math: fold every remaining pymath primitive through a raw helper table (27 functions), plus isclose/comb/perm - #1378

Merged

youknowone merged 31 commits into

mainfrom

jitcode

Aug 22, 2026

Commits on Aug 22, 2026

objspace: NaN and complex take Python 3.14 pointer identity

kyokuping
authored and
youknowone
committed
f15a5a7
jit: pin w_class on the float list-store fast paths

kyokuping
authored and
youknowone
committed
a10a27f
math: call the __floor__/__ceil__/__trunc__ descriptor without binding it
youknowone
committed
128e605
math: reduce floor/ceil's __float__ fallback through newlong_from_float
youknowone
committed
a4570cc
math: reduce a machine-word gcd pair without rbigint
youknowone
committed
9090c34
jit: read a plain residual builtin call's positionals from the shadow slots
youknowone
committed
1212c9f
jit: specialize math.floor, math.ceil, math.trunc and math.fabs
youknowone
committed
3e8f379
bench(synth): record the wasm jit-stats baselines for the two math fold fixtures
youknowone
committed
83b7d11
math: fold every remaining pymath primitive through a raw helper table
youknowone
committed
bce4595
jit: specialize the generic math float folds and math.isclose
youknowone
committed
3b693d6
bench(synth): merge the four math fold fixtures into one
youknowone
committed
8f08e16
jit: read a residual call's roots through the scope's cached cell
youknowone
committed
73d28c6
interpreter, jit: fold a table of builtins out of their residual call
youknowone
committed
e903efb
bench(synth): add the builtin fold fixture and its native jit-stats baselines
youknowone
committed
13e1e7e
jit: decline a fold before it runs the builtin, and cross-check every float helper
youknowone
committed
ae672fc
jit: give the two-argument builtin fold its result value before it guards
youknowone
committed
ddd9de6
math: convert isclose's operands before checking the tolerances
youknowone
committed
ee2fc92
bench(snippets): make the min/max tie assertion observable
youknowone
committed
e14adca
jit: keep the min/max fold off bigint operands
youknowone
committed
906135d
bench(snippets): cover the bigint pair whose address order is deterministic
youknowone
committed
ad5e124
jit: keep the abs fold off int subclasses
youknowone
committed
dc6b634
builtins: keep bigints out of the all-int sorter
youknowone
committed
9230a8e
builtins: dispatch abs() through the receiver's __abs__
youknowone
committed
e62dfad
comments: cite this branch's upstream references by symbol
youknowone
committed
876b5d4
builtins: give int's __float__ and the __round__ slots a structural body
youknowone
committed
276bbea
portal: key the pypyjit green on the running profile state
youknowone
committed
06ac64a
check: band the collection-schedule guard_failures on inline_freevar_after_mayforce
youknowone
committed
a35309a
check: record builtin_folds_hot's wasm baseline and refit both its ceilings
youknowone
committed
07937df
check: state a wasm ceiling for str_getitem_len_hot, and correct the constant's comment
youknowone
committed
52d28ff
check: double builtin_folds_hot's loop counts and refit its wasm ceiling to 13
youknowone
committed
1422b28
Merge branch 'main' into jitcode
youknowone
authored
2ff0deb

Navigation Menu

math: fold every remaining pymath primitive through a raw helper table (27 functions), plus isclose/comb/perm - #1378

math: fold every remaining pymath primitive through a raw helper table (27 functions), plus isclose/comb/perm#1378
youknowone merged 31 commits into
mainfrom
jitcode

Commits on Aug 22, 2026

objspace: NaN and complex take Python 3.14 pointer identity

jit: pin w_class on the float list-store fast paths

math: call the floor/ceil/trunc descriptor without binding it

math: reduce floor/ceil's float fallback through newlong_from_float

math: reduce a machine-word gcd pair without rbigint

jit: read a plain residual builtin call's positionals from the shadow slots

jit: specialize math.floor, math.ceil, math.trunc and math.fabs

bench(synth): record the wasm jit-stats baselines for the two math fold fixtures

math: fold every remaining pymath primitive through a raw helper table

jit: specialize the generic math float folds and math.isclose

bench(synth): merge the four math fold fixtures into one

jit: read a residual call's roots through the scope's cached cell

interpreter, jit: fold a table of builtins out of their residual call

bench(synth): add the builtin fold fixture and its native jit-stats baselines

jit: decline a fold before it runs the builtin, and cross-check every float helper

jit: give the two-argument builtin fold its result value before it guards

math: convert isclose's operands before checking the tolerances

bench(snippets): make the min/max tie assertion observable

jit: keep the min/max fold off bigint operands

bench(snippets): cover the bigint pair whose address order is deterministic

jit: keep the abs fold off int subclasses

builtins: keep bigints out of the all-int sorter

builtins: dispatch abs() through the receiver's abs

comments: cite this branch's upstream references by symbol

builtins: give int's float and the round slots a structural body

portal: key the pypyjit green on the running profile state

check: band the collection-schedule guard_failures on inline_freevar_after_mayforce

check: record builtin_folds_hot's wasm baseline and refit both its ceilings

check: state a wasm ceiling for str_getitem_len_hot, and correct the constant's comment

check: double builtin_folds_hot's loop counts and refit its wasm ceiling to 13

Merge branch 'main' into jitcode