perf(filecoin-proofs): speed up Fr32Reader

Switch to processing bit padding in blocks of 127bytes using u128s

filecoin-proofs/Cargo.toml

@@ -36,7 +36,6 @@ anyhow = "1.0.23"
 rand_xorshift = "0.2.0"
 sha2 = "0.9.1"
 typenum = "1.11.2"
-bitintr = "0.3.0"
 gperftools = { version = "0.2", optional = true }
 generic-array = "0.14.4"
 structopt = "0.3.12"
@@ -45,6 +44,7 @@ indicatif = "0.15.0"
 groupy = "0.3.0"
 dialoguer = "0.7.1"
 clap = "2.33.3"
+byte-slice-cast = "1.0.0"
 
 [dependencies.reqwest]
 version = "0.10"

filecoin-proofs/benches/preprocessing.rs

@@ -2,7 +2,7 @@ use std::io::{self, Read};
 use std::time::Duration;
 
 use criterion::{criterion_group, criterion_main, Criterion, ParameterizedBenchmark, Throughput};
-use filecoin_proofs::fr32_reader::Fr32Reader;
+use filecoin_proofs::{add_piece, fr32_reader::Fr32Reader, PaddedBytesAmount, UnpaddedBytesAmount};
 use rand::{thread_rng, Rng};
 
 #[cfg(feature = "cpu-profile")]
@@ -52,6 +52,7 @@ fn preprocessing_benchmark(c: &mut Criterion) {
                     let mut reader = Fr32Reader::new(io::Cursor::new(&data));
                     reader.read_to_end(&mut buf).expect("in memory read error");
                     assert!(buf.len() >= data.len());
+                    buf.clear();
                 });
                 stop_profile();
             },
@@ -63,5 +64,37 @@ fn preprocessing_benchmark(c: &mut Criterion) {
     );
 }
 
-criterion_group!(benches, preprocessing_benchmark);
+fn add_piece_benchmark(c: &mut Criterion) {
+    c.bench(
+        "preprocessing",
+        ParameterizedBenchmark::new(
+            "add_piece",
+            |b, size| {
+                let padded_size = PaddedBytesAmount(*size as u64);
+                let unpadded_size: UnpaddedBytesAmount = padded_size.into();
+                let data = random_data(unpadded_size.0 as usize);
+                let mut buf = Vec::with_capacity(*size);
+
+                start_profile(&format!("add_piece_{}", *size));
+                b.iter(|| {
+                    add_piece(
+                        io::Cursor::new(&data),
+                        &mut buf,
+                        unpadded_size,
+                        &[unpadded_size][..],
+                    )
+                    .unwrap();
+                    buf.clear();
+                });
+                stop_profile();
+            },
+            vec![512, 256 * 1024, 512 * 1024, 1024 * 1024, 2 * 1024 * 1024],
+        )
+        .sample_size(10)
+        .throughput(|s| Throughput::Bytes(*s as u64))
+        .warm_up_time(Duration::from_secs(1)),
+    );
+}
+
+criterion_group!(benches, preprocessing_benchmark, add_piece_benchmark);
 criterion_main!(benches);