Unroll Sha3 inner loop

Issue #699 since fixed. Nearly a x3 perf improvement.

Using --release-fast.

Sha3_256 (before): 96 Mb/s
Sha3_256  (after): 267 Mb/s

Sha3_512 (before): 53 Mb/s
Sha3_512  (after): 142 Mb/s

No real gains from unrolling other initialization loops in crypto
functions so have been left as is.

3 files changed, 10 insertions, 14 deletions

diff --git a/std/crypto/md5.zig b/std/crypto/md5.zig
index 313bff91bf..26700cd65b 100644
--- a/std/crypto/md5.zig
+++ b/std/crypto/md5.zig
@@ -108,7 +108,6 @@ pub const Md5 = struct {
 
         var s: [16]u32 = undefined;
 
-        // ERROR: cannot unroll this at comptime
         var i: usize = 0;
         while (i < 16) : (i += 1) {
             // NOTE: Performing or's separately improves perf by ~10%
diff --git a/std/crypto/sha2.zig b/std/crypto/sha2.zig
index bfb604d456..113bab926b 100644
--- a/std/crypto/sha2.zig
+++ b/std/crypto/sha2.zig
@@ -156,7 +156,6 @@ fn Sha2_32(comptime params: Sha2Params32) type { return struct {
 
         var s: [64]u32 = undefined;
 
-        // ERROR: Cannot unroll at compile-time.
         var i: usize = 0;
         while (i < 16) : (i += 1) {
             s[i] = 0;
@@ -472,7 +471,6 @@ fn Sha2_64(comptime params: Sha2Params64) type { return struct {
 
         var s: [80]u64 = undefined;
 
-        // ERROR: Cannot unroll at compile-time.
         var i: usize = 0;
         while (i < 16) : (i += 1) {
             s[i] = 0;
diff --git a/std/crypto/sha3.zig b/std/crypto/sha3.zig
index 759ddf43aa..6e6a86b3d5 100644
--- a/std/crypto/sha3.zig
+++ b/std/crypto/sha3.zig
@@ -123,35 +123,34 @@ fn keccak_f(comptime F: usize, d: []u8) void {
         *r = mem.readIntLE(u64, d[8*i .. 8*i + 8]);
     }
 
-    var x: usize = 0;
-    var y: usize = 0;
-    // TODO: Cannot unroll all loops here due to comptime differences.
-    inline for (RC[0..no_rounds]) |round| {
+    comptime var x: usize = 0;
+    comptime var y: usize = 0;
+    for (RC[0..no_rounds]) |round| {
         // theta
-        x = 0; while (x < 5) : (x += 1) {
+        x = 0; inline while (x < 5) : (x += 1) {
             c[x] = s[x] ^ s[x+5] ^ s[x+10] ^ s[x+15] ^ s[x+20];
         }
-        x = 0; while (x < 5) : (x += 1) {
+        x = 0; inline while (x < 5) : (x += 1) {
             t[0] = c[M5[x+4]] ^ math.rotl(u64, c[M5[x+1]], usize(1));
-            y = 0; while (y < 5) : (y += 1) {
+            y = 0; inline while (y < 5) : (y += 1) {
                 s[x + y*5] ^= t[0];
             }
         }
 
         // rho+pi
         t[0] = s[1];
-        x = 0; while (x < 24) : (x += 1) {
+        x = 0; inline while (x < 24) : (x += 1) {
             c[0] = s[PIL[x]];
             s[PIL[x]] = math.rotl(u64, t[0], ROTC[x]);
             t[0] = c[0];
         }
 
         // chi
-        y = 0; while (y < 5) : (y += 1) {
-            x = 0; while (x < 5) : (x += 1) {
+        y = 0; inline while (y < 5) : (y += 1) {
+            x = 0; inline while (x < 5) : (x += 1) {
                 c[x] = s[x + y*5];
             }
-            x = 0; while (x < 5) : (x += 1) {
+            x = 0; inline while (x < 5) : (x += 1) {
                 s[x + y*5] = c[x] ^ (~c[M5[x+1]] & c[M5[x+2]]);
             }
         }