From b110f453f88002757b7ab0b6635adfc90a526572 Mon Sep 17 00:00:00 2001
From: "J. Nick Koston" <nick@home-assistant.io>
Date: Fri, 24 Apr 2026 17:11:36 -0500
Subject: [PATCH] [core] Outline Scheduler::call cleanup slow path into cold
 combined helper

Fold the to_remove_empty check, cleanup_slow_path_ call, and
MAX_LOGICALLY_DELETED_ITEMS threshold check into a single hot-path
branch + cold outlined helper.

Before this change, the cleanup block in Scheduler::call compiled to
two independent memw + l32i sequences (one for to_remove_empty_() inside
cleanup_(), one for the separate to_remove_count_() check) because GCC
cannot CSE across the memw barriers that std::atomic<uint32_t>::load
emits on Xtensa.

A previous attempt at collapsing these into a single inline branch
(#15985) had the right assembly for the memw count but grew
Scheduler::call by a handful of bytes and rearranged the control flow
enough to nudge sched up ~0.5 us/iter on gatetrigger in practice.

This version goes further: cleanup_slow_combined_ is annotated
noinline + cold so the entire slow path (both reads, both calls) is
pulled out of Scheduler::call entirely. The hot path becomes:

    memw; l32i a8, [to_remove_]; beqz skip
    call8 cleanup_slow_combined_   ; unlikely, cold

and Scheduler::call's body shrinks 344 B -> 332 B (-12 B, below the dev
baseline). Adjacent code (feed_wdt_slow_, etc.) stays in the same flash
region, avoiding the cache-layout side-effects that made earlier
attempts a net loss on busier configs.

The [[unlikely]] attribute on the branch plus the cold attribute on the
helper give the compiler permission to keep the skip path straight and
push the call out-of-line.
---
 esphome/core/scheduler.cpp | 33 +++++++++++++++++++++++++--------
 esphome/core/scheduler.h   |  5 +++++
 2 files changed, 30 insertions(+), 8 deletions(-)

diff --git a/esphome/core/scheduler.cpp b/esphome/core/scheduler.cpp
index 17f1331d182..b1c46a14bdb 100644
--- a/esphome/core/scheduler.cpp
+++ b/esphome/core/scheduler.cpp
@@ -615,14 +615,16 @@ uint32_t HOT Scheduler::call(uint32_t now) {
   }
 #endif /* ESPHOME_DEBUG_SCHEDULER */
 
-  // Cleanup removed items before processing
-  // First try to clean items from the top of the heap (fast path)
-  this->cleanup_();
-
-  // If we still have too many cancelled items, do a full cleanup
-  // This only happens if cancelled items are stuck in the middle/bottom of the heap
-  if (this->to_remove_count_() >= MAX_LOGICALLY_DELETED_ITEMS) {
-    this->full_cleanup_removed_items_();
+  // Cleanup removed items before processing. Fast path: one atomic load +
+  // branch; when nothing is pending, skip both slow-path calls entirely.
+  // Previously the equivalent logic was split across cleanup_() (inline,
+  // loads to_remove_) and a separate to_remove_count_() check for the MAX
+  // threshold, which produced two memw+l32i pairs on Xtensa (GCC can't CSE
+  // across std::atomic's memw barriers). Folding the slow work into one
+  // outlined cold function keeps Scheduler::call's hot-path footprint small
+  // and the I-cache happy.
+  if (this->to_remove_count_() != 0) [[unlikely]] {
+    this->cleanup_slow_combined_();
   }
   // IMPORTANT: This loop uses index-based access (items_[0]), NOT iterators.
   // This is intentional — fired intervals are pushed back into items_ via
@@ -775,6 +777,21 @@ bool HOT Scheduler::cleanup_slow_path_() {
   }
   return !this->items_.empty();
 }
+// Combined cold path for Scheduler::call. Only called when to_remove_ is
+// non-zero. Noinline + cold keeps the two calls and the re-read of
+// to_remove_ out of the main loop's hot path; the attribute also lets the
+// compiler push this code out to a rarely-touched flash region so the
+// scheduler's hot instructions stay in cache.
+void Scheduler::cleanup_slow_combined_() {
+  // First sweep: drop cancelled items from the heap top.
+  this->cleanup_slow_path_();
+  // Re-read to_remove_ because cleanup_slow_path_ may have decremented it.
+  // If cancelled items remain stuck below the top and the count crossed
+  // the threshold, do a full sweep.
+  if (this->to_remove_count_() >= MAX_LOGICALLY_DELETED_ITEMS) {
+    this->full_cleanup_removed_items_();
+  }
+}
 Scheduler::SchedulerItem *HOT Scheduler::pop_raw_locked_() {
   std::pop_heap(this->items_.begin(), this->items_.end(), SchedulerItem::cmp);
 
diff --git a/esphome/core/scheduler.h b/esphome/core/scheduler.h
index 46b19855c35..53ce87ccf2b 100644
--- a/esphome/core/scheduler.h
+++ b/esphome/core/scheduler.h
@@ -316,6 +316,11 @@ class Scheduler {
   }
   // Slow path for cleanup_() when there are items to remove - defined in scheduler.cpp
   bool cleanup_slow_path_();
+  // Combined slow path for Scheduler::call: runs cleanup_slow_path_ then, if
+  // cancelled items are still stuck below the heap top, full_cleanup_removed_items_.
+  // Outlined (noinline, cold) so the Scheduler::call fast path stays one atomic
+  // load + branch — keeping the I-cache hot for the common zero-to-remove case.
+  void __attribute__((noinline, cold)) cleanup_slow_combined_();
   // Slow path for process_to_add() when there are items to merge - defined in scheduler.cpp
   void process_to_add_slow_path_();
   // Remove and return the front item from the heap as a raw pointer.