Compare commits

...

4 Commits

Author SHA1 Message Date
J. Nick Koston
63184e95a2 [api] Only apply -O2 to libsodium, not noise-c
CodSpeed results show only Noise benchmarks improved with -O2 —
the speedup comes from libsodium's crypto primitives (Curve25519,
ChaCha20, Poly1305), not noise-c's protocol layer. Narrow the
optimization to libsodium only.
2026-04-12 19:15:06 -10:00
J. Nick Koston
5a250cc74f [api] Compile noise-c and libsodium with -O2 for speed
Crypto libraries are CPU-bound and benefit significantly from speed
optimization over the default -Os. Add a post: extra_script that
appends -O2 to noise-c and libsodium build flags when API noise
encryption is enabled. GCC uses the last -O flag, so this overrides
the global -Os for these libraries only.
2026-04-12 19:03:21 -10:00
J. Nick Koston
02f828fcbf [benchmark] Use -Os to match firmware optimization level
CodSpeed benchmarks were building with -O2, while all firmware
targets (ESP8266, ESP32, LibreTiny) use -Os. This mismatch means
the benchmarks cannot detect inlining regressions that affect real
devices — GCC under -O2 inlines functions that -Os outlines due to
its size-conscious cost model.

Switch to -Os with -ffunction-sections/-fdata-sections for proper
dead-code stripping (needed because -Os preserves references that
-O2 optimizes away at compile time).
2026-04-12 18:37:50 -10:00
J. Nick Koston
ab64916c37 [benchmark] Use -Os to match firmware optimization level
CodSpeed benchmarks were building with -O2, while all firmware
targets (ESP8266, ESP32, LibreTiny) use -Os. This mismatch means
the benchmarks cannot detect inlining regressions that affect real
devices — GCC under -O2 inlines functions that -Os outlines due to
its size-conscious cost model.

Remove the -Os unflag and -O2 override so benchmarks use the
platform default -Os, matching what actually runs on devices.
2026-04-12 18:32:03 -10:00
3 changed files with 28 additions and 4 deletions

View File

@@ -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)})

View File

@@ -0,0 +1,9 @@
# Compile libsodium with -O2 for speed instead of the default -Os.
# libsodium provides the crypto primitives (Curve25519, ChaCha20, Poly1305)
# used by the Noise protocol and benefits significantly from speed optimization.
# GCC uses the last -O flag, so appending -O2 overrides the global -Os.
Import("env")
for lb in env.GetLibBuilders():
if lb.name == "libsodium":
lb.env.Append(CCFLAGS=["-O2"])

View File

@@ -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
],