[api] Add speed_optimized proto option for hot encode paths

Add a new (speed_optimized) message option that emits
__attribute__((optimize("O2"))) on the generated encode() and
calculate_size() methods. Under -Os, GCC does not inline the small
ProtoEncode helpers (write_raw_byte, encode_varint, etc.) into the
generated methods, causing significant overhead on hot paths.

Apply to SensorStateResponse and BluetoothLERawAdvertisementsResponse
which are the highest-frequency encode paths.

esphome/components/api/__init__.py

@@ -1,5 +1,6 @@
 import base64
 import logging
+import pathlib
 
 from esphome import automation
 from esphome.automation import Condition
@@ -458,6 +459,10 @@ async def to_code(config: ConfigType) -> None:
         # Enable optimized memzero/memcmp in libsodium instead of volatile byte loops
         cg.add_build_flag("-DHAVE_WEAK_SYMBOLS=1")
         cg.add_build_flag("-DHAVE_INLINE_ASM=1")
+        # Compile crypto libraries with -O2 for speed instead of -Os.
+        # Crypto is CPU-bound and benefits significantly from speed optimization.
+        # GCC uses the last -O flag, so appending -O2 overrides the global -Os.
+        _write_crypto_optimize_script()
     else:
         cg.add_define("USE_API_PLAINTEXT")
 
@@ -465,6 +470,17 @@ async def to_code(config: ConfigType) -> None:
     cg.add_global(api_ns.using)
 
 
+_CRYPTO_OPTIMIZE_SCRIPT = "crypto_optimize.py"
+
+
+def _write_crypto_optimize_script() -> None:
+    from esphome.helpers import copy_file_if_changed
+
+    script_src = pathlib.Path(__file__).parent / f"{_CRYPTO_OPTIMIZE_SCRIPT}.script"
+    copy_file_if_changed(script_src, CORE.relative_build_path(_CRYPTO_OPTIMIZE_SCRIPT))
+    cg.add_platformio_option("extra_scripts", [f"post:{_CRYPTO_OPTIMIZE_SCRIPT}"])
+
+
 KEY_VALUE_SCHEMA = cv.Schema({cv.string: cv.templatable(cv.string_strict)})
 
 

esphome/components/api/api.proto

@@ -671,6 +671,7 @@ message SensorStateResponse {
   option (source) = SOURCE_SERVER;
   option (ifdef) = "USE_SENSOR";
   option (no_delay) = true;
+  option (speed_optimized) = true;
 
   fixed32 key = 1 [(force) = true];
   float state = 2;
@@ -1638,6 +1639,7 @@ message BluetoothLERawAdvertisementsResponse {
   option (source) = SOURCE_SERVER;
   option (ifdef) = "USE_BLUETOOTH_PROXY";
   option (no_delay) = true;
+  option (speed_optimized) = true;
 
   repeated BluetoothLERawAdvertisement advertisements = 1 [(fixed_array_with_length_define) = "BLUETOOTH_PROXY_ADVERTISEMENT_BATCH_SIZE"];
 }

esphome/components/api/api_options.proto

@@ -23,6 +23,7 @@ extend google.protobuf.MessageOptions {
     optional bool no_delay = 1040 [default=false];
     optional string base_class = 1041;
     optional bool inline_encode = 1042 [default=false];
+    optional bool speed_optimized = 1043 [default=false];
 }
 
 extend google.protobuf.FieldOptions {

esphome/components/api/api_pb2.cpp

@@ -745,7 +745,8 @@ uint32_t ListEntitiesSensorResponse::calculate_size() const {
 #endif
   return size;
 }
-uint8_t *SensorStateResponse::encode(ProtoWriteBuffer &buffer PROTO_ENCODE_DEBUG_PARAM) const {
+__attribute__((optimize("O2"))) uint8_t *SensorStateResponse::encode(
+    ProtoWriteBuffer &buffer PROTO_ENCODE_DEBUG_PARAM) const {
   uint8_t *__restrict__ pos = buffer.get_pos();
   ProtoEncode::write_tag_and_fixed32(pos PROTO_ENCODE_DEBUG_ARG, 13, this->key);
   ProtoEncode::encode_float(pos PROTO_ENCODE_DEBUG_ARG, 2, this->state);
@@ -755,7 +756,7 @@ uint8_t *SensorStateResponse::encode(ProtoWriteBuffer &buffer PROTO_ENCODE_DEBUG
 #endif
   return pos;
 }
-uint32_t SensorStateResponse::calculate_size() const {
+__attribute__((optimize("O2"))) uint32_t SensorStateResponse::calculate_size() const {
   uint32_t size = 0;
   size += 5;
   size += ProtoSize::calc_float(1, this->state);
@@ -2328,7 +2329,8 @@ bool SubscribeBluetoothLEAdvertisementsRequest::decode_varint(uint32_t field_id,
   }
   return true;
 }
-uint8_t *BluetoothLERawAdvertisementsResponse::encode(ProtoWriteBuffer &buffer PROTO_ENCODE_DEBUG_PARAM) const {
+__attribute__((optimize("O2"))) uint8_t *BluetoothLERawAdvertisementsResponse::encode(
+    ProtoWriteBuffer &buffer PROTO_ENCODE_DEBUG_PARAM) const {
   uint8_t *__restrict__ pos = buffer.get_pos();
   for (uint16_t i = 0; i < this->advertisements_len; i++) {
     auto &sub_msg = this->advertisements[i];
@@ -2350,7 +2352,7 @@ uint8_t *BluetoothLERawAdvertisementsResponse::encode(ProtoWriteBuffer &buffer P
   }
   return pos;
 }
-uint32_t BluetoothLERawAdvertisementsResponse::calculate_size() const {
+__attribute__((optimize("O2"))) uint32_t BluetoothLERawAdvertisementsResponse::calculate_size() const {
   uint32_t size = 0;
   for (uint16_t i = 0; i < this->advertisements_len; i++) {
     auto &sub_msg = this->advertisements[i];

esphome/components/api/crypto_optimize.py.script

@@ -0,0 +1,9 @@
+# Compile crypto libraries with -O2 for speed instead of the default -Os.
+# Crypto is CPU-bound and benefits significantly from speed optimization.
+# GCC uses the last -O flag, so appending -O2 overrides the global -Os
+# for these libraries only.
+Import("env")
+
+for lb in env.GetLibBuilders():
+    if lb.name in ("noise-c", "libsodium"):
+        lb.env.Append(CCFLAGS=["-O2"])

script/api_protobuf/api_protobuf.py

@@ -2679,6 +2679,13 @@ def build_message_type(
         and get_opt(desc, inline_opt, False)
     )
 
+    # Check if this message wants speed-optimized encode/calculate_size.
+    # When set, __attribute__((optimize("O2"))) is added to the definitions
+    # so GCC inlines the small ProtoEncode helpers even under -Os.
+    speed_opt = getattr(pb, "speed_optimized", None)
+    is_speed_optimized = speed_opt is not None and get_opt(desc, speed_opt, False)
+    speed_attr = '__attribute__((optimize("O2"))) ' if is_speed_optimized else ""
+
     # Only generate encode method if this message needs encoding and has fields
     if needs_encode and encode and not is_inline_only:
         # Add PROTO_ENCODE_DEBUG_ARG after pos in all proto_* calls
@@ -2688,7 +2695,7 @@ def build_message_type(
             )
             for line in encode
         ]
-        o = f"uint8_t *{desc.name}::encode(ProtoWriteBuffer &buffer PROTO_ENCODE_DEBUG_PARAM) const {{\n"
+        o = f"{speed_attr}uint8_t *{desc.name}::encode(ProtoWriteBuffer &buffer PROTO_ENCODE_DEBUG_PARAM) const {{\n"
         o += "  uint8_t *__restrict__ pos = buffer.get_pos();\n"
         o += indent("\n".join(encode_debug)) + "\n"
         o += "  return pos;\n"
@@ -2702,7 +2709,7 @@ def build_message_type(
 
     # Add calculate_size method only if this message needs encoding and has fields
     if needs_encode and size_calc and not is_inline_only:
-        o = f"uint32_t {desc.name}::calculate_size() const {{\n"
+        o = f"{speed_attr}uint32_t {desc.name}::calculate_size() const {{\n"
         o += "  uint32_t size = 0;\n"
         o += indent("\n".join(size_calc)) + "\n"
         o += "  return size;\n"

script/cpp_benchmark.py

@@ -26,12 +26,11 @@ CORE_BENCHMARKS_DIR: Path = Path(root_path) / "tests" / "benchmarks" / "core"
 STUBS_DIR: Path = Path(root_path) / "tests" / "benchmarks" / "stubs"
 
 PLATFORMIO_OPTIONS = {
-    "build_unflags": [
-        "-Os",  # remove default size-opt
-    ],
     "build_flags": [
-        "-O2",  # optimize for speed (CodSpeed recommends RelWithDebInfo)
+        "-Os",  # match firmware optimization level (detects inlining regressions)
         "-g",  # debug symbols for profiling
+        "-ffunction-sections",  # required for dead-code stripping with -Os
+        "-fdata-sections",  # required for dead-code stripping with -Os
         "-DUSE_BENCHMARK",  # disable WarnIfComponentBlockingGuard in finish()
         f"-I{STUBS_DIR}",  # stub headers for ESP32-only components
     ],