Merge pull request #102 from LebedevRI/develop

Benchmark attempt

Author: LebedevRI

.github/workflows/CI-linux.yml

@@ -595,7 +595,7 @@ jobs:
           cd "${{ env.SRC_DIR }}"
           cargo miri nextest run --target ${{ env.RUST_TARGET }}
       - name: Run tests under Sanitizer
-        timeout-minutes: 3
+        timeout-minutes: 5
         if: inputs.job-flavor == 'ASAN' || inputs.job-flavor == 'MSAN' || inputs.job-flavor == 'TSAN'
         run: |
           set -xe
@@ -617,6 +617,14 @@ jobs:
           export ${{ inputs.job-flavor }}_OPTIONS="allocator_may_return_null=1"
           cd "${{ env.SRC_DIR }}"
           cargo test --all-targets $BUILD_ARGS --target ${{ env.RUST_TARGET }} --profile ${{ inputs.build-profile }}-sanitized
+      - name: Dry-run benchmarks
+        timeout-minutes: 1
+        if: inputs.job-flavor == 'plain'
+        run: |
+          set -xe
+          . "$HOME/.cargo/env"
+          cd "${{ env.SRC_DIR }}"
+          cargo bench --target ${{ env.RUST_TARGET }} --profile ${{ inputs.build-profile }} --benches -- --quick --warm-up-time 1e-9 --measurement-time 1e-9 --noplot
       - name: Perform SonarQube static analysis
         timeout-minutes: 1
         if: inputs.job-flavor == 'SonarQube'

.github/workflows/CI-macOS.yml

@@ -4,6 +4,9 @@ on:
       os:
         required: true
         type: string
+      os-ver:
+        required: true
+        type: string
       runs-on:
         required: true
         type: string
@@ -32,6 +35,7 @@ env:
   RUSTFLAGS: "-Dwarnings"
   SRC_DIR: ${{ github.workspace }}/rawspeed.rs
   CODECOV_TOKEN_EXISTS: ${{ secrets.CODECOV_TOKEN != '' }}
+  MACOSX_DEPLOYMENT_TARGET: ${{ inputs.os-ver }}
 
 jobs:
   macOS:
@@ -251,3 +255,10 @@ jobs:
           files: ${{ env.SRC_DIR }}/codecov.json
           root_dir: ${{ env.SRC_DIR }}
           fail_ci_if_error: true
+      - name: Dry-run benchmarks
+        timeout-minutes: 1
+        run: |
+          set -xe
+          . "$HOME/.cargo/env"
+          cd "$SRC_DIR"
+          cargo bench --profile ${{ inputs.build-profile }} --benches -- --quick --warm-up-time 1e-9 --measurement-time 1e-9 --noplot

.github/workflows/CI.yml

@@ -126,6 +126,7 @@ jobs:
     uses: ./.github/workflows/CI-macOS.yml
     with:
       os: ${{ matrix.compiler.os }}
+      os-ver: ${{ matrix.compiler.os-ver }}
       runs-on: ${{ matrix.compiler.os }}-${{ matrix.compiler.os-ver }}
       builder-host: ${{ matrix.compiler.host }}
       builder-target: ${{ matrix.compiler.target }}

Cargo.toml

@@ -34,6 +34,7 @@ members = [
     "src/metadata/xmltokendesparsifier",
     "src/metadata/xmltokenizer",
     "src/misc/md5",
+    "benches/misc/md5",
     "src/parsers",
     "src/parsers/rawparser",
     "src/std",

benches/misc/md5/Cargo.toml

@@ -0,0 +1,22 @@
+[package]
+name = "rawspeed-bench-md5"
+version.workspace = true
+authors.workspace = true
+edition.workspace = true
+rust-version.workspace = true
+documentation.workspace = true
+homepage.workspace = true
+repository.workspace = true
+license.workspace = true
+
+[lints]
+workspace = true
+
+[dev-dependencies]
+rawspeed-misc-md5 = { path = "../../../src/misc/md5" }
+criterion = { version = "0.8.2", default-features = false, features = [] }
+
+[[bench]]
+name = "rawspeed-bench-md5"
+path = "mod.rs"
+harness = false

benches/misc/md5/mod.rs

@@ -0,0 +1,47 @@
+use criterion::{
+    AxisScale, BenchmarkId, Criterion, PlotConfiguration, Throughput,
+    criterion_group, criterion_main,
+};
+use rawspeed_misc_md5::md5::{self, MD5};
+
+#[inline(never)]
+fn md5_slice(input: &[u8]) -> md5::MD5State {
+    let mut hasher = MD5::default();
+    hasher.extend(input);
+    hasher.flush()
+}
+
+fn md5_benchmark(c: &mut Criterion) {
+    static KIB: usize = 1024;
+
+    let sizes: Box<dyn Iterator<Item = _>> = if true {
+        let sizes = [4 * KIB].into_iter();
+        Box::new(sizes)
+    } else {
+        let sizes =
+            core::iter::successors(Some(1_usize), |&prev| prev.checked_mul(2))
+                .take_while(|s| *s <= 2 * KIB * KIB);
+        let sizes = [0].into_iter().chain(sizes);
+        Box::new(sizes)
+    };
+
+    let mut group = c.benchmark_group("md5");
+    group.plot_config(
+        PlotConfiguration::default().summary_scale(AxisScale::Logarithmic),
+    );
+    for size in sizes {
+        group.throughput(Throughput::Bytes(size.try_into().unwrap()));
+        group.bench_with_input(
+            BenchmarkId::from_parameter(size),
+            &size,
+            |b, size| {
+                let input = core::hint::black_box(vec![0; *size]);
+                b.iter(|| md5_slice(input.as_slice()));
+            },
+        );
+    }
+    group.finish();
+}
+
+criterion_group!(benches, md5_benchmark);
+criterion_main!(benches);

src/misc/md5/md5/mod.rs

@@ -168,29 +168,33 @@ impl MD5State {
     }
 
     #[inline]
-    fn md5_compress(&mut self, block: &MD5Block) {
-        let mut schedule: [u32; 16] = Default::default();
-
-        for (s, bytes) in schedule.iter_mut().zip(block.0.chunks_exact(32 / 8))
-        {
-            *s = u32::from_le_bytes(bytes.try_into().unwrap());
-        }
+    #[expect(clippy::many_single_char_names)]
+    fn round<S>(&mut self, r: MD5Round, step: &StepParams, schedule: S)
+    where
+        S: Fn(usize) -> u32,
+    {
+        let [a, b, c, d] = self.0;
+        let expr = r.get_expr(*self);
+        let a = a
+            .wrapping_add(expr)
+            .wrapping_add(step.t)
+            .wrapping_add(schedule(step.k))
+            .rotate_left(step.s)
+            .wrapping_add(b);
+        self.0 = [d, a, b, c];
+    }
 
-        let round_impl = |r: MD5Round, tmp: MD5State, step: &StepParams| {
-            let [a, b, _rest @ ..] = tmp.0;
-            let expr = r.get_expr(tmp);
-            let a = a
-                .wrapping_add(expr)
-                .wrapping_add(step.t)
-                .wrapping_add(*schedule.get(step.k).unwrap());
-            b.wrapping_add(a.rotate_left(step.s))
+    #[inline]
+    fn md5_compress(&mut self, block: &MD5Block) {
+        let schedule = |k: usize| -> u32 {
+            let bytes = block.0.chunks_exact(32 / 8).nth(k).unwrap();
+            u32::from_le_bytes(bytes.try_into().unwrap())
         };
 
         let mut tmp = *self;
         for (round, steps) in STAGES.iter().enumerate() {
             for step in steps {
-                tmp.0[0] = round_impl(MD5Round(round), tmp, step);
-                tmp.0.rotate_right(1);
+                tmp.round(MD5Round(round), step, schedule);
             }
         }
 

src/std/wrapping_numerics_arith/tests/add_unsigned_and_signed.rs

@@ -35,7 +35,7 @@ fn run_test(lhs: BoundUnsigned<T>, rhs: U) {
 }
 
 #[test]
-#[cfg(not(miri))]
+#[cfg_attr(miri, ignore)]
 fn exhaustive_test() {
     for bound in T::MIN..=T::MAX {
         let Some(bound) = Bound::new(bound) else {

src/std/wrapping_numerics_arith/tests/add_unsigned_and_unsigned.rs

@@ -23,7 +23,7 @@ fn run_test(lhs: BoundUnsigned<T>, rhs: T) {
 }
 
 #[test]
-#[cfg(not(miri))]
+#[cfg_attr(miri, ignore)]
 fn add_test_exhaustive() {
     for bound in u8::MIN..=u8::MAX {
         let Some(bound) = Bound::new(bound) else {

tests/metadata/camerasxml_parser/xmllint.rs

@@ -1,7 +1,7 @@
 #[cfg(test)]
 mod tests {
     #[test]
-    #[cfg(not(miri))]
+    #[cfg_attr(miri, ignore)]
     fn xmllint() -> Result<(), Box<dyn core::error::Error>> {
         use rawspeed_metadata_camerasxml_parser::camerasxml_parser;
         let camerasxml_path =