Block arithmetic on restricted floats - #286

Draft

maleadt wants to merge 15 commits into

mainfrom

tb/restricted_float_arithmetic

Commits on Jul 28, 2026

Make `is_restricted_float` extensible

maleadt
and
claude
committed
17e8029
Reject restricted floats in direct tile arithmetic

maleadt
and
claude
committed
9130d56
Block implicit FP8/FP4 arithmetic in kernels

maleadt
and
claude
committed
2fac994
Cast explicitly in the FP8 multiply-add tests

maleadt
and
claude
committed
654e60c
Test that restricted floats reject arithmetic

maleadt
and
claude
committed
c00c205
Document how restricted-float arithmetic is rejected

maleadt
and
claude
committed
1aa7efd
Test that reduce and scan reject extension-registered restricted floats

maleadt
and
claude
committed
415cc42
Overlay DLFP8Types comparisons as Float32 upcasts

maleadt
and
claude
committed
4cdafbe
Drop the FP8 multiply-add execution tests

maleadt
and
claude
committed
9dc8d2f
Drop the operator argument from check_arithmetic

maleadt
and
claude
committed
871c692
Gate restricted floats at the broadcast and map layer

maleadt
and
claude
committed
4ee2b7d
Drop the restricted-float blocklist overlays

maleadt
and
claude
committed
c50ea85
Test what the restricted-float gate lets through

maleadt
and
claude
committed
89ec986
Document what numeric floats still allow

maleadt
and
claude
committed
70c02ca
Describe the actual broadcast/map relationship

maleadt
and
claude
committed
595aca6