diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index aea7af7..38496f2 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -13,19 +13,43 @@ env:
 
 jobs:
   build_and_test:
-    name: Rust project - latest
-    runs-on: ubuntu-latest
+    name: ${{ matrix.os }} ${{ matrix.toolchain }}
+    runs-on: ${{ matrix.os }}
     strategy:
       matrix:
         toolchain:
           - stable
           - beta
           - nightly
+        os:
+          - ubuntu-latest
+          - windows-latest
+          - macos-latest
+          - macos-13
     steps:
       - uses: actions/checkout@v4
-      - run: rustup update ${{ matrix.toolchain }} && rustup default ${{ matrix.toolchain }}
+      - name: Rustup update
+        if: matrix.os != 'windows-latest'
+        run: rustup update ${{ matrix.toolchain }}
+      - run: rustup default ${{ matrix.toolchain }}
+      - uses: Swatinem/rust-cache@v2
+        with:
+          cache-all-crates: true
+          key: ${{ runner.os }}-${{ matrix.toolchain }}
       - run: cargo build --verbose
       - run: cargo test --verbose
-      - run: ./create_measurements.sh 1_000_000_000
+      - name: Create measurements
+        run: cargo r --quiet --bin create_measurements --features=generate --release 1_000_000_000
       - run: cargo install hyperfine
-      - run: ./bench.sh
+      - name: Build release
+        run: cargo build --release
+      - name: Benchmark
+        if: matrix.os != 'windows-latest'
+        run: |
+          python -c "import platform, os; print('OS:', platform.system(), platform.release(), 'Architecture:', platform.machine(), 'CPU threads:', os.cpu_count())"
+          hyperfine --warmup 0 --runs 5 "./target/release/brc-rs"
+      - name: Benchmark
+        if: matrix.os == 'windows-latest'
+        run: |
+          python -c "import platform, os; print('OS:', platform.system(), platform.release(), 'Architecture:', platform.machine(), 'CPU threads:', os.cpu_count())"
+          hyperfine --warmup 0 --runs 5 "target\release\brc-rs.exe"
diff --git a/src/bin/create_measurements.rs b/src/bin/create_measurements.rs
index 0958f94..3008a0d 100644
--- a/src/bin/create_measurements.rs
+++ b/src/bin/create_measurements.rs
@@ -46,6 +46,10 @@ fn generate_measurements<W: Write + Send + 'static>(
 ) -> std::io::Result<()> {
     let start = Instant::now();
     let par_count = std::thread::available_parallelism().unwrap();
+    println!(
+        "Starting generating {} measurements with {} threads",
+        size, par_count
+    );
     let task_size = size / par_count;
     let mut tasks = vec![task_size; par_count.into()];
     for i in 0..(size % par_count) {
diff --git a/src/main.rs b/src/main.rs
index 79626df..954b7e5 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -38,13 +38,17 @@ impl Add<&mut Self> for WeatherStationStats {
         }
     }
 }
-fn parse_line(line: &[u8]) -> (&[u8], i64) {
+fn parse_line(mut line: &[u8]) -> (&[u8], i64) {
     // we know that the measurement is pure ASCII and is at max 5 characters long
     // based on this we can find the semicolon faster by doing at most 6 byte comparisons by iterating the reversed bytes
     // At the same time, we _are_ iterating through the measurement from the least significant character to the biggest
     let mut semicolon_idx = 0;
     let mut is_negative = false;
     let mut measurement = 0;
+    // stupid Windows check
+    if cfg!(windows) && line[line.len() - 1] == b'\r' {
+        line = &line[..line.len() - 1];
+    }
     for (idx, b) in line.into_iter().rev().take(6).enumerate() {
         match (b, idx) {
             (b';', _) => {
@@ -311,7 +315,7 @@ mod tests {
                     .split(",")
                     .zip(calc(Some(format!("{}.txt", $file_name))).split(","))
                 {
-                    assert_eq!(val, expected);
+                    assert_eq!(val.trim(), expected.trim());
                 }
             }
         };