Speed up light_compute by ~9%

Cargo.lock

@@ -254,6 +254,11 @@ name = "crossbeam"
 version = "0.2.10"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 
+[[package]]
+name = "crunchy"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+
 [[package]]
 name = "crypt32-sys"
 version = "0.2.0"
@@ -360,6 +365,7 @@ dependencies = [
 name = "ethash"
 version = "1.7.0"
 dependencies = [
+ "crunchy 0.1.0 (registry+https://github.com/rust-lang/crates.io-index)",
  "log 0.3.7 (registry+https://github.com/rust-lang/crates.io-index)",
  "parking_lot 0.4.0 (registry+https://github.com/rust-lang/crates.io-index)",
  "primal 0.2.3 (registry+https://github.com/rust-lang/crates.io-index)",
@@ -2943,6 +2949,7 @@ dependencies = [
 "checksum core-foundation 0.2.2 (registry+https://github.com/rust-lang/crates.io-index)" = "20a6d0448d3a99d977ae4a2aa5a98d886a923e863e81ad9ff814645b6feb3bbd"
 "checksum core-foundation-sys 0.2.2 (registry+https://github.com/rust-lang/crates.io-index)" = "05eed248dc504a5391c63794fe4fb64f46f071280afaa1b73308f3c0ce4574c5"
 "checksum crossbeam 0.2.10 (registry+https://github.com/rust-lang/crates.io-index)" = "0c5ea215664ca264da8a9d9c3be80d2eaf30923c259d03e870388eb927508f97"
+"checksum crunchy 0.1.0 (registry+https://github.com/rust-lang/crates.io-index)" = "b56bdac907d4b64254aed43964b11dd334bc46e0826a0e3add75caddfaf90175"
 "checksum crypt32-sys 0.2.0 (registry+https://github.com/rust-lang/crates.io-index)" = "e34988f7e069e0b2f3bfc064295161e489b2d4e04a2e4248fb94360cdf00b4ec"
 "checksum ctrlc 1.1.1 (git+https://github.com/paritytech/rust-ctrlc.git)" = "<none>"
 "checksum daemonize 0.2.2 (registry+https://github.com/rust-lang/crates.io-index)" = "271ec51b7e0bee92f0d04601422c73eb76ececf197026711c97ad25038a010cf"

ethash/Cargo.toml

@@ -10,6 +10,7 @@ log = "0.3"
 sha3 = { path = "../util/sha3" }
 primal = "0.2.3"
 parking_lot = "0.4"
+crunchy = "0.1.0"
 
 [features]
 benches = []

ethash/src/compute.rs

@@ -287,12 +287,39 @@ fn hash_compute(light: &Light, full_size: usize, header_hash: &H256, nonce: u64)
 		let num_full_pages = (full_size / page_size) as u32;
 		let cache: &[Node] = &light.cache;  // deref once for better performance
 
-		for i in 0..(ETHASH_ACCESSES as u32) {
+		debug_assert_eq!(ETHASH_ACCESSES, 64);
-			let index = fnv_hash(f_mix.get_unchecked(0).as_words().get_unchecked(0) ^ i, *mix.get_unchecked(0).as_words().get_unchecked((i as usize) % MIX_WORDS)) % num_full_pages;
+		debug_assert_eq!(MIX_NODES, 2);
-			for n in 0..MIX_NODES {
+		debug_assert_eq!(NODE_WORDS, 16);
-				let tmp_node = calculate_dag_item(index * MIX_NODES as u32 + n as u32, cache);
+
-				for w in 0..NODE_WORDS {
+		unroll! {
-					*mix.get_unchecked_mut(n).as_words_mut().get_unchecked_mut(w) = fnv_hash(*mix.get_unchecked(n).as_words().get_unchecked(w), *tmp_node.as_words().get_unchecked(w));
+			// ETHASH_ACCESSES
+			for i_usize in 0..64 {
+				let i = i_usize as u32;
+
+				let index = fnv_hash(
+					f_mix.get_unchecked(0).as_words().get_unchecked(0) ^ i,
+					*mix.get_unchecked(0).as_words().get_unchecked(i_usize % MIX_WORDS)
+				) % num_full_pages;
+
+				unroll! {
+					// MIX_NODES
+					for n in 0..2 {
+						let tmp_node = calculate_dag_item(
+							index * MIX_NODES as u32 + n as u32,
+							cache,
+						);
+
+						unroll! {
+							// NODE_WORDS
+							for w in 0..16 {
+								*mix.get_unchecked_mut(n).as_words_mut().get_unchecked_mut(w) =
+									fnv_hash(
+										*mix.get_unchecked(n).as_words().get_unchecked(w),
+										*tmp_node.as_words().get_unchecked(w),
+									);
+							}
+						}
+					}
 				}
 			}
 		}

ethash/src/lib.rs

@@ -22,6 +22,8 @@ extern crate primal;
 extern crate sha3;
 extern crate parking_lot;
 
+#[macro_use]
+extern crate crunchy;
 #[macro_use]
 extern crate log;
 mod compute;