sha256

commit 3060645abfec415f8dae8c08a613f7c51cc9c36c
parent c832ff090cd2f301d18c8f04886a2124d392a75f
Author: Jared Tobin <jared@jtobin.io>
Date:   Sun,  1 Feb 2026 14:26:34 +0400

lib: export list and noinline pragma

The NOINLINE pragma is required with the explicit export list to stop
GHC from mis-optimizing the unsaf efunctions.

Diffstat:
M
README.md
 | 
9
++++-----
M
lib/Crypto/Hash/SHA256/Arm.hs
 | 
9
++++++++-

2 files changed, 12 insertions(+), 6 deletions(-)
diff --git a/README.md b/README.md
@@ -64,11 +64,10 @@ benchmark suite):
 
 ```
   benchmarking ppad-sha256/SHA256 (32B input)/hash
-  time                 67.70 ns   (66.08 ns .. 71.11 ns)
-                       0.992 R²   (0.979 R² .. 1.000 R²)
-  mean                 66.55 ns   (65.91 ns .. 68.79 ns)
-  std dev              3.919 ns   (720.2 ps .. 8.253 ns)
-  variance introduced by outliers: 78% (severely inflated)
+  time                 65.85 ns   (65.76 ns .. 65.95 ns)
+                       1.000 R²   (1.000 R² .. 1.000 R²)
+  mean                 65.90 ns   (65.80 ns .. 65.99 ns)
+  std dev              307.5 ps   (256.0 ps .. 392.9 ps)
 
   benchmarking ppad-sha256/HMAC-SHA256 (32B input)/hmac
   time                 135.7 ns   (135.2 ns .. 136.1 ns)
diff --git a/lib/Crypto/Hash/SHA256/Arm.hs b/lib/Crypto/Hash/SHA256/Arm.hs
@@ -46,6 +46,7 @@ fi :: (Integral a, Num b) => a -> b
 fi = fromIntegral
 {-# INLINE fi #-}
 
+
 peek_registers
   :: Ptr Word32
   -> Registers
@@ -174,7 +175,7 @@ _hmac rp bp k m = do
   poke_registers rp (iv ())
   update rp bp (xor k (Exts.wordToWord32# 0x5C5C5C5C##))
   update rp bp block
-{-# INLINABLE _hmac #-}
+{-# NOINLINE _hmac #-}
 
 _hmac_rr
   :: Ptr Word32 -- ^ register state
@@ -204,6 +205,8 @@ _hmac_bb rp bp k m = do
   update rp bp inner
 {-# INLINABLE _hmac_bb #-}
 
+-- | HMAC(key, v || sep || data) using ARM crypto extensions.
+-- Writes result to destination pointer.
 _hmac_rsb
   :: Ptr Word32    -- ^ destination (8 Word32s)
   -> Ptr Word32    -- ^ scratch block buffer (16 Word32s)
@@ -223,6 +226,8 @@ _hmac_rsb rp bp k v sep dat = do
   update rp bp inner
 {-# INLINABLE _hmac_rsb #-}
 
+-- | Hash (v || sep || dat) with ARM crypto extensions.
+-- Assumes register state already initialized at rp.
 _hash_vsb
   :: Ptr Word32    -- ^ register state
   -> Ptr Word32    -- ^ block buffer
@@ -236,9 +241,11 @@ _hash_vsb rp bp el v sep dat@(BI.PS _ _ l)
       -- first block is complete: v || sep || dat[0:31]
       let !b0 = parse_vsb v sep dat
       update rp bp b0
+      -- hash remaining complete blocks from dat[31:]
       let !rest    = BU.unsafeDrop 31 dat
           !restLen = l - 31
       hash_blocks rp bp rest
+      -- handle final padding
       let !finLen = restLen `rem` 64
           !fin    = BU.unsafeDrop (restLen - finLen) rest
           !total  = el + 33 + fi l