Add support to benchmark fdct functions

Author: vstroebel

Cargo.toml

@@ -16,6 +16,9 @@ default = ["std"]
 simd = ["std"]
 std = []
 
+# DO NOT USE THIS IN PRODUCTION. Expose several internal functions for benchmark purposes.
+benchmark = []
+
 [dependencies]
 
 [dev-dependencies]

criterion/Cargo.toml

@@ -9,8 +9,11 @@ license = "MIT OR Apache-2.0"
 [workspace]
 members = ["."]
 
+[features]
+simd = ["jpeg-encoder/simd"]
+
 [dependencies]
-jpeg-encoder = {path = ".."}
+jpeg-encoder = {path = "..", features = ["benchmark"]}
 
 [dev-dependencies]
 criterion = { version = "0.5", default-features = false, features = ["plotters", "cargo_bench_support"]}
@@ -21,3 +24,7 @@ opt-level = 1
 [[bench]]
 name = "encode"
 harness = false
+
+[[bench]]
+name = "fdct"
+harness = false

criterion/benches/fdct.rs

@@ -0,0 +1,46 @@
+use criterion::{black_box, criterion_group, criterion_main, Criterion};
+
+use jpeg_encoder::fdct;
+use std::time::Duration;
+
+const INPUT1: [i16; 64] = [
+    -70, -71, -70, -68, -67, -67, -67, -67, -72, -73, -72, -70, -69, -69, -68, -69, -75, -76,
+    -74, -73, -73, -72, -71, -70, -77, -78, -77, -75, -76, -75, -73, -71, -78, -77, -77, -76,
+    -79, -77, -76, -75, -78, -78, -77, -77, -77, -77, -78, -77, -79, -79, -78, -78, -78, -78,
+    -79, -78, -80, -79, -78, -78, -81, -80, -78, -76,
+];
+
+fn criterion_benchmark(c: &mut Criterion) {
+
+    let mut group = c.benchmark_group("fdct");
+    group.measurement_time(Duration::from_secs(60));
+    group.warm_up_time(Duration::from_secs(10));
+
+    group.bench_function("default fdct", |b| {
+        b.iter(|| {
+            let mut input = INPUT1.clone();
+            fdct(
+                black_box(&mut input),
+            );
+            black_box(&input);
+        })
+    });
+
+    #[cfg(all(feature = "simd", any(target_arch = "x86", target_arch = "x86_64")))]
+    group.bench_function("fdct avx2", |b| {
+        b.iter(|| {
+            use jpeg_encoder::fdct_avx2;
+
+            let mut input = INPUT1.clone();
+            fdct_avx2(
+                black_box(&mut input),
+            );
+            black_box(&input);
+        })
+    });
+
+    group.finish();
+}
+
+criterion_group!(benches, criterion_benchmark);
+criterion_main!(benches);

src/avx2.rs

@@ -2,7 +2,7 @@ mod fdct;
 mod ycbcr;
 
 use crate::encoder::Operations;
-pub(crate) use fdct::fdct_avx2;
+pub use fdct::fdct_avx2;
 pub(crate) use ycbcr::*;
 
 pub(crate) struct AVX2Operations;

src/fdct.rs

@@ -102,7 +102,7 @@ fn into_el(v: i32) -> i16 {
 
 #[allow(clippy::erasing_op)]
 #[allow(clippy::identity_op)]
-pub(crate) fn fdct(data: &mut [i16; 64]) {
+pub fn fdct(data: &mut [i16; 64]) {
     /* Pass 1: process rows. */
     /* Note results are scaled up by sqrt(8) compared to a true DCT; */
     /* furthermore, we scale the results by 2**PASS1_BITS. */

src/lib.rs

@@ -48,6 +48,11 @@ pub use image_buffer::{cmyk_to_ycck, rgb_to_ycbcr, ImageBuffer};
 pub use quantization::QuantizationTableType;
 pub use writer::{Density, JfifWrite};
 
+#[cfg(feature = "benchmark")]
+pub use fdct::fdct;
+#[cfg(all(feature = "benchmark", feature = "simd", any(target_arch = "x86", target_arch = "x86_64")))]
+pub use avx2::fdct_avx2;
+
 #[cfg(test)]
 mod tests {
     use crate::image_buffer::rgb_to_ycbcr;