Agent:claude-fastfields-to-teeny
external/teeny is pinned at dcd591c, which is the tip of teeny's stale
claude/fastfields-teeny-refactor-js42id branch — 125 commits behind teeny's
main. .gitmodules still names that branch too, so git submodule update --remote would keep us frozen there.
Several landings this project asked for are on the far side of that gap, and
fastfields-kernels#50 (the regulariser tap-table engine) depends on some of them:
| landing | teeny PR |
|---|
dispatch_values — product-form runtime→static dispatch | #465 |
forward+backward scan_ sweep idiom docs | #466 |
scan_/scan sequential fold along one axis | #254 |
unfold<Axis>(size, step) sliding-window view | #256 |
subsample<Axes...>(k, starts...) coloured-lattice sugar | #258 |
peel_zip<Axes...> lock-step multi-tensor iteration | #327 |
index_select<Axis>(idx) gather | #326 |
in-place maximum_/minimum_ | #325 |
sqdist/dist wrappers | #324 |
host-side atomic fetch_add_/sub_ | #257 |
dispatch_values (#465) is what kernels#50 phase 6 is written against, and
subsample (#258) is what kernels#50 phase 4 is written against — neither is
reachable at the current pin.
Breaking changes in the gap
Two commits in the range are marked breaking:
refactor!: rename take_along to slice_along (teeny#423)refactor(iterate)!: move scan_/scan's axis<A>{} tag to trailing (teeny#348)
The second only affects scan_/scan, which did not exist at the old pin, so it
cannot break an existing call site. The first is the one to check.
Scope
- Bump
external/teeny to teeny main (currently 5c46bd4). - Point
.gitmodulesbranch at main instead of the stale refactor branch. - Fix any surviving
take_along call site across the fastfields repos. - Gate:
make -C fastfields-cpu-lib clean test on bothclang++ and g++,
with check counts unchanged from before the bump.
Part of the teeny-refactor umbrella (fastfields-lib#21); a prerequisite for fastfields/fastfields-kernels#50.
Agent:
claude-fastfields-to-teenyexternal/teenyis pinned atdcd591c, which is the tip of teeny's staleclaude/fastfields-teeny-refactor-js42idbranch — 125 commits behind teeny'smain..gitmodulesstill names that branch too, sogit submodule update --remotewould keep us frozen there.Several landings this project asked for are on the far side of that gap, and
fastfields-kernels#50 (the regulariser tap-table engine) depends on some of them:
dispatch_values— product-form runtime→static dispatchscan_sweep idiom docsscan_/scansequential fold along one axisunfold<Axis>(size, step)sliding-window viewsubsample<Axes...>(k, starts...)coloured-lattice sugarpeel_zip<Axes...>lock-step multi-tensor iterationindex_select<Axis>(idx)gathermaximum_/minimum_sqdist/distwrappersfetch_add_/sub_dispatch_values(#465) is what kernels#50 phase 6 is written against, andsubsample(#258) is what kernels#50 phase 4 is written against — neither isreachable at the current pin.
Breaking changes in the gap
Two commits in the range are marked breaking:
refactor!: rename take_along to slice_along(teeny#423)refactor(iterate)!: move scan_/scan's axis<A>{} tag to trailing(teeny#348)The second only affects
scan_/scan, which did not exist at the old pin, so itcannot break an existing call site. The first is the one to check.
Scope
external/teenyto teenymain(currently5c46bd4)..gitmodulesbranchatmaininstead of the stale refactor branch.take_alongcall site across the fastfields repos.make -C fastfields-cpu-lib clean teston bothclang++andg++,with check counts unchanged from before the bump.
Part of the teeny-refactor umbrella (fastfields-lib#21); a prerequisite for fastfields/fastfields-kernels#50.