fix(ep-cpu): check the spin deadline on every yield at the two sites #1825 missed

crates/onnx-runtime-ep-cpu/src/decode_spmd.rs

@@ -154,10 +154,6 @@ pub const WORKER_PROFILE_ENV: &str = "ONNX_GENAI_CPU_DECODE_WORKER_PROFILE";
 /// pure spinning and only genuinely idle gaps ramp into yielding then parking.
 const SPIN_LOOP_BUDGET: u32 = 1 << 12;
 
-/// Spin iterations between wall-clock checks, so `Instant::now()` (a vDSO read,
-/// ~20 ns) is amortised over the hot spin loop rather than read every iteration.
-const CLOCK_CHECK_STRIDE: u32 = 1 << 6;
-
 /// How long [`SpmdDecodePools::build_with_schedule`]'s readiness barrier waits
 /// for every spawned worker to announce itself before declaring the pool
 /// unbuildable.
@@ -859,7 +855,28 @@ impl SharedState {
                 std::hint::spin_loop();
             } else {
                 thread::yield_now();
-                if spins.is_multiple_of(CLOCK_CHECK_STRIDE) && start.elapsed() >= blocktime {
+                // Check the clock on *every* yield, not on a stride -- the same
+                // correction #1825 made to the readiness barrier below, which
+                // missed this site. The clock is never read during the pure
+                // spin phase here, so a stride amortised nothing: its only
+                // effect was to multiply the granularity of the blocktime
+                // deadline by 64 yields. `SPIN_LOOP_BUDGET` (4096) is itself a
+                // multiple of that removed 64-iteration stride, so the yield
+                // phase began exactly on a stride boundary -- the deadline was
+                // evaluated once, on the first yield, and then not again for 64
+                // more. A yield costs microseconds to milliseconds under
+                // contention, and contention is exactly when a worker holding a
+                // core past the window it was told to release at does the most
+                // damage. `Instant::now()` is a vDSO read against a yield that
+                // costs orders of magnitude more -- measured on this host,
+                // 32ns per read against 1214ns for an *uncontended*
+                // `yield_now`, i.e. 2.6%, and the fraction only shrinks as
+                // contention makes the yield slower. Checking every time is
+                // free exactly where it matters.
+                // With the stride gone this file has no clock-stride constant
+                // left: its only use was in the phase its own doc comment said
+                // it did not apply to.
+                if start.elapsed() >= blocktime {
                     break;
                 }
             }

crates/onnx-runtime-ep-cpu/src/task_runtime/pool.rs

@@ -98,8 +98,11 @@ const MAX_SPIN: Duration = Duration::from_micros(500);
 /// Pure `spin_loop` iterations before a spinning worker starts yielding.
 const SPIN_LOOP_BUDGET: u32 = 1 << 12;
 
-/// Spin iterations between wall-clock reads, so `Instant::now()` (~20 ns) is
-/// amortised rather than paid every iteration.
+/// Spin iterations between wall-clock reads *during the pure-`spin_loop` phase*,
+/// so `Instant::now()` (~20 ns) is amortised rather than paid every iteration.
+/// It deliberately does not gate the yield phase: a yield costs microseconds to
+/// milliseconds under contention, so a stride there would multiply the spin
+/// window's granularity by 64 yields of an already-starved thread (#1825).
 const CLOCK_CHECK_STRIDE: u32 = 1 << 6;
 
 /// Dispatcher spins between `yield_now` calls while waiting for stragglers.
@@ -395,11 +398,34 @@ impl Shared {
                 spins = spins.wrapping_add(1);
                 if spins < SPIN_LOOP_BUDGET {
                     std::hint::spin_loop();
+                    // The stride belongs here and only here: a `spin_loop`
+                    // iteration costs nanoseconds, so an unamortised
+                    // `Instant::now()` would dominate the phase. And the check
+                    // has to run here at all, because at the converged idle
+                    // window (`MIN_SPIN`, 20us) the deadline expires *before*
+                    // the spin phase ends: 4096 `spin_loop`s measure 128us on
+                    // this host.
+                    if spins.is_multiple_of(CLOCK_CHECK_STRIDE) && start.elapsed() >= spin_window {
+                        break;
+                    }
                 } else {
                     thread::yield_now();
-                }
-                if spins.is_multiple_of(CLOCK_CHECK_STRIDE) && start.elapsed() >= spin_window {
-                    break;
+                    // ...and must not apply here. A yield costs microseconds to
+                    // milliseconds under contention, so a stride of 64
+                    // multiplies the window's granularity by 64 yields of an
+                    // already-starved thread -- see #1825, which made this
+                    // correction in `decode_spmd`'s readiness barrier. It bites
+                    // whenever the window outlasts the spin phase, which is the
+                    // whole grown range: 4096 `spin_loop`s measure 128us on
+                    // this host, well under `MAX_SPIN` (500us), so every window
+                    // above the floor reaches the yield phase. `MAX_SPIN`'s
+                    // contract is that a process which stops inferencing
+                    // returns to ~0% CPU in under a millisecond; a
+                    // stride-gated check cannot honour that under exactly the
+                    // load that makes it matter.
+                    if start.elapsed() >= spin_window {
+                        break;
+                    }
                 }
             }
             if caught {