct-safe? mt) (mark-struct (nd mr) mt) (nd mr)) msub (if (struct-safe? mt) (mark-struct (nd mr) mt) (nd mr))
kr (infer (nth args 1) tenv env) kr (infer (nth args 1) tenv env)
ft (field-type mt (get (nth args 1) :val)) ft (field-type mt (get (nth args 1) :val))
dr (when (= n 3) (infer (nth args 2) tenv env))] dr (when (= n 3) (infer (nth args 2) tenv env))
[(if dr (join ft (ty dr)) ft) rt (if dr (join ft (ty dr)) ft)
(assoc node :args (if dr [msub (nd kr) (nd dr)] [msub (nd kr)]))])) node' (assoc node :args (if dr [msub (nd kr) (nd dr)] [msub (nd kr)]))]
[rt (if (= rt :double) (assoc node' :num-read :double) node')]))
(defn- infer-reduce-hof (defn- infer-reduce-hof
"reduce over a typed vector with a fn-literal: seed the closure's accumulator "reduce over a typed vector with a fn-literal: seed the closure's accumulator