Fix benchmark, add more inlines

* These benchmarks were not run before the last release, but due to recent stable
  rustc versions the benchmarks were more in the 7ms range. This moves the needle
  back to acceptable levels by forcing more functions to be inlined.

CHANGELOG.md

@@ -6,6 +6,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## Unreleased
 
+## [v0.8.1] 2024-09-03
+- Restore performance for recent rustc versions, force more functions to be inlined
+
 ## [v0.8.0] 2024-09-02
 - Update MSRV to `1.74` to [!130](https://github.com/rsadsb/dump1090_rs/pull/130)
 - Properly decode short ADSB messages, thanks [@gariac](https://github.com/gariac) for finding. [!130](https://github.com/rsadsb/dump1090_rs/pull/130)

README.md

@@ -104,9 +104,9 @@ Reading from a 512KB iq sample to ADS-B bytes takes ~3.0 ms, but feel free to ru
 
 ### Intel i7-7700K CPU @ 4.20GHz
 ```
-01                      time:   [2.9850 ms 2.9859 ms 2.9868 ms]
-02                      time:   [2.9260 ms 2.9279 ms 2.9308 ms]
-03                      time:   [2.8449 ms 2.8525 ms 2.8653 ms]
+01                      time:   [3.6691 ms 3.6950 ms 3.7264 ms]
+02                      time:   [3.5941 ms 3.5987 ms 3.6040 ms]
+03                      time:   [3.4930 ms 3.4961 ms 3.4994 ms]
 ```
 
 # Changes

benches/demod_benchmark.rs

@@ -1,56 +1,28 @@
-// third-party
-use assert_hex::assert_eq_hex;
+use std::hint::black_box;
+
 use criterion::{criterion_group, criterion_main, Criterion};
-use hexlit::hex;
-// crate
 use libdump1090_rs::{demod_2400::demodulate2400, icao_filter::icao_flush, utils};
 use num_complex::Complex;
 
-fn routine(data: [Complex<i16>; 0x20000], expected_data: &Vec<Vec<u8>>) {
+fn routine(data: [Complex<i16>; 0x20000]) {
     // make sure icao starts in a deterministic position
     icao_flush();
     let outbuf = utils::to_mag(&data);
-    let data = demodulate2400(&outbuf).unwrap();
-    for (a, b) in data.iter().zip(expected_data.iter()) {
-        assert_eq_hex!(a.buffer(), *b);
-    }
+    let _ = black_box(demodulate2400(&outbuf).unwrap());
 }
 
 fn criterion_benchmark(c: &mut Criterion) {
     let filename = "test_iq/test_1641427457780.iq";
     let data_01 = utils::read_test_data(filename);
-    let expected_data_01 = Vec::from([
-        hex!("8dad929358b9c6273f002169c02e").to_vec(),
-        hex!("8daa2bc4f82100020049b8db9449").to_vec(),
-        hex!("8daa2bc4f82100020049b8db9449").to_vec(),
-        hex!("02e1971ce17c84").to_vec(),
-        hex!("8da0aaa058bf163fcf860013e840").to_vec(),
-    ]);
 
     let filename = "test_iq/test_1641428165033.iq";
     let data_02 = utils::read_test_data(filename);
-    let expected_data_02 = Vec::from([
-        hex!("8da79de99909932f780c9e2f2f8f").to_vec(),
-        hex!("8dac04d358a7820a86ac3709e689").to_vec(),
-        hex!("8dac04d3ea4288669b5c082751d4").to_vec(),
-        hex!("8da79de958bdf59c85104874adad").to_vec(),
-        hex!("5dad92936265f5").to_vec(),
-        hex!("5dad92936265f525be017735997b").to_vec(),
-    ]);
 
     let filename = "test_iq/test_1641428106243.iq";
     let data_03 = utils::read_test_data(filename);
-    let expected_data_03 = Vec::from([
-        hex!("8da8aac8990c30b51808aa24e573").to_vec(),
-        hex!("02e19838bff1d9").to_vec(),
-        hex!("8dada6b9990cf61e4848af2a8656").to_vec(),
-        hex!("8da4ba025885462008fa0a4a6eb2").to_vec(),
-        hex!("8da4ba025885462008fa0a4a6eb2").to_vec(),
-        hex!("8da4ba0299115f301074a72db6ff").to_vec(),
-    ]);
-    c.bench_function("01", |b| b.iter(|| routine(data_01, &expected_data_01)));
-    c.bench_function("02", |b| b.iter(|| routine(data_02, &expected_data_02)));
-    c.bench_function("03", |b| b.iter(|| routine(data_03, &expected_data_03)));
+    c.bench_function("01", |b| b.iter(|| routine(data_01)));
+    c.bench_function("02", |b| b.iter(|| routine(data_02)));
+    c.bench_function("03", |b| b.iter(|| routine(data_03)));
 }
 
 criterion_group!(benches, criterion_benchmark);

src/crc.rs

@@ -259,6 +259,7 @@ pub const CRC_TABLE: [u32; 256] = [
     0x00fa_0480,
 ];
 
+#[inline]
 pub fn modes_checksum(message: &[u8], bits: usize) -> u32 {
     let mut rem: u32 = 0;
     let n = bits / 8;

src/demod_2400.rs

@@ -19,6 +19,7 @@ enum Phase {
 }
 
 impl From<usize> for Phase {
+    #[inline(always)]
     fn from(num: usize) -> Self {
         match num % 5 {
             0 => Self::Zero,
@@ -33,6 +34,7 @@ impl From<usize> for Phase {
 
 impl Phase {
     /// Increment from 0..4 for incrementing the starting phase
+    #[inline(always)]
     fn next_start(self) -> Self {
         match self {
             Self::Zero => Self::One,
@@ -44,6 +46,7 @@ impl Phase {
     }
 
     /// Increment by expected next phase transition for bit denoting
+    #[inline(always)]
     fn next(self) -> Self {
         match self {
             Self::Zero => Self::Two,
@@ -55,6 +58,7 @@ impl Phase {
     }
 
     /// Amount of mag indexs used, for adding to the next start index
+    #[inline(always)]
     fn increment_index(self, index: usize) -> usize {
         index
             + match self {
@@ -98,6 +102,7 @@ pub struct ModeSMessage {
 }
 
 impl ModeSMessage {
+    #[inline(always)]
     pub fn buffer(&self) -> &[u8] {
         match self.msglen {
             MsgLen::Short => &self.msg[..MODES_SHORT_MSG_BYTES],
@@ -106,6 +111,7 @@ impl ModeSMessage {
     }
 }
 
+#[inline(always)]
 pub fn demodulate2400(mag: &MagnitudeBuffer) -> Result<Vec<ModeSMessage>, &'static str> {
     let mut results = vec![];
 
@@ -205,6 +211,7 @@ pub fn demodulate2400(mag: &MagnitudeBuffer) -> Result<Vec<ModeSMessage>, &'stat
     Ok(results)
 }
 
+#[inline(always)]
 fn check_preamble(preamble: &[u16]) -> Option<(i32, u32, u32)> {
     // This gets rid of the 3 core::panicking::panic_bounds_check calls,
     // but doesn't look to improve performance

src/mode_s/mod.rs

@@ -10,6 +10,7 @@ use super::{crc::modes_checksum, icao_filter::icao_filter_test};
 
 // mode_s.c:215
 #[must_use]
+#[inline(always)]
 pub fn getbits(data: &[u8], firstbit_1idx: usize, lastbit_1idx: usize) -> usize {
     let mut ans: usize = 0;
 
@@ -33,7 +34,7 @@ pub fn getbits(data: &[u8], firstbit_1idx: usize, lastbit_1idx: usize) -> usize
 pub fn score_modes_message(msg: &[u8]) -> Option<(MsgLen, i32)> {
     let validbits = msg.len() * 8;
 
-    if validbits < 56 {
+    if validbits < MODES_SHORT_MSG_BYTES * 8 {
         return None;
     }