Repository navigation

perf(jit): cumulative improvement of JIT performance - #2

Merged

fffonion merged 19 commits into

masterfrom

codex/perf-experiment

Jul 16, 2026

Commits on Jul 16, 2026
- perf(jit): avoid redundant write-protect on trace entry
  fffonion
  committed
  b773443
- perf(jit): index native traces by dense id
  fffonion
  committed
  c3fa7cb
- perf(jit): use dense trace entry lookup
  fffonion
  committed
  d991211
- perf(jit): use dense loop header lookup
  fffonion
  committed
  4438670
- perf(jit): keep non-yielding host calls in native traces
  fffonion
  committed
  960fb2a
- perf(jit): link hot non-yielding side exits
  fffonion
  committed
  4fcd2df
- perf(jit): inline scalar-only side-exit restores
  fffonion
  committed
  c509091
- perf(jit): dispatch linked traces from the outer loop
  fffonion
  committed
  7e6e7dd
- perf(jit): move owned exit values into locals
  fffonion
  committed
  cf05a12
- perf(jit): initialize owned temps inline
  fffonion
  committed
  f636bf8
- perf(jit): fast-path compiled trace handoffs
  fffonion
  committed
  b824f03
- perf(jit): recognize array length set as push
  fffonion
  committed
  680d6af
- perf(jit): reject non-loop entries before blocked lookup
  fffonion
  committed
  93261c2
- perf(jit): borrow array reads into host calls
  fffonion
  committed
  1bdcf9f
- perf(jit): specialize collection mutation bridges
  fffonion
  committed
  4396120
- perf(jit): borrow array reads into collection ops
  fffonion
  committed
  126ac00
- fix(vm): enforce non-yielding host return contract
  fffonion
  committed
  481bb97
- fix(jit): isolate native bridge errors per thread
  fffonion
  committed
  7ba306a
- chore(jit): remove redundant terminal clones
  fffonion
  committed
  1a71f65

Navigation Menu

Repository navigation

perf(jit): cumulative improvement of JIT performance - #2

perf(jit): cumulative improvement of JIT performance#2
fffonion merged 19 commits into
masterfrom
codex/perf-experiment

Commits on Jul 16, 2026

perf(jit): avoid redundant write-protect on trace entry

perf(jit): index native traces by dense id

perf(jit): use dense trace entry lookup

perf(jit): use dense loop header lookup

perf(jit): keep non-yielding host calls in native traces

perf(jit): link hot non-yielding side exits

perf(jit): inline scalar-only side-exit restores

perf(jit): dispatch linked traces from the outer loop

perf(jit): move owned exit values into locals

perf(jit): initialize owned temps inline

perf(jit): fast-path compiled trace handoffs

perf(jit): recognize array length set as push

perf(jit): reject non-loop entries before blocked lookup

perf(jit): borrow array reads into host calls

perf(jit): specialize collection mutation bridges

perf(jit): borrow array reads into collection ops

fix(vm): enforce non-yielding host return contract

fix(jit): isolate native bridge errors per thread

chore(jit): remove redundant terminal clones