diff --git a/CHANGES.md b/CHANGES.md --- a/CHANGES.md +++ b/CHANGES.md @@ -1,3 +1,160 @@ +See also https://pvp.haskell.org/faq + +## Version 1.5.1.0 + + * Update xxHash to version 0.8.3 + +## Version 1.5.0.0 + + * Add `QuantifiedConstraints` superclasses to `Hashable1/2`: + +```haskell +class (Eq1 t, forall a. Hashable a => Hashable (t a)) => Hashable1 t where +class (Eq2 t, forall a. Hashable a => Hashable1 (t a)) => Hashable2 t where +``` + + * Change contexts of `Compose`, `Product` and `Sum` instances. + This and above is the similar change as [CLC proposal #10](https://github.com/haskell/core-libraries-committee/issues/10) + + * The above changes require `base-4.18.0.0`, so we drop support for GHC prior GHC-9.6.5 + (The `hashable-1.4` branch will be maintained for time being for older GHC users). + + * Make `Arg a b` instance behave as `Hashable a` instance. + +## Version 1.4.7.0 + + * Make `arch-native` disabled by default. + +## Version 1.4.6.0 + + * Use GND&DerivingVia to derive `newtype` intances (`Data.Semigroup`, `Data.Monoid`, `Identity` etc). + +## Version 1.4.5.0 + + * Drop support for GHCs prior 8.6.5 + * Use xxhash for hashing bytestrings and bytearrays. + Note: when compiling binaries for distribution, you may need to disable + `arch-native` flag. + +## Version 1.4.4.0 + + * Depend on `os-string-2` for GHC-9.2+ + * Support `filepath-1.5` + +## Version 1.4.3.0 + + * Export `defaultHashWithSalt` and `defaultHash`. + * Fix issue of tuples with 0 first component causing all-zero state. + * Change `hashInt` to mix bits more. + +## Version 1.4.2.0 + + * Fix the foreign signature of `getThreadId` + https://github.com/haskell-unordered-containers/hashable/pull/263 + * Drop support for GHCs prior GHC-8.2 + The recent `unordered-containers` releases support only GHC-8.2+ + * Add instance for `OsString`, `PosixString`, `WindowsString` from `filepath-1.4.100.1` + * Add `Hashable ByteArray` instance using `data-array-byte` compat package + +## Version 1.4.1.0 + + * Add instance for `Data.Array.Byte.ByteArray`. + +## Version 1.4.0.2 + + * Restore older GHC support + * Support GHC-9.0.2 + +## Version 1.4.0.1 + + * `text-2.0` compatibility + +## Version 1.4.0.0 + + * `Eq` is now a superclass of `Hashable`. + Also `Eq1` is a superclass of `Hashable1` and `Eq2` is a superclass of `Hashable2` + when exists. + + * Remove `Hashable1 Fixed` instance + * Remove `Hashable1 Semi.Min/Max/...` instances as they don't have `Eq1` instance. + +## Version 1.3.5.0 + + * Add `Solo` instance (base-4.15+, GHC-9+) + +## Version 1.3.4.1 + + * Fix compilation on 32 bit platforms + * Fix `Tree` instance + +## Version 1.3.4.0 + * `Text` and `ByteString` hashes include length. + This fixes a variant of https://github.com/haskell-unordered-containers/hashable/issues/74 + for texts and bytestrings. + https://github.com/haskell-unordered-containers/hashable/pull/223 + * Use correct prime in `combine`. + This should improve the hash quality of compound structures on 64bit systems. + https://github.com/haskell-unordered-containers/hashable/pull/224 + * Add instance for types in `containers` package + https://github.com/haskell-unordered-containers/hashable/pull/226 + * Change `Int`, `Int64` and `Word64` `hashWithSalt` slightly. + https://github.com/haskell-unordered-containers/hashable/pull/227 + +## Version 1.3.3.0 + + * `Text` hashing uses 64-bit FNV prime + * Don't truncate Text hashvalues on 64bit Windows: + https://github.com/haskell-unordered-containers/hashable/pull/211 + +## Version 1.3.2.0 + + * Add `Hashable (Fixed a)` for `base <4.7` versions. + * Add documentation: + - `hashable` is not a stable hash + - `hashWithSalt` may return negative values + - there is `time-compat` with `Hashable` instances for `time` types. + * Add `random-initial-seed` flag causing the initial seed + to be randomized on each start of an executable using `hashable`. + +## Version 1.3.1.0 + + * Add `Hashable1` instances to `semigroups` types. + + * Use `ghc-bignum` with GHC-9.0 + + * Use FNV-1 constants. + + * Make `hashable-examples` a test-suite + +## Version 1.3.0.0 + + * Semantic change of `Hashable Arg` instance to *not* hash the second + argument of `Arg` in order to be consistent with `Eq Arg` (#171) + + * Semantic change of `Hashable Float` and `Hashable Double` instances + to hash `-0.0` and `0.0` to the same value (#173) + + * Add `Hashable` instance for `Fingerprint` (#156) + + * Add new `Data.Hashable.Generic` module providing the default + implementations `genericHashWithSalt` and `genericLiftHashWithSalt` + together with other Generics support helpers (#148, #178) + + * Bump minimum version requirement of `base` to `base-4.5` (i.e. GHC >= 7.4) + +---- + +## Version 1.2.7.0 + + * Add `Hashable` and `Hashable1` instances for `Complex` + + * Fix undefined behavior in `hashable_fn_hash()` implementation + due to signed integer overflow (#152) + + * Mark `Data.Hashable.Lifted` as `Trustworthy` (re SafeHaskell) + + * Support GHC 8.4 + ## Version 1.2.6.1 * Use typeRepFingerprint from Type.Reflection.Unsafe @@ -100,14 +257,14 @@ * Fix bug where code relied on rewrite rules firing for correctness. -## Version1.2.0.4 +## Version 1.2.0.4 * Update docs to match code. * Work around bug in GHCi runtime linker, which never call static initializers. -## Version1.2.0.3 +## Version 1.2.0.3 * Make building of SSE 4.1 code conditional, as it doesn't work on all platforms. @@ -115,17 +272,17 @@ * Use a fixed salt, but allow random salting. Random salting by default broke people's code. -## Version1.2.0.2 +## Version 1.2.0.2 * Work around ghci linker bug on Windows. -## Version1.2.0.1 +## Version 1.2.0.1 * Fix performance bug in SSE implementation of SipHash. * Fix segfault due to incorrect stack alignment on Windows. -## Version1.2.0.0 +## Version 1.2.0.0 * Switch string hashing from FNV-1 to SipHash, in an effort to prevent collision attacks. @@ -141,11 +298,13 @@ * Add instance for Ordering. -## Version1.1.2.5 +---- +## Version 1.1.2.5 + * Bug fix for bytestring < 0.10.0. -## Version1.1.2.4 +## Version 1.1.2.4 * Switch string hashing from Bernstein to FNV-1 @@ -155,48 +314,44 @@ * Now works with GHC 7.6. -## Version1.1.2.3 +## Version 1.1.2.3 * Add instance for TypeRep. * Update dependency on test-framework. -## Version1.1.2.2 +## Version 1.1.2.2 * Bug fix for GHC 7.4 -## Version1.1.2.1 +## Version 1.1.2.1 * Update dependency on test-framework. * Improve documentation of combine. -## Version1.1.2.0 - - * Add instances for Interger, Ratio, Float, Double, and StableName. +## Version 1.1.2.0 * Fix hash collision issues for lists and tuples when using a user-specified salt. -## Version1.1.1.0 + * Add instances for `Integer`, `Ratio`, `Float`, `Double`, and `StableName`. * Improved instances for tuples and lists. - * Add instances for StableName, Float, Double, Integer, and Ratio. - -## Version1.1.1.0 +## Version 1.1.1.0 - * Add hashWithSalt, which allows the user to create different hash + * Add `hashWithSalt`, which allows the user to create different hash values for the same input by providing different seeds. This is useful for application like Cuckoo hashing which need a family of hash functions. - * Fix a bug in the Hashable instance for Int64/Word64 on 32-bit + * Fix a bug in the `Hashable` instance for `Int64`/`Word64` on 32-bit platforms. * Improved resilience to leading zero in the input being hashed. -## Version1.1.0.0 +## Version 1.1.0.0 * Add instance for: strict and lazy Texts, ThreadId @@ -209,17 +364,19 @@ * Fix ByteString hashing to work correctly on both 32 and 64-bit platforms. -## Version1.0.1.1 +## Version 1.0.1.1 * Fix bug in Hashable instance for lazy ByteStrings where differences in the internal structure of the ByteString could cause different hash values for ByteStrings that are equal according to ==. -## Version1.0.1.0 +## Version 1.0.1.0 * Add two helpers for creating Hashable instances: hashPtr and hashByteArray. -## Version1.0.0 +---- + +## Version 1.0.0 * Separate Hashable class to its own package from hashmap 1.0.0.3. diff --git a/Data/Hashable.hs b/Data/Hashable.hs deleted file mode 100644 --- a/Data/Hashable.hs +++ /dev/null @@ -1,213 +0,0 @@ -{-# LANGUAGE CPP #-} -{-# LANGUAGE DeriveDataTypeable #-} -#if defined(__GLASGOW_HASKELL__) && __GLASGOW_HASKELL__ >= 702 -{-# LANGUAGE Trustworthy #-} -#endif - ------------------------------------------------------------------------- --- | --- Module : Data.Hashable --- Copyright : (c) Milan Straka 2010 --- (c) Johan Tibell 2011 --- (c) Bryan O'Sullivan 2011, 2012 --- License : BSD-style --- Maintainer : johan.tibell@gmail.com --- Stability : provisional --- Portability : portable --- --- This module defines a class, 'Hashable', for types that can be --- converted to a hash value. This class exists for the benefit of --- hashing-based data structures. The module provides instances for --- most standard types. Efficient instances for other types can be --- generated automatically and effortlessly using the generics support --- in GHC 7.2 and above. --- --- The easiest way to get started is to use the 'hash' function. Here --- is an example session with @ghci@. --- --- > ghci> import Data.Hashable --- > ghci> hash "foo" --- > 60853164 - -module Data.Hashable - ( - -- * Hashing and security - -- $security - - -- * Computing hash values - Hashable(..) - - -- * Creating new instances - -- | There are two ways to create new instances: by deriving - -- instances automatically using GHC's generic programming - -- support or by writing instances manually. - - -- ** Generic instances - -- $generics - - -- *** Understanding a compiler error - -- $generic_err - - -- ** Writing instances by hand - -- $blocks - - -- *** Hashing contructors with multiple fields - -- $multiple-fields - - -- *** Hashing types with multiple constructors - -- $multiple-ctors - - , hashUsing - , hashPtr - , hashPtrWithSalt -#if defined(__GLASGOW_HASKELL__) - , hashByteArray - , hashByteArrayWithSalt -#endif - -- * Caching hashes - , Hashed - , hashed - , unhashed - , mapHashed - , traverseHashed - ) where - -import Data.Hashable.Class - -#ifdef GENERICS -import Data.Hashable.Generic () -#endif - --- $security --- #security# --- --- Applications that use hash-based data structures to store input --- from untrusted users can be susceptible to \"hash DoS\", a class of --- denial-of-service attack that uses deliberately chosen colliding --- inputs to force an application into unexpectedly behaving with --- quadratic time complexity. --- --- At this time, the string hashing functions used in this library are --- susceptible to such attacks and users are recommended to either use --- a 'Data.Map' to store keys derived from untrusted input or to use a --- hash function (e.g. SipHash) that's resistant to such attacks. A --- future version of this library might ship with such hash functions. - --- $generics --- --- Beginning with GHC 7.2, the recommended way to make instances of --- 'Hashable' for most types is to use the compiler's support for --- automatically generating default instances. --- --- > {-# LANGUAGE DeriveGeneric #-} --- > --- > import GHC.Generics (Generic) --- > import Data.Hashable --- > --- > data Foo a = Foo a String --- > deriving (Eq, Generic) --- > --- > instance Hashable a => Hashable (Foo a) --- > --- > data Colour = Red | Green | Blue --- > deriving Generic --- > --- > instance Hashable Colour --- --- If you omit a body for the instance declaration, GHC will generate --- a default instance that correctly and efficiently hashes every --- constructor and parameter. - --- $generic_err --- --- Suppose you intend to use the generic machinery to automatically --- generate a 'Hashable' instance. --- --- > data Oops = Oops --- > -- forgot to add "deriving Generic" here! --- > --- > instance Hashable Oops --- --- And imagine that, as in the example above, you forget to add a --- \"@deriving 'Generic'@\" clause to your data type. At compile time, --- you will get an error message from GHC that begins roughly as --- follows: --- --- > No instance for (GHashable (Rep Oops)) --- --- This error can be confusing, as 'GHashable' is not exported (it is --- an internal typeclass used by this library's generics machinery). --- The correct fix is simply to add the missing \"@deriving --- 'Generic'@\". - --- $blocks --- --- To maintain high quality hashes, new 'Hashable' instances should be --- built using existing 'Hashable' instances, combinators, and hash --- functions. --- --- The functions below can be used when creating new instances of --- 'Hashable'. For example, for many string-like types the --- 'hashWithSalt' method can be defined in terms of either --- 'hashPtrWithSalt' or 'hashByteArrayWithSalt'. Here's how you could --- implement an instance for the 'B.ByteString' data type, from the --- @bytestring@ package: --- --- > import qualified Data.ByteString as B --- > import qualified Data.ByteString.Internal as B --- > import qualified Data.ByteString.Unsafe as B --- > import Data.Hashable --- > import Foreign.Ptr (castPtr) --- > --- > instance Hashable B.ByteString where --- > hashWithSalt salt bs = B.inlinePerformIO $ --- > B.unsafeUseAsCStringLen bs $ \(p, len) -> --- > hashPtrWithSalt p (fromIntegral len) salt - --- $multiple-fields --- --- Hash constructors with multiple fields by chaining 'hashWithSalt': --- --- > data Date = Date Int Int Int --- > --- > instance Hashable Date where --- > hashWithSalt s (Date yr mo dy) = --- > s `hashWithSalt` --- > yr `hashWithSalt` --- > mo `hashWithSalt` dy --- --- If you need to chain hashes together, use 'hashWithSalt' and follow --- this recipe: --- --- > combineTwo h1 h2 = h1 `hashWithSalt` h2 - --- $multiple-ctors --- --- For a type with several value constructors, there are a few --- possible approaches to writing a 'Hashable' instance. --- --- If the type is an instance of 'Enum', the easiest path is to --- convert it to an 'Int', and use the existing 'Hashable' instance --- for 'Int'. --- --- > data Color = Red | Green | Blue --- > deriving Enum --- > --- > instance Hashable Color where --- > hashWithSalt = hashUsing fromEnum --- --- If the type's constructors accept parameters, it is important to --- distinguish the constructors. To distinguish the constructors, add --- a different integer to the hash computation of each constructor: --- --- > data Time = Days Int --- > | Weeks Int --- > | Months Int --- > --- > instance Hashable Time where --- > hashWithSalt s (Days n) = s `hashWithSalt` --- > (0::Int) `hashWithSalt` n --- > hashWithSalt s (Weeks n) = s `hashWithSalt` --- > (1::Int) `hashWithSalt` n --- > hashWithSalt s (Months n) = s `hashWithSalt` --- > (2::Int) `hashWithSalt` n diff --git a/Data/Hashable/Class.hs b/Data/Hashable/Class.hs deleted file mode 100644 --- a/Data/Hashable/Class.hs +++ /dev/null @@ -1,911 +0,0 @@ -{-# LANGUAGE BangPatterns, CPP, ForeignFunctionInterface, MagicHash, - ScopedTypeVariables, UnliftedFFITypes #-} - -#if __GLASGOW_HASKELL__ < 710 -{-# LANGUAGE DeriveDataTypeable #-} -#endif - -#if __GLASGOW_HASKELL__ >= 801 -{-# LANGUAGE PolyKinds #-} -- For TypeRep instances -#endif - -#ifdef GENERICS -{-# LANGUAGE DefaultSignatures, FlexibleContexts, GADTs, - MultiParamTypeClasses, EmptyDataDecls #-} -#endif - ------------------------------------------------------------------------- --- | --- Module : Data.Hashable.Class --- Copyright : (c) Milan Straka 2010 --- (c) Johan Tibell 2011 --- (c) Bryan O'Sullivan 2011, 2012 --- License : BSD-style --- Maintainer : johan.tibell@gmail.com --- Stability : provisional --- Portability : portable --- --- This module defines a class, 'Hashable', for types that can be --- converted to a hash value. This class exists for the benefit of --- hashing-based data structures. The module provides instances for --- most standard types. - -module Data.Hashable.Class - ( - -- * Computing hash values - Hashable(..) - , Hashable1(..) - , Hashable2(..) -#ifdef GENERICS - -- ** Support for generics - , GHashable(..) - , HashArgs(..) - , Zero - , One -#endif - - -- * Creating new instances - , hashUsing - , hashPtr - , hashPtrWithSalt - , hashByteArray - , hashByteArrayWithSalt - , defaultHashWithSalt - -- * Higher Rank Functions - , hashWithSalt1 - , hashWithSalt2 - , defaultLiftHashWithSalt - -- * Caching hashes - , Hashed - , hashed - , unhashed - , mapHashed - , traverseHashed - ) where - -import Control.Applicative (Const(..)) -import Control.Exception (assert) -import Control.DeepSeq (NFData(rnf)) -import Data.Bits (shiftL, shiftR, xor) -import qualified Data.ByteString as B -import qualified Data.ByteString.Lazy as BL -import qualified Data.ByteString.Unsafe as B -import Data.Int (Int8, Int16, Int32, Int64) -import Data.List (foldl') -import Data.Ratio (Ratio, denominator, numerator) -import qualified Data.Text as T -import qualified Data.Text.Array as TA -import qualified Data.Text.Internal as T -import qualified Data.Text.Lazy as TL -import Data.Version (Version(..)) -import Data.Word (Word8, Word16, Word32, Word64) -import Foreign.C (CString) -import Foreign.Marshal.Utils (with) -import Foreign.Ptr (Ptr, FunPtr, IntPtr, WordPtr, castPtr, castFunPtrToPtr, ptrToIntPtr) -import Foreign.Storable (alignment, peek, sizeOf) -import GHC.Base (ByteArray#) -import GHC.Conc (ThreadId(..)) -import GHC.Prim (ThreadId#) -import System.IO.Unsafe (unsafeDupablePerformIO) -import System.Mem.StableName -import Data.Unique (Unique, hashUnique) - --- As we use qualified F.Foldable, we don't get warnings with newer base -import qualified Data.Foldable as F - -#if MIN_VERSION_base(4,7,0) -import Data.Proxy (Proxy) -#endif - -#if MIN_VERSION_base(4,7,0) -import Data.Fixed (Fixed(..)) -#endif - -#if MIN_VERSION_base(4,8,0) -import Data.Functor.Identity (Identity(..)) -#endif - -#ifdef GENERICS -import GHC.Generics -#endif - -#if __GLASGOW_HASKELL__ >= 801 -import Type.Reflection (Typeable, TypeRep, SomeTypeRep(..)) -import Type.Reflection.Unsafe (typeRepFingerprint) -import GHC.Fingerprint.Type(Fingerprint(..)) -#elif __GLASGOW_HASKELL__ >= 710 -import Data.Typeable (typeRepFingerprint, Typeable, TypeRep) -import GHC.Fingerprint.Type(Fingerprint(..)) -#elif __GLASGOW_HASKELL__ >= 702 -import Data.Typeable.Internal (Typeable, TypeRep (..)) -import GHC.Fingerprint.Type(Fingerprint(..)) -#elif __GLASGOW_HASKELL__ >= 606 -import Data.Typeable (typeRepKey, Typeable, TypeRep) -#endif - -#if __GLASGOW_HASKELL__ >= 703 -import Foreign.C (CLong(..)) -import Foreign.C.Types (CInt(..)) -#else -import Foreign.C (CLong) -import Foreign.C.Types (CInt) -#endif - -#if !(MIN_VERSION_base(4,8,0)) -import Data.Word (Word) -#endif - -#if MIN_VERSION_base(4,7,0) -import Data.Bits (finiteBitSize) -#else -import Data.Bits (bitSize) -#endif - -#if !(MIN_VERSION_bytestring(0,10,0)) -import qualified Data.ByteString.Lazy.Internal as BL -- foldlChunks -#endif - -#if MIN_VERSION_bytestring(0,10,4) -import qualified Data.ByteString.Short.Internal as BSI -#endif - -#ifdef VERSION_integer_gmp - -# if MIN_VERSION_integer_gmp(1,0,0) -# define MIN_VERSION_integer_gmp_1_0_0 -# endif - -import GHC.Exts (Int(..)) -import GHC.Integer.GMP.Internals (Integer(..)) -# if defined(MIN_VERSION_integer_gmp_1_0_0) -import GHC.Exts (sizeofByteArray#) -import GHC.Integer.GMP.Internals (BigNat(BN#)) -# endif -#endif - -#if MIN_VERSION_base(4,8,0) -import Data.Void (Void, absurd) -import GHC.Natural (Natural(..)) -import GHC.Exts (Word(..)) -#endif - -#if MIN_VERSION_base(4,9,0) -import qualified Data.List.NonEmpty as NE -import Data.Semigroup -import Data.Functor.Classes (Eq1(..),Ord1(..),Show1(..),showsUnaryWith) - -import Data.Functor.Compose (Compose(..)) -import qualified Data.Functor.Product as FP -import qualified Data.Functor.Sum as FS -#endif - -import Data.String (IsString(..)) - -#include "MachDeps.h" - -infixl 0 `hashWithSalt` - ------------------------------------------------------------------------- --- * Computing hash values - --- | A default salt used in the implementation of 'hash'. -defaultSalt :: Int -#if WORD_SIZE_IN_BITS == 64 -defaultSalt = -2578643520546668380 -- 0xdc36d1615b7400a4 -#else -defaultSalt = 0x087fc72c -#endif -{-# INLINE defaultSalt #-} - --- | The class of types that can be converted to a hash value. --- --- Minimal implementation: 'hashWithSalt'. -class Hashable a where - -- | Return a hash value for the argument, using the given salt. - -- - -- The general contract of 'hashWithSalt' is: - -- - -- * If two values are equal according to the '==' method, then - -- applying the 'hashWithSalt' method on each of the two values - -- /must/ produce the same integer result if the same salt is - -- used in each case. - -- - -- * It is /not/ required that if two values are unequal - -- according to the '==' method, then applying the - -- 'hashWithSalt' method on each of the two values must produce - -- distinct integer results. However, the programmer should be - -- aware that producing distinct integer results for unequal - -- values may improve the performance of hashing-based data - -- structures. - -- - -- * This method can be used to compute different hash values for - -- the same input by providing a different salt in each - -- application of the method. This implies that any instance - -- that defines 'hashWithSalt' /must/ make use of the salt in - -- its implementation. - hashWithSalt :: Int -> a -> Int - - -- | Like 'hashWithSalt', but no salt is used. The default - -- implementation uses 'hashWithSalt' with some default salt. - -- Instances might want to implement this method to provide a more - -- efficient implementation than the default implementation. - hash :: a -> Int - hash = hashWithSalt defaultSalt - -#ifdef GENERICS - default hashWithSalt :: (Generic a, GHashable Zero (Rep a)) => Int -> a -> Int - hashWithSalt salt = ghashWithSalt HashArgs0 salt . from - -data Zero -data One - -data HashArgs arity a where - HashArgs0 :: HashArgs Zero a - HashArgs1 :: (Int -> a -> Int) -> HashArgs One a - --- | The class of types that can be generically hashed. -class GHashable arity f where - ghashWithSalt :: HashArgs arity a -> Int -> f a -> Int - -#endif - -class Hashable1 t where - -- | Lift a hashing function through the type constructor. - liftHashWithSalt :: (Int -> a -> Int) -> Int -> t a -> Int -#ifdef GENERICS - default liftHashWithSalt :: (Generic1 t, GHashable One (Rep1 t)) => (Int -> a -> Int) -> Int -> t a -> Int - liftHashWithSalt h salt = ghashWithSalt (HashArgs1 h) salt . from1 -#endif - -class Hashable2 t where - -- | Lift a hashing function through the binary type constructor. - liftHashWithSalt2 :: (Int -> a -> Int) -> (Int -> b -> Int) -> Int -> t a b -> Int - --- | Lift the 'hashWithSalt' function through the type constructor. --- --- > hashWithSalt1 = liftHashWithSalt hashWithSalt -hashWithSalt1 :: (Hashable1 f, Hashable a) => Int -> f a -> Int -hashWithSalt1 = liftHashWithSalt hashWithSalt - --- | Lift the 'hashWithSalt' function through the type constructor. --- --- > hashWithSalt2 = liftHashWithSalt2 hashWithSalt hashWithSalt -hashWithSalt2 :: (Hashable2 f, Hashable a, Hashable b) => Int -> f a b -> Int -hashWithSalt2 = liftHashWithSalt2 hashWithSalt hashWithSalt - --- | Lift the 'hashWithSalt' function halfway through the type constructor. --- This function makes a suitable default implementation of 'liftHashWithSalt', --- given that the type constructor @t@ in question can unify with @f a@. -defaultLiftHashWithSalt :: (Hashable2 f, Hashable a) => (Int -> b -> Int) -> Int -> f a b -> Int -defaultLiftHashWithSalt h = liftHashWithSalt2 hashWithSalt h - --- | Since we support a generic implementation of 'hashWithSalt' we --- cannot also provide a default implementation for that method for --- the non-generic instance use case. Instead we provide --- 'defaultHashWith'. -defaultHashWithSalt :: Hashable a => Int -> a -> Int -defaultHashWithSalt salt x = salt `combine` hash x - --- | Transform a value into a 'Hashable' value, then hash the --- transformed value using the given salt. --- --- This is a useful shorthand in cases where a type can easily be --- mapped to another type that is already an instance of 'Hashable'. --- Example: --- --- > data Foo = Foo | Bar --- > deriving (Enum) --- > --- > instance Hashable Foo where --- > hashWithSalt = hashUsing fromEnum -hashUsing :: (Hashable b) => - (a -> b) -- ^ Transformation function. - -> Int -- ^ Salt. - -> a -- ^ Value to transform. - -> Int -hashUsing f salt x = hashWithSalt salt (f x) -{-# INLINE hashUsing #-} - -instance Hashable Int where - hash = id - hashWithSalt = defaultHashWithSalt - -instance Hashable Int8 where - hash = fromIntegral - hashWithSalt = defaultHashWithSalt - -instance Hashable Int16 where - hash = fromIntegral - hashWithSalt = defaultHashWithSalt - -instance Hashable Int32 where - hash = fromIntegral - hashWithSalt = defaultHashWithSalt - -instance Hashable Int64 where - hash n -#if MIN_VERSION_base(4,7,0) - | finiteBitSize (undefined :: Int) == 64 = fromIntegral n -#else - | bitSize (undefined :: Int) == 64 = fromIntegral n -#endif - | otherwise = fromIntegral (fromIntegral n `xor` - (fromIntegral n `shiftR` 32 :: Word64)) - hashWithSalt = defaultHashWithSalt - -instance Hashable Word where - hash = fromIntegral - hashWithSalt = defaultHashWithSalt - -instance Hashable Word8 where - hash = fromIntegral - hashWithSalt = defaultHashWithSalt - -instance Hashable Word16 where - hash = fromIntegral - hashWithSalt = defaultHashWithSalt - -instance Hashable Word32 where - hash = fromIntegral - hashWithSalt = defaultHashWithSalt - -instance Hashable Word64 where - hash n -#if MIN_VERSION_base(4,7,0) - | finiteBitSize (undefined :: Int) == 64 = fromIntegral n -#else - | bitSize (undefined :: Int) == 64 = fromIntegral n -#endif - | otherwise = fromIntegral (n `xor` (n `shiftR` 32)) - hashWithSalt = defaultHashWithSalt - -instance Hashable () where - hash = fromEnum - hashWithSalt = defaultHashWithSalt - -instance Hashable Bool where - hash = fromEnum - hashWithSalt = defaultHashWithSalt - -instance Hashable Ordering where - hash = fromEnum - hashWithSalt = defaultHashWithSalt - -instance Hashable Char where - hash = fromEnum - hashWithSalt = defaultHashWithSalt - -#if defined(MIN_VERSION_integer_gmp_1_0_0) -instance Hashable BigNat where - hashWithSalt salt (BN# ba) = hashByteArrayWithSalt ba 0 numBytes salt - `hashWithSalt` size - where - size = numBytes `quot` SIZEOF_HSWORD - numBytes = I# (sizeofByteArray# ba) -#endif - -#if MIN_VERSION_base(4,8,0) -instance Hashable Natural where -# if defined(MIN_VERSION_integer_gmp_1_0_0) - hash (NatS# n) = hash (W# n) - hash (NatJ# bn) = hash bn - - hashWithSalt salt (NatS# n) = hashWithSalt salt (W# n) - hashWithSalt salt (NatJ# bn) = hashWithSalt salt bn -# else - hash (Natural n) = hash n - - hashWithSalt salt (Natural n) = hashWithSalt salt n -# endif -#endif - -instance Hashable Integer where -#if defined(VERSION_integer_gmp) -# if defined(MIN_VERSION_integer_gmp_1_0_0) - hash (S# n) = (I# n) - hash (Jp# bn) = hash bn - hash (Jn# bn) = negate (hash bn) - - hashWithSalt salt (S# n) = hashWithSalt salt (I# n) - hashWithSalt salt (Jp# bn) = hashWithSalt salt bn - hashWithSalt salt (Jn# bn) = negate (hashWithSalt salt bn) -# else - hash (S# int) = I# int - hash n@(J# size# byteArray) - | n >= minInt && n <= maxInt = fromInteger n :: Int - | otherwise = let size = I# size# - numBytes = SIZEOF_HSWORD * abs size - in hashByteArrayWithSalt byteArray 0 numBytes defaultSalt - `hashWithSalt` size - where minInt = fromIntegral (minBound :: Int) - maxInt = fromIntegral (maxBound :: Int) - - hashWithSalt salt (S# n) = hashWithSalt salt (I# n) - hashWithSalt salt n@(J# size# byteArray) - | n >= minInt && n <= maxInt = hashWithSalt salt (fromInteger n :: Int) - | otherwise = let size = I# size# - numBytes = SIZEOF_HSWORD * abs size - in hashByteArrayWithSalt byteArray 0 numBytes salt - `hashWithSalt` size - where minInt = fromIntegral (minBound :: Int) - maxInt = fromIntegral (maxBound :: Int) -# endif -#else - hashWithSalt salt = foldl' hashWithSalt salt . go - where - go n | inBounds n = [fromIntegral n :: Int] - | otherwise = fromIntegral n : go (n `shiftR` WORD_SIZE_IN_BITS) - maxInt = fromIntegral (maxBound :: Int) - inBounds x = x >= fromIntegral (minBound :: Int) && x <= maxInt -#endif - -#if MIN_VERSION_base(4,9,0) --- Starting with base-4.9, numerator/denominator don't need 'Integral' anymore -instance Hashable a => Hashable (Ratio a) where -#else -instance (Integral a, Hashable a) => Hashable (Ratio a) where -#endif - {-# SPECIALIZE instance Hashable (Ratio Integer) #-} - hash a = hash (numerator a) `hashWithSalt` denominator a - hashWithSalt s a = s `hashWithSalt` numerator a `hashWithSalt` denominator a - -instance Hashable Float where - hash x - | isIEEE x = - assert (sizeOf x >= sizeOf (0::Word32) && - alignment x >= alignment (0::Word32)) $ - hash ((unsafeDupablePerformIO $ with x $ peek . castPtr) :: Word32) - | otherwise = hash (show x) - hashWithSalt = defaultHashWithSalt - -instance Hashable Double where - hash x - | isIEEE x = - assert (sizeOf x >= sizeOf (0::Word64) && - alignment x >= alignment (0::Word64)) $ - hash ((unsafeDupablePerformIO $ with x $ peek . castPtr) :: Word64) - | otherwise = hash (show x) - hashWithSalt = defaultHashWithSalt - --- | A value with bit pattern (01)* (or 5* in hexa), for any size of Int. --- It is used as data constructor distinguisher. GHC computes its value during --- compilation. -distinguisher :: Int -distinguisher = fromIntegral $ (maxBound :: Word) `quot` 3 -{-# INLINE distinguisher #-} - -instance Hashable a => Hashable (Maybe a) where - hash Nothing = 0 - hash (Just a) = distinguisher `hashWithSalt` a - hashWithSalt = hashWithSalt1 - -instance Hashable1 Maybe where - liftHashWithSalt _ s Nothing = s `combine` 0 - liftHashWithSalt h s (Just a) = s `combine` distinguisher `h` a - -instance (Hashable a, Hashable b) => Hashable (Either a b) where - hash (Left a) = 0 `hashWithSalt` a - hash (Right b) = distinguisher `hashWithSalt` b - hashWithSalt = hashWithSalt1 - -instance Hashable a => Hashable1 (Either a) where - liftHashWithSalt = defaultLiftHashWithSalt - -instance Hashable2 Either where - liftHashWithSalt2 h _ s (Left a) = s `combine` 0 `h` a - liftHashWithSalt2 _ h s (Right b) = s `combine` distinguisher `h` b - -instance (Hashable a1, Hashable a2) => Hashable (a1, a2) where - hash (a1, a2) = hash a1 `hashWithSalt` a2 - hashWithSalt = hashWithSalt1 - -instance Hashable a1 => Hashable1 ((,) a1) where - liftHashWithSalt = defaultLiftHashWithSalt - -instance Hashable2 (,) where - liftHashWithSalt2 h1 h2 s (a1, a2) = s `h1` a1 `h2` a2 - -instance (Hashable a1, Hashable a2, Hashable a3) => Hashable (a1, a2, a3) where - hash (a1, a2, a3) = hash a1 `hashWithSalt` a2 `hashWithSalt` a3 - hashWithSalt = hashWithSalt1 - -instance (Hashable a1, Hashable a2) => Hashable1 ((,,) a1 a2) where - liftHashWithSalt = defaultLiftHashWithSalt - -instance Hashable a1 => Hashable2 ((,,) a1) where - liftHashWithSalt2 h1 h2 s (a1, a2, a3) = - (s `hashWithSalt` a1) `h1` a2 `h2` a3 - -instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4) => - Hashable (a1, a2, a3, a4) where - hash (a1, a2, a3, a4) = hash a1 `hashWithSalt` a2 - `hashWithSalt` a3 `hashWithSalt` a4 - hashWithSalt = hashWithSalt1 - -instance (Hashable a1, Hashable a2, Hashable a3) => Hashable1 ((,,,) a1 a2 a3) where - liftHashWithSalt = defaultLiftHashWithSalt - -instance (Hashable a1, Hashable a2) => Hashable2 ((,,,) a1 a2) where - liftHashWithSalt2 h1 h2 s (a1, a2, a3, a4) = - (s `hashWithSalt` a1 `hashWithSalt` a2) `h1` a3 `h2` a4 - -instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4, Hashable a5) - => Hashable (a1, a2, a3, a4, a5) where - hash (a1, a2, a3, a4, a5) = - hash a1 `hashWithSalt` a2 `hashWithSalt` a3 - `hashWithSalt` a4 `hashWithSalt` a5 - hashWithSalt = hashWithSalt1 - -instance (Hashable a1, Hashable a2, Hashable a3, - Hashable a4) => Hashable1 ((,,,,) a1 a2 a3 a4) where - liftHashWithSalt = defaultLiftHashWithSalt - -instance (Hashable a1, Hashable a2, Hashable a3) - => Hashable2 ((,,,,) a1 a2 a3) where - liftHashWithSalt2 h1 h2 s (a1, a2, a3, a4, a5) = - (s `hashWithSalt` a1 `hashWithSalt` a2 - `hashWithSalt` a3) `h1` a4 `h2` a5 - - -instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4, Hashable a5, - Hashable a6) => Hashable (a1, a2, a3, a4, a5, a6) where - hash (a1, a2, a3, a4, a5, a6) = - hash a1 `hashWithSalt` a2 `hashWithSalt` a3 - `hashWithSalt` a4 `hashWithSalt` a5 `hashWithSalt` a6 - hashWithSalt = hashWithSalt1 - -instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4, - Hashable a5) => Hashable1 ((,,,,,) a1 a2 a3 a4 a5) where - liftHashWithSalt = defaultLiftHashWithSalt - -instance (Hashable a1, Hashable a2, Hashable a3, - Hashable a4) => Hashable2 ((,,,,,) a1 a2 a3 a4) where - liftHashWithSalt2 h1 h2 s (a1, a2, a3, a4, a5, a6) = - (s `hashWithSalt` a1 `hashWithSalt` a2 `hashWithSalt` a3 - `hashWithSalt` a4) `h1` a5 `h2` a6 - - -instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4, Hashable a5, - Hashable a6, Hashable a7) => - Hashable (a1, a2, a3, a4, a5, a6, a7) where - hash (a1, a2, a3, a4, a5, a6, a7) = - hash a1 `hashWithSalt` a2 `hashWithSalt` a3 - `hashWithSalt` a4 `hashWithSalt` a5 `hashWithSalt` a6 `hashWithSalt` a7 - hashWithSalt s (a1, a2, a3, a4, a5, a6, a7) = - s `hashWithSalt` a1 `hashWithSalt` a2 `hashWithSalt` a3 - `hashWithSalt` a4 `hashWithSalt` a5 `hashWithSalt` a6 `hashWithSalt` a7 - -instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4, Hashable a5, Hashable a6) => Hashable1 ((,,,,,,) a1 a2 a3 a4 a5 a6) where - liftHashWithSalt = defaultLiftHashWithSalt - -instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4, - Hashable a5) => Hashable2 ((,,,,,,) a1 a2 a3 a4 a5) where - liftHashWithSalt2 h1 h2 s (a1, a2, a3, a4, a5, a6, a7) = - (s `hashWithSalt` a1 `hashWithSalt` a2 `hashWithSalt` a3 - `hashWithSalt` a4 `hashWithSalt` a5) `h1` a6 `h2` a7 - -instance Hashable (StableName a) where - hash = hashStableName - hashWithSalt = defaultHashWithSalt - --- Auxillary type for Hashable [a] definition -data SPInt = SP !Int !Int - -instance Hashable a => Hashable [a] where - {-# SPECIALIZE instance Hashable [Char] #-} - hashWithSalt = hashWithSalt1 - -instance Hashable1 [] where - liftHashWithSalt h salt arr = finalise (foldl' step (SP salt 0) arr) - where - finalise (SP s l) = hashWithSalt s l - step (SP s l) x = SP (h s x) (l + 1) - -instance Hashable B.ByteString where - hashWithSalt salt bs = unsafeDupablePerformIO $ - B.unsafeUseAsCStringLen bs $ \(p, len) -> - hashPtrWithSalt p (fromIntegral len) salt - -instance Hashable BL.ByteString where - hashWithSalt = BL.foldlChunks hashWithSalt - -#if MIN_VERSION_bytestring(0,10,4) -instance Hashable BSI.ShortByteString where -#if MIN_VERSION_base(4,3,0) - hashWithSalt salt sbs@(BSI.SBS ba) = -#else - hashWithSalt salt sbs@(BSI.SBS ba _) = -#endif - hashByteArrayWithSalt ba 0 (BSI.length sbs) salt -#endif - -instance Hashable T.Text where - hashWithSalt salt (T.Text arr off len) = - hashByteArrayWithSalt (TA.aBA arr) (off `shiftL` 1) (len `shiftL` 1) - salt - -instance Hashable TL.Text where - hashWithSalt = TL.foldlChunks hashWithSalt - --- | Compute the hash of a ThreadId. -hashThreadId :: ThreadId -> Int -hashThreadId (ThreadId t) = hash (fromIntegral (getThreadId t) :: Int) - -foreign import ccall unsafe "rts_getThreadId" getThreadId - :: ThreadId# -> CInt - -instance Hashable ThreadId where - hash = hashThreadId - hashWithSalt = defaultHashWithSalt - -instance Hashable (Ptr a) where - hashWithSalt salt p = hashWithSalt salt $ ptrToIntPtr p - -instance Hashable (FunPtr a) where - hashWithSalt salt p = hashWithSalt salt $ castFunPtrToPtr p - -instance Hashable IntPtr where - hash n = fromIntegral n - hashWithSalt = defaultHashWithSalt - -instance Hashable WordPtr where - hash n = fromIntegral n - hashWithSalt = defaultHashWithSalt - -#if __GLASGOW_HASKELL__ < 801 --- | Compute the hash of a TypeRep, in various GHC versions we can do this quickly. -hashTypeRep :: TypeRep -> Int -{-# INLINE hashTypeRep #-} -#if __GLASGOW_HASKELL__ >= 710 --- Fingerprint is just the MD5, so taking any Int from it is fine -hashTypeRep tr = let Fingerprint x _ = typeRepFingerprint tr in fromIntegral x -#elif __GLASGOW_HASKELL__ >= 702 --- Fingerprint is just the MD5, so taking any Int from it is fine -hashTypeRep (TypeRep (Fingerprint x _) _ _) = fromIntegral x -#elif __GLASGOW_HASKELL__ >= 606 -hashTypeRep = unsafeDupablePerformIO . typeRepKey -#else -hashTypeRep = hash . show -#endif - -instance Hashable TypeRep where - hash = hashTypeRep - hashWithSalt = defaultHashWithSalt - {-# INLINE hash #-} - -#else - -hashTypeRep :: Type.Reflection.TypeRep a -> Int -hashTypeRep tr = - let Fingerprint x _ = typeRepFingerprint tr in fromIntegral x - -instance Hashable Type.Reflection.SomeTypeRep where - hash (Type.Reflection.SomeTypeRep r) = hashTypeRep r - hashWithSalt = defaultHashWithSalt - {-# INLINE hash #-} - -instance Hashable (Type.Reflection.TypeRep a) where - hash = hashTypeRep - hashWithSalt = defaultHashWithSalt - {-# INLINE hash #-} -#endif - -#if MIN_VERSION_base(4,8,0) -instance Hashable Void where - hashWithSalt _ = absurd -#endif - --- | Compute a hash value for the content of this pointer. -hashPtr :: Ptr a -- ^ pointer to the data to hash - -> Int -- ^ length, in bytes - -> IO Int -- ^ hash value -hashPtr p len = hashPtrWithSalt p len defaultSalt - --- | Compute a hash value for the content of this pointer, using an --- initial salt. --- --- This function can for example be used to hash non-contiguous --- segments of memory as if they were one contiguous segment, by using --- the output of one hash as the salt for the next. -hashPtrWithSalt :: Ptr a -- ^ pointer to the data to hash - -> Int -- ^ length, in bytes - -> Int -- ^ salt - -> IO Int -- ^ hash value -hashPtrWithSalt p len salt = - fromIntegral `fmap` c_hashCString (castPtr p) (fromIntegral len) - (fromIntegral salt) - -foreign import ccall unsafe "hashable_fnv_hash" c_hashCString - :: CString -> CLong -> CLong -> IO CLong - --- | Compute a hash value for the content of this 'ByteArray#', --- beginning at the specified offset, using specified number of bytes. -hashByteArray :: ByteArray# -- ^ data to hash - -> Int -- ^ offset, in bytes - -> Int -- ^ length, in bytes - -> Int -- ^ hash value -hashByteArray ba0 off len = hashByteArrayWithSalt ba0 off len defaultSalt -{-# INLINE hashByteArray #-} - --- | Compute a hash value for the content of this 'ByteArray#', using --- an initial salt. --- --- This function can for example be used to hash non-contiguous --- segments of memory as if they were one contiguous segment, by using --- the output of one hash as the salt for the next. -hashByteArrayWithSalt - :: ByteArray# -- ^ data to hash - -> Int -- ^ offset, in bytes - -> Int -- ^ length, in bytes - -> Int -- ^ salt - -> Int -- ^ hash value -hashByteArrayWithSalt ba !off !len !h = - fromIntegral $ c_hashByteArray ba (fromIntegral off) (fromIntegral len) - (fromIntegral h) - -foreign import ccall unsafe "hashable_fnv_hash_offset" c_hashByteArray - :: ByteArray# -> CLong -> CLong -> CLong -> CLong - --- | Combine two given hash values. 'combine' has zero as a left --- identity. -combine :: Int -> Int -> Int -combine h1 h2 = (h1 * 16777619) `xor` h2 - -instance Hashable Unique where - hash = hashUnique - hashWithSalt = defaultHashWithSalt - -instance Hashable Version where - hashWithSalt salt (Version branch tags) = - salt `hashWithSalt` branch `hashWithSalt` tags - -#if MIN_VERSION_base(4,7,0) --- Using hashWithSalt1 would cause needless constraint -instance Hashable (Fixed a) where - hashWithSalt salt (MkFixed i) = hashWithSalt salt i -instance Hashable1 Fixed where - liftHashWithSalt _ salt (MkFixed i) = hashWithSalt salt i -#endif - -#if MIN_VERSION_base(4,8,0) -instance Hashable a => Hashable (Identity a) where - hashWithSalt = hashWithSalt1 -instance Hashable1 Identity where - liftHashWithSalt h salt (Identity x) = h salt x -#endif - --- Using hashWithSalt1 would cause needless constraint -instance Hashable a => Hashable (Const a b) where - hashWithSalt salt (Const x) = hashWithSalt salt x - -instance Hashable a => Hashable1 (Const a) where - liftHashWithSalt = defaultLiftHashWithSalt - -instance Hashable2 Const where - liftHashWithSalt2 f _ salt (Const x) = f salt x - -#if MIN_VERSION_base(4,7,0) -instance Hashable (Proxy a) where - hash _ = 0 - hashWithSalt s _ = s - -instance Hashable1 Proxy where - liftHashWithSalt _ s _ = s -#endif - --- instances formerly provided by 'semigroups' package -#if MIN_VERSION_base(4,9,0) -instance Hashable a => Hashable (NE.NonEmpty a) where - hashWithSalt p (a NE.:| as) = p `hashWithSalt` a `hashWithSalt` as - -instance Hashable a => Hashable (Min a) where - hashWithSalt p (Min a) = hashWithSalt p a - -instance Hashable a => Hashable (Max a) where - hashWithSalt p (Max a) = hashWithSalt p a - -instance (Hashable a, Hashable b) => Hashable (Arg a b) where - hashWithSalt p (Arg a b) = hashWithSalt p a `hashWithSalt` b - -instance Hashable a => Hashable (First a) where - hashWithSalt p (First a) = hashWithSalt p a - -instance Hashable a => Hashable (Last a) where - hashWithSalt p (Last a) = hashWithSalt p a - -instance Hashable a => Hashable (WrappedMonoid a) where - hashWithSalt p (WrapMonoid a) = hashWithSalt p a - -instance Hashable a => Hashable (Option a) where - hashWithSalt p (Option a) = hashWithSalt p a -#endif - --- instances for @Data.Functor.{Product,Sum,Compose}@, present --- in base-4.9 and onward. -#if MIN_VERSION_base(4,9,0) --- | In general, @hash (Compose x) ≠ hash x@. However, @hashWithSalt@ satisfies --- its variant of this equivalence. -instance (Hashable1 f, Hashable1 g, Hashable a) => Hashable (Compose f g a) where - hashWithSalt = hashWithSalt1 - -instance (Hashable1 f, Hashable1 g) => Hashable1 (Compose f g) where - liftHashWithSalt h s = liftHashWithSalt (liftHashWithSalt h) s . getCompose - -instance (Hashable1 f, Hashable1 g) => Hashable1 (FP.Product f g) where - liftHashWithSalt h s (FP.Pair a b) = liftHashWithSalt h (liftHashWithSalt h s a) b - -instance (Hashable1 f, Hashable1 g, Hashable a) => Hashable (FP.Product f g a) where - hashWithSalt = hashWithSalt1 - -instance (Hashable1 f, Hashable1 g) => Hashable1 (FS.Sum f g) where - liftHashWithSalt h s (FS.InL a) = liftHashWithSalt h (s `combine` 0) a - liftHashWithSalt h s (FS.InR a) = liftHashWithSalt h (s `combine` distinguisher) a - -instance (Hashable1 f, Hashable1 g, Hashable a) => Hashable (FS.Sum f g a) where - hashWithSalt = hashWithSalt1 -#endif - --- | A hashable value along with the result of the 'hash' function. -data Hashed a = Hashed a {-# UNPACK #-} !Int - deriving (Typeable) - --- | Wrap a hashable value, caching the 'hash' function result. -hashed :: Hashable a => a -> Hashed a -hashed a = Hashed a (hash a) - --- | Unwrap hashed value. -unhashed :: Hashed a -> a -unhashed (Hashed a _) = a - --- | Uses precomputed hash to detect inequality faster -instance Eq a => Eq (Hashed a) where - Hashed a ha == Hashed b hb = ha == hb && a == b - -instance Ord a => Ord (Hashed a) where - Hashed a _ `compare` Hashed b _ = a `compare` b - -instance Show a => Show (Hashed a) where - showsPrec d (Hashed a _) = showParen (d > 10) $ - showString "hashed" . showChar ' ' . showsPrec 11 a - -instance Hashable (Hashed a) where - hashWithSalt = defaultHashWithSalt - hash (Hashed _ h) = h - --- This instance is a little unsettling. It is unusal for --- 'liftHashWithSalt' to ignore its first argument when a --- value is actually available for it to work on. -instance Hashable1 Hashed where - liftHashWithSalt _ s (Hashed _ h) = defaultHashWithSalt s h - -instance (IsString a, Hashable a) => IsString (Hashed a) where - fromString s = let r = fromString s in Hashed r (hash r) - -instance F.Foldable Hashed where - foldr f acc (Hashed a _) = f a acc - -instance NFData a => NFData (Hashed a) where - rnf = rnf . unhashed - --- | 'Hashed' cannot be 'Functor' -mapHashed :: Hashable b => (a -> b) -> Hashed a -> Hashed b -mapHashed f (Hashed a _) = hashed (f a) - --- | 'Hashed' cannot be 'Traversable' -traverseHashed :: (Hashable b, Functor f) => (a -> f b) -> Hashed a -> f (Hashed b) -traverseHashed f (Hashed a _) = fmap hashed (f a) - --- instances for @Data.Functor.Classes@ higher rank typeclasses --- in base-4.9 and onward. -#if MIN_VERSION_base(4,9,0) -instance Eq1 Hashed where - liftEq f (Hashed a ha) (Hashed b hb) = ha == hb && f a b - -instance Ord1 Hashed where - liftCompare f (Hashed a _) (Hashed b _) = f a b - -instance Show1 Hashed where - liftShowsPrec sp _ d (Hashed a _) = showsUnaryWith sp "hashed" d a -#endif - - diff --git a/Data/Hashable/Generic.hs b/Data/Hashable/Generic.hs deleted file mode 100644 --- a/Data/Hashable/Generic.hs +++ /dev/null @@ -1,125 +0,0 @@ -{-# LANGUAGE BangPatterns, FlexibleInstances, KindSignatures, - ScopedTypeVariables, TypeOperators, - MultiParamTypeClasses, GADTs, FlexibleContexts #-} -{-# OPTIONS_GHC -fno-warn-orphans #-} - ------------------------------------------------------------------------- --- | --- Module : Data.Hashable.Generic --- Copyright : (c) Bryan O'Sullivan 2012 --- License : BSD-style --- Maintainer : bos@serpentine.com --- Stability : provisional --- Portability : GHC >= 7.2 --- --- Hashable support for GHC generics. - -module Data.Hashable.Generic - ( - ) where - -import Data.Bits (shiftR) -import Data.Hashable.Class -import GHC.Generics - --- Type without constructors -instance GHashable arity V1 where - ghashWithSalt _ salt _ = hashWithSalt salt () - --- Constructor without arguments -instance GHashable arity U1 where - ghashWithSalt _ salt U1 = hashWithSalt salt () - -instance (GHashable arity a, GHashable arity b) => GHashable arity (a :*: b) where - ghashWithSalt toHash salt (x :*: y) = - (ghashWithSalt toHash (ghashWithSalt toHash salt x) y) - --- Metadata (constructor name, etc) -instance GHashable arity a => GHashable arity (M1 i c a) where - ghashWithSalt targs salt = ghashWithSalt targs salt . unM1 - --- Constants, additional parameters, and rank-1 recursion -instance Hashable a => GHashable arity (K1 i a) where - ghashWithSalt _ = hashUsing unK1 - -instance GHashable One Par1 where - ghashWithSalt (HashArgs1 h) salt = h salt . unPar1 - -instance Hashable1 f => GHashable One (Rec1 f) where - ghashWithSalt (HashArgs1 h) salt = liftHashWithSalt h salt . unRec1 - -instance (Hashable1 f, GHashable One g) => GHashable One (f :.: g) where - ghashWithSalt targs salt = liftHashWithSalt (ghashWithSalt targs) salt . unComp1 - -class SumSize f => GSum arity f where - hashSum :: HashArgs arity a -> Int -> Int -> f a -> Int - -- hashSum args salt index value = ... - --- [Note: Hashing a sum type] --- --- The tree structure is used in GHC.Generics to represent the sum (and --- product) part of the generic represention of the type, e.g.: --- --- (C0 ... :+: C1 ...) :+: (C2 ... :+: (C3 ... :+: C4 ...)) --- --- The value constructed with C2 constructor is represented as (R1 (L1 ...)). --- Yet, if we think that this tree is a flat (heterogenous) list: --- --- [C0 ..., C1 ..., C2 ..., C3 ..., C4... ] --- --- then the value constructed with C2 is a (dependent) pair (2, ...), and --- hashing it is simple: --- --- salt `hashWithSalt` (2 :: Int) `hashWithSalt` ... --- --- This is what we do below. When drilling down the tree, we count how many --- leafs are to the left (`index` variable). At the leaf case C1, we'll have an --- actual index into the sum. --- --- This works well for balanced data. However for recursive types like: --- --- data Nat = Z | S Nat --- --- the `hashWithSalt salt (S (S (S Z)))` is --- --- salt `hashWithSalt` (1 :: Int) -- first S --- `hashWithSalt` (1 :: Int) -- second S --- `hashWithSalt` (1 :: Int) -- third S --- `hashWithSalt` (0 :: Int) -- Z --- `hashWithSalt` () -- U1 --- --- For that type the manual implementation: --- --- instance Hashable Nat where --- hashWithSalt salt n = hashWithSalt salt (natToInteger n) --- --- would be better performing CPU and hash-quality wise (assuming that --- Integer's Hashable is of high quality). --- -instance (GSum arity a, GSum arity b) => GHashable arity (a :+: b) where - ghashWithSalt toHash salt = hashSum toHash salt 0 - -instance (GSum arity a, GSum arity b) => GSum arity (a :+: b) where - hashSum toHash !salt !index s = case s of - L1 x -> hashSum toHash salt index x - R1 x -> hashSum toHash salt (index + sizeL) x - where - sizeL = unTagged (sumSize :: Tagged a) - {-# INLINE hashSum #-} - -instance GHashable arity a => GSum arity (C1 c a) where - hashSum toHash !salt !index (M1 x) = ghashWithSalt toHash (hashWithSalt salt index) x - {-# INLINE hashSum #-} - -class SumSize f where - sumSize :: Tagged f - -newtype Tagged (s :: * -> *) = Tagged {unTagged :: Int} - -instance (SumSize a, SumSize b) => SumSize (a :+: b) where - sumSize = Tagged $ unTagged (sumSize :: Tagged a) + - unTagged (sumSize :: Tagged b) - -instance SumSize (C1 c a) where - sumSize = Tagged 1 - diff --git a/Data/Hashable/Lifted.hs b/Data/Hashable/Lifted.hs deleted file mode 100644 --- a/Data/Hashable/Lifted.hs +++ /dev/null @@ -1,96 +0,0 @@ ------------------------------------------------------------------------- --- | --- Module : Data.Hashable.Class --- Copyright : (c) Milan Straka 2010 --- (c) Johan Tibell 2011 --- (c) Bryan O'Sullivan 2011, 2012 --- License : BSD-style --- Maintainer : johan.tibell@gmail.com --- Stability : provisional --- Portability : portable --- --- Lifting of the 'Hashable' class to unary and binary type constructors. --- These classes are needed to express the constraints on arguments of --- types that are parameterized by type constructors. Fixed-point data --- types and monad transformers are such types. - -module Data.Hashable.Lifted - ( -- * Type Classes - Hashable1(..) - , Hashable2(..) - -- * Auxiliary Functions - , hashWithSalt1 - , hashWithSalt2 - , defaultLiftHashWithSalt - -- * Motivation - -- $motivation - ) where - -import Data.Hashable.Class - --- $motivation --- --- This type classes provided in this module are used to express constraints --- on type constructors in a Haskell98-compatible fashion. As an example, consider --- the following two types (Note that these instances are not actually provided --- because @hashable@ does not have @transformers@ or @free@ as a dependency): --- --- > newtype WriterT w m a = WriterT { runWriterT :: m (a, w) } --- > data Free f a = Pure a | Free (f (Free f a)) --- --- The 'Hashable1' instances for @WriterT@ and @Free@ could be written as: --- --- > instance (Hashable w, Hashable1 m) => Hashable1 (WriterT w m) where --- > liftHashWithSalt h s (WriterT m) = --- > liftHashWithSalt (liftHashWithSalt2 h hashWithSalt) s m --- > instance Hashable1 f => Hashable1 (Free f) where --- > liftHashWithSalt h = go where --- > go s x = case x of --- > Pure a -> h s a --- > Free p -> liftHashWithSalt go s p --- --- The 'Hashable' instances for these types can be trivially recovered with --- 'hashWithSalt1': --- --- > instance (Hashable w, Hashable1 m, Hashable a) => Hashable (WriterT w m a) where --- > hashWithSalt = hashWithSalt1 --- > instance (Hashable1 f, Hashable a) => Hashable (Free f a) where --- > hashWithSalt = hashWithSalt1 - --- --- $discussion --- --- Regardless of whether 'hashWithSalt1' is used to provide an implementation --- of 'hashWithSalt', they should produce the same hash when called with --- the same arguments. This is the only law that 'Hashable1' and 'Hashable2' --- are expected to follow. --- --- The typeclasses in this module only provide lifting for 'hashWithSalt', not --- for 'hash'. This is because such liftings cannot be defined in a way that --- would satisfy the @liftHash@ variant of the above law. As an illustration --- of the problem we run into, let us assume that 'Hashable1' were --- given a 'liftHash' method: --- --- > class Hashable1 t where --- > liftHash :: (Int -> a) -> t a -> Int --- > liftHashWithSalt :: (Int -> a -> Int) -> Int -> t a -> Int --- --- Even for a type as simple as 'Maybe', the problem manifests itself. The --- 'Hashable' instance for 'Maybe' is: --- --- > distinguisher :: Int --- > distinguisher = ... --- > --- > instance Hashable a => Hashable (Maybe a) where --- > hash Nothing = 0 --- > hash (Just a) = distinguisher `hashWithSalt` a --- > hashWithSalt s Nothing = ... --- > hashWithSalt s (Just a) = ... --- --- The implementation of 'hash' calls 'hashWithSalt' on @a@. The hypothetical --- @liftHash@ defined earlier only accepts an argument that corresponds to --- the implementation of 'hash' for @a@. Consequently, this formulation of --- @liftHash@ would not provide a way to match the current behavior of 'hash' --- for 'Maybe'. This problem gets worse when 'Either' and @[]@ are considered. --- The solution adopted in this library is to omit @liftHash@ entirely. - diff --git a/Data/Hashable/RandomSource.hs b/Data/Hashable/RandomSource.hs deleted file mode 100644 --- a/Data/Hashable/RandomSource.hs +++ /dev/null @@ -1,32 +0,0 @@ -{-# LANGUAGE CPP, ForeignFunctionInterface #-} -#if defined(__GLASGOW_HASKELL__) && __GLASGOW_HASKELL__ >= 702 -{-# LANGUAGE Trustworthy #-} -#endif - -module Data.Hashable.RandomSource - ( - getRandomBytes - , getRandomBytes_ - ) where - -import Data.ByteString as B -import Data.ByteString.Internal (create) -import Foreign.C.Error (throwErrnoIfMinus1_) -#if MIN_VERSION_base(4,5,0) -import Foreign.C.Types (CInt(CInt)) -#else -import Foreign.C.Types (CInt) -#endif -import Foreign.Ptr (Ptr) - -getRandomBytes :: Int -> IO ByteString -getRandomBytes nbytes - | nbytes <= 0 = return B.empty - | otherwise = create nbytes $ flip (getRandomBytes_ "getRandomBytes") nbytes - -getRandomBytes_ :: String -> Ptr a -> Int -> IO () -getRandomBytes_ what ptr nbytes = do - throwErrnoIfMinus1_ what $ c_getRandomBytes ptr (fromIntegral nbytes) - -foreign import ccall unsafe "hashable_getRandomBytes" c_getRandomBytes - :: Ptr a -> CInt -> IO CInt diff --git a/Data/Hashable/SipHash.hs b/Data/Hashable/SipHash.hs deleted file mode 100644 --- a/Data/Hashable/SipHash.hs +++ /dev/null @@ -1,159 +0,0 @@ -{-# LANGUAGE BangPatterns, CPP, GeneralizedNewtypeDeriving, RecordWildCards #-} -{-# OPTIONS_GHC -fno-warn-name-shadowing #-} - -module Data.Hashable.SipHash - ( - LE64 - , Sip - , fromWord64 - , fullBlock - , lastBlock - , finalize - , hashByteString - ) where - -#include "MachDeps.h" - -import Data.Bits ((.|.), (.&.), rotateL, shiftL, xor) -#if MIN_VERSION_base(4,5,0) -import Data.Bits (unsafeShiftL) -#endif -import Data.Word (Word8, Word64) -import Foreign.ForeignPtr (withForeignPtr) -import Foreign.Ptr (Ptr, castPtr, plusPtr) -import Data.ByteString.Internal (ByteString(PS), inlinePerformIO) -import Foreign.Storable (peek) -import Numeric (showHex) - -newtype LE64 = LE64 { fromLE64 :: Word64 } - deriving (Eq) - -instance Show LE64 where - show (LE64 !v) = let s = showHex v "" - in "0x" ++ replicate (16 - length s) '0' ++ s - -data Sip = Sip { - v0 :: {-# UNPACK #-} !Word64, v1 :: {-# UNPACK #-} !Word64 - , v2 :: {-# UNPACK #-} !Word64, v3 :: {-# UNPACK #-} !Word64 - } - -fromWord64 :: Word64 -> LE64 -#ifndef WORDS_BIGENDIAN -fromWord64 = LE64 -#else -#error big endian support TBD -#endif - -initState :: (Sip -> r) -> Word64 -> Word64 -> r -initState k k0 k1 = k (Sip s0 s1 s2 s3) - where !s0 = (k0 `xor` 0x736f6d6570736575) - !s1 = (k1 `xor` 0x646f72616e646f6d) - !s2 = (k0 `xor` 0x6c7967656e657261) - !s3 = (k1 `xor` 0x7465646279746573) - -sipRound :: (Sip -> r) -> Sip -> r -sipRound k Sip{..} = k (Sip v0_c v1_d v2_c v3_d) - where v0_a = v0 + v1 - v2_a = v2 + v3 - v1_a = v1 `rotateL` 13 - v3_a = v3 `rotateL` 16 - v1_b = v1_a `xor` v0_a - v3_b = v3_a `xor` v2_a - v0_b = v0_a `rotateL` 32 - v2_b = v2_a + v1_b - v0_c = v0_b + v3_b - v1_c = v1_b `rotateL` 17 - v3_c = v3_b `rotateL` 21 - v1_d = v1_c `xor` v2_b - v3_d = v3_c `xor` v0_c - v2_c = v2_b `rotateL` 32 - -fullBlock :: Int -> LE64 -> (Sip -> r) -> Sip -> r -fullBlock c m k st@Sip{..} - | c == 2 = sipRound (sipRound k') st' - | otherwise = runRounds c k' st' - where k' st1@Sip{..} = k st1{ v0 = v0 `xor` fromLE64 m } - st' = st{ v3 = v3 `xor` fromLE64 m } -{-# INLINE fullBlock #-} - -runRounds :: Int -> (Sip -> r) -> Sip -> r -runRounds c k = go 0 - where go i st - | i < c = sipRound (go (i+1)) st - | otherwise = k st -{-# INLINE runRounds #-} - -lastBlock :: Int -> Int -> LE64 -> (Sip -> r) -> Sip -> r -lastBlock !c !len !m k st = -#ifndef WORDS_BIGENDIAN - fullBlock c (LE64 m') k st -#else -#error big endian support TBD -#endif - where m' = fromLE64 m .|. ((fromIntegral len .&. 0xff) `shiftL` 56) -{-# INLINE lastBlock #-} - -finalize :: Int -> (Word64 -> r) -> Sip -> r -finalize d k st@Sip{..} - | d == 4 = sipRound (sipRound (sipRound (sipRound k'))) st' - | otherwise = runRounds d k' st' - where k' Sip{..} = k $! v0 `xor` v1 `xor` v2 `xor` v3 - st' = st{ v2 = v2 `xor` 0xff } -{-# INLINE finalize #-} - -hashByteString :: Int -> Int -> Word64 -> Word64 -> ByteString -> Word64 -hashByteString !c !d k0 k1 (PS fp off len) = - inlinePerformIO . withForeignPtr fp $ \basePtr -> - let ptr0 = basePtr `plusPtr` off - scant = len .&. 7 - endBlocks = ptr0 `plusPtr` (len - scant) - go !ptr st - | ptr == endBlocks = readLast ptr - | otherwise = do - m <- peekLE64 ptr - fullBlock c m (go (ptr `plusPtr` 8)) st - where - zero !m _ _ = lastBlock c len (LE64 m) (finalize d return) st - one k m p s = do - w <- fromIntegral `fmap` peekByte p - k (m .|. (w `unsafeShiftL` s)) (p `plusPtr` 1) (s+8) - readLast p = - case scant of - 0 -> zero 0 p (0::Int) - 1 -> one zero 0 p 0 - 2 -> one (one zero) 0 p 0 - 3 -> one (one (one zero)) 0 p 0 - 4 -> one (one (one (one zero))) 0 p 0 - 5 -> one (one (one (one (one zero)))) 0 p 0 - 6 -> one (one (one (one (one (one zero))))) 0 p 0 - _ -> one (one (one (one (one (one (one zero)))))) 0 p 0 - in initState (go ptr0) k0 k1 - -peekByte :: Ptr Word8 -> IO Word8 -peekByte = peek - -peekLE64 :: Ptr Word8 -> IO LE64 -#if defined(x86_64_HOST_ARCH) || defined(i386_HOST_ARCH) --- platforms on which unaligned loads are legal and usually fast -peekLE64 p = LE64 `fmap` peek (castPtr p) -#else -peekLE64 p = do - let peek8 d = fromIntegral `fmap` peekByte (p `plusPtr` d) - b0 <- peek8 0 - b1 <- peek8 1 - b2 <- peek8 2 - b3 <- peek8 3 - b4 <- peek8 4 - b5 <- peek8 5 - b6 <- peek8 6 - b7 <- peek8 7 - let !w = (b7 `shiftL` 56) .|. (b6 `shiftL` 48) .|. (b5 `shiftL` 40) .|. - (b4 `shiftL` 32) .|. (b3 `shiftL` 24) .|. (b2 `shiftL` 16) .|. - (b1 `shiftL` 8) .|. b0 - return (fromWord64 w) -#endif - -#if !(MIN_VERSION_base(4,5,0)) -unsafeShiftL :: Word64 -> Int -> Word64 -unsafeShiftL = shiftL -#endif diff --git a/benchmarks/Benchmarks.hs b/benchmarks/Benchmarks.hs deleted file mode 100644 --- a/benchmarks/Benchmarks.hs +++ /dev/null @@ -1,314 +0,0 @@ -{-# LANGUAGE BangPatterns, CPP, ForeignFunctionInterface, MagicHash, - UnboxedTuples, DeriveGeneric #-} - -module Main (main) where - -import Control.Monad.ST -import Criterion.Main -import Data.Hashable -import Data.Hashable.SipHash -import Data.Int -import Foreign.ForeignPtr -import GHC.Exts -import GHC.ST (ST(..)) -import Data.Word -import Foreign.C.Types (CInt(..), CLong(..), CSize(..)) -import Foreign.Ptr -import Data.ByteString.Internal -import GHC.Generics (Generic) -import qualified Data.ByteString.Lazy as BL -import qualified Data.Text as T -import qualified Data.Text.Lazy as TL -import qualified Crypto.MAC.SipHash as HS -import qualified Data.ByteString.Char8 as B8 - --- Benchmark English words (5 and 8), base64 encoded integers (11), --- SHA1 hashes as hex (40), and large blobs (1 Mb). -main :: IO () -main = do - -- We do not actually care about the contents of these pointers. - fp5 <- mallocForeignPtrBytes 5 - fp8 <- mallocForeignPtrBytes 8 - fp11 <- mallocForeignPtrBytes 11 - fp40 <- mallocForeignPtrBytes 40 - fp128 <- mallocForeignPtrBytes 128 - fp512 <- mallocForeignPtrBytes 512 - let !mb = 2^(20 :: Int) -- 1 Mb - fp1Mb <- mallocForeignPtrBytes mb - - let exP = P 22.0203 234.19 'x' 6424 - exS = S3 - exPS = PS3 'z' 7715 - - -- We don't care about the contents of these either. - let !ba5 = new 5; !ba8 = new 8; !ba11 = new 11; !ba40 = new 40 - !ba128 = new 128; !ba512 = new 512; !ba1Mb = new mb - - s5 = ['\0'..'\4']; s8 = ['\0'..'\7']; s11 = ['\0'..'\10'] - s40 = ['\0'..'\39']; s128 = ['\0'..'\127']; s512 = ['\0'..'\511'] - s1Mb = ['\0'..'\999999'] - - !bs5 = B8.pack s5; !bs8 = B8.pack s8; !bs11 = B8.pack s11 - !bs40 = B8.pack s40; !bs128 = B8.pack s128; !bs512 = B8.pack s512 - !bs1Mb = B8.pack s1Mb - - blmeg = BL.take (fromIntegral mb) . BL.fromChunks . repeat - bl5 = BL.fromChunks [bs5]; bl8 = BL.fromChunks [bs8] - bl11 = BL.fromChunks [bs11]; bl40 = BL.fromChunks [bs40] - bl128 = BL.fromChunks [bs128]; bl512 = BL.fromChunks [bs512] - bl1Mb_40 = blmeg bs40; bl1Mb_128 = blmeg bs128 - bl1Mb_64k = blmeg (B8.take 65536 bs1Mb) - - !t5 = T.pack s5; !t8 = T.pack s8; !t11 = T.pack s11 - !t40 = T.pack s40; !t128 = T.pack s128; !t512 = T.pack s512 - !t1Mb = T.pack s1Mb - - tlmeg = TL.take (fromIntegral mb) . TL.fromChunks . repeat - tl5 = TL.fromStrict t5; tl8 = TL.fromStrict t8 - tl11 = TL.fromStrict t11; tl40 = TL.fromStrict t40 - tl128 = TL.fromStrict t128; tl512 = TL.fromChunks (replicate 4 t128) - tl1Mb_40 = tlmeg t40; tl1Mb_128 = tlmeg t128 - tl1Mb_64k = tlmeg (T.take 65536 t1Mb) - - let k0 = 0x4a7330fae70f52e8 - k1 = 0x919ea5953a9a1ec9 - sipHash = hashByteString 2 4 k0 k1 - hsSipHash = HS.hash (HS.SipKey k0 k1) - cSipHash (PS fp off len) = - inlinePerformIO . withForeignPtr fp $ \ptr -> - return $! c_siphash 2 4 k0 k1 (ptr `plusPtr` off) (fromIntegral len) - cSipHash24 (PS fp off len) = - inlinePerformIO . withForeignPtr fp $ \ptr -> - return $! c_siphash24 k0 k1 (ptr `plusPtr` off) (fromIntegral len) - fnvHash (PS fp off len) = - inlinePerformIO . withForeignPtr fp $ \ptr -> - return $! fnv_hash (ptr `plusPtr` off) (fromIntegral len) 2166136261 -#ifdef HAVE_SSE2 - sse2SipHash (PS fp off len) = - inlinePerformIO . withForeignPtr fp $ \ptr -> - return $! sse2_siphash k0 k1 (ptr `plusPtr` off) (fromIntegral len) -#endif -#ifdef HAVE_SSE41 - sse41SipHash (PS fp off len) = - inlinePerformIO . withForeignPtr fp $ \ptr -> - return $! sse41_siphash k0 k1 (ptr `plusPtr` off) (fromIntegral len) -#endif - - withForeignPtr fp5 $ \ p5 -> - withForeignPtr fp8 $ \ p8 -> - withForeignPtr fp11 $ \ p11 -> - withForeignPtr fp40 $ \ p40 -> - withForeignPtr fp128 $ \ p128 -> - withForeignPtr fp512 $ \ p512 -> - withForeignPtr fp1Mb $ \ p1Mb -> - defaultMain - [ bgroup "hashPtr" - [ bench "5" $ whnfIO $ hashPtr p5 5 - , bench "8" $ whnfIO $ hashPtr p8 8 - , bench "11" $ whnfIO $ hashPtr p11 11 - , bench "40" $ whnfIO $ hashPtr p40 40 - , bench "128" $ whnfIO $ hashPtr p128 128 - , bench "512" $ whnfIO $ hashPtr p512 512 - , bench "2^20" $ whnfIO $ hashPtr p1Mb mb - ] - , bgroup "hashByteArray" - [ bench "5" $ whnf (hashByteArray ba5 0) 5 - , bench "8" $ whnf (hashByteArray ba8 0) 8 - , bench "11" $ whnf (hashByteArray ba11 0) 11 - , bench "40" $ whnf (hashByteArray ba40 0) 40 - , bench "128" $ whnf (hashByteArray ba128 0) 128 - , bench "512" $ whnf (hashByteArray ba512 0) 512 - , bench "2^20" $ whnf (hashByteArray ba1Mb 0) mb - ] - , bgroup "hash" - [ bgroup "ByteString" - [ bgroup "strict" - [ bench "5" $ whnf hash bs5 - , bench "8" $ whnf hash bs8 - , bench "11" $ whnf hash bs11 - , bench "40" $ whnf hash bs40 - , bench "128" $ whnf hash bs128 - , bench "512" $ whnf hash bs512 - , bench "2^20" $ whnf hash bs1Mb - ] - , bgroup "lazy" - [ bench "5" $ whnf hash bl5 - , bench "8" $ whnf hash bl8 - , bench "11" $ whnf hash bl11 - , bench "40" $ whnf hash bl40 - , bench "128" $ whnf hash bl128 - , bench "512" $ whnf hash bl512 - , bench "2^20_40" $ whnf hash bl1Mb_40 - , bench "2^20_128" $ whnf hash bl1Mb_128 - , bench "2^20_64k" $ whnf hash bl1Mb_64k - ] - ] - , bgroup "String" - [ bench "5" $ whnf hash s5 - , bench "8" $ whnf hash s8 - , bench "11" $ whnf hash s11 - , bench "40" $ whnf hash s40 - , bench "128" $ whnf hash s128 - , bench "512" $ whnf hash s512 - , bench "2^20" $ whnf hash s1Mb - ] - , bgroup "Text" - [ bgroup "strict" - [ bench "5" $ whnf hash t5 - , bench "8" $ whnf hash t8 - , bench "11" $ whnf hash t11 - , bench "40" $ whnf hash t40 - , bench "128" $ whnf hash t128 - , bench "512" $ whnf hash t512 - , bench "2^20" $ whnf hash t1Mb - ] - , bgroup "lazy" - [ bench "5" $ whnf hash tl5 - , bench "8" $ whnf hash tl8 - , bench "11" $ whnf hash tl11 - , bench "40" $ whnf hash tl40 - , bench "128" $ whnf hash tl128 - , bench "512" $ whnf hash tl512 - , bench "2^20_40" $ whnf hash tl1Mb_40 - , bench "2^20_128" $ whnf hash tl1Mb_128 - , bench "2^20_64k" $ whnf hash tl1Mb_64k - ] - ] - , bench "Int8" $ whnf hash (0xef :: Int8) - , bench "Int16" $ whnf hash (0x7eef :: Int16) - , bench "Int32" $ whnf hash (0x7eadbeef :: Int32) - , bench "Int" $ whnf hash (0x7eadbeefdeadbeef :: Int) - , bench "Int64" $ whnf hash (0x7eadbeefdeadbeef :: Int64) - , bench "Double" $ whnf hash (0.3780675796601578 :: Double) - ] - , bgroup "sipHash" - [ bench "5" $ whnf sipHash bs5 - , bench "8" $ whnf sipHash bs8 - , bench "11" $ whnf sipHash bs11 - , bench "40" $ whnf sipHash bs40 - , bench "128" $ whnf sipHash bs128 - , bench "512" $ whnf sipHash bs512 - , bench "2^20" $ whnf sipHash bs1Mb - ] - , bgroup "cSipHash" - [ bench "5" $ whnf cSipHash bs5 - , bench "8" $ whnf cSipHash bs8 - , bench "11" $ whnf cSipHash bs11 - , bench "40" $ whnf cSipHash bs40 - , bench "128" $ whnf cSipHash bs128 - , bench "512" $ whnf cSipHash bs512 - , bench "2^20" $ whnf cSipHash bs1Mb - ] - , bgroup "cSipHash24" - [ bench "5" $ whnf cSipHash24 bs5 - , bench "8" $ whnf cSipHash24 bs8 - , bench "11" $ whnf cSipHash24 bs11 - , bench "40" $ whnf cSipHash24 bs40 - , bench "128" $ whnf cSipHash24 bs128 - , bench "512" $ whnf cSipHash24 bs512 - , bench "2^20" $ whnf cSipHash24 bs1Mb - ] -#ifdef HAVE_SSE2 - , bgroup "sse2SipHash" - [ bench "5" $ whnf sse2SipHash bs5 - , bench "8" $ whnf sse2SipHash bs8 - , bench "11" $ whnf sse2SipHash bs11 - , bench "40" $ whnf sse2SipHash bs40 - , bench "128" $ whnf sse2SipHash bs128 - , bench "512" $ whnf sse2SipHash bs512 - , bench "2^20" $ whnf sse2SipHash bs1Mb - ] -#endif -#ifdef HAVE_SSE41 - , bgroup "sse41SipHash" - [ bench "5" $ whnf sse41SipHash bs5 - , bench "8" $ whnf sse41SipHash bs8 - , bench "11" $ whnf sse41SipHash bs11 - , bench "40" $ whnf sse41SipHash bs40 - , bench "128" $ whnf sse41SipHash bs128 - , bench "512" $ whnf sse41SipHash bs512 - , bench "2^20" $ whnf sse41SipHash bs1Mb - ] -#endif - , bgroup "pkgSipHash" - [ bench "5" $ whnf hsSipHash bs5 - , bench "8" $ whnf hsSipHash bs8 - , bench "11" $ whnf hsSipHash bs11 - , bench "40" $ whnf hsSipHash bs40 - , bench "128" $ whnf hsSipHash bs128 - , bench "512" $ whnf hsSipHash bs512 - , bench "2^20" $ whnf hsSipHash bs1Mb - ] - , bgroup "fnv" - [ bench "5" $ whnf fnvHash bs5 - , bench "8" $ whnf fnvHash bs8 - , bench "11" $ whnf fnvHash bs11 - , bench "40" $ whnf fnvHash bs40 - , bench "128" $ whnf fnvHash bs128 - , bench "512" $ whnf fnvHash bs512 - , bench "2^20" $ whnf fnvHash bs1Mb - ] - , bgroup "Int" - [ bench "id32" $ whnf id (0x7eadbeef :: Int32) - , bench "id64" $ whnf id (0x7eadbeefdeadbeef :: Int64) - , bench "wang32" $ whnf hash_wang_32 0xdeadbeef - , bench "wang64" $ whnf hash_wang_64 0xdeadbeefdeadbeef - , bench "jenkins32a" $ whnf hash_jenkins_32a 0xdeadbeef - , bench "jenkins32b" $ whnf hash_jenkins_32b 0xdeadbeef - ] - , bgroup "Generic" - [ bench "product" $ whnf hash exP - , bench "sum" $ whnf hash exS - , bench "product and sum" $ whnf hash exPS - ] - ] - -data ByteArray = BA { unBA :: !ByteArray# } - -new :: Int -> ByteArray# -new (I# n#) = unBA (runST $ ST $ \s1 -> - case newByteArray# n# s1 of - (# s2, ary #) -> case unsafeFreezeByteArray# ary s2 of - (# s3, ba #) -> (# s3, BA ba #)) - -foreign import ccall unsafe "hashable_siphash" c_siphash - :: CInt -> CInt -> Word64 -> Word64 -> Ptr Word8 -> CSize -> Word64 -foreign import ccall unsafe "hashable_siphash24" c_siphash24 - :: Word64 -> Word64 -> Ptr Word8 -> CSize -> Word64 -#ifdef HAVE_SSE2 -foreign import ccall unsafe "hashable_siphash24_sse2" sse2_siphash - :: Word64 -> Word64 -> Ptr Word8 -> CSize -> Word64 -#endif -#ifdef HAVE_SSE41 -foreign import ccall unsafe "hashable_siphash24_sse41" sse41_siphash - :: Word64 -> Word64 -> Ptr Word8 -> CSize -> Word64 -#endif - -foreign import ccall unsafe "hashable_fnv_hash" fnv_hash - :: Ptr Word8 -> CLong -> CLong -> CLong - -foreign import ccall unsafe "hashable_wang_32" hash_wang_32 - :: Word32 -> Word32 -foreign import ccall unsafe "hashable_wang_64" hash_wang_64 - :: Word64 -> Word64 -foreign import ccall unsafe "hash_jenkins_32a" hash_jenkins_32a - :: Word32 -> Word32 -foreign import ccall unsafe "hash_jenkins_32b" hash_jenkins_32b - :: Word32 -> Word32 - -data PS - = PS1 Int Char Bool - | PS2 String () - | PS3 Char Int - deriving (Generic) - -data P = P Double Float Char Int - deriving (Generic) - -data S = S1 | S2 | S3 | S4 | S5 - deriving (Generic) - -instance Hashable PS -instance Hashable P -instance Hashable S - diff --git a/benchmarks/cbits/inthash.c b/benchmarks/cbits/inthash.c deleted file mode 100644 --- a/benchmarks/cbits/inthash.c +++ /dev/null @@ -1,28 +0,0 @@ -#include - -/* - * 32-bit hashes by Bob Jenkins. - */ - -uint32_t hash_jenkins_32a(uint32_t a) -{ - a = (a+0x7ed55d16) + (a<<12); - a = (a^0xc761c23c) ^ (a>>19); - a = (a+0x165667b1) + (a<<5); - a = (a+0xd3a2646c) ^ (a<<9); - a = (a+0xfd7046c5) + (a<<3); - a = (a^0xb55a4f09) ^ (a>>16); - return a; -} - -uint32_t hash_jenkins_32b(uint32_t a) -{ - a -= (a<<6); - a ^= (a>>17); - a -= (a<<9); - a ^= (a<<4); - a -= (a<<3); - a ^= (a<<10); - a ^= (a>>15); - return a; -} diff --git a/benchmarks/cbits/siphash-sse2.c b/benchmarks/cbits/siphash-sse2.c deleted file mode 100644 --- a/benchmarks/cbits/siphash-sse2.c +++ /dev/null @@ -1,129 +0,0 @@ -/* - * The original code was developed by Samuel Neves, and has been - * only lightly modified. - * - * Used with permission. - */ -#pragma GCC target("sse2") - -#include -#include "siphash.h" - -#define _mm_roti_epi64(x, c) ((16 == (c)) ? _mm_shufflelo_epi16((x), _MM_SHUFFLE(2,1,0,3)) : _mm_xor_si128(_mm_slli_epi64((x), (c)), _mm_srli_epi64((x), 64-(c)))) - -u64 hashable_siphash24_sse2(u64 ik0, u64 ik1, const u8 *m, size_t n) -{ - __m128i v0, v1, v2, v3; - __m128i k0, k1; - __m128i mi, mask, len; - size_t i, k; - union { u64 gpr; __m128i xmm; } hash; - const u8 *p; - - /* We used to use the _mm_seti_epi32 intrinsic to initialize - SSE2 registers. This compiles to a movdqa instruction, - which requires 16-byte alignment. On 32-bit Windows, it - looks like ghc's runtime linker doesn't align ".rdata" - sections as requested, so we got segfaults for our trouble. - - Now we use an intrinsic that cares less about alignment - (_mm_loadu_si128, aka movdqu) instead, and all seems - happy. */ - - static const u32 const iv[6][4] = { - { 0x70736575, 0x736f6d65, 0, 0 }, - { 0x6e646f6d, 0x646f7261, 0, 0 }, - { 0x6e657261, 0x6c796765, 0, 0 }, - { 0x79746573, 0x74656462, 0, 0 }, - { -1, -1, 0, 0 }, - { 255, 0, 0, 0 }, - }; - - k0 = _mm_loadl_epi64((__m128i*)(&ik0)); - k1 = _mm_loadl_epi64((__m128i*)(&ik1)); - - v0 = _mm_xor_si128(k0, _mm_loadu_si128((__m128i*) &iv[0])); - v1 = _mm_xor_si128(k1, _mm_loadu_si128((__m128i*) &iv[1])); - v2 = _mm_xor_si128(k0, _mm_loadu_si128((__m128i*) &iv[2])); - v3 = _mm_xor_si128(k1, _mm_loadu_si128((__m128i*) &iv[3])); - -#define HALF_ROUND(a,b,c,d,s,t) \ - do \ - { \ - a = _mm_add_epi64(a, b); c = _mm_add_epi64(c, d); \ - b = _mm_roti_epi64(b, s); d = _mm_roti_epi64(d, t); \ - b = _mm_xor_si128(b, a); d = _mm_xor_si128(d, c); \ - } while(0) - -#define COMPRESS(v0,v1,v2,v3) \ - do \ - { \ - HALF_ROUND(v0,v1,v2,v3,13,16); \ - v0 = _mm_shufflelo_epi16(v0, _MM_SHUFFLE(1,0,3,2)); \ - HALF_ROUND(v2,v1,v0,v3,17,21); \ - v2 = _mm_shufflelo_epi16(v2, _MM_SHUFFLE(1,0,3,2)); \ - } while(0) - - for(i = 0; i < (n-n%8); i += 8) - { - mi = _mm_loadl_epi64((__m128i*)(m + i)); - v3 = _mm_xor_si128(v3, mi); - if (SIPHASH_ROUNDS == 2) { - COMPRESS(v0,v1,v2,v3); COMPRESS(v0,v1,v2,v3); - } else { - for (k = 0; k < SIPHASH_ROUNDS; ++k) - COMPRESS(v0,v1,v2,v3); - } - v0 = _mm_xor_si128(v0, mi); - } - - p = m + n; - - /* We must be careful to not trigger a segfault by reading an - unmapped page. So where is the end of our input? */ - - if (((uintptr_t) p & 4095) == 0) - /* Exactly at a page boundary: do not read past the end. */ - mi = _mm_setzero_si128(); - else if (((uintptr_t) p & 4095) <= 4088) - /* Inside a page: safe to read past the end, as we'll - mask out any bits we shouldn't have looked at below. */ - mi = _mm_loadl_epi64((__m128i*)(m + i)); - else - /* Within 8 bytes of the end of a page: ensure that - our final read re-reads some bytes so that we do - not cross the page boundary, then shift our result - right so that the re-read bytes vanish. */ - mi = _mm_srli_epi64(_mm_loadl_epi64((__m128i*)(((uintptr_t) m + i) & ~7)), - 8 * (((uintptr_t) m + i) % 8)); - - len = _mm_set_epi32(0, 0, (n&0xff) << 24, 0); - mask = _mm_srli_epi64(_mm_loadu_si128((__m128i*) &iv[4]), 8*(8-n%8)); - mi = _mm_xor_si128(_mm_and_si128(mi, mask), len); - - v3 = _mm_xor_si128(v3, mi); - if (SIPHASH_ROUNDS == 2) { - COMPRESS(v0,v1,v2,v3); COMPRESS(v0,v1,v2,v3); - } else { - for (k = 0; k < SIPHASH_ROUNDS; ++k) - COMPRESS(v0,v1,v2,v3); - } - v0 = _mm_xor_si128(v0, mi); - - v2 = _mm_xor_si128(v2, _mm_loadu_si128((__m128i*) &iv[5])); - if (SIPHASH_FINALROUNDS == 4) { - COMPRESS(v0,v1,v2,v3); COMPRESS(v0,v1,v2,v3); - COMPRESS(v0,v1,v2,v3); COMPRESS(v0,v1,v2,v3); - } else { - for (k = 0; k < SIPHASH_FINALROUNDS; ++k) - COMPRESS(v0,v1,v2,v3); - } - - v0 = _mm_xor_si128(_mm_xor_si128(v0, v1), _mm_xor_si128(v2, v3)); - hash.xmm = v0; - -#undef COMPRESS -#undef HALF_ROUND - //return _mm_extract_epi32(v0, 0) | (((u64)_mm_extract_epi32(v0, 1)) << 32); - return hash.gpr; -} diff --git a/benchmarks/cbits/siphash-sse41.c b/benchmarks/cbits/siphash-sse41.c deleted file mode 100644 --- a/benchmarks/cbits/siphash-sse41.c +++ /dev/null @@ -1,86 +0,0 @@ -/* - * The original code was developed by Samuel Neves, and has been - * only lightly modified. - * - * Used with permission. - */ -#pragma GCC target("sse4.1") - -#include -#include "siphash.h" - -// Specialized for siphash, do not reuse -#define rotate16(x) _mm_shufflehi_epi16((x), _MM_SHUFFLE(2,1,0,3)) - -#define _mm_roti_epi64(x, c) (((c) == 16) ? rotate16((x)) : _mm_xor_si128(_mm_slli_epi64((x), (c)), _mm_srli_epi64((x), 64-(c)))) -//#define _mm_roti_epi64(x, c) _mm_xor_si128(_mm_slli_epi64((x), (c)), _mm_srli_epi64((x), 64-(c))) - - -u64 hashable_siphash24_sse41(u64 _k0, u64 _k1, const unsigned char *m, size_t n) -{ - __m128i v0, v1, v02, v13; - __m128i k0; - __m128i mi, mask, len, h; - const __m128i zero = _mm_setzero_si128(); - size_t i, k; - union { u64 gpr; __m128i xmm; } hash; - unsigned char key[16]; - - ((u64 *)key)[0] = _k0; - ((u64 *)key)[1] = _k1; - - k0 = _mm_loadu_si128((__m128i*)(key + 0)); - - v0 = _mm_xor_si128(k0, _mm_set_epi32(0x646f7261, 0x6e646f6d, 0x736f6d65, 0x70736575)); - v1 = _mm_xor_si128(k0, _mm_set_epi32(0x74656462, 0x79746573, 0x6c796765, 0x6e657261)); - - v02 = _mm_unpacklo_epi64(v0, v1); - v13 = _mm_unpackhi_epi64(v0, v1); - -#define HALF_ROUND(a,b,s,t) \ -do \ -{ \ - __m128i b1,b2; \ - a = _mm_add_epi64(a, b); \ - b1 = _mm_roti_epi64(b, s); b2 = _mm_roti_epi64(b, t); b = _mm_blend_epi16(b1, b2, 0xF0); \ - b = _mm_xor_si128(b, a); \ -} while(0) - -#define COMPRESS(v02,v13) \ - do \ - { \ - HALF_ROUND(v02,v13,13,16); \ - v02 = _mm_shuffle_epi32(v02, _MM_SHUFFLE(0,1,3,2)); \ - HALF_ROUND(v02,v13,17,21); \ - v02 = _mm_shuffle_epi32(v02, _MM_SHUFFLE(0,1,3,2)); \ - } while(0) - - for(i = 0; i < (n-n%8); i += 8) - { - mi = _mm_loadl_epi64((__m128i*)(m + i)); - v13 = _mm_xor_si128(v13, _mm_unpacklo_epi64(zero, mi)); - for(k = 0; k < SIPHASH_ROUNDS; ++k) COMPRESS(v02,v13); - v02 = _mm_xor_si128(v02, mi); - } - - mi = _mm_loadl_epi64((__m128i*)(m + i)); - len = _mm_set_epi32(0, 0, (n&0xff) << 24, 0); - mask = _mm_srli_epi64(_mm_set_epi32(0, 0, 0xffffffff, 0xffffffff), 8*(8-n%8)); - mi = _mm_xor_si128(_mm_and_si128(mi, mask), len); - - v13 = _mm_xor_si128(v13, _mm_unpacklo_epi64(zero, mi)); - for(k = 0; k < SIPHASH_ROUNDS; ++k) COMPRESS(v02,v13); - v02 = _mm_xor_si128(v02, mi); - - v02 = _mm_xor_si128(v02, _mm_set_epi32(0, 0xff, 0, 0)); - for(k = 0; k < SIPHASH_FINALROUNDS; ++k) COMPRESS(v02,v13); - - v0 = _mm_xor_si128(v02, v13); - v0 = _mm_xor_si128(v0, _mm_castps_si128(_mm_movehl_ps(_mm_castsi128_ps(zero), _mm_castsi128_ps(v0)))); - hash.xmm = v0; - -#undef COMPRESS -#undef HALF_ROUND - //return _mm_extract_epi32(v0, 0) | (((u64)_mm_extract_epi32(v0, 1)) << 32); - return hash.gpr; -} diff --git a/benchmarks/cbits/siphash.c b/benchmarks/cbits/siphash.c deleted file mode 100644 --- a/benchmarks/cbits/siphash.c +++ /dev/null @@ -1,262 +0,0 @@ -/* Almost a verbatim copy of the reference implementation. */ - -#include -#include "siphash.h" - -#define ROTL(x,b) (u64)(((x) << (b)) | ((x) >> (64 - (b)))) - -#define SIPROUND \ - do { \ - v0 += v1; v1=ROTL(v1,13); v1 ^= v0; v0=ROTL(v0,32); \ - v2 += v3; v3=ROTL(v3,16); v3 ^= v2; \ - v0 += v3; v3=ROTL(v3,21); v3 ^= v0; \ - v2 += v1; v1=ROTL(v1,17); v1 ^= v2; v2=ROTL(v2,32); \ - } while(0) - -#if defined(__i386) -# define _siphash24 plain_siphash24 -#endif - -static inline u64 odd_read(const u8 *p, int count, u64 val, int shift) -{ - switch (count) { - case 7: val |= ((u64)p[6]) << (shift + 48); - case 6: val |= ((u64)p[5]) << (shift + 40); - case 5: val |= ((u64)p[4]) << (shift + 32); - case 4: val |= ((u64)p[3]) << (shift + 24); - case 3: val |= ((u64)p[2]) << (shift + 16); - case 2: val |= ((u64)p[1]) << (shift + 8); - case 1: val |= ((u64)p[0]) << shift; - } - return val; -} - -static inline u64 _siphash(int c, int d, u64 k0, u64 k1, - const u8 *str, size_t len) -{ - u64 v0 = 0x736f6d6570736575ull ^ k0; - u64 v1 = 0x646f72616e646f6dull ^ k1; - u64 v2 = 0x6c7967656e657261ull ^ k0; - u64 v3 = 0x7465646279746573ull ^ k1; - const u8 *end, *p; - u64 b; - int i; - - for (p = str, end = str + (len & ~7); p < end; p += 8) { - u64 m = peek_u64le((u64 *) p); - v3 ^= m; - if (c == 2) { - SIPROUND; - SIPROUND; - } else { - for (i = 0; i < c; i++) - SIPROUND; - } - v0 ^= m; - } - - b = odd_read(p, len & 7, ((u64) len) << 56, 0); - - v3 ^= b; - if (c == 2) { - SIPROUND; - SIPROUND; - } else { - for (i = 0; i < c; i++) - SIPROUND; - } - v0 ^= b; - - v2 ^= 0xff; - if (d == 4) { - SIPROUND; - SIPROUND; - SIPROUND; - SIPROUND; - } else { - for (i = 0; i < d; i++) - SIPROUND; - } - b = v0 ^ v1 ^ v2 ^ v3; - return b; -} - - -static inline u64 _siphash24(u64 k0, u64 k1, const u8 *str, size_t len) -{ - return _siphash(2, 4, k0, k1, str, len); -} - -#if defined(__i386) -# undef _siphash24 - -static u64 (*_siphash24)(u64 k0, u64 k1, const u8 *, size_t); - -static void maybe_use_sse() - __attribute__((constructor)); - -static void maybe_use_sse() -{ - uint32_t eax = 1, ebx, ecx, edx; - - __asm volatile - ("mov %%ebx, %%edi;" /* 32bit PIC: don't clobber ebx */ - "cpuid;" - "mov %%ebx, %%esi;" - "mov %%edi, %%ebx;" - :"+a" (eax), "=S" (ebx), "=c" (ecx), "=d" (edx) - : :"edi"); - -#if defined(HAVE_SSE2) - if (edx & (1 << 26)) - _siphash24 = hashable_siphash24_sse2; -#if defined(HAVE_SSE41) - else if (ecx & (1 << 19)) - _siphash24 = hashable_siphash24_sse41; -#endif - else -#endif - _siphash24 = plain_siphash24; -} - -#endif - -/* ghci's linker fails to call static initializers. */ -static inline void ensure_sse_init() -{ -#if defined(__i386) - if (_siphash24 == NULL) - maybe_use_sse(); -#endif -} - -u64 hashable_siphash(int c, int d, u64 k0, u64 k1, const u8 *str, size_t len) -{ - return _siphash(c, d, k0, k1, str, len); -} - -u64 hashable_siphash24(u64 k0, u64 k1, const u8 *str, size_t len) -{ - ensure_sse_init(); - return _siphash24(k0, k1, str, len); -} - -/* Used for ByteArray#s. We can't treat them like pointers in - native Haskell, but we can in unsafe FFI calls. - */ -u64 hashable_siphash24_offset(u64 k0, u64 k1, - const u8 *str, size_t off, size_t len) -{ - ensure_sse_init(); - return _siphash24(k0, k1, str + off, len); -} - -static int _siphash_chunk(int c, int d, int buffered, u64 v[5], - const u8 *str, size_t len, size_t totallen) -{ - u64 v0 = v[0], v1 = v[1], v2 = v[2], v3 = v[3], m, b; - const u8 *p, *end; - u64 carry = 0; - int i; - - if (buffered > 0) { - int unbuffered = 8 - buffered; - int tobuffer = unbuffered > len ? len : unbuffered; - int shift = buffered << 3; - - m = odd_read(str, tobuffer, v[4], shift); - str += tobuffer; - buffered += tobuffer; - len -= tobuffer; - - if (buffered < 8) - carry = m; - else { - v3 ^= m; - if (c == 2) { - SIPROUND; - SIPROUND; - } else { - for (i = 0; i < c; i++) - SIPROUND; - } - v0 ^= m; - buffered = 0; - m = 0; - } - } - - for (p = str, end = str + (len & ~7); p < end; p += 8) { - m = peek_u64le((u64 *) p); - v3 ^= m; - if (c == 2) { - SIPROUND; - SIPROUND; - } else { - for (i = 0; i < c; i++) - SIPROUND; - } - v0 ^= m; - } - - b = odd_read(p, len & 7, 0, 0); - - if (totallen == -1) { - v[0] = v0; - v[1] = v1; - v[2] = v2; - v[3] = v3; - v[4] = b | carry; - - return buffered + (len & 7); - } - - b |= ((u64) totallen) << 56; - - v3 ^= b; - if (c == 2) { - SIPROUND; - SIPROUND; - } else { - for (i = 0; i < c; i++) - SIPROUND; - } - v0 ^= b; - - v2 ^= 0xff; - if (d == 4) { - SIPROUND; - SIPROUND; - SIPROUND; - SIPROUND; - } else { - for (i = 0; i < d; i++) - SIPROUND; - } - v[4] = v0 ^ v1 ^ v2 ^ v3; - return 0; -} - -void hashable_siphash_init(u64 k0, u64 k1, u64 *v) -{ - v[0] = 0x736f6d6570736575ull ^ k0; - v[1] = 0x646f72616e646f6dull ^ k1; - v[2] = 0x6c7967656e657261ull ^ k0; - v[3] = 0x7465646279746573ull ^ k1; - v[4] = 0; -} - -int hashable_siphash24_chunk(int buffered, u64 v[5], const u8 *str, - size_t len, size_t totallen) -{ - return _siphash_chunk(2, 4, buffered, v, str, len, totallen); -} - -/* - * Used for ByteArray#. - */ -int hashable_siphash24_chunk_offset(int buffered, u64 v[5], const u8 *str, - size_t off, size_t len, size_t totallen) -{ - return _siphash_chunk(2, 4, buffered, v, str + off, len, totallen); -} diff --git a/benchmarks/cbits/siphash.h b/benchmarks/cbits/siphash.h deleted file mode 100644 --- a/benchmarks/cbits/siphash.h +++ /dev/null @@ -1,68 +0,0 @@ -#ifndef _hashable_siphash_h -#define _hashable_siphash_h - -#include -#include - -typedef uint64_t u64; -typedef uint32_t u32; -typedef uint16_t u16; -typedef uint8_t u8; - -#define SIPHASH_ROUNDS 2 -#define SIPHASH_FINALROUNDS 4 - -u64 hashable_siphash(int, int, u64, u64, const u8 *, size_t); -u64 hashable_siphash24(u64, u64, const u8 *, size_t); - -#if defined(__i386) - -/* To use SSE instructions, we have to adjust the stack from its - default of 4-byte alignment to use 16-byte alignment. */ - -# define ALIGNED_STACK __attribute__((force_align_arg_pointer)) - -u64 hashable_siphash24_sse2(u64, u64, const u8 *, size_t) ALIGNED_STACK; -u64 hashable_siphash24_sse41(u64, u64, const u8 *, size_t) ALIGNED_STACK; -#endif - -#if defined(_WIN32) -# define __LITTLE_ENDIAN 1234 -# define __BIG_ENDIAN 4321 -# define __BYTE_ORDER __LITTLE_ENDIAN - -#elif (defined(__FreeBSD__) && __FreeBSD_version >= 470000) || defined(__OpenBSD__) || defined(__NetBSD__) -# include -# define __BIG_ENDIAN BIG_ENDIAN -# define __LITTLE_ENDIAN LITTLE_ENDIAN -# define __BYTE_ORDER BYTE_ORDER - -#elif (defined(BSD) && (BSD >= 199103)) || defined(__APPLE__) -# include -# define __BIG_ENDIAN BIG_ENDIAN -# define __LITTLE_ENDIAN LITTLE_ENDIAN -# define __BYTE_ORDER BYTE_ORDER - -#elif defined(__linux__) -# include -#endif - -static inline u64 peek_u64le(const u64 *p) -{ - u64 x = *p; - -#if __BYTE_ORDER == __BIG_ENDIAN - x = ((x & 0xff00000000000000ull) >> 56) | - ((x & 0x00ff000000000000ull) >> 40) | - ((x & 0x0000ff0000000000ull) >> 24) | - ((x & 0x000000ff00000000ull) >> 8) | - ((x & 0x00000000ff000000ull) << 8) | - ((x & 0x0000000000ff0000ull) << 24) | - ((x & 0x000000000000ff00ull) << 40) | - ((x & 0x00000000000000ffull) << 56); -#endif - - return x; -} - -#endif /* _hashable_siphash_h */ diff --git a/benchmarks/cbits/wang.c b/benchmarks/cbits/wang.c deleted file mode 100644 --- a/benchmarks/cbits/wang.c +++ /dev/null @@ -1,29 +0,0 @@ -/* - * These hash functions were developed by Thomas Wang. - * - * http://www.concentric.net/~ttwang/tech/inthash.htm - */ - -#include - -uint32_t hashable_wang_32(uint32_t a) -{ - a = (a ^ 61) ^ (a >> 16); - a = a + (a << 3); - a = a ^ (a >> 4); - a = a * 0x27d4eb2d; - a = a ^ (a >> 15); - return a; -} - -uint64_t hashable_wang_64(uint64_t key) -{ - key = (~key) + (key << 21); // key = (key << 21) - key - 1; - key = key ^ ((key >> 24) | (key << 40)); - key = (key + (key << 3)) + (key << 8); // key * 265 - key = key ^ ((key >> 14) | (key << 50)); - key = (key + (key << 2)) + (key << 4); // key * 21 - key = key ^ ((key >> 28) | (key << 36)); - key = key + (key << 31); - return key; -} diff --git a/cbits-unix/init.c b/cbits-unix/init.c new file mode 100644 --- /dev/null +++ b/cbits-unix/init.c @@ -0,0 +1,39 @@ +#include +#include +#include +#include +#include +#include + +uint64_t hs_hashable_init() { + + /* if there is /dev/urandom, read from it */ + FILE *urandom = fopen("/dev/urandom", "r"); + if (urandom) { + uint64_t result = 0; + size_t r = fread(&result, sizeof(uint64_t), 1, urandom); + fclose(urandom); + + if (r == 1) { + return result; + } else { + return 0xfeed1000; + } + + } else { + /* time of day */ + struct timeval tp = {0, 0}; + gettimeofday(&tp, NULL); + + /* cputime */ + clock_t c = clock(); + + /* process id */ + pid_t p = getpid(); + + return ((uint64_t) tp.tv_sec) + ^ ((uint64_t) tp.tv_usec) + ^ ((uint64_t) c << 16) + ^ ((uint64_t) p << 32); + } +} diff --git a/cbits-win/init.c b/cbits-win/init.c new file mode 100644 --- /dev/null +++ b/cbits-win/init.c @@ -0,0 +1,28 @@ +#include + +#include + +uint64_t hs_hashable_init() { + /* Handy list at https://stackoverflow.com/a/3487338/1308058 */ + + uint64_t a = GetCurrentProcessId(); /* DWORD */ + uint64_t b = GetCurrentThreadId(); /* DWORD */ + uint64_t c = GetTickCount(); /* DWORD */ + + SYSTEMTIME t = {0,0,0,0,0,0,0,0}; + GetSystemTime(&t); + + LARGE_INTEGER i; + QueryPerformanceCounter(&i); + + return a ^ (b << 32) ^ (c << 16) + ^ ((uint64_t) t.wYear << 56) + ^ ((uint64_t) t.wMonth << 48) + ^ ((uint64_t) t.wDayOfWeek << 40) + ^ ((uint64_t) t.wDay << 32) + ^ ((uint64_t) t.wHour << 24) + ^ ((uint64_t) t.wMinute << 16) + ^ ((uint64_t) t.wSecond << 8) + ^ ((uint64_t) t.wMilliseconds << 0) + ^ ((uint64_t) i.QuadPart); +} diff --git a/cbits/fnv.c b/cbits/fnv.c deleted file mode 100644 --- a/cbits/fnv.c +++ /dev/null @@ -1,53 +0,0 @@ -/* -Copyright Johan Tibell 2011 - -All rights reserved. - -Redistribution and use in source and binary forms, with or without -modification, are permitted provided that the following conditions are met: - - * Redistributions of source code must retain the above copyright - notice, this list of conditions and the following disclaimer. - - * Redistributions in binary form must reproduce the above - copyright notice, this list of conditions and the following - disclaimer in the documentation and/or other materials provided - with the distribution. - - * Neither the name of Johan Tibell nor the names of other - contributors may be used to endorse or promote products derived - from this software without specific prior written permission. - -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -*/ - -/* FNV-1 hash - * - * The FNV-1 hash description: http://isthe.com/chongo/tech/comp/fnv/ - * The FNV-1 hash is public domain: http://isthe.com/chongo/tech/comp/fnv/#public_domain - */ -long hashable_fnv_hash(const unsigned char* str, long len, long hash) { - - while (len--) { - hash = (hash * 16777619) ^ *str++; - } - - return hash; -} - -/* Used for ByteArray#s. We can't treat them like pointers in - native Haskell, but we can in unsafe FFI calls. - */ -long hashable_fnv_hash_offset(const unsigned char* str, long offset, long len, long hash) { - return hashable_fnv_hash(str + offset, len, hash); -} diff --git a/cbits/getRandomBytes.c b/cbits/getRandomBytes.c deleted file mode 100644 --- a/cbits/getRandomBytes.c +++ /dev/null @@ -1,93 +0,0 @@ -/* -Copyright Bryan O'Sullivan 2012 - -All rights reserved. - -Redistribution and use in source and binary forms, with or without -modification, are permitted provided that the following conditions are met: - - * Redistributions of source code must retain the above copyright - notice, this list of conditions and the following disclaimer. - - * Redistributions in binary form must reproduce the above - copyright notice, this list of conditions and the following - disclaimer in the documentation and/or other materials provided - with the distribution. - - * Neither the name of Johan Tibell nor the names of other - contributors may be used to endorse or promote products derived - from this software without specific prior written permission. - -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -*/ - -#include "MachDeps.h" - -int hashable_getRandomBytes(unsigned char *dest, int nbytes); - -#if defined(mingw32_HOST_OS) || defined(__MINGW32__) - -#include -#include - -int hashable_getRandomBytes(unsigned char *dest, int nbytes) -{ - HCRYPTPROV hCryptProv; - int ret; - - if (!CryptAcquireContextA(&hCryptProv, NULL, NULL, PROV_RSA_FULL, - CRYPT_VERIFYCONTEXT)) - return -1; - - ret = CryptGenRandom(hCryptProv, (DWORD) nbytes, (BYTE *) dest) ? nbytes : -1; - - CryptReleaseContext(hCryptProv, 0); - - bail: - return ret; -} - -#else - -#include -#include -#include - -/* Assumptions: /dev/urandom exists and does something sane, and does - not block. */ - -int hashable_getRandomBytes(unsigned char *dest, int nbytes) -{ - ssize_t off, nread; - int fd; - - fd = open("/dev/urandom", O_RDONLY); - if (fd == -1) - return -1; - - for (off = 0; nbytes > 0; nbytes -= nread) { - nread = read(fd, dest + off, nbytes); - off += nread; - if (nread == -1) { - off = -1; - break; - } - } - - bail: - close(fd); - - return off; -} - -#endif diff --git a/examples/Main.hs b/examples/Main.hs --- a/examples/Main.hs +++ b/examples/Main.hs @@ -6,12 +6,12 @@ data Foo = Foo1 Int Char Bool | Foo2 String () - deriving (Generic) + deriving (Eq, Generic) instance Hashable Foo data Bar = Bar Double Float - deriving (Generic) + deriving (Eq, Generic) instance Hashable Bar @@ -31,6 +31,8 @@ -- Higher Rank Hashable Examples -- ----------------------------------- +{- TODO: + newtype WriterT w m a = WriterT { runWriterT :: m (a, w) } data Free f a = Pure a | Free (f (Free f a)) @@ -47,4 +49,4 @@ hashWithSalt = hashWithSalt1 instance (Hashable1 f, Hashable a) => Hashable (Free f a) where hashWithSalt = hashWithSalt1 - +-} diff --git a/hashable.cabal b/hashable.cabal --- a/hashable.cabal +++ b/hashable.cabal @@ -1,171 +1,212 @@ -Name: hashable -Version: 1.2.6.1 -Synopsis: A class for types that can be converted to a hash value -Description: This package defines a class, 'Hashable', for types that - can be converted to a hash value. This class - exists for the benefit of hashing-based data - structures. The package provides instances for - basic types and a way to combine hash values. -Homepage: http://github.com/tibbe/hashable -License: BSD3 -License-file: LICENSE -Author: Milan Straka - Johan Tibell -Maintainer: johan.tibell@gmail.com -bug-reports: https://github.com/tibbe/hashable/issues -Stability: Provisional -Category: Data -Build-type: Simple -Cabal-version: >=1.8 --- tests/Properties.hs shouldn't have to go here, but the source files --- for the test-suite stanzas don't get picked up by `cabal sdist`. -Extra-source-files: - CHANGES.md, README.md, tests/Properties.hs, - benchmarks/Benchmarks.hs, benchmarks/cbits/*.c, benchmarks/cbits/*.h +cabal-version: 2.2 +name: hashable +version: 1.5.1.0 +synopsis: A class for types that can be converted to a hash value +description: + This package defines a class, 'Hashable', for types that can be converted to a hash value. + This class exists for the benefit of hashing-based data structures. + The package provides instances for basic types and a way to combine hash values. + . + 'Hashable' is intended exclusively for use in in-memory data structures. + . + 'Hashable' does /not/ have a fixed standard. + This allows it to improve over time. + . + Because it does not have a fixed standard, different computers or computers on different versions of the code will observe different hash values. + As such, 'hashable' is not recommended for use other than in-memory datastructures. + Specifically, 'hashable' is not intended for network use or in applications which persist hashed values. + For stable hashing use named hashes: sha256, crc32, xxhash etc. -Flag integer-gmp - Description: Are we using integer-gmp to provide fast Integer instances? - Default: True +homepage: http://github.com/haskell-unordered-containers/hashable +license: BSD-3-Clause +license-file: LICENSE +author: + Milan Straka + Johan Tibell -Flag sse2 - Description: Do we want to assume that a target supports SSE 2? - Default: True - Manual: True +maintainer: Oleg Grenrus +bug-reports: + https://github.com/haskell-unordered-containers/hashable/issues -Flag sse41 - Description: Do we want to assume that a target supports SSE 4.1? - Default: False - Manual: True +stability: Provisional +category: Data +build-type: Simple +tested-with: + GHC ==9.6.5 || ==9.8.2 || ==9.8.3 || ==9.10.2 || ==9.12.2 || ==9.14.1 -Flag examples - Description: Build example modules - Default: False - Manual: True +extra-source-files: + CHANGES.md + include/HsHashable.h + include/HsXXHash.h + README.md + xxHash-0.8.3/xxhash.h -Library - Exposed-modules: Data.Hashable - Data.Hashable.Lifted - Other-modules: Data.Hashable.Class - Build-depends: base >= 4.4 && < 4.11, - bytestring >= 0.9 && < 0.11, - deepseq >= 1.3 - if impl(ghc) - Build-depends: ghc-prim, - text >= 0.11.0.5 - if impl(ghc) && flag(integer-gmp) - Build-depends: integer-gmp >= 0.2 +flag arch-native + description: + Use @-march=native@ when compiling C sources. + Portable implementation is 15-50% slower. + Consider enabling this flag if hashing performance is important. - if impl(ghc >= 7.2.1) - CPP-Options: -DGENERICS - Other-modules: Data.Hashable.Generic + manual: True + default: False - C-sources: - cbits/fnv.c +flag random-initial-seed + description: + Randomly initialize the initial seed on each final executable invocation + This is useful for catching cases when you rely on (non-existent) + stability of hashable's hash functions. + This is not a security feature. - Ghc-options: -Wall - if impl(ghc >= 6.8) - Ghc-options: -fwarn-tabs - else - c-sources: cbits/getRandomBytes.c - other-modules: Data.Hashable.RandomSource + manual: True + default: False + +library + exposed-modules: + Data.Hashable + Data.Hashable.Generic + Data.Hashable.Lifted + + other-modules: + Data.Hashable.Class + Data.Hashable.FFI + Data.Hashable.Generic.Instances + Data.Hashable.Imports + Data.Hashable.LowLevel + Data.Hashable.Mix + Data.Hashable.XXH3 + + include-dirs: include xxHash-0.8.3 + includes: + HsHashable.h + HsXXHash.h + xxhash.h + + hs-source-dirs: src + build-depends: + , base >=4.18.0.0 && <4.23 + , bytestring >=0.11.5.3 && <0.13 + , containers >=0.6.7 && <0.9 + , deepseq >=1.4.8.1 && <1.6 + , text >=2.0.2 && <2.2 + + if impl(ghc <9.8) + build-depends: ghc-prim + + -- depend on os-string on newer GHCs only. + -- os-string has tight lower bound on bytestring, which prevents + -- using bundled version on older GHCs. + build-depends: os-string >=2.0.2 && <2.1 + + -- we also ensure that we can get filepath-1.5 only with GHC-9.2 + -- therefore there is else-branch with stricter upper bound. + build-depends: filepath >=1.4.200.1 && <1.6 + + if (flag(random-initial-seed) && impl(ghc)) + cpp-options: -DHASHABLE_RANDOM_SEED=1 + if os(windows) - extra-libraries: advapi32 + c-sources: cbits-win/init.c -Test-suite tests - Type: exitcode-stdio-1.0 - Hs-source-dirs: tests - Main-is: Main.hs - Other-modules: Properties Regress - Build-depends: base >= 4.0 && < 5.0, - bytestring, - ghc-prim, - hashable, - test-framework >= 0.3.3, - test-framework-hunit, - test-framework-quickcheck2 >= 0.2.9, - HUnit, - QuickCheck >= 2.4.0.1, - random >= 1.0 && < 1.2, - text >= 0.11.0.5 - if !os(windows) - Build-depends: unix - CPP-options: -DHAVE_MMAP - Other-modules: Regress.Mmap + else + c-sources: cbits-unix/init.c - Ghc-options: -Wall -fno-warn-orphans - if impl(ghc >= 7.2.1) - CPP-Options: -DGENERICS + default-language: Haskell2010 + other-extensions: + BangPatterns + CPP + DeriveDataTypeable + FlexibleContexts + FlexibleInstances + GADTs + KindSignatures + MagicHash + MultiParamTypeClasses + QuantifiedConstraints + ScopedTypeVariables + Trustworthy + TypeOperators + UnliftedFFITypes -benchmark benchmarks - -- We cannot depend on the hashable library directly as that creates - -- a dependency cycle. - hs-source-dirs: . benchmarks + ghc-options: -Wall - main-is: Benchmarks.hs + if flag(arch-native) + -- Cabal doesn't pass cc-options to "ordinary" Haskell source compilation + -- https://github.com/haskell/cabal/issues/9801 + ghc-options: -optc=-march=native -optc-mtune=native + + if impl(ghc >=9.0) + -- these flags may abort compilation with GHC-8.10 + -- https://gitlab.haskell.org/ghc/ghc/-/merge_requests/3295 + ghc-options: -Winferred-safe-imports -Wmissing-safe-haskell-mode + +test-suite hashable-tests + type: exitcode-stdio-1.0 + hs-source-dirs: tests + main-is: Main.hs other-modules: - Data.Hashable - Data.Hashable.Class - Data.Hashable.RandomSource - Data.Hashable.SipHash - type: exitcode-stdio-1.0 + Properties + Regress build-depends: - base, - bytestring, - criterion >= 1.0, - ghc-prim, - siphash, - text - - if impl(ghc) - Build-depends: ghc-prim, - text >= 0.11.0.5 - if impl(ghc) && flag(integer-gmp) - Build-depends: integer-gmp >= 0.2 + , base + , bytestring + , filepath + , ghc-prim + , hashable + , QuickCheck >=2.15 + , tasty ^>=1.5 + , tasty-hunit ^>=0.10.1 + , tasty-quickcheck ^>=0.10.3 || ^>=0.11 + , text >=0.11.0.5 - if impl(ghc >= 7.2.1) - CPP-Options: -DGENERICS + if impl(ghc >=9.2) + build-depends: os-string - include-dirs: - benchmarks/cbits + if !os(windows) + cpp-options: -DHAVE_MMAP + other-modules: Regress.Mmap + other-extensions: CApiFFI - includes: - siphash.h + ghc-options: -Wall -fno-warn-orphans + default-language: Haskell2010 - c-sources: - benchmarks/cbits/inthash.c - benchmarks/cbits/siphash.c - benchmarks/cbits/wang.c - cbits/fnv.c +test-suite xxhash-tests + type: exitcode-stdio-1.0 + hs-source-dirs: tests src + main-is: xxhash-tests.hs + other-modules: + Data.Hashable.FFI + Data.Hashable.XXH3 - if (arch(i386) || arch(x86_64)) && flag(sse2) - cpp-options: -DHAVE_SSE2 - c-sources: - benchmarks/cbits/siphash-sse2.c + default-language: Haskell2010 + build-depends: + , base <5 + , bytestring + , primitive ^>=0.9.0.0 + , tasty ^>=1.5 + , tasty-hunit ^>=0.10.1 + , tasty-quickcheck ^>=0.10.3 || ^>=0.11 - if flag(sse41) - cpp-options: -DHAVE_SSE41 - c-sources: - benchmarks/cbits/siphash-sse41.c + include-dirs: include xxHash-0.8.3 + includes: + HsXXHash.h + xxhash.h - Ghc-options: -Wall -O2 - if impl(ghc >= 6.8) - Ghc-options: -fwarn-tabs - else - c-sources: cbits/getRandomBytes.c - other-modules: Data.Hashable.RandomSource - if os(windows) - extra-libraries: advapi32 + if !impl(ghc >=9.4) + build-depends: data-array-byte >=0.1.0.1 && <0.2 +test-suite hashable-examples + type: exitcode-stdio-1.0 + build-depends: + , base + , ghc-prim + , hashable -Executable hashable-examples - if flag(examples) - build-depends: base, hashable - else - buildable: False - hs-source-dirs: examples - main-is: Main.hs + hs-source-dirs: examples + main-is: Main.hs + default-language: Haskell2010 source-repository head type: git - location: https://github.com/tibbe/hashable.git + location: + https://github.com/haskell-unordered-containers/hashable.git diff --git a/include/HsHashable.h b/include/HsHashable.h new file mode 100644 --- /dev/null +++ b/include/HsHashable.h @@ -0,0 +1,8 @@ +#ifndef HS_HASHABLE_H +#define HS_HASHABLE_H + +#include + +uint64_t hs_hashable_init(); + +#endif diff --git a/include/HsXXHash.h b/include/HsXXHash.h new file mode 100644 --- /dev/null +++ b/include/HsXXHash.h @@ -0,0 +1,35 @@ +#ifndef HS_XXHASH_H +#define HS_XXHASH_H + +#include + +#define XXH_INLINE_ALL +#include "xxhash.h" + +#define hs_XXH3_sizeof_state_s sizeof(struct XXH3_state_s) + +static inline uint64_t hs_XXH3_64bits_withSeed_offset(const uint8_t *ptr, size_t off, size_t len, uint64_t seed) { + return XXH3_64bits_withSeed(ptr + off, len, seed); +} + +static inline uint64_t hs_XXH3_64bits_withSeed_u64(uint64_t val, uint64_t seed) { + return XXH3_64bits_withSeed(&val, sizeof(val), seed); +} + +static inline uint64_t hs_XXH3_64bits_withSeed_u32(uint32_t val, uint64_t seed) { + return XXH3_64bits_withSeed(&val, sizeof(val), seed); +} + +static inline void hs_XXH3_64bits_update_offset(XXH3_state_t *statePtr, const uint8_t *ptr, size_t off, size_t len) { + XXH3_64bits_update(statePtr, ptr + off, len); +} + +static inline void hs_XXH3_64bits_update_u64(XXH3_state_t *statePtr, uint64_t val) { + XXH3_64bits_update(statePtr, &val, sizeof(val)); +} + +static inline void hs_XXH3_64bits_update_u32(XXH3_state_t *statePtr, uint32_t val) { + XXH3_64bits_update(statePtr, &val, sizeof(val)); +} + +#endif /* HS_XXHASH_H */ diff --git a/src/Data/Hashable.hs b/src/Data/Hashable.hs new file mode 100644 --- /dev/null +++ b/src/Data/Hashable.hs @@ -0,0 +1,215 @@ +{-# LANGUAGE DeriveDataTypeable #-} +{-# LANGUAGE Safe #-} + +------------------------------------------------------------------------ +-- | +-- Module : Data.Hashable +-- Copyright : (c) Milan Straka 2010 +-- (c) Johan Tibell 2011 +-- (c) Bryan O'Sullivan 2011, 2012 +-- SPDX-License-Identifier : BSD-3-Clause +-- Maintainer : johan.tibell@gmail.com +-- Stability : provisional +-- Portability : portable +-- +-- This module defines a class, 'Hashable', for types that can be +-- converted to a hash value. This class exists for the benefit of +-- hashing-based data structures. The module provides instances for +-- most standard types. Efficient instances for other types can be +-- generated automatically and effortlessly using the generics support +-- in GHC 7.4 and above. +-- +-- The easiest way to get started is to use the 'hash' function. Here +-- is an example session with @ghci@. +-- +-- > ghci> import Data.Hashable +-- > ghci> hash "foo" +-- > 60853164 + +module Data.Hashable + ( + -- * Hashing and security + -- $security + + -- * Computing hash values + Hashable(..) + + -- * Creating new instances + -- | There are two ways to create new instances: by deriving + -- instances automatically using GHC's generic programming + -- support or by writing instances manually. + + -- ** Generic instances + -- $generics + + -- *** Understanding a compiler error + -- $generic_err + + -- ** Writing instances by hand + -- $blocks + + -- *** Hashing contructors with multiple fields + -- $multiple-fields + + -- *** Hashing types with multiple constructors + -- $multiple-ctors + + , hashUsing + , hashPtr + , hashPtrWithSalt + , hashByteArray + , hashByteArrayWithSalt + + , defaultHashWithSalt + , defaultHash + + -- * Caching hashes + , Hashed + , hashed + , hashedHash + , unhashed + , mapHashed + , traverseHashed + ) where + +import Data.Hashable.Class +import Data.Hashable.Generic () + +-- $security +-- #security# +-- +-- Applications that use hash-based data structures to store input +-- from untrusted users can be susceptible to \"hash DoS\", a class of +-- denial-of-service attack that uses deliberately chosen colliding +-- inputs to force an application into unexpectedly behaving with +-- quadratic time complexity. +-- +-- At this time, the string hashing functions used in this library are +-- susceptible to such attacks and users are recommended to either use +-- a 'Data.Map' to store keys derived from untrusted input or to use a +-- hash function (e.g. SipHash) that's resistant to such attacks. A +-- future version of this library might ship with such hash functions. + +-- $generics +-- +-- The recommended way to make instances of +-- 'Hashable' for most types is to use the compiler's support for +-- automatically generating default instances using "GHC.Generics". +-- +-- > {-# LANGUAGE DeriveGeneric #-} +-- > +-- > import GHC.Generics (Generic) +-- > import Data.Hashable +-- > +-- > data Foo a = Foo a String +-- > deriving (Eq, Generic) +-- > +-- > instance Hashable a => Hashable (Foo a) +-- > +-- > data Colour = Red | Green | Blue +-- > deriving Generic +-- > +-- > instance Hashable Colour +-- +-- If you omit a body for the instance declaration, GHC will generate +-- a default instance that correctly and efficiently hashes every +-- constructor and parameter. +-- +-- The default implementations are provided by +-- 'genericHashWithSalt' and 'genericLiftHashWithSalt'; those together with +-- the generic type class 'GHashable' and auxiliary functions are exported +-- from the "Data.Hashable.Generic" module. + +-- $generic_err +-- +-- Suppose you intend to use the generic machinery to automatically +-- generate a 'Hashable' instance. +-- +-- > data Oops = Oops +-- > -- forgot to add "deriving Generic" here! +-- > +-- > instance Hashable Oops +-- +-- And imagine that, as in the example above, you forget to add a +-- \"@deriving 'Generic'@\" clause to your data type. At compile time, +-- you will get an error message from GHC that begins roughly as +-- follows: +-- +-- > No instance for (GHashable (Rep Oops)) +-- +-- This error can be confusing, as 'GHashable' is not exported (it is +-- an internal typeclass used by this library's generics machinery). +-- The correct fix is simply to add the missing \"@deriving +-- 'Generic'@\". + +-- $blocks +-- +-- To maintain high quality hashes, new 'Hashable' instances should be +-- built using existing 'Hashable' instances, combinators, and hash +-- functions. +-- +-- The functions below can be used when creating new instances of +-- 'Hashable'. For example, for many string-like types the +-- 'hashWithSalt' method can be defined in terms of either +-- 'hashPtrWithSalt' or 'hashByteArrayWithSalt'. Here's how you could +-- implement an instance for the 'B.ByteString' data type, from the +-- @bytestring@ package: +-- +-- > import qualified Data.ByteString as B +-- > import qualified Data.ByteString.Internal as B +-- > import qualified Data.ByteString.Unsafe as B +-- > import Data.Hashable +-- > import Foreign.Ptr (castPtr) +-- > +-- > instance Hashable B.ByteString where +-- > hashWithSalt salt bs = B.inlinePerformIO $ +-- > B.unsafeUseAsCStringLen bs $ \(p, len) -> +-- > hashPtrWithSalt p (fromIntegral len) salt + +-- $multiple-fields +-- +-- Hash constructors with multiple fields by chaining 'hashWithSalt': +-- +-- > data Date = Date Int Int Int +-- > +-- > instance Hashable Date where +-- > hashWithSalt s (Date yr mo dy) = +-- > s `hashWithSalt` +-- > yr `hashWithSalt` +-- > mo `hashWithSalt` dy +-- +-- If you need to chain hashes together, use 'hashWithSalt' and follow +-- this recipe: +-- +-- > combineTwo h1 h2 = h1 `hashWithSalt` h2 + +-- $multiple-ctors +-- +-- For a type with several value constructors, there are a few +-- possible approaches to writing a 'Hashable' instance. +-- +-- If the type is an instance of 'Enum', the easiest path is to +-- convert it to an 'Int', and use the existing 'Hashable' instance +-- for 'Int'. +-- +-- > data Color = Red | Green | Blue +-- > deriving Enum +-- > +-- > instance Hashable Color where +-- > hashWithSalt = hashUsing fromEnum +-- +-- If the type's constructors accept parameters, it is important to +-- distinguish the constructors. To distinguish the constructors, add +-- a different integer to the hash computation of each constructor: +-- +-- > data Time = Days Int +-- > | Weeks Int +-- > | Months Int +-- > +-- > instance Hashable Time where +-- > hashWithSalt s (Days n) = s `hashWithSalt` +-- > (0::Int) `hashWithSalt` n +-- > hashWithSalt s (Weeks n) = s `hashWithSalt` +-- > (1::Int) `hashWithSalt` n +-- > hashWithSalt s (Months n) = s `hashWithSalt` +-- > (2::Int) `hashWithSalt` n diff --git a/src/Data/Hashable/Class.hs b/src/Data/Hashable/Class.hs new file mode 100644 --- /dev/null +++ b/src/Data/Hashable/Class.hs @@ -0,0 +1,1002 @@ +{-# LANGUAGE BangPatterns #-} +{-# LANGUAGE CApiFFI #-} +{-# LANGUAGE CPP #-} +{-# LANGUAGE DefaultSignatures #-} +{-# LANGUAGE DerivingStrategies #-} +{-# LANGUAGE FlexibleContexts #-} +{-# LANGUAGE GeneralizedNewtypeDeriving #-} +{-# LANGUAGE MagicHash #-} +{-# LANGUAGE MultiParamTypeClasses #-} +{-# LANGUAGE PackageImports #-} +{-# LANGUAGE PolyKinds #-} +{-# LANGUAGE QuantifiedConstraints #-} +{-# LANGUAGE ScopedTypeVariables #-} +{-# LANGUAGE StandaloneDeriving #-} +{-# LANGUAGE Trustworthy #-} +{-# LANGUAGE TypeFamilies #-} +{-# LANGUAGE UnliftedFFITypes #-} +{-# OPTIONS_GHC -fno-warn-deprecations #-} + +------------------------------------------------------------------------ +-- | +-- Module : Data.Hashable.Class +-- Copyright : (c) Milan Straka 2010 +-- (c) Johan Tibell 2011 +-- (c) Bryan O'Sullivan 2011, 2012 +-- SPDX-License-Identifier : BSD-3-Clause +-- Maintainer : johan.tibell@gmail.com +-- Stability : provisional +-- Portability : portable +-- +-- This module defines a class, 'Hashable', for types that can be +-- converted to a hash value. This class exists for the benefit of +-- hashing-based data structures. The module provides instances for +-- most standard types. + +module Data.Hashable.Class + ( + -- * Computing hash values + Hashable(..) + , Hashable1(..) + , Hashable2(..) + + -- ** Support for generics + , genericHashWithSalt + , genericLiftHashWithSalt + , GHashable(..) + , HashArgs(..) + , Zero + , One + + -- * Creating new instances + , hashUsing + , hashPtr + , hashPtrWithSalt + , hashByteArray + , hashByteArrayWithSalt + , defaultHashWithSalt + , defaultHash + -- * Higher Rank Functions + , hashWithSalt1 + , hashWithSalt2 + , defaultLiftHashWithSalt + -- * Caching hashes + , Hashed + , hashed + , hashedHash + , unhashed + , mapHashed + , traverseHashed + ) where + +import Control.Applicative (Const (..)) +import Control.DeepSeq (NFData (rnf)) +import Control.Exception (assert) +import Control.Monad.ST (runST) +import Data.Array.Byte (ByteArray (..)) +import Data.Complex (Complex (..)) +import Data.Fixed (Fixed (..)) +import Data.Functor.Classes (Eq1 (..), Eq2 (..), Ord1 (..), Show1 (..)) +import Data.Functor.Compose (Compose (..)) +import Data.Functor.Identity (Identity (..)) +import Data.Int (Int16, Int8) +import Data.Kind (Type) +import Data.List (foldl') +import Data.Proxy (Proxy) +import Data.Ratio (Ratio, denominator, numerator) +import Data.String (IsString (..)) +import Data.Tuple (Solo (..)) +import Data.Unique (Unique, hashUnique) +import Data.Version (Version (..)) +import Data.Void (Void, absurd) +import Data.Word (Word16, Word8) +import Foreign.Ptr (FunPtr, IntPtr, Ptr, WordPtr, castFunPtrToPtr, ptrToIntPtr) +import Foreign.Storable (alignment, sizeOf) +import GHC.Base (ByteArray#) +import GHC.Conc (ThreadId (..)) +import GHC.Fingerprint.Type (Fingerprint (..)) +import GHC.Word (Word (..)) +import System.Mem.StableName (StableName, hashStableName) +import Type.Reflection (SomeTypeRep (..), TypeRep) +import Type.Reflection.Unsafe (typeRepFingerprint) + +import qualified Data.Array.Byte as AB +import qualified Data.ByteString as B +import qualified Data.ByteString.Lazy as BL +import qualified Data.ByteString.Short.Internal as BSI +import qualified Data.Functor.Product as FP +import qualified Data.Functor.Sum as FS +import qualified Data.IntMap as IntMap +import qualified Data.IntSet as IntSet +import qualified Data.List.NonEmpty as NE +import qualified Data.Map as Map +import qualified Data.Semigroup as Semi +import qualified Data.Sequence as Seq +import qualified Data.Set as Set +import qualified Data.Text as T +import qualified Data.Text.Array as TA +import qualified Data.Text.Internal as T +import qualified Data.Text.Lazy as TL +import qualified Data.Tree as Tree + +import GHC.Generics + +#if MIN_VERSION_base(4,19,0) +import GHC.Conc.Sync (fromThreadId) +#else +import GHC.Prim (ThreadId#) +#if __GLASGOW_HASKELL__ >= 904 +import Foreign.C.Types (CULLong (..)) +#elif __GLASGOW_HASKELL__ >= 900 +import Foreign.C.Types (CLong (..)) +#else +import Foreign.C.Types (CInt (..)) +#endif +#endif + +import GHC.Exts (Int (..), sizeofByteArray#) +import GHC.Num.BigNat (BigNat (..)) +import GHC.Num.Integer (Integer (..)) +import GHC.Num.Natural (Natural (..)) + + +import GHC.Float (castDoubleToWord64, castFloatToWord32) + +-- filepath >=1.4.100 && <1.5 has System.OsString.Internal.Types module +#if MIN_VERSION_filepath(1,4,100) && !(MIN_VERSION_filepath(1,5,0)) +#define HAS_OS_STRING_filepath 1 +#else +#define HAS_OS_STRING_filepath 0 +#endif + +-- if we depend on os_string module, then it has System.OsString.Internal.Types +-- module as well +#ifdef MIN_VERSION_os_string +#define HAS_OS_STRING_os_string 1 +#else +#define HAS_OS_STRING_os_string 0 +#endif + +#if HAS_OS_STRING_filepath && HAS_OS_STRING_os_string +import "os-string" System.OsString.Internal.Types (OsString (..), PosixString (..), WindowsString (..)) +import qualified "filepath" System.OsString.Internal.Types as FP (OsString (..), PosixString (..), WindowsString (..)) +#elif HAS_OS_STRING_filepath || HAS_OS_STRING_os_string +import System.OsString.Internal.Types (OsString (..), PosixString (..), WindowsString (..)) +#endif + +import Data.Hashable.Imports +import Data.Hashable.LowLevel +import Data.Hashable.XXH3 + +#include "MachDeps.h" + +-- sometimes we need dependency of filepath, sometimes we dont +import System.FilePath () + +infixl 0 `hashWithSalt` + +------------------------------------------------------------------------ +-- * Computing hash values + +-- | The class of types that can be converted to a hash value. +-- +-- Minimal implementation: 'hashWithSalt'. +-- +-- 'Hashable' is intended exclusively for use in in-memory data structures. +-- . +-- 'Hashable' does /not/ have a fixed standard. +-- This allows it to improve over time. +-- . +-- Because it does not have a fixed standard, different computers or computers on different versions of the code will observe different hash values. +-- As such, 'Hashable' is not recommended for use other than in-memory datastructures. +-- Specifically, 'Hashable' is not intended for network use or in applications which persist hashed values. +-- For stable hashing use named hashes: sha256, crc32, xxhash etc. +-- +-- If you are looking for 'Hashable' instance in @time@ package, +-- check [time-compat](https://hackage.haskell.org/package/time-compat) +-- +class Eq a => Hashable a where + -- | Return a hash value for the argument, using the given salt. + -- + -- The general contract of 'hashWithSalt' is: + -- + -- * If two values are equal according to the '==' method, then + -- applying the 'hashWithSalt' method on each of the two values + -- /must/ produce the same integer result if the same salt is + -- used in each case. + -- + -- * It is /not/ required that if two values are unequal + -- according to the '==' method, then applying the + -- 'hashWithSalt' method on each of the two values must produce + -- distinct integer results. However, the programmer should be + -- aware that producing distinct integer results for unequal + -- values may improve the performance of hashing-based data + -- structures. + -- + -- * This method can be used to compute different hash values for + -- the same input by providing a different salt in each + -- application of the method. This implies that any instance + -- that defines 'hashWithSalt' /must/ make use of the salt in + -- its implementation. + -- + -- * 'hashWithSalt' may return negative 'Int' values. + -- + hashWithSalt :: Int -> a -> Int + + -- | Like 'hashWithSalt', but no salt is used. The default + -- implementation uses 'hashWithSalt' with some default salt. + -- Instances might want to implement this method to provide a more + -- efficient implementation than the default implementation. + hash :: a -> Int + hash = defaultHash + + default hashWithSalt :: (Generic a, GHashable Zero (Rep a)) => Int -> a -> Int + hashWithSalt = genericHashWithSalt + {-# INLINE hashWithSalt #-} + +-- | Generic 'hashWithSalt'. +-- +-- @since 1.3.0.0 +genericHashWithSalt :: (Generic a, GHashable Zero (Rep a)) => Int -> a -> Int +genericHashWithSalt = \salt -> ghashWithSalt HashArgs0 salt . from +{-# INLINE genericHashWithSalt #-} + +data Zero +data One + +data family HashArgs arity a :: Type +data instance HashArgs Zero a = HashArgs0 +newtype instance HashArgs One a = HashArgs1 (Int -> a -> Int) + +-- | The class of types that can be generically hashed. +class GHashable arity f where + ghashWithSalt :: HashArgs arity a -> Int -> f a -> Int + +class (Eq1 t, forall a. Hashable a => Hashable (t a)) => Hashable1 t where + -- | Lift a hashing function through the type constructor. + liftHashWithSalt :: (Int -> a -> Int) -> Int -> t a -> Int + + default liftHashWithSalt :: (Generic1 t, GHashable One (Rep1 t)) => (Int -> a -> Int) -> Int -> t a -> Int + liftHashWithSalt = genericLiftHashWithSalt + {-# INLINE liftHashWithSalt #-} + +-- | Generic 'liftHashWithSalt'. +-- +-- @since 1.3.0.0 +genericLiftHashWithSalt :: (Generic1 t, GHashable One (Rep1 t)) => (Int -> a -> Int) -> Int -> t a -> Int +genericLiftHashWithSalt = \h salt -> ghashWithSalt (HashArgs1 h) salt . from1 +{-# INLINE genericLiftHashWithSalt #-} + +class (Eq2 t, forall a. Hashable a => Hashable1 (t a)) => Hashable2 t where + -- | Lift a hashing function through the binary type constructor. + liftHashWithSalt2 :: (Int -> a -> Int) -> (Int -> b -> Int) -> Int -> t a b -> Int + +-- | Lift the 'hashWithSalt' function through the type constructor. +-- +-- > hashWithSalt1 = liftHashWithSalt hashWithSalt +hashWithSalt1 :: (Hashable1 f, Hashable a) => Int -> f a -> Int +hashWithSalt1 = liftHashWithSalt hashWithSalt + +-- | Lift the 'hashWithSalt' function through the type constructor. +-- +-- > hashWithSalt2 = liftHashWithSalt2 hashWithSalt hashWithSalt +hashWithSalt2 :: (Hashable2 f, Hashable a, Hashable b) => Int -> f a b -> Int +hashWithSalt2 = liftHashWithSalt2 hashWithSalt hashWithSalt + +-- | Lift the 'hashWithSalt' function halfway through the type constructor. +-- This function makes a suitable default implementation of 'liftHashWithSalt', +-- given that the type constructor @t@ in question can unify with @f a@. +defaultLiftHashWithSalt :: (Hashable2 f, Hashable a) => (Int -> b -> Int) -> Int -> f a b -> Int +defaultLiftHashWithSalt h = liftHashWithSalt2 hashWithSalt h + +-- | Since we support a generic implementation of 'hashWithSalt' we +-- cannot also provide a default implementation for that method for +-- the non-generic instance use case. Instead we provide +-- 'defaultHashWith'. +-- +-- @since 1.4.3.0 +-- +defaultHashWithSalt :: Hashable a => Int -> a -> Int +defaultHashWithSalt salt x = salt `hashInt` hash x + +-- | Default implementation of 'hash' based on 'hashWithSalt'. +-- +-- @since 1.4.3.0 +-- +defaultHash :: Hashable a => a -> Int +defaultHash = hashWithSalt defaultSalt + +-- | Transform a value into a 'Hashable' value, then hash the +-- transformed value using the given salt. +-- +-- This is a useful shorthand in cases where a type can easily be +-- mapped to another type that is already an instance of 'Hashable'. +-- Example: +-- +-- > data Foo = Foo | Bar +-- > deriving (Enum) +-- > +-- > instance Hashable Foo where +-- > hashWithSalt = hashUsing fromEnum +-- +-- @since 1.2.0.0 +hashUsing :: (Hashable b) => + (a -> b) -- ^ Transformation function. + -> Int -- ^ Salt. + -> a -- ^ Value to transform. + -> Int +hashUsing f salt x = hashWithSalt salt (f x) +{-# INLINE hashUsing #-} + +instance Hashable Int where + hash = id + hashWithSalt = hashInt + +instance Hashable Int8 where + hash = fromIntegral + hashWithSalt = defaultHashWithSalt + +instance Hashable Int16 where + hash = fromIntegral + hashWithSalt = defaultHashWithSalt + +instance Hashable Int32 where + hash = fromIntegral + hashWithSalt = defaultHashWithSalt + +instance Hashable Int64 where + hash = fromIntegral + hashWithSalt = hashInt64 + +instance Hashable Word where + hash = fromIntegral + hashWithSalt = defaultHashWithSalt + +instance Hashable Word8 where + hash = fromIntegral + hashWithSalt = defaultHashWithSalt + +instance Hashable Word16 where + hash = fromIntegral + hashWithSalt = defaultHashWithSalt + +instance Hashable Word32 where + hash = fromIntegral + hashWithSalt = defaultHashWithSalt + +instance Hashable Word64 where + hash = fromIntegral + hashWithSalt = hashWord64 + +instance Hashable () where + hash = fromEnum + hashWithSalt = defaultHashWithSalt + +instance Hashable Bool where + hash = fromEnum + hashWithSalt = defaultHashWithSalt + +instance Hashable Ordering where + hash = fromEnum + hashWithSalt = defaultHashWithSalt + +instance Hashable Char where + hash = fromEnum + hashWithSalt = defaultHashWithSalt + +instance Hashable BigNat where + hashWithSalt salt (BN# ba) = hashWithSalt salt (ByteArray ba) + +instance Hashable Natural where + hash (NS n) = hash (W# n) + hash (NB bn) = hash (BN# bn) + + hashWithSalt salt (NS n) = hashWithSalt salt (W# n) + hashWithSalt salt (NB bn) = hashWithSalt salt (BN# bn) + +instance Hashable Integer where + hash (IS n) = I# n + hash (IP bn) = hash (BN# bn) + hash (IN bn) = negate (hash (BN# bn)) + + hashWithSalt salt (IS n) = hashWithSalt salt (I# n) + hashWithSalt salt (IP bn) = hashWithSalt salt (BN# bn) + hashWithSalt salt (IN bn) = negate (hashWithSalt salt (BN# bn)) + +instance Hashable a => Hashable (Complex a) where + {-# SPECIALIZE instance Hashable (Complex Double) #-} + {-# SPECIALIZE instance Hashable (Complex Float) #-} + hash (r :+ i) = hash r `hashWithSalt` i + hashWithSalt = hashWithSalt1 +instance Hashable1 Complex where + liftHashWithSalt h s (r :+ i) = s `h` r `h` i + +instance Hashable a => Hashable (Ratio a) where + {-# SPECIALIZE instance Hashable (Ratio Integer) #-} + hash a = hash (numerator a) `hashWithSalt` denominator a + hashWithSalt s a = s `hashWithSalt` numerator a `hashWithSalt` denominator a + +-- | __Note__: prior to @hashable-1.3.0.0@, @hash 0.0 /= hash (-0.0)@ +-- +-- The 'hash' of NaN is not well defined. +-- +-- @since 1.3.0.0 +instance Hashable Float where + hash x + | x == -0.0 || x == 0.0 = 0 -- see note in 'Hashable Double' + | isIEEE x = + assert (sizeOf x >= sizeOf (0::Word32) && + alignment x >= alignment (0::Word32)) $ + hash (castFloatToWord32 x) + | otherwise = hash (show x) + hashWithSalt = defaultHashWithSalt + +-- | __Note__: prior to @hashable-1.3.0.0@, @hash 0.0 /= hash (-0.0)@ +-- +-- The 'hash' of NaN is not well defined. +-- +-- @since 1.3.0.0 +instance Hashable Double where + hash x + | x == -0.0 || x == 0.0 = 0 -- s.t. @hash -0.0 == hash 0.0@ ; see #173 + | isIEEE x = + assert (sizeOf x >= sizeOf (0::Word64) && + alignment x >= alignment (0::Word64)) $ + hash (castDoubleToWord64 x) + | otherwise = hash (show x) + hashWithSalt = defaultHashWithSalt + +-- | A value with bit pattern (01)* (or 5* in hexa), for any size of Int. +-- It is used as data constructor distinguisher. GHC computes its value during +-- compilation. +distinguisher :: Int +distinguisher = fromIntegral $ (maxBound :: Word) `quot` 3 +{-# INLINE distinguisher #-} + +instance Hashable a => Hashable (Maybe a) where + hash Nothing = 0 + hash (Just a) = distinguisher `hashWithSalt` a + hashWithSalt = hashWithSalt1 + +instance Hashable1 Maybe where + liftHashWithSalt _ s Nothing = s `hashInt` 0 + liftHashWithSalt h s (Just a) = s `hashInt` distinguisher `h` a + +instance (Hashable a, Hashable b) => Hashable (Either a b) where + hash (Left a) = 0 `hashWithSalt` a + hash (Right b) = distinguisher `hashWithSalt` b + hashWithSalt = hashWithSalt1 + +instance Hashable a => Hashable1 (Either a) where + liftHashWithSalt = defaultLiftHashWithSalt + +instance Hashable2 Either where + liftHashWithSalt2 h _ s (Left a) = s `hashInt` 0 `h` a + liftHashWithSalt2 _ h s (Right b) = s `hashInt` distinguisher `h` b + +instance (Hashable a1, Hashable a2) => Hashable (a1, a2) where + hashWithSalt = hashWithSalt1 + +instance Hashable a1 => Hashable1 ((,) a1) where + liftHashWithSalt = defaultLiftHashWithSalt + +instance Hashable2 (,) where + liftHashWithSalt2 h1 h2 s (a1, a2) = s `h1` a1 `h2` a2 + +instance (Hashable a1, Hashable a2, Hashable a3) => Hashable (a1, a2, a3) where + hashWithSalt = hashWithSalt1 + +instance (Hashable a1, Hashable a2) => Hashable1 ((,,) a1 a2) where + liftHashWithSalt = defaultLiftHashWithSalt + +instance Hashable a1 => Hashable2 ((,,) a1) where + liftHashWithSalt2 h1 h2 s (a1, a2, a3) = + (s `hashWithSalt` a1) `h1` a2 `h2` a3 + +instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4) => + Hashable (a1, a2, a3, a4) where + hashWithSalt = hashWithSalt1 + +instance (Hashable a1, Hashable a2, Hashable a3) => Hashable1 ((,,,) a1 a2 a3) where + liftHashWithSalt = defaultLiftHashWithSalt + +instance (Hashable a1, Hashable a2) => Hashable2 ((,,,) a1 a2) where + liftHashWithSalt2 h1 h2 s (a1, a2, a3, a4) = + (s `hashWithSalt` a1 `hashWithSalt` a2) `h1` a3 `h2` a4 + +instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4, Hashable a5) + => Hashable (a1, a2, a3, a4, a5) where + hashWithSalt s (a1, a2, a3, a4, a5) = + s `hashWithSalt` a1 `hashWithSalt` a2 `hashWithSalt` a3 + `hashWithSalt` a4 `hashWithSalt` a5 + +{- +instance (Hashable a1, Hashable a2, Hashable a3, + Hashable a4) => Hashable1 ((,,,,) a1 a2 a3 a4) where + liftHashWithSalt = defaultLiftHashWithSalt + +instance (Hashable a1, Hashable a2, Hashable a3) + => Hashable2 ((,,,,) a1 a2 a3) where + liftHashWithSalt2 h1 h2 s (a1, a2, a3, a4, a5) = + (s `hashWithSalt` a1 `hashWithSalt` a2 + `hashWithSalt` a3) `h1` a4 `h2` a5 +-} + +instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4, Hashable a5, + Hashable a6) => Hashable (a1, a2, a3, a4, a5, a6) where + hashWithSalt s (a1, a2, a3, a4, a5, a6) = + s `hashWithSalt` a1 `hashWithSalt` a2 `hashWithSalt` a3 + `hashWithSalt` a4 `hashWithSalt` a5 `hashWithSalt` a6 + +{- +instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4, + Hashable a5) => Hashable1 ((,,,,,) a1 a2 a3 a4 a5) where + liftHashWithSalt = defaultLiftHashWithSalt + +instance (Hashable a1, Hashable a2, Hashable a3, + Hashable a4) => Hashable2 ((,,,,,) a1 a2 a3 a4) where + liftHashWithSalt2 h1 h2 s (a1, a2, a3, a4, a5, a6) = + (s `hashWithSalt` a1 `hashWithSalt` a2 `hashWithSalt` a3 + `hashWithSalt` a4) `h1` a5 `h2` a6 +-} + +instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4, Hashable a5, + Hashable a6, Hashable a7) => + Hashable (a1, a2, a3, a4, a5, a6, a7) where + hashWithSalt s (a1, a2, a3, a4, a5, a6, a7) = + s `hashWithSalt` a1 `hashWithSalt` a2 `hashWithSalt` a3 + `hashWithSalt` a4 `hashWithSalt` a5 `hashWithSalt` a6 `hashWithSalt` a7 + +{- +instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4, Hashable a5, Hashable a6) => Hashable1 ((,,,,,,) a1 a2 a3 a4 a5 a6) where + liftHashWithSalt = defaultLiftHashWithSalt + +instance (Hashable a1, Hashable a2, Hashable a3, Hashable a4, + Hashable a5) => Hashable2 ((,,,,,,) a1 a2 a3 a4 a5) where + liftHashWithSalt2 h1 h2 s (a1, a2, a3, a4, a5, a6, a7) = + (s `hashWithSalt` a1 `hashWithSalt` a2 `hashWithSalt` a3 + `hashWithSalt` a4 `hashWithSalt` a5) `h1` a6 `h2` a7 +-} + +instance Hashable (StableName a) where + hash = hashStableName + hashWithSalt = defaultHashWithSalt + +-- Auxiliary type for Hashable [a] definition +data SPInt = SP !Int !Int + +instance Hashable a => Hashable [a] where + {-# SPECIALIZE instance Hashable [Char] #-} + hashWithSalt = hashWithSalt1 + +instance Hashable1 [] where + liftHashWithSalt h salt arr = finalise (foldl' step (SP salt 0) arr) + where + finalise (SP s l) = hashWithSalt s l + step (SP s l) x = SP (h s x) (l + 1) + +instance Hashable B.ByteString where + hash bs = fromIntegral (xxh3_64bit_withSeed_bs bs 0) + + hashWithSalt salt bs = + fromIntegral (xxh3_64bit_withSeed_bs bs (fromIntegral (hashWithSalt salt len))) + where + len = B.length bs + +instance Hashable BL.ByteString where + hashWithSalt salt lbs = runST $ do + s <- xxh3_64bit_createState + xxh3_64bit_reset_withSeed s (fromIntegral salt) + len <- BL.foldrChunks (step s) return lbs 0 + xxh3_64bit_update_w64 s len + digest <- xxh3_64bit_digest s + return (fromIntegral digest) + where + step s bs next !acc = do + xxh3_64bit_update_bs s bs + next (acc + fromIntegral (B.length bs)) + +instance Hashable BSI.ShortByteString where + hash (BSI.SBS ba) = hash (ByteArray ba) + hashWithSalt salt (BSI.SBS ba) = hashWithSalt salt (ByteArray ba) + +#if HAS_OS_STRING_filepath || HAS_OS_STRING_os_string +-- | @since 1.4.2.0 +deriving newtype instance Hashable PosixString + +-- | @since 1.4.2.0 +deriving newtype instance Hashable WindowsString + +-- | @since 1.4.2.0 +deriving newtype instance Hashable OsString +#endif + +#if HAS_OS_STRING_filepath && HAS_OS_STRING_os_string +deriving newtype instance Hashable FP.PosixString +deriving newtype instance Hashable FP.WindowsString +deriving newtype instance Hashable FP.OsString +#endif + +#if MIN_VERSION_text(2,0,0) + +instance Hashable T.Text where + hash (T.Text (TA.ByteArray arr) off len) = + fromIntegral (xxh3_64bit_withSeed_ba (ByteArray arr) off len 0) + hashWithSalt salt (T.Text (TA.ByteArray arr) off len) = + fromIntegral (xxh3_64bit_withSeed_ba (ByteArray arr) off len (fromIntegral (hashWithSalt salt len))) + +instance Hashable TL.Text where + hashWithSalt salt lt = runST $ do + s <- xxh3_64bit_createState + xxh3_64bit_reset_withSeed s (fromIntegral salt) + len <- TL.foldrChunks (step s) return lt 0 + xxh3_64bit_update_w64 s len + digest <- xxh3_64bit_digest s + return (fromIntegral digest) + where + step s (T.Text (TA.ByteArray arr) off len) next !acc = do + xxh3_64bit_update_ba s (ByteArray arr) off len + next (acc + fromIntegral len) + +#else + +instance Hashable T.Text where + hash (T.Text arr off len) = + fromIntegral (xxh3_64bit_withSeed_ba (ByteArray (TA.aBA arr)) (unsafeShiftL off 1) (unsafeShiftL len 1) 0) + hashWithSalt salt (T.Text arr off len) = + fromIntegral (xxh3_64bit_withSeed_ba (ByteArray (TA.aBA arr)) (unsafeShiftL off 1) (unsafeShiftL len 1) (fromIntegral (hashWithSalt salt len))) + +instance Hashable TL.Text where + hashWithSalt salt lt = runST $ do + s <- xxh3_64bit_createState + xxh3_64bit_reset_withSeed s (fromIntegral salt) + len <- TL.foldrChunks (step s) return lt 0 + xxh3_64bit_update_w64 s len + digest <- xxh3_64bit_digest s + return (fromIntegral digest) + where + step s (T.Text arr off len) next !acc = do + xxh3_64bit_update_ba s (ByteArray (TA.aBA arr)) (unsafeShiftL off 1) (unsafeShiftL len 1) + next (acc + fromIntegral len) + +#endif + +#if !MIN_VERSION_base(4,19,0) +fromThreadId :: ThreadId -> Word64 +fromThreadId (ThreadId t) = fromIntegral (getThreadId t) + +-- this cannot be capi, as GHC panics. +foreign import ccall unsafe "rts_getThreadId" getThreadId +#if __GLASGOW_HASKELL__ >= 904 + -- https://gitlab.haskell.org/ghc/ghc/-/merge_requests/6163 + :: ThreadId# -> CULLong +#elif __GLASGOW_HASKELL__ >= 900 + -- https://gitlab.haskell.org/ghc/ghc/-/merge_requests/1254 + :: ThreadId# -> CLong +#else + :: ThreadId# -> CInt +#endif +#endif + +instance Hashable ThreadId where + hash = hash . fromThreadId + hashWithSalt = defaultHashWithSalt + +instance Hashable (Ptr a) where + hashWithSalt salt p = hashWithSalt salt $ ptrToIntPtr p + +instance Hashable (FunPtr a) where + hashWithSalt salt p = hashWithSalt salt $ castFunPtrToPtr p + +instance Hashable IntPtr where + hash n = fromIntegral n + hashWithSalt = defaultHashWithSalt + +instance Hashable WordPtr where + hash n = fromIntegral n + hashWithSalt = defaultHashWithSalt + +---------------------------------------------------------------------------- +-- Fingerprint & TypeRep instances + +-- | @since 1.3.0.0 +instance Hashable Fingerprint where + hash (Fingerprint x _) = fromIntegral x + hashWithSalt = defaultHashWithSalt + {-# INLINE hash #-} + +hashTypeRep :: Type.Reflection.TypeRep a -> Int +hashTypeRep tr = + let Fingerprint x _ = typeRepFingerprint tr in fromIntegral x + +instance Hashable Type.Reflection.SomeTypeRep where + hash (Type.Reflection.SomeTypeRep r) = hashTypeRep r + hashWithSalt = defaultHashWithSalt + {-# INLINE hash #-} + +instance Hashable (Type.Reflection.TypeRep a) where + hash = hashTypeRep + hashWithSalt = defaultHashWithSalt + {-# INLINE hash #-} + +---------------------------------------------------------------------------- + +instance Hashable Void where + hashWithSalt _ = absurd + +-- | Compute a hash value for the content of this pointer. +hashPtr :: Ptr a -- ^ pointer to the data to hash + -> Int -- ^ length, in bytes + -> IO Int -- ^ hash value +hashPtr p len = hashPtrWithSalt p len defaultSalt + +-- | Compute a hash value for the content of this 'ByteArray#', +-- beginning at the specified offset, using specified number of bytes. +hashByteArray :: ByteArray# -- ^ data to hash + -> Int -- ^ offset, in bytes + -> Int -- ^ length, in bytes + -> Int -- ^ hash value +hashByteArray ba0 off len = hashByteArrayWithSalt ba0 off len defaultSalt +{-# INLINE hashByteArray #-} + +instance Hashable Unique where + hash = hashUnique + hashWithSalt = defaultHashWithSalt + +instance Hashable Version where + hashWithSalt salt (Version branch tags) = + salt `hashWithSalt` branch `hashWithSalt` tags + +deriving newtype instance Hashable (Fixed a) + +deriving newtype instance Hashable a => Hashable (Identity a) +instance Hashable1 Identity where + liftHashWithSalt h salt (Identity x) = h salt x + +-- Using hashWithSalt1 would cause needless constraint +deriving newtype instance Hashable a => Hashable (Const a b) + +instance Hashable a => Hashable1 (Const a) where + liftHashWithSalt = defaultLiftHashWithSalt + +instance Hashable2 Const where + liftHashWithSalt2 f _ salt (Const x) = f salt x + +instance Hashable (Proxy a) where + hash _ = 0 + hashWithSalt s _ = s + +instance Hashable1 Proxy where + liftHashWithSalt _ s _ = s + +instance Hashable a => Hashable (NE.NonEmpty a) where + hashWithSalt p (a NE.:| as) = p `hashWithSalt` a `hashWithSalt` as + +-- | @since 1.3.1.0 +instance Hashable1 NE.NonEmpty where + liftHashWithSalt h salt (a NE.:| as) = liftHashWithSalt h (h salt a) as + +deriving newtype instance Hashable a => Hashable (Semi.Min a) +deriving newtype instance Hashable a => Hashable (Semi.Max a) + +-- | __Note__: Prior to @hashable-1.3.0.0@ the hash computation included the second argument of 'Arg' which wasn't consistent with its 'Eq' instance. +-- +-- Since @hashable-1.5.0.0@, @hash (Semi.arg a _) = hash a@ +-- +-- @since 1.3.0.0 +instance Hashable a => Hashable (Semi.Arg a b) where + hash (Semi.Arg a _) = hash a + hashWithSalt p (Semi.Arg a _) = hashWithSalt p a + +deriving newtype instance Hashable a => Hashable (Semi.First a) +deriving newtype instance Hashable a => Hashable (Semi.Last a) +deriving newtype instance Hashable a => Hashable (Semi.WrappedMonoid a) + +#if !MIN_VERSION_base(4,16,0) +deriving newtype instance Hashable a => Hashable (Semi.Option a) +#endif + +-- TODO: this instance is removed as there isn't Eq1 Min/Max, ... + +#if 0 +-- | @since 1.3.1.0 +-- instance Hashable1 Min where liftHashWithSalt h salt (Min a) = h salt a + +-- | @since 1.3.1.0 +-- instance Hashable1 Max where liftHashWithSalt h salt (Max a) = h salt a + +-- | @since 1.3.1.0 +-- instance Hashable1 First where liftHashWithSalt h salt (First a) = h salt a + +-- | @since 1.3.1.0 +-- instance Hashable1 Last where liftHashWithSalt h salt (Last a) = h salt a + + +-- | @since 1.3.1.0 +-- instance Hashable1 WrappedMonoid where liftHashWithSalt h salt (WrapMonoid a) = h salt a + +-- | @since 1.3.1.0 +-- instance Hashable1 Option where liftHashWithSalt h salt (Option a) = liftHashWithSalt h salt a +#endif + +instance (Hashable (f (g a))) => Hashable (Compose f g a) where + hash (Compose x) = hash x + hashWithSalt p (Compose x) = hashWithSalt p x + +instance (Hashable1 f, Hashable1 g) => Hashable1 (Compose f g) where + liftHashWithSalt h s = liftHashWithSalt (liftHashWithSalt h) s . getCompose + +instance (Hashable1 f, Hashable1 g) => Hashable1 (FP.Product f g) where + liftHashWithSalt h s (FP.Pair a b) = liftHashWithSalt h (liftHashWithSalt h s a) b + +instance (Hashable (f a), Hashable (g a)) => Hashable (FP.Product f g a) where + hashWithSalt s (FP.Pair a b) = s `hashWithSalt` a `hashWithSalt` b + +instance (Hashable1 f, Hashable1 g) => Hashable1 (FS.Sum f g) where + liftHashWithSalt h s (FS.InL a) = liftHashWithSalt h (s `hashInt` 0) a + liftHashWithSalt h s (FS.InR a) = liftHashWithSalt h (s `hashInt` distinguisher) a + +instance (Hashable (f a), Hashable (g a)) => Hashable (FS.Sum f g a) where + hashWithSalt s (FS.InL a) = hashWithSalt (s `hashInt` 0) a + hashWithSalt s (FS.InR a) = hashWithSalt (s `hashInt` distinguisher) a + +-- | This instance was available since 1.4.1.0 only for GHC-9.4+ +-- +-- @since 1.4.2.0 +-- +instance Hashable AB.ByteArray where + hash ba@(AB.ByteArray ba') = + fromIntegral (xxh3_64bit_withSeed_ba ba 0 len 0) + where + !len = I# (sizeofByteArray# ba') + + hashWithSalt salt ba@(AB.ByteArray ba') = + fromIntegral (xxh3_64bit_withSeed_ba ba 0 len (fromIntegral (hashWithSalt salt len))) + where + !len = I# (sizeofByteArray# ba') + +------------------------------------------------------------------------------- +-- Hashed +------------------------------------------------------------------------------- + +-- | A hashable value along with the result of the 'hash' function. +data Hashed a = Hashed a {-# UNPACK #-} !Int + +-- | Wrap a hashable value, caching the 'hash' function result. +hashed :: Hashable a => a -> Hashed a +hashed a = Hashed a (hash a) + +-- | Unwrap hashed value. +unhashed :: Hashed a -> a +unhashed (Hashed a _) = a + +-- | 'hash' has 'Eq' requirement. +-- +-- @since 1.4.0.0 +hashedHash :: Hashed a -> Int +hashedHash (Hashed _ h) = h + +-- | Uses precomputed hash to detect inequality faster +instance Eq a => Eq (Hashed a) where + Hashed a ha == Hashed b hb = ha == hb && a == b + +instance Ord a => Ord (Hashed a) where + Hashed a _ `compare` Hashed b _ = a `compare` b + +instance Show a => Show (Hashed a) where + showsPrec d (Hashed a _) = showParen (d > 10) $ + showString "hashed" . showChar ' ' . showsPrec 11 a + +instance Eq a => Hashable (Hashed a) where + hashWithSalt = defaultHashWithSalt + hash = hashedHash + + +-- This instance is a little unsettling. It is unusal for +-- 'liftHashWithSalt' to ignore its first argument when a +-- value is actually available for it to work on. +instance Hashable1 Hashed where + liftHashWithSalt _ s (Hashed _ h) = defaultHashWithSalt s h + +instance (IsString a, Hashable a) => IsString (Hashed a) where + fromString s = let r = fromString s in Hashed r (hash r) + +instance Foldable Hashed where + foldMap f (Hashed a _) = f a + foldr f acc (Hashed a _) = f a acc + +instance NFData a => NFData (Hashed a) where + rnf = rnf . unhashed + +-- | 'Hashed' cannot be 'Functor' +mapHashed :: Hashable b => (a -> b) -> Hashed a -> Hashed b +mapHashed f (Hashed a _) = hashed (f a) + +-- | 'Hashed' cannot be 'Traversable' +traverseHashed :: (Hashable b, Functor f) => (a -> f b) -> Hashed a -> f (Hashed b) +traverseHashed f (Hashed a _) = fmap hashed (f a) + +-- instances for @Data.Functor.Classes@ higher rank typeclasses +-- in base-4.9 and onward. +instance Eq1 Hashed where + liftEq f (Hashed a ha) (Hashed b hb) = ha == hb && f a b + +instance Ord1 Hashed where + liftCompare f (Hashed a _) (Hashed b _) = f a b + +instance Show1 Hashed where + liftShowsPrec sp _ d (Hashed a _) = showParen (d > 10) $ + showString "hashed " . sp 11 a + +------------------------------------------------------------------------------- +-- containers +------------------------------------------------------------------------------- + +-- | @since 1.3.4.0 +instance Hashable2 Map.Map where + liftHashWithSalt2 hk hv s m = Map.foldlWithKey' + (\s' k v -> hv (hk s' k) v) + (hashWithSalt s (Map.size m)) + m + +-- | @since 1.3.4.0 +instance Hashable k => Hashable1 (Map.Map k) where + liftHashWithSalt h s m = Map.foldlWithKey' + (\s' k v -> h (hashWithSalt s' k) v) + (hashWithSalt s (Map.size m)) + m + +-- | @since 1.3.4.0 +instance (Hashable k, Hashable v) => Hashable (Map.Map k v) where + hashWithSalt = hashWithSalt2 + +-- | @since 1.3.4.0 +instance Hashable1 IntMap.IntMap where + liftHashWithSalt h s m = IntMap.foldlWithKey' + (\s' k v -> h (hashWithSalt s' k) v) + (hashWithSalt s (IntMap.size m)) + m + +-- | @since 1.3.4.0 +instance Hashable v => Hashable (IntMap.IntMap v) where + hashWithSalt = hashWithSalt1 + +-- | @since 1.3.4.0 +instance Hashable1 Set.Set where + liftHashWithSalt h s x = Set.foldl' h (hashWithSalt s (Set.size x)) x + +-- | @since 1.3.4.0 +instance Hashable v => Hashable (Set.Set v) where + hashWithSalt = hashWithSalt1 + +-- | @since 1.3.4.0 +instance Hashable IntSet.IntSet where + hashWithSalt salt x = IntSet.foldl' hashWithSalt + (hashWithSalt salt (IntSet.size x)) + x + +-- | @since 1.3.4.0 +instance Hashable1 Seq.Seq where + liftHashWithSalt h s x = foldl' h (hashWithSalt s (Seq.length x)) x + +-- | @since 1.3.4.0 +instance Hashable v => Hashable (Seq.Seq v) where + hashWithSalt = hashWithSalt1 + +-- | @since 1.3.4.0 +instance Hashable1 Tree.Tree where + liftHashWithSalt h = go where + go s (Tree.Node x xs) = liftHashWithSalt go (h s x) xs + +-- | @since 1.3.4.0 +instance Hashable v => Hashable (Tree.Tree v) where + hashWithSalt = hashWithSalt1 + +------------------------------------------------------------------------------- +-- Solo +------------------------------------------------------------------------------- + +instance Hashable a => Hashable (Solo a) where + hashWithSalt = hashWithSalt1 +instance Hashable1 Solo where + liftHashWithSalt h salt (Solo x) = h salt x diff --git a/src/Data/Hashable/FFI.hs b/src/Data/Hashable/FFI.hs new file mode 100644 --- /dev/null +++ b/src/Data/Hashable/FFI.hs @@ -0,0 +1,76 @@ +{-# LANGUAGE CApiFFI #-} +{-# LANGUAGE MagicHash #-} +{-# LANGUAGE Trustworthy #-} +{-# LANGUAGE UnliftedFFITypes #-} +module Data.Hashable.FFI ( + -- * One shot + unsafe_xxh3_64bit_withSeed_ptr, + unsafe_xxh3_64bit_withSeed_ba, + unsafe_xxh3_64bit_withSeed_u64, + unsafe_xxh3_64bit_withSeed_u32, + -- * Incremental + unsafe_xxh3_sizeof_state, + unsafe_xxh3_initState, + unsafe_xxh3_64bit_reset_withSeed, + unsafe_xxh3_64bit_digest, + unsafe_xxh3_64bit_update_ptr, + unsafe_xxh3_64bit_update_ba, + unsafe_xxh3_64bit_update_u64, + unsafe_xxh3_64bit_update_u32, +) where + +import Data.Word (Word32, Word64, Word8) +import Foreign.C.Types (CSize (..)) +import Foreign.Ptr (Ptr) +import GHC.Exts (ByteArray#, MutableByteArray#) + +-- Note: we use unsafe FFI calls, as we expect our use case to be hashing only small data (<1kb, at most 4k). + +------------------------------------------------------------------------------- +-- OneShot +------------------------------------------------------------------------------- + +foreign import capi unsafe "HsXXHash.h XXH3_64bits_withSeed" + unsafe_xxh3_64bit_withSeed_ptr :: Ptr Word8 -> CSize -> Word64 -> IO Word64 + +foreign import capi unsafe "HsXXHash.h hs_XXH3_64bits_withSeed_offset" + unsafe_xxh3_64bit_withSeed_ba :: ByteArray# -> CSize -> CSize -> Word64 -> Word64 + +foreign import capi unsafe "HsXXHash.h hs_XXH3_64bits_withSeed_u64" + unsafe_xxh3_64bit_withSeed_u64 :: Word64 -> Word64 -> Word64 + +foreign import capi unsafe "HsXXHash.h hs_XXH3_64bits_withSeed_u32" + unsafe_xxh3_64bit_withSeed_u32 :: Word32 -> Word64 -> Word64 + +------------------------------------------------------------------------------- +-- Incremental +------------------------------------------------------------------------------- + +-- reset and update functions return OK/Error +-- we ignore that: +-- * reset errors only on NULL state +-- * update cannot even error + +foreign import capi unsafe "HsXXHash.h value hs_XXH3_sizeof_state_s" + unsafe_xxh3_sizeof_state :: Int + +foreign import capi unsafe "HsXXHash.h XXH3_INITSTATE" + unsafe_xxh3_initState :: MutableByteArray# s -> IO () + +foreign import capi unsafe "HsXXHash.h XXH3_64bits_reset_withSeed" + unsafe_xxh3_64bit_reset_withSeed :: MutableByteArray# s -> Word64 -> IO () + +foreign import capi unsafe "HsXXHash.h XXH3_64bits_digest" + unsafe_xxh3_64bit_digest :: MutableByteArray# s -> IO Word64 + +foreign import capi unsafe "HsXXHash.h XXH3_64bits_update" + unsafe_xxh3_64bit_update_ptr :: MutableByteArray# s -> Ptr Word8 -> CSize -> IO () + +foreign import capi unsafe "HsXXHash.h hs_XXH3_64bits_update_offset" + unsafe_xxh3_64bit_update_ba :: MutableByteArray# s -> ByteArray# -> CSize -> CSize -> IO () + +foreign import capi unsafe "HsXXHash.h hs_XXH3_64bits_update_u64" + unsafe_xxh3_64bit_update_u64 :: MutableByteArray# s -> Word64 -> IO () + +foreign import capi unsafe "HsXXHash.h hs_XXH3_64bits_update_u32" + unsafe_xxh3_64bit_update_u32 :: MutableByteArray# s -> Word32 -> IO () diff --git a/src/Data/Hashable/Generic.hs b/src/Data/Hashable/Generic.hs new file mode 100644 --- /dev/null +++ b/src/Data/Hashable/Generic.hs @@ -0,0 +1,25 @@ +{-# LANGUAGE Safe #-} + +-- | +-- Module : Data.Hashable.Generic +-- SPDX-License-Identifier : BSD-3-Clause +-- Stability : provisional +-- Portability : GHC >= 7.4 +-- +-- Hashable support for GHC generics. +-- +-- @since 1.3.0.0 +module Data.Hashable.Generic + ( + -- * Implementation using Generics. + genericHashWithSalt + , genericLiftHashWithSalt + -- * Constraints + , GHashable (..) + , One + , Zero + , HashArgs (..) + ) where + +import Data.Hashable.Generic.Instances () +import Data.Hashable.Class diff --git a/src/Data/Hashable/Generic/Instances.hs b/src/Data/Hashable/Generic/Instances.hs new file mode 100644 --- /dev/null +++ b/src/Data/Hashable/Generic/Instances.hs @@ -0,0 +1,124 @@ +{-# LANGUAGE BangPatterns, FlexibleInstances, KindSignatures, + ScopedTypeVariables, TypeOperators, + MultiParamTypeClasses, GADTs, FlexibleContexts #-} +{-# OPTIONS_GHC -fno-warn-orphans #-} +{-# LANGUAGE Trustworthy #-} + +------------------------------------------------------------------------ +-- | +-- Module : Data.Hashable.Generic.Instances +-- Copyright : (c) Bryan O'Sullivan 2012 +-- SPDX-License-Identifier : BSD-3-Clause +-- Maintainer : bos@serpentine.com +-- Stability : provisional +-- Portability : GHC >= 7.4 +-- +-- Internal module defining orphan instances for "GHC.Generics" +-- +module Data.Hashable.Generic.Instances () where + +import Data.Hashable.Class +import GHC.Generics +import Data.Kind (Type) + + +-- Type without constructors +instance GHashable arity V1 where + ghashWithSalt _ salt _ = hashWithSalt salt () + +-- Constructor without arguments +instance GHashable arity U1 where + ghashWithSalt _ salt U1 = hashWithSalt salt () + +instance (GHashable arity a, GHashable arity b) => GHashable arity (a :*: b) where + ghashWithSalt toHash salt (x :*: y) = + (ghashWithSalt toHash (ghashWithSalt toHash salt x) y) + +-- Metadata (constructor name, etc) +instance GHashable arity a => GHashable arity (M1 i c a) where + ghashWithSalt targs salt = ghashWithSalt targs salt . unM1 + +-- Constants, additional parameters, and rank-1 recursion +instance Hashable a => GHashable arity (K1 i a) where + ghashWithSalt _ = hashUsing unK1 + +instance GHashable One Par1 where + ghashWithSalt (HashArgs1 h) salt = h salt . unPar1 + +instance Hashable1 f => GHashable One (Rec1 f) where + ghashWithSalt (HashArgs1 h) salt = liftHashWithSalt h salt . unRec1 + +instance (Hashable1 f, GHashable One g) => GHashable One (f :.: g) where + ghashWithSalt targs salt = liftHashWithSalt (ghashWithSalt targs) salt . unComp1 + +class SumSize f => GSum arity f where + hashSum :: HashArgs arity a -> Int -> Int -> f a -> Int + -- hashSum args salt index value = ... + +-- [Note: Hashing a sum type] +-- +-- The tree structure is used in GHC.Generics to represent the sum (and +-- product) part of the generic representation of the type, e.g.: +-- +-- (C0 ... :+: C1 ...) :+: (C2 ... :+: (C3 ... :+: C4 ...)) +-- +-- The value constructed with C2 constructor is represented as (R1 (L1 ...)). +-- Yet, if we think that this tree is a flat (heterogeneous) list: +-- +-- [C0 ..., C1 ..., C2 ..., C3 ..., C4... ] +-- +-- then the value constructed with C2 is a (dependent) pair (2, ...), and +-- hashing it is simple: +-- +-- salt `hashWithSalt` (2 :: Int) `hashWithSalt` ... +-- +-- This is what we do below. When drilling down the tree, we count how many +-- leafs are to the left (`index` variable). At the leaf case C1, we'll have an +-- actual index into the sum. +-- +-- This works well for balanced data. However for recursive types like: +-- +-- data Nat = Z | S Nat +-- +-- the `hashWithSalt salt (S (S (S Z)))` is +-- +-- salt `hashWithSalt` (1 :: Int) -- first S +-- `hashWithSalt` (1 :: Int) -- second S +-- `hashWithSalt` (1 :: Int) -- third S +-- `hashWithSalt` (0 :: Int) -- Z +-- `hashWithSalt` () -- U1 +-- +-- For that type the manual implementation: +-- +-- instance Hashable Nat where +-- hashWithSalt salt n = hashWithSalt salt (natToInteger n) +-- +-- would be better performing CPU and hash-quality wise (assuming that +-- Integer's Hashable is of high quality). +-- +instance (GSum arity a, GSum arity b) => GHashable arity (a :+: b) where + ghashWithSalt toHash salt = hashSum toHash salt 0 + +instance (GSum arity a, GSum arity b) => GSum arity (a :+: b) where + hashSum toHash !salt !index s = case s of + L1 x -> hashSum toHash salt index x + R1 x -> hashSum toHash salt (index + sizeL) x + where + sizeL = unTagged (sumSize :: Tagged a) + {-# INLINE hashSum #-} + +instance GHashable arity a => GSum arity (C1 c a) where + hashSum toHash !salt !index (M1 x) = ghashWithSalt toHash (hashWithSalt salt index) x + {-# INLINE hashSum #-} + +class SumSize f where + sumSize :: Tagged f + +newtype Tagged (s :: Type -> Type) = Tagged {unTagged :: Int} + +instance (SumSize a, SumSize b) => SumSize (a :+: b) where + sumSize = Tagged $ unTagged (sumSize :: Tagged a) + + unTagged (sumSize :: Tagged b) + +instance SumSize (C1 c a) where + sumSize = Tagged 1 diff --git a/src/Data/Hashable/Imports.hs b/src/Data/Hashable/Imports.hs new file mode 100644 --- /dev/null +++ b/src/Data/Hashable/Imports.hs @@ -0,0 +1,14 @@ +-- | This module exists to avoid conditional imports +-- and unused import warnings. +{-# LANGUAGE Safe #-} +module Data.Hashable.Imports ( + Int64, Int32, + Word64, Word32, + xor, shiftR, shiftL, unsafeShiftL, unsafeShiftR, + (.&.), +) where + +import Data.Bits (shiftL, shiftR, unsafeShiftL, unsafeShiftR, xor, (.&.)) +import Data.Int (Int32, Int64) +import Data.Word (Word32, Word64) +import Prelude () diff --git a/src/Data/Hashable/Lifted.hs b/src/Data/Hashable/Lifted.hs new file mode 100644 --- /dev/null +++ b/src/Data/Hashable/Lifted.hs @@ -0,0 +1,98 @@ +{-# LANGUAGE Safe #-} + +------------------------------------------------------------------------ +-- | +-- Module : Data.Hashable.Lifted +-- Copyright : (c) Milan Straka 2010 +-- (c) Johan Tibell 2011 +-- (c) Bryan O'Sullivan 2011, 2012 +-- SPDX-License-Identifier : BSD-3-Clause +-- Maintainer : johan.tibell@gmail.com +-- Stability : provisional +-- Portability : portable +-- +-- Lifting of the 'Hashable' class to unary and binary type constructors. +-- These classes are needed to express the constraints on arguments of +-- types that are parameterized by type constructors. Fixed-point data +-- types and monad transformers are such types. + +module Data.Hashable.Lifted + ( -- * Type Classes + Hashable1(..) + , Hashable2(..) + -- * Auxiliary Functions + , hashWithSalt1 + , hashWithSalt2 + , defaultLiftHashWithSalt + -- * Motivation + -- $motivation + ) where + +import Data.Hashable.Class + +-- $motivation +-- +-- This type classes provided in this module are used to express constraints +-- on type constructors in a Haskell98-compatible fashion. As an example, consider +-- the following two types (Note that these instances are not actually provided +-- because @hashable@ does not have @transformers@ or @free@ as a dependency): +-- +-- > newtype WriterT w m a = WriterT { runWriterT :: m (a, w) } +-- > data Free f a = Pure a | Free (f (Free f a)) +-- +-- The 'Hashable1' instances for @WriterT@ and @Free@ could be written as: +-- +-- > instance (Hashable w, Hashable1 m) => Hashable1 (WriterT w m) where +-- > liftHashWithSalt h s (WriterT m) = +-- > liftHashWithSalt (liftHashWithSalt2 h hashWithSalt) s m +-- > instance Hashable1 f => Hashable1 (Free f) where +-- > liftHashWithSalt h = go where +-- > go s x = case x of +-- > Pure a -> h s a +-- > Free p -> liftHashWithSalt go s p +-- +-- The 'Hashable' instances for these types can be trivially recovered with +-- 'hashWithSalt1': +-- +-- > instance (Hashable w, Hashable1 m, Hashable a) => Hashable (WriterT w m a) where +-- > hashWithSalt = hashWithSalt1 +-- > instance (Hashable1 f, Hashable a) => Hashable (Free f a) where +-- > hashWithSalt = hashWithSalt1 + +-- +-- $discussion +-- +-- Regardless of whether 'hashWithSalt1' is used to provide an implementation +-- of 'hashWithSalt', they should produce the same hash when called with +-- the same arguments. This is the only law that 'Hashable1' and 'Hashable2' +-- are expected to follow. +-- +-- The typeclasses in this module only provide lifting for 'hashWithSalt', not +-- for 'hash'. This is because such liftings cannot be defined in a way that +-- would satisfy the @liftHash@ variant of the above law. As an illustration +-- of the problem we run into, let us assume that 'Hashable1' were +-- given a 'liftHash' method: +-- +-- > class Hashable1 t where +-- > liftHash :: (a -> Int) -> t a -> Int +-- > liftHashWithSalt :: (Int -> a -> Int) -> Int -> t a -> Int +-- +-- Even for a type as simple as 'Maybe', the problem manifests itself. The +-- 'Hashable' instance for 'Maybe' is: +-- +-- > distinguisher :: Int +-- > distinguisher = ... +-- > +-- > instance Hashable a => Hashable (Maybe a) where +-- > hash Nothing = 0 +-- > hash (Just a) = distinguisher `hashWithSalt` a +-- > hashWithSalt s Nothing = ... +-- > hashWithSalt s (Just a) = ... +-- +-- The implementation of 'hash' calls 'hashWithSalt' on @a@. The hypothetical +-- @liftHash@ defined earlier only accepts an argument that corresponds to +-- the implementation of 'hash' for @a@. Consequently, this formulation of +-- @liftHash@ would not provide a way to match the current behavior of 'hash' +-- for 'Maybe'. This problem gets worse when 'Either' and @[]@ are considered. +-- The solution adopted in this library is to omit @liftHash@ entirely. + diff --git a/src/Data/Hashable/LowLevel.hs b/src/Data/Hashable/LowLevel.hs new file mode 100644 --- /dev/null +++ b/src/Data/Hashable/LowLevel.hs @@ -0,0 +1,103 @@ +{-# LANGUAGE CPP, BangPatterns, MagicHash, CApiFFI, UnliftedFFITypes #-} +{-# LANGUAGE Trustworthy #-} +-- | A module containing low-level hash primitives. +module Data.Hashable.LowLevel ( + Salt, + defaultSalt, + hashInt, + hashInt64, + hashWord64, + hashPtrWithSalt, + hashByteArrayWithSalt, +) where + +#include "MachDeps.h" + +import Data.Array.Byte (ByteArray (..)) +import Foreign.Ptr (Ptr, castPtr) +import GHC.Base (ByteArray#) + +#ifdef HASHABLE_RANDOM_SEED +import System.IO.Unsafe (unsafePerformIO) +#endif + +import Data.Hashable.Imports +import Data.Hashable.Mix +import Data.Hashable.XXH3 + +------------------------------------------------------------------------------- +-- Initial seed +------------------------------------------------------------------------------- + +#ifdef HASHABLE_RANDOM_SEED +initialSeed :: Word64 +initialSeed = unsafePerformIO initialSeedC +{-# NOINLINE initialSeed #-} + +foreign import capi "HsHashable.h hs_hashable_init" initialSeedC :: IO Word64 +#endif + +-- | A default salt used in the implementation of 'hash'. +defaultSalt :: Salt +#ifdef HASHABLE_RANDOM_SEED +defaultSalt = hashInt defaultSalt' (fromIntegral initialSeed) +#else +defaultSalt = defaultSalt' +#endif +{-# INLINE defaultSalt #-} + +defaultSalt' :: Salt +#if WORD_SIZE_IN_BITS == 64 +defaultSalt' = -3750763034362895579 -- 14695981039346656037 :: Int64 +#else +defaultSalt' = -2128831035 -- 2166136261 :: Int32 +#endif +{-# INLINE defaultSalt' #-} + +------------------------------------------------------------------------------- +-- Hash primitives +------------------------------------------------------------------------------- + +-- | Hash 'Int'. First argument is a salt, second argument is an 'Int'. +-- The result is new salt / hash value. +hashInt :: Salt -> Int -> Salt +hashInt !s !x = fromIntegral (mixHash (fromIntegral s) (fromIntegral x)) + +hashInt64 :: Salt -> Int64 -> Salt +hashWord64 :: Salt -> Word64 -> Salt + +#if WORD_SIZE_IN_BITS == 64 +hashInt64 !s !x = hashInt s (fromIntegral x) +hashWord64 !s !x = hashInt s (fromIntegral x) +#else +hashInt64 !s !x = hashInt (hashInt s (fromIntegral x)) (fromIntegral (x `unsafeShiftR` 32)) +hashWord64 !s !x = hashInt (hashInt s (fromIntegral x)) (fromIntegral (x `unsafeShiftR` 32)) +#endif + +-- | Compute a hash value for the content of this pointer, using an +-- initial salt. +-- +-- This function can for example be used to hash non-contiguous +-- segments of memory as if they were one contiguous segment, by using +-- the output of one hash as the salt for the next. +hashPtrWithSalt :: Ptr a -- ^ pointer to the data to hash + -> Int -- ^ length, in bytes + -> Salt -- ^ salt + -> IO Salt -- ^ hash value +hashPtrWithSalt ptr len salt = + fromIntegral `fmap` xxh3_64bit_withSeed_ptr (castPtr ptr) len (fromIntegral salt) + +-- | Compute a hash value for the content of this 'ByteArray#', using +-- an initial salt. +-- +-- This function can for example be used to hash non-contiguous +-- segments of memory as if they were one contiguous segment, by using +-- the output of one hash as the salt for the next. +hashByteArrayWithSalt + :: ByteArray# -- ^ data to hash + -> Int -- ^ offset, in bytes + -> Int -- ^ length, in bytes + -> Salt -- ^ salt + -> Salt -- ^ hash value +hashByteArrayWithSalt ba !off !len !salt = + fromIntegral (xxh3_64bit_withSeed_ba (ByteArray ba) off len (fromIntegral salt)) diff --git a/src/Data/Hashable/Mix.hs b/src/Data/Hashable/Mix.hs new file mode 100644 --- /dev/null +++ b/src/Data/Hashable/Mix.hs @@ -0,0 +1,48 @@ +{-# LANGUAGE CPP #-} +{-# LANGUAGE MagicHash #-} +{-# LANGUAGE Trustworthy #-} +{-# LANGUAGE UnboxedTuples #-} +module Data.Hashable.Mix ( + Salt, + mixHash, +) where + +#include "MachDeps.h" + +import Data.Bits (unsafeShiftR, xor) +import GHC.Exts (Word (..), byteSwap#, timesWord2#, xor#) + +type Salt = Int + +mulFold :: Word -> Word -> Word +mulFold (W# x) (W# y) = case timesWord2# x y of + (# hi, lo #) -> W# (xor# hi lo) + +byteSwap :: Word -> Word +byteSwap (W# w) = W# (byteSwap# w) + +avalanche :: Word -> Word +avalanche z0 = +#if WORD_SIZE_IN_BITS == 64 + -- MurmurHash3Mixer + let z1 = shiftXorMultiply 33 0xff51afd7ed558ccd z0 + z2 = shiftXorMultiply 33 0xc4ceb9fe1a85ec53 z1 + z3 = shiftXor 33 z2 + in z3 +#else + -- MurmurHash3Mixer 32bit + let z1 = shiftXorMultiply 16 0x85ebca6b z0 + z2 = shiftXorMultiply 13 0xc2b2ae35 z1 + z3 = shiftXor 16 z2 + in z3 +#endif + +shiftXor :: Int -> Word -> Word +shiftXor n w = w `xor` (w `unsafeShiftR` n) + +shiftXorMultiply :: Int -> Word -> Word -> Word +shiftXorMultiply n k w = shiftXor n w * k + +-- | Mix hash is inspired by how xxh3 works on small (<=16byte) inputs. +mixHash :: Word -> Word -> Word +mixHash hi lo = avalanche (byteSwap lo + hi + mulFold hi lo) diff --git a/src/Data/Hashable/XXH3.hs b/src/Data/Hashable/XXH3.hs new file mode 100644 --- /dev/null +++ b/src/Data/Hashable/XXH3.hs @@ -0,0 +1,126 @@ +{-# LANGUAGE BangPatterns #-} +{-# LANGUAGE CPP #-} +{-# LANGUAGE MagicHash #-} +{-# LANGUAGE PatternSynonyms #-} +{-# LANGUAGE ScopedTypeVariables #-} +{-# LANGUAGE Trustworthy #-} +{-# LANGUAGE UnboxedTuples #-} +module Data.Hashable.XXH3 ( + -- * One shot + xxh3_64bit_withSeed_ptr, + xxh3_64bit_withSeed_bs, + xxh3_64bit_withSeed_ba, + xxh3_64bit_withSeed_w64, + xxh3_64bit_withSeed_w32, + -- * Incremental + XXH3_State, + xxh3_64bit_createState, + xxh3_64bit_reset_withSeed, + xxh3_64bit_digest, + xxh3_64bit_update_bs, + xxh3_64bit_update_ba, + xxh3_64bit_update_w64, + xxh3_64bit_update_w32, +) where + +import Control.Monad.ST.Unsafe (unsafeIOToST) +import Data.Array.Byte (ByteArray (..), MutableByteArray (..)) +import Data.ByteString.Internal (ByteString (..), accursedUnutterablePerformIO) +import Data.Word (Word32, Word64, Word8) +import Foreign (Ptr) +import GHC.Exts (Int (..), MutableByteArray#, newAlignedPinnedByteArray#) +import GHC.ForeignPtr (unsafeWithForeignPtr) +import GHC.ST (ST (..)) + +import Data.Hashable.FFI + +------------------------------------------------------------------------------- +-- OneShot +------------------------------------------------------------------------------- + +-- | Hash 'Ptr' +xxh3_64bit_withSeed_ptr :: Ptr Word8 -> Int -> Word64 -> IO Word64 +xxh3_64bit_withSeed_ptr !ptr !len !salt = + unsafe_xxh3_64bit_withSeed_ptr ptr (fromIntegral len) salt + +-- | Hash 'ByteString'. +xxh3_64bit_withSeed_bs :: ByteString -> Word64 -> Word64 +xxh3_64bit_withSeed_bs (BS fptr len) !salt = accursedUnutterablePerformIO $ + unsafeWithForeignPtr fptr $ \ptr -> + unsafe_xxh3_64bit_withSeed_ptr ptr (fromIntegral len) salt + +-- | Hash (part of) 'ByteArray'. +xxh3_64bit_withSeed_ba :: ByteArray -> Int -> Int -> Word64 -> Word64 +xxh3_64bit_withSeed_ba (ByteArray ba) !off !len !salt = + unsafe_xxh3_64bit_withSeed_ba ba (fromIntegral off) (fromIntegral len) salt + +-- | Hash 'Word64'. +xxh3_64bit_withSeed_w64 :: Word64 -> Word64 -> Word64 +xxh3_64bit_withSeed_w64 !x !salt = + unsafe_xxh3_64bit_withSeed_u64 x salt + +-- | Hash 'Word32'. +xxh3_64bit_withSeed_w32 :: Word32 -> Word64 -> Word64 +xxh3_64bit_withSeed_w32 !x !salt = + unsafe_xxh3_64bit_withSeed_u32 x salt + +------------------------------------------------------------------------------- +-- Incremental +------------------------------------------------------------------------------- + +-- | Mutable XXH3 state. +data XXH3_State s = XXH3 (MutableByteArray# s) + +-- | Create 'XXH3_State'. +xxh3_64bit_createState :: forall s. ST s (XXH3_State s) +xxh3_64bit_createState = do + -- aligned alloc, otherwise we get segfaults. + -- see XXH3_createState implementation + MutableByteArray ba <- newAlignedPinnedByteArray unsafe_xxh3_sizeof_state 64 + unsafeIOToST (unsafe_xxh3_initState ba) + return (XXH3 ba) + +-- | Reset 'XXH3_State' with a seed. +xxh3_64bit_reset_withSeed :: XXH3_State s -> Word64 -> ST s () +xxh3_64bit_reset_withSeed (XXH3 s) seed = do + unsafeIOToST (unsafe_xxh3_64bit_reset_withSeed s seed) + +-- | Return a hash value from a 'XXH3_State'. +-- +-- Doesn't mutate given state, so you can update, digest and update again. +xxh3_64bit_digest :: XXH3_State s -> ST s Word64 +xxh3_64bit_digest (XXH3 s) = + unsafeIOToST (unsafe_xxh3_64bit_digest s) + +-- | Update 'XXH3_State' with 'ByteString'. +xxh3_64bit_update_bs :: XXH3_State s -> ByteString -> ST s () +xxh3_64bit_update_bs (XXH3 s) (BS fptr len) = unsafeIOToST $ + unsafeWithForeignPtr fptr $ \ptr -> + unsafe_xxh3_64bit_update_ptr s ptr (fromIntegral len) + +-- | Update 'XXH3_State' with (part of) 'ByteArray' +xxh3_64bit_update_ba :: XXH3_State s -> ByteArray -> Int -> Int -> ST s () +xxh3_64bit_update_ba (XXH3 s) (ByteArray ba) !off !len = unsafeIOToST $ + unsafe_xxh3_64bit_update_ba s ba (fromIntegral off) (fromIntegral len) + +-- | Update 'XXH3_State' with 'Word64'. +xxh3_64bit_update_w64 :: XXH3_State s -> Word64 -> ST s () +xxh3_64bit_update_w64 (XXH3 s) w64 = unsafeIOToST $ + unsafe_xxh3_64bit_update_u64 s w64 + +-- | Update 'XXH3_State' with 'Word32'. +xxh3_64bit_update_w32 :: XXH3_State s -> Word32 -> ST s () +xxh3_64bit_update_w32 (XXH3 s) w32 = unsafeIOToST $ + unsafe_xxh3_64bit_update_u32 s w32 + +------------------------------------------------------------------------------- +-- mini-primitive +------------------------------------------------------------------------------- + +newAlignedPinnedByteArray + :: Int -- ^ size + -> Int -- ^ alignment + -> ST s (MutableByteArray s) +{-# INLINE newAlignedPinnedByteArray #-} +newAlignedPinnedByteArray (I# n) (I# k) = + ST (\s -> case newAlignedPinnedByteArray# n k s of (# s', arr #) -> (# s', MutableByteArray arr #)) diff --git a/tests/Main.hs b/tests/Main.hs --- a/tests/Main.hs +++ b/tests/Main.hs @@ -5,10 +5,10 @@ import Properties (properties) import Regress (regressions) -import Test.Framework (defaultMain, testGroup) +import Test.Tasty (defaultMain, testGroup) main :: IO () -main = defaultMain [ - testGroup "properties" properties - , testGroup "regressions" regressions - ] +main = defaultMain $ testGroup "hashable" + [ testGroup "properties" properties + , testGroup "regressions" regressions + ] diff --git a/tests/Properties.hs b/tests/Properties.hs --- a/tests/Properties.hs +++ b/tests/Properties.hs @@ -1,8 +1,6 @@ {-# LANGUAGE BangPatterns, CPP, GeneralizedNewtypeDeriving, MagicHash, Rank2Types, UnboxedTuples #-} -#ifdef GENERICS -{-# LANGUAGE DeriveGeneric, ScopedTypeVariables #-} -#endif +{-# LANGUAGE DeriveGeneric, ScopedTypeVariables, PackageImports #-} -- | QuickCheck tests for the 'Data.Hashable' module. We test -- functions by comparing the C and Haskell implementations. @@ -11,6 +9,7 @@ import Data.Hashable (Hashable, hash, hashByteArray, hashPtr, Hashed, hashed, unhashed, hashWithSalt) +import Data.Hashable.Generic (genericHashWithSalt) import Data.Hashable.Lifted (hashWithSalt1) import qualified Data.ByteString as B import qualified Data.ByteString.Lazy as BL @@ -20,21 +19,25 @@ import Control.Monad (ap, liftM) import System.IO.Unsafe (unsafePerformIO) import Foreign.Marshal.Array (withArray) -import GHC.Base (ByteArray#, Int(..), newByteArray#, unsafeCoerce#, - writeWord8Array#) +import GHC.Base (ByteArray#, Int(..), newByteArray#, writeWord8Array#) +import GHC.Exts (unsafeCoerce#) import GHC.ST (ST(..), runST) import GHC.Word (Word8(..)) import Test.QuickCheck hiding ((.&.)) -import Test.Framework (Test, testGroup) -import Test.Framework.Providers.QuickCheck2 (testProperty) -#ifdef GENERICS +import Test.Tasty (TestTree, testGroup) +import Test.Tasty.QuickCheck (testProperty) import GHC.Generics -#endif -#if MIN_VERSION_bytestring(0,10,4) import qualified Data.ByteString.Short as BS + +#if MIN_VERSION_filepath(1,4,100) && !(MIN_VERSION_filepath(1,5,0)) +import qualified "filepath" System.OsString.Internal.Types as FP #endif +#ifdef MIN_VERSION_os_string +import qualified "os-string" System.OsString.Internal.Types as OS +#endif + ------------------------------------------------------------------------ -- * Properties @@ -52,10 +55,8 @@ ((BL.fromChunks . map (B.pack . nonEmpty)) `fmap` arbitrary) where nonEmpty (NonEmpty a) = a -#if MIN_VERSION_bytestring(0,10,4) instance Arbitrary BS.ShortByteString where arbitrary = BS.pack `fmap` arbitrary -#endif -- | Validate the implementation by comparing the C and Haskell -- versions. @@ -85,16 +86,17 @@ -- | Ensure that the rechunk function causes a rechunked string to -- still match its original form. -pTextRechunk :: T.Text -> NonEmptyList ChunkSize -> Bool -pTextRechunk t cs = TL.fromStrict t == rechunkText t cs +pTextRechunk :: T.Text -> NonEmptyList ChunkSize -> Property +pTextRechunk t cs = TL.fromStrict t === rechunkText t cs -- | Lazy strings must hash to the same value no matter how they are -- chunked. -pTextLazyRechunked :: T.Text - -> NonEmptyList ChunkSize -> NonEmptyList ChunkSize -> Bool -pTextLazyRechunked t cs0 cs1 = - hash (rechunkText t cs0) == hash (rechunkText t cs1) +pTextLazyRechunked :: T.Text -> NonEmptyList ChunkSize -> NonEmptyList ChunkSize -> Property +pTextLazyRechunked t cs0 cs1 = hash (rechunkText t cs0) === hash (rechunkText t cs1) +pTextLazyRechunked' :: T.Text -> Int -> NonEmptyList ChunkSize -> NonEmptyList ChunkSize -> Property +pTextLazyRechunked' t salt cs0 cs1 = hashWithSalt salt (rechunkText t cs0) === hashWithSalt salt (rechunkText t cs1) + -- | Break up a string into chunks of different sizes. rechunkText :: T.Text -> NonEmptyList ChunkSize -> TL.Text rechunkText t0 (NonEmpty cs0) = TL.fromChunks . go t0 . cycle $ cs0 @@ -104,11 +106,9 @@ where (a,b) = T.splitAt (unCS c) t go _ [] = error "Properties.rechunk - The 'impossible' happened!" -#if MIN_VERSION_bytestring(0,10,4) -- | Content equality implies hash equality. pBSShort :: BS.ShortByteString -> BS.ShortByteString -> Bool pBSShort a b = if (a == b) then (hash a == hash b) else True -#endif -- | Content equality implies hash equality. pBS :: B.ByteString -> B.ByteString -> Bool @@ -134,10 +134,12 @@ -- | Lazy bytestrings must hash to the same value no matter how they -- are chunked. -pBSLazyRechunked :: B.ByteString - -> NonEmptyList ChunkSize -> NonEmptyList ChunkSize -> Bool -pBSLazyRechunked t cs1 cs2 = hash (rechunkBS t cs1) == hash (rechunkBS t cs2) +pBSLazyRechunked :: B.ByteString -> NonEmptyList ChunkSize -> NonEmptyList ChunkSize -> Property +pBSLazyRechunked t cs1 cs2 = hash (rechunkBS t cs1) === hash (rechunkBS t cs2) +pBSLazyRechunked' :: B.ByteString -> Int -> NonEmptyList ChunkSize -> NonEmptyList ChunkSize -> Property +pBSLazyRechunked' t salt cs1 cs2 = hashWithSalt salt (rechunkBS t cs1) === hashWithSalt salt (rechunkBS t cs2) + -- This wrapper is required by 'runST'. data ByteArray = BA { unBA :: ByteArray# } @@ -156,10 +158,8 @@ -- Generics -#ifdef GENERICS - data Product2 a b = Product2 a b - deriving (Generic) + deriving (Eq, Generic) instance (Arbitrary a, Arbitrary b) => Arbitrary (Product2 a b) where arbitrary = Product2 `liftM` arbitrary `ap` arbitrary @@ -167,7 +167,7 @@ instance (Hashable a, Hashable b) => Hashable (Product2 a b) data Product3 a b c = Product3 a b c - deriving (Generic) + deriving (Eq, Generic) instance (Arbitrary a, Arbitrary b, Arbitrary c) => Arbitrary (Product3 a b c) where @@ -191,6 +191,13 @@ data Sum3 a b c = S3a a | S3b b | S3c c deriving (Eq, Ord, Show, Generic) +instance (Arbitrary a, Arbitrary b, Arbitrary c) => Arbitrary (Sum3 a b c) where + arbitrary = oneof + [ fmap S3a arbitrary + , fmap S3b arbitrary + , fmap S3c arbitrary + ] + instance (Hashable a, Hashable b, Hashable c) => Hashable (Sum3 a b c) -- Hashes of the same parameter, but with different sum constructors, @@ -208,7 +215,8 @@ , hash (S3b x :: Sum3 Int Int Int) , hash (S3c x :: Sum3 Int Int Int) ] -#endif +pGeneric :: Sum3 Int Bool String -> Int -> Bool +pGeneric x salt = hashWithSalt salt x == genericHashWithSalt salt x instance (Arbitrary a, Hashable a) => Arbitrary (Hashed a) where arbitrary = fmap hashed arbitrary @@ -217,7 +225,7 @@ pLiftedHashed :: Int -> Hashed (Either Int String) -> Bool pLiftedHashed s h = hashWithSalt s h == hashWithSalt1 s h -properties :: [Test] +properties :: [TestTree] properties = [ testProperty "bernstein" pHash , testGroup "text" @@ -225,17 +233,16 @@ , testProperty "text/lazy" pTextLazy , testProperty "text/rechunk" pTextRechunk , testProperty "text/rechunked" pTextLazyRechunked + , testProperty "text/rechunked-salt" pTextLazyRechunked' ] , testGroup "bytestring" [ testProperty "bytestring/strict" pBS , testProperty "bytestring/lazy" pBSLazy -#if MIN_VERSION_bytestring(0,10,4) , testProperty "bytestring/short" pBSShort -#endif , testProperty "bytestring/rechunk" pBSRechunk , testProperty "bytestring/rechunked" pBSLazyRechunked + , testProperty "bytestring/rechunked-salt" pBSLazyRechunked' ] -#ifdef GENERICS , testGroup "generics" [ -- Note: "product2" and "product3" have been temporarily @@ -246,8 +253,8 @@ -- "product3" pProduct3 testProperty "sum2_differ" pSum2_differ , testProperty "sum3_differ" pSum3_differ + , testProperty "genericHashWithSalt" pGeneric ] -#endif , testGroup "lifted law" [ testProperty "Hashed" pLiftedHashed ] @@ -257,8 +264,24 @@ -- Utilities fromStrict :: B.ByteString -> BL.ByteString -#if MIN_VERSION_bytestring(0,10,0) fromStrict = BL.fromStrict -#else -fromStrict b = BL.fromChunks [b] + +------------------------------------------------------------------------ +-- test that instances exist + +instanceExists :: Hashable a => a -> () +instanceExists _ = () + +#if MIN_VERSION_filepath(1,4,100) && !(MIN_VERSION_filepath(1,5,0)) +_fp1, _fp2, _fp3 :: () +_fp1 = instanceExists (undefined :: FP.OsString) +_fp2 = instanceExists (undefined :: FP.WindowsString) +_fp3 = instanceExists (undefined :: FP.PosixString) +#endif + +#ifdef MIN_VERSION_os_string +_os1, _os2, _os3 :: () +_os1 = instanceExists (undefined :: OS.OsString) +_os2 = instanceExists (undefined :: OS.WindowsString) +_os3 = instanceExists (undefined :: OS.PosixString) #endif diff --git a/tests/Regress.hs b/tests/Regress.hs --- a/tests/Regress.hs +++ b/tests/Regress.hs @@ -1,36 +1,117 @@ {-# LANGUAGE CPP #-} {-# LANGUAGE DeriveGeneric #-} +{-# LANGUAGE OverloadedStrings #-} module Regress (regressions) where -import qualified Test.Framework as F -import Test.Framework.Providers.HUnit (testCase) -import Test.HUnit ((@=?)) +import Test.Tasty (TestTree, testGroup) +import Control.Monad (when) +import Test.Tasty.HUnit (testCase, Assertion, assertFailure, (@?=)) +import Test.Tasty.QuickCheck (testProperty) import GHC.Generics (Generic) import Data.List (nub) +import Data.Fixed (Pico) +import Data.Text (Text) +import Data.ByteString (ByteString) +import qualified Data.Text.Lazy as TL +import qualified Data.ByteString.Char8 as BS8 +import qualified Data.ByteString.Lazy as BSL +import qualified Data.ByteString.Lazy.Char8 as BSL8 + #ifdef HAVE_MMAP import qualified Regress.Mmap as Mmap #endif import Data.Hashable -regressions :: [F.Test] +#include "MachDeps.h" + +assertInequal :: Eq a => String -> a -> a -> Assertion +assertInequal msg x y + | x == y = assertFailure msg + | otherwise = return () + +regressions :: [TestTree] regressions = [] ++ #ifdef HAVE_MMAP Mmap.regressions ++ + [ testCase "Fixed" $ do + (hash (1 :: Pico) == hash (2 :: Pico)) @?= False + ] ++ #endif - [ F.testGroup "Generic: sum of nullary constructors" + [ testGroup "Generic: sum of nullary constructors" [ testCase "0" $ nullaryCase 0 S0 , testCase "1" $ nullaryCase 1 S1 , testCase "2" $ nullaryCase 2 S2 , testCase "3" $ nullaryCase 3 S3 , testCase "4" $ nullaryCase 4 S4 ] + + , testCase "Zero tuples: issue 271" $ do + assertInequal "Hash of (0,0) != 0" (hash (0 :: Int, 0 :: Int)) 0 + assertInequal "Hash of (0,0,0) != 0" (hash (0 :: Int, 0 :: Int, 0 :: Int)) 0 + + , testProperty "odd, odd: issue 271" $ \x' y' -> + let x = if odd x' then x' else x' + 1 :: Int + y = if odd y' then y' else y' + 1 :: Int + in hash (x, y) /= hash (negate x, negate y) + , testCase "Generic: Peano https://github.com/tibbe/hashable/issues/135" $ do let ns = take 20 $ iterate S Z let hs = map hash ns - hs @=? nub hs + hs @?= nub hs +#if WORD_SIZE_IN_BITS == 64 + , testCase "64 bit Text" $ do + let expected = +#if MIN_VERSION_text(2,0,0) + -3150353794653054837 +#else + 660667291861873677 +#endif + hash ("hello world" :: Text) @?= expected +#endif + , testGroup "concatenation" + [ testCase "String" $ do + let lhs, rhs :: (String, String) + lhs = ("foo", "bar") + rhs = ("foobar", "") + + when (hash lhs == hash rhs) $ do + assertFailure "Should have different hashes" + + , testCase "Text" $ do + let lhs, rhs :: (Text, Text) + lhs = ("foo", "bar") + rhs = ("foobar", "") + + when (hash lhs == hash rhs) $ do + assertFailure "Should have different hashes" + + , testCase "Lazy Text" $ do + let lhs, rhs :: (TL.Text, TL.Text) + lhs = ("foo", "bar") + rhs = ("foobar", "") + + when (hash lhs == hash rhs) $ do + assertFailure "Should have different hashes" + + , testCase "ByteString" $ do + let lhs, rhs :: (ByteString, ByteString) + lhs = (BS8.pack "foo", BS8.pack "bar") + rhs = (BS8.pack "foobar", BS8.empty) + + when (hash lhs == hash rhs) $ do + assertFailure "Should have different hashes" + + , testCase "Lazy ByteString" $ do + let lhs, rhs :: (BSL.ByteString, BSL.ByteString) + lhs = (BSL8.pack "foo", BSL8.pack "bar") + rhs = (BSL8.pack "foobar", BSL.empty) + + when (hash lhs == hash rhs) $ do + assertFailure "Should have different hashes" + ] ] where nullaryCase :: Int -> SumOfNullary -> IO () @@ -38,10 +119,10 @@ let salt = 42 let expected = salt `hashWithSalt` n `hashWithSalt` () let actual = hashWithSalt salt s - expected @=? actual + actual @?= expected -data SumOfNullary = S0 | S1 | S2 | S3 | S4 deriving (Generic) +data SumOfNullary = S0 | S1 | S2 | S3 | S4 deriving (Eq, Generic) instance Hashable SumOfNullary -data Nat = Z | S Nat deriving (Generic) +data Nat = Z | S Nat deriving (Eq, Generic) instance Hashable Nat diff --git a/tests/Regress/Mmap.hsc b/tests/Regress/Mmap.hsc --- a/tests/Regress/Mmap.hsc +++ b/tests/Regress/Mmap.hsc @@ -1,4 +1,4 @@ -{-# LANGUAGE ForeignFunctionInterface #-} +{-# LANGUAGE CApiFFI #-} module Regress.Mmap (regressions) where @@ -14,8 +14,8 @@ import Foreign.Ptr (Ptr, intPtrToPtr, nullPtr, plusPtr) import GHC.ForeignPtr (newForeignPtr_) import System.Posix.Types (COff(..)) -import Test.Framework (Test) -import Test.Framework.Providers.HUnit (testCase) +import Test.Tasty (TestTree) +import Test.Tasty.HUnit (testCase) import qualified Data.ByteString as B withMapping :: (Ptr a -> Int -> IO ()) -> IO () @@ -43,7 +43,7 @@ forM_ (B.tails bs0) $ \bs -> do evaluate (hash bs) -regressions :: [Test] +regressions :: [TestTree] regressions = [ testCase "hashNearPageBoundary" hashNearPageBoundary ] @@ -60,14 +60,14 @@ mprotect addr len prot = throwErrnoIfMinus1_ "mprotect" $ c_mprotect addr len prot -foreign import ccall unsafe "sys/mman.h mmap" +foreign import capi unsafe "sys/mman.h mmap" c_mmap :: Ptr a -> CSize -> CInt -> CInt -> CInt -> COff -> IO (Ptr a) -foreign import ccall unsafe "sys/mman.h munmap" +foreign import capi unsafe "sys/mman.h munmap" c_munmap :: Ptr a -> CSize -> IO CInt -foreign import ccall unsafe "sys/mman.h mprotect" +foreign import capi unsafe "sys/mman.h mprotect" c_mprotect :: Ptr a -> CSize -> CInt -> IO CInt -foreign import ccall unsafe "unistd.h getpagesize" +foreign import capi unsafe "unistd.h getpagesize" getPageSize :: IO CInt diff --git a/tests/xxhash-tests.hs b/tests/xxhash-tests.hs new file mode 100644 --- /dev/null +++ b/tests/xxhash-tests.hs @@ -0,0 +1,63 @@ +{-# LANGUAGE NumericUnderscores #-} +module Main (main) where + +import Control.Monad.ST (runST) +import qualified Data.ByteString as BS +import qualified Data.Primitive as P +import Data.Word (Word32, Word64) +import Test.Tasty (defaultMain, testGroup) +import Test.Tasty.HUnit (testCase, (@=?)) +import Test.Tasty.QuickCheck (testProperty, (===)) + +import Data.Hashable.XXH3 + +main :: IO () +main = defaultMain $ testGroup "xxhash" + [ testGroup "oneshot" + [ testProperty "w64-ref" $ \w salt -> + xxh3_64bit_withSeed_w64 w salt === xxh3_64bit_withSeed_w64_ref w salt + , testCase "w64-examples" $ do + xxh3_64bit_withSeed_w64 0 0 @=? 0xc77b_3abb_6f87_acd9 + xxh3_64bit_withSeed_w64 0x12 1 @=? 0xbba4_8522_c425_46b2 + xxh3_64bit_withSeed_w64 0x2100_0000_0000_0000 0 @=? 0xb7cb_e42a_e127_8055 + xxh3_64bit_withSeed_w64 0x1eb6e9 0 @=? 0x8e_adc3_1b56 + + , testProperty "w32-ref" $ \w salt -> + xxh3_64bit_withSeed_w32 w salt === xxh3_64bit_withSeed_w32_ref w salt + + , testCase "w32-examples" $ do + xxh3_64bit_withSeed_w32 0 0 @=? 0x48b2_c926_16fc_193d + xxh3_64bit_withSeed_w32 0x12 1 @=? 0x2870_1df3_2a21_6ad3 + + ] + + , testGroup "incremental" + [ testProperty "empty" $ \seed -> do + let expected = xxh3_64bit_withSeed_bs BS.empty seed + let actual = runST $ do + s <- xxh3_64bit_createState + xxh3_64bit_reset_withSeed s seed + xxh3_64bit_digest s + + actual === expected + + , testProperty "bs" $ \w8s seed -> do + let bs = BS.pack w8s + let expected = xxh3_64bit_withSeed_bs bs seed + let actual = runST $ do + s <- xxh3_64bit_createState + xxh3_64bit_reset_withSeed s seed + xxh3_64bit_update_bs s bs + xxh3_64bit_digest s + + actual === expected + ] + ] + +xxh3_64bit_withSeed_w64_ref :: Word64 -> Word64 -> Word64 +xxh3_64bit_withSeed_w64_ref w salt = case P.primArrayFromList [w] of + P.PrimArray ba -> xxh3_64bit_withSeed_ba (P.ByteArray ba) 0 8 salt + +xxh3_64bit_withSeed_w32_ref :: Word32 -> Word64 -> Word64 +xxh3_64bit_withSeed_w32_ref w salt = case P.primArrayFromList [w] of + P.PrimArray ba -> xxh3_64bit_withSeed_ba (P.ByteArray ba) 0 4 salt diff --git a/xxHash-0.8.3/xxhash.h b/xxHash-0.8.3/xxhash.h new file mode 100644 --- /dev/null +++ b/xxHash-0.8.3/xxhash.h @@ -0,0 +1,7238 @@ +/* + * xxHash - Extremely Fast Hash algorithm + * Header File + * Copyright (C) 2012-2023 Yann Collet + * + * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php) + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following disclaimer + * in the documentation and/or other materials provided with the + * distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * You can contact the author at: + * - xxHash homepage: https://www.xxhash.com + * - xxHash source repository: https://github.com/Cyan4973/xxHash + */ + +/*! + * @mainpage xxHash + * + * xxHash is an extremely fast non-cryptographic hash algorithm, working at RAM speed + * limits. + * + * It is proposed in four flavors, in three families: + * 1. @ref XXH32_family + * - Classic 32-bit hash function. Simple, compact, and runs on almost all + * 32-bit and 64-bit systems. + * 2. @ref XXH64_family + * - Classic 64-bit adaptation of XXH32. Just as simple, and runs well on most + * 64-bit systems (but _not_ 32-bit systems). + * 3. @ref XXH3_family + * - Modern 64-bit and 128-bit hash function family which features improved + * strength and performance across the board, especially on smaller data. + * It benefits greatly from SIMD and 64-bit without requiring it. + * + * Benchmarks + * --- + * The reference system uses an Intel i7-9700K CPU, and runs Ubuntu x64 20.04. + * The open source benchmark program is compiled with clang v10.0 using -O3 flag. + * + * | Hash Name | ISA ext | Width | Large Data Speed | Small Data Velocity | + * | -------------------- | ------- | ----: | ---------------: | ------------------: | + * | XXH3_64bits() | @b AVX2 | 64 | 59.4 GB/s | 133.1 | + * | MeowHash | AES-NI | 128 | 58.2 GB/s | 52.5 | + * | XXH3_128bits() | @b AVX2 | 128 | 57.9 GB/s | 118.1 | + * | CLHash | PCLMUL | 64 | 37.1 GB/s | 58.1 | + * | XXH3_64bits() | @b SSE2 | 64 | 31.5 GB/s | 133.1 | + * | XXH3_128bits() | @b SSE2 | 128 | 29.6 GB/s | 118.1 | + * | RAM sequential read | | N/A | 28.0 GB/s | N/A | + * | ahash | AES-NI | 64 | 22.5 GB/s | 107.2 | + * | City64 | | 64 | 22.0 GB/s | 76.6 | + * | T1ha2 | | 64 | 22.0 GB/s | 99.0 | + * | City128 | | 128 | 21.7 GB/s | 57.7 | + * | FarmHash | AES-NI | 64 | 21.3 GB/s | 71.9 | + * | XXH64() | | 64 | 19.4 GB/s | 71.0 | + * | SpookyHash | | 64 | 19.3 GB/s | 53.2 | + * | Mum | | 64 | 18.0 GB/s | 67.0 | + * | CRC32C | SSE4.2 | 32 | 13.0 GB/s | 57.9 | + * | XXH32() | | 32 | 9.7 GB/s | 71.9 | + * | City32 | | 32 | 9.1 GB/s | 66.0 | + * | Blake3* | @b AVX2 | 256 | 4.4 GB/s | 8.1 | + * | Murmur3 | | 32 | 3.9 GB/s | 56.1 | + * | SipHash* | | 64 | 3.0 GB/s | 43.2 | + * | Blake3* | @b SSE2 | 256 | 2.4 GB/s | 8.1 | + * | HighwayHash | | 64 | 1.4 GB/s | 6.0 | + * | FNV64 | | 64 | 1.2 GB/s | 62.7 | + * | Blake2* | | 256 | 1.1 GB/s | 5.1 | + * | SHA1* | | 160 | 0.8 GB/s | 5.6 | + * | MD5* | | 128 | 0.6 GB/s | 7.8 | + * @note + * - Hashes which require a specific ISA extension are noted. SSE2 is also noted, + * even though it is mandatory on x64. + * - Hashes with an asterisk are cryptographic. Note that MD5 is non-cryptographic + * by modern standards. + * - Small data velocity is a rough average of algorithm's efficiency for small + * data. For more accurate information, see the wiki. + * - More benchmarks and strength tests are found on the wiki: + * https://github.com/Cyan4973/xxHash/wiki + * + * Usage + * ------ + * All xxHash variants use a similar API. Changing the algorithm is a trivial + * substitution. + * + * @pre + * For functions which take an input and length parameter, the following + * requirements are assumed: + * - The range from [`input`, `input + length`) is valid, readable memory. + * - The only exception is if the `length` is `0`, `input` may be `NULL`. + * - For C++, the objects must have the *TriviallyCopyable* property, as the + * functions access bytes directly as if it was an array of `unsigned char`. + * + * @anchor single_shot_example + * **Single Shot** + * + * These functions are stateless functions which hash a contiguous block of memory, + * immediately returning the result. They are the easiest and usually the fastest + * option. + * + * XXH32(), XXH64(), XXH3_64bits(), XXH3_128bits() + * + * @code{.c} + * #include + * #include "xxhash.h" + * + * // Example for a function which hashes a null terminated string with XXH32(). + * XXH32_hash_t hash_string(const char* string, XXH32_hash_t seed) + * { + * // NULL pointers are only valid if the length is zero + * size_t length = (string == NULL) ? 0 : strlen(string); + * return XXH32(string, length, seed); + * } + * @endcode + * + * + * @anchor streaming_example + * **Streaming** + * + * These groups of functions allow incremental hashing of unknown size, even + * more than what would fit in a size_t. + * + * XXH32_reset(), XXH64_reset(), XXH3_64bits_reset(), XXH3_128bits_reset() + * + * @code{.c} + * #include + * #include + * #include "xxhash.h" + * // Example for a function which hashes a FILE incrementally with XXH3_64bits(). + * XXH64_hash_t hashFile(FILE* f) + * { + * // Allocate a state struct. Do not just use malloc() or new. + * XXH3_state_t* state = XXH3_createState(); + * assert(state != NULL && "Out of memory!"); + * // Reset the state to start a new hashing session. + * XXH3_64bits_reset(state); + * char buffer[4096]; + * size_t count; + * // Read the file in chunks + * while ((count = fread(buffer, 1, sizeof(buffer), f)) != 0) { + * // Run update() as many times as necessary to process the data + * XXH3_64bits_update(state, buffer, count); + * } + * // Retrieve the finalized hash. This will not change the state. + * XXH64_hash_t result = XXH3_64bits_digest(state); + * // Free the state. Do not use free(). + * XXH3_freeState(state); + * return result; + * } + * @endcode + * + * Streaming functions generate the xxHash value from an incremental input. + * This method is slower than single-call functions, due to state management. + * For small inputs, prefer `XXH32()` and `XXH64()`, which are better optimized. + * + * An XXH state must first be allocated using `XXH*_createState()`. + * + * Start a new hash by initializing the state with a seed using `XXH*_reset()`. + * + * Then, feed the hash state by calling `XXH*_update()` as many times as necessary. + * + * The function returns an error code, with 0 meaning OK, and any other value + * meaning there is an error. + * + * Finally, a hash value can be produced anytime, by using `XXH*_digest()`. + * This function returns the nn-bits hash as an int or long long. + * + * It's still possible to continue inserting input into the hash state after a + * digest, and generate new hash values later on by invoking `XXH*_digest()`. + * + * When done, release the state using `XXH*_freeState()`. + * + * + * @anchor canonical_representation_example + * **Canonical Representation** + * + * The default return values from XXH functions are unsigned 32, 64 and 128 bit + * integers. + * This the simplest and fastest format for further post-processing. + * + * However, this leaves open the question of what is the order on the byte level, + * since little and big endian conventions will store the same number differently. + * + * The canonical representation settles this issue by mandating big-endian + * convention, the same convention as human-readable numbers (large digits first). + * + * When writing hash values to storage, sending them over a network, or printing + * them, it's highly recommended to use the canonical representation to ensure + * portability across a wider range of systems, present and future. + * + * The following functions allow transformation of hash values to and from + * canonical format. + * + * XXH32_canonicalFromHash(), XXH32_hashFromCanonical(), + * XXH64_canonicalFromHash(), XXH64_hashFromCanonical(), + * XXH128_canonicalFromHash(), XXH128_hashFromCanonical(), + * + * @code{.c} + * #include + * #include "xxhash.h" + * + * // Example for a function which prints XXH32_hash_t in human readable format + * void printXxh32(XXH32_hash_t hash) + * { + * XXH32_canonical_t cano; + * XXH32_canonicalFromHash(&cano, hash); + * size_t i; + * for(i = 0; i < sizeof(cano.digest); ++i) { + * printf("%02x", cano.digest[i]); + * } + * printf("\n"); + * } + * + * // Example for a function which converts XXH32_canonical_t to XXH32_hash_t + * XXH32_hash_t convertCanonicalToXxh32(XXH32_canonical_t cano) + * { + * XXH32_hash_t hash = XXH32_hashFromCanonical(&cano); + * return hash; + * } + * @endcode + * + * + * @file xxhash.h + * xxHash prototypes and implementation + */ + +#if defined (__cplusplus) +extern "C" { +#endif + +/* **************************** + * INLINE mode + ******************************/ +/*! + * @defgroup public Public API + * Contains details on the public xxHash functions. + * @{ + */ +#ifdef XXH_DOXYGEN +/*! + * @brief Gives access to internal state declaration, required for static allocation. + * + * Incompatible with dynamic linking, due to risks of ABI changes. + * + * Usage: + * @code{.c} + * #define XXH_STATIC_LINKING_ONLY + * #include "xxhash.h" + * @endcode + */ +# define XXH_STATIC_LINKING_ONLY +/* Do not undef XXH_STATIC_LINKING_ONLY for Doxygen */ + +/*! + * @brief Gives access to internal definitions. + * + * Usage: + * @code{.c} + * #define XXH_STATIC_LINKING_ONLY + * #define XXH_IMPLEMENTATION + * #include "xxhash.h" + * @endcode + */ +# define XXH_IMPLEMENTATION +/* Do not undef XXH_IMPLEMENTATION for Doxygen */ + +/*! + * @brief Exposes the implementation and marks all functions as `inline`. + * + * Use these build macros to inline xxhash into the target unit. + * Inlining improves performance on small inputs, especially when the length is + * expressed as a compile-time constant: + * + * https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html + * + * It also keeps xxHash symbols private to the unit, so they are not exported. + * + * Usage: + * @code{.c} + * #define XXH_INLINE_ALL + * #include "xxhash.h" + * @endcode + * Do not compile and link xxhash.o as a separate object, as it is not useful. + */ +# define XXH_INLINE_ALL +# undef XXH_INLINE_ALL +/*! + * @brief Exposes the implementation without marking functions as inline. + */ +# define XXH_PRIVATE_API +# undef XXH_PRIVATE_API +/*! + * @brief Emulate a namespace by transparently prefixing all symbols. + * + * If you want to include _and expose_ xxHash functions from within your own + * library, but also want to avoid symbol collisions with other libraries which + * may also include xxHash, you can use @ref XXH_NAMESPACE to automatically prefix + * any public symbol from xxhash library with the value of @ref XXH_NAMESPACE + * (therefore, avoid empty or numeric values). + * + * Note that no change is required within the calling program as long as it + * includes `xxhash.h`: Regular symbol names will be automatically translated + * by this header. + */ +# define XXH_NAMESPACE /* YOUR NAME HERE */ +# undef XXH_NAMESPACE +#endif + +#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)) \ + && !defined(XXH_INLINE_ALL_31684351384) + /* this section should be traversed only once */ +# define XXH_INLINE_ALL_31684351384 + /* give access to the advanced API, required to compile implementations */ +# undef XXH_STATIC_LINKING_ONLY /* avoid macro redef */ +# define XXH_STATIC_LINKING_ONLY + /* make all functions private */ +# undef XXH_PUBLIC_API +# if defined(__GNUC__) +# define XXH_PUBLIC_API static __inline __attribute__((__unused__)) +# elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) +# define XXH_PUBLIC_API static inline +# elif defined(_MSC_VER) +# define XXH_PUBLIC_API static __inline +# else + /* note: this version may generate warnings for unused static functions */ +# define XXH_PUBLIC_API static +# endif + + /* + * This part deals with the special case where a unit wants to inline xxHash, + * but "xxhash.h" has previously been included without XXH_INLINE_ALL, + * such as part of some previously included *.h header file. + * Without further action, the new include would just be ignored, + * and functions would effectively _not_ be inlined (silent failure). + * The following macros solve this situation by prefixing all inlined names, + * avoiding naming collision with previous inclusions. + */ + /* Before that, we unconditionally #undef all symbols, + * in case they were already defined with XXH_NAMESPACE. + * They will then be redefined for XXH_INLINE_ALL + */ +# undef XXH_versionNumber + /* XXH32 */ +# undef XXH32 +# undef XXH32_createState +# undef XXH32_freeState +# undef XXH32_reset +# undef XXH32_update +# undef XXH32_digest +# undef XXH32_copyState +# undef XXH32_canonicalFromHash +# undef XXH32_hashFromCanonical + /* XXH64 */ +# undef XXH64 +# undef XXH64_createState +# undef XXH64_freeState +# undef XXH64_reset +# undef XXH64_update +# undef XXH64_digest +# undef XXH64_copyState +# undef XXH64_canonicalFromHash +# undef XXH64_hashFromCanonical + /* XXH3_64bits */ +# undef XXH3_64bits +# undef XXH3_64bits_withSecret +# undef XXH3_64bits_withSeed +# undef XXH3_64bits_withSecretandSeed +# undef XXH3_createState +# undef XXH3_freeState +# undef XXH3_copyState +# undef XXH3_64bits_reset +# undef XXH3_64bits_reset_withSeed +# undef XXH3_64bits_reset_withSecret +# undef XXH3_64bits_update +# undef XXH3_64bits_digest +# undef XXH3_generateSecret + /* XXH3_128bits */ +# undef XXH128 +# undef XXH3_128bits +# undef XXH3_128bits_withSeed +# undef XXH3_128bits_withSecret +# undef XXH3_128bits_reset +# undef XXH3_128bits_reset_withSeed +# undef XXH3_128bits_reset_withSecret +# undef XXH3_128bits_reset_withSecretandSeed +# undef XXH3_128bits_update +# undef XXH3_128bits_digest +# undef XXH128_isEqual +# undef XXH128_cmp +# undef XXH128_canonicalFromHash +# undef XXH128_hashFromCanonical + /* Finally, free the namespace itself */ +# undef XXH_NAMESPACE + + /* employ the namespace for XXH_INLINE_ALL */ +# define XXH_NAMESPACE XXH_INLINE_ + /* + * Some identifiers (enums, type names) are not symbols, + * but they must nonetheless be renamed to avoid redeclaration. + * Alternative solution: do not redeclare them. + * However, this requires some #ifdefs, and has a more dispersed impact. + * Meanwhile, renaming can be achieved in a single place. + */ +# define XXH_IPREF(Id) XXH_NAMESPACE ## Id +# define XXH_OK XXH_IPREF(XXH_OK) +# define XXH_ERROR XXH_IPREF(XXH_ERROR) +# define XXH_errorcode XXH_IPREF(XXH_errorcode) +# define XXH32_canonical_t XXH_IPREF(XXH32_canonical_t) +# define XXH64_canonical_t XXH_IPREF(XXH64_canonical_t) +# define XXH128_canonical_t XXH_IPREF(XXH128_canonical_t) +# define XXH32_state_s XXH_IPREF(XXH32_state_s) +# define XXH32_state_t XXH_IPREF(XXH32_state_t) +# define XXH64_state_s XXH_IPREF(XXH64_state_s) +# define XXH64_state_t XXH_IPREF(XXH64_state_t) +# define XXH3_state_s XXH_IPREF(XXH3_state_s) +# define XXH3_state_t XXH_IPREF(XXH3_state_t) +# define XXH128_hash_t XXH_IPREF(XXH128_hash_t) + /* Ensure the header is parsed again, even if it was previously included */ +# undef XXHASH_H_5627135585666179 +# undef XXHASH_H_STATIC_13879238742 +#endif /* XXH_INLINE_ALL || XXH_PRIVATE_API */ + +/* **************************************************************** + * Stable API + *****************************************************************/ +#ifndef XXHASH_H_5627135585666179 +#define XXHASH_H_5627135585666179 1 + +/*! @brief Marks a global symbol. */ +#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API) +# if defined(_WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT)) +# ifdef XXH_EXPORT +# define XXH_PUBLIC_API __declspec(dllexport) +# elif XXH_IMPORT +# define XXH_PUBLIC_API __declspec(dllimport) +# endif +# else +# define XXH_PUBLIC_API /* do nothing */ +# endif +#endif + +#ifdef XXH_NAMESPACE +# define XXH_CAT(A,B) A##B +# define XXH_NAME2(A,B) XXH_CAT(A,B) +# define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber) +/* XXH32 */ +# define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32) +# define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState) +# define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState) +# define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset) +# define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update) +# define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest) +# define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState) +# define XXH32_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash) +# define XXH32_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical) +/* XXH64 */ +# define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64) +# define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState) +# define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState) +# define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset) +# define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update) +# define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest) +# define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState) +# define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash) +# define XXH64_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical) +/* XXH3_64bits */ +# define XXH3_64bits XXH_NAME2(XXH_NAMESPACE, XXH3_64bits) +# define XXH3_64bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecret) +# define XXH3_64bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSeed) +# define XXH3_64bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecretandSeed) +# define XXH3_createState XXH_NAME2(XXH_NAMESPACE, XXH3_createState) +# define XXH3_freeState XXH_NAME2(XXH_NAMESPACE, XXH3_freeState) +# define XXH3_copyState XXH_NAME2(XXH_NAMESPACE, XXH3_copyState) +# define XXH3_64bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset) +# define XXH3_64bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSeed) +# define XXH3_64bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecret) +# define XXH3_64bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecretandSeed) +# define XXH3_64bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_update) +# define XXH3_64bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_digest) +# define XXH3_generateSecret XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret) +# define XXH3_generateSecret_fromSeed XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret_fromSeed) +/* XXH3_128bits */ +# define XXH128 XXH_NAME2(XXH_NAMESPACE, XXH128) +# define XXH3_128bits XXH_NAME2(XXH_NAMESPACE, XXH3_128bits) +# define XXH3_128bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSeed) +# define XXH3_128bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecret) +# define XXH3_128bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecretandSeed) +# define XXH3_128bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset) +# define XXH3_128bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSeed) +# define XXH3_128bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecret) +# define XXH3_128bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecretandSeed) +# define XXH3_128bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_update) +# define XXH3_128bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_digest) +# define XXH128_isEqual XXH_NAME2(XXH_NAMESPACE, XXH128_isEqual) +# define XXH128_cmp XXH_NAME2(XXH_NAMESPACE, XXH128_cmp) +# define XXH128_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH128_canonicalFromHash) +# define XXH128_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH128_hashFromCanonical) +#endif + + +/* ************************************* +* Compiler specifics +***************************************/ + +/* specific declaration modes for Windows */ +#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API) +# if defined(_WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT)) +# ifdef XXH_EXPORT +# define XXH_PUBLIC_API __declspec(dllexport) +# elif XXH_IMPORT +# define XXH_PUBLIC_API __declspec(dllimport) +# endif +# else +# define XXH_PUBLIC_API /* do nothing */ +# endif +#endif + +#if defined (__GNUC__) +# define XXH_CONSTF __attribute__((__const__)) +# define XXH_PUREF __attribute__((__pure__)) +# define XXH_MALLOCF __attribute__((__malloc__)) +#else +# define XXH_CONSTF /* disable */ +# define XXH_PUREF +# define XXH_MALLOCF +#endif + +/* ************************************* +* Version +***************************************/ +#define XXH_VERSION_MAJOR 0 +#define XXH_VERSION_MINOR 8 +#define XXH_VERSION_RELEASE 3 +/*! @brief Version number, encoded as two digits each */ +#define XXH_VERSION_NUMBER (XXH_VERSION_MAJOR *100*100 + XXH_VERSION_MINOR *100 + XXH_VERSION_RELEASE) + +/*! + * @brief Obtains the xxHash version. + * + * This is mostly useful when xxHash is compiled as a shared library, + * since the returned value comes from the library, as opposed to header file. + * + * @return @ref XXH_VERSION_NUMBER of the invoked library. + */ +XXH_PUBLIC_API XXH_CONSTF unsigned XXH_versionNumber (void); + + +/* **************************** +* Common basic types +******************************/ +#include /* size_t */ +/*! + * @brief Exit code for the streaming API. + */ +typedef enum { + XXH_OK = 0, /*!< OK */ + XXH_ERROR /*!< Error */ +} XXH_errorcode; + + +/*-********************************************************************** +* 32-bit hash +************************************************************************/ +#if defined(XXH_DOXYGEN) /* Don't show include */ +/*! + * @brief An unsigned 32-bit integer. + * + * Not necessarily defined to `uint32_t` but functionally equivalent. + */ +typedef uint32_t XXH32_hash_t; + +#elif !defined (__VMS) \ + && (defined (__cplusplus) \ + || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) +# ifdef _AIX +# include +# else +# include +# endif + typedef uint32_t XXH32_hash_t; + +#else +# include +# if UINT_MAX == 0xFFFFFFFFUL + typedef unsigned int XXH32_hash_t; +# elif ULONG_MAX == 0xFFFFFFFFUL + typedef unsigned long XXH32_hash_t; +# else +# error "unsupported platform: need a 32-bit type" +# endif +#endif + +/*! + * @} + * + * @defgroup XXH32_family XXH32 family + * @ingroup public + * Contains functions used in the classic 32-bit xxHash algorithm. + * + * @note + * XXH32 is useful for older platforms, with no or poor 64-bit performance. + * Note that the @ref XXH3_family provides competitive speed for both 32-bit + * and 64-bit systems, and offers true 64/128 bit hash results. + * + * @see @ref XXH64_family, @ref XXH3_family : Other xxHash families + * @see @ref XXH32_impl for implementation details + * @{ + */ + +/*! + * @brief Calculates the 32-bit hash of @p input using xxHash32. + * + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * @param seed The 32-bit seed to alter the hash's output predictably. + * + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return The calculated 32-bit xxHash32 value. + * + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32 (const void* input, size_t length, XXH32_hash_t seed); + +#ifndef XXH_NO_STREAM +/*! + * @typedef struct XXH32_state_s XXH32_state_t + * @brief The opaque state struct for the XXH32 streaming API. + * + * @see XXH32_state_s for details. + * @see @ref streaming_example "Streaming Example" + */ +typedef struct XXH32_state_s XXH32_state_t; + +/*! + * @brief Allocates an @ref XXH32_state_t. + * + * @return An allocated pointer of @ref XXH32_state_t on success. + * @return `NULL` on failure. + * + * @note Must be freed with XXH32_freeState(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_MALLOCF XXH32_state_t* XXH32_createState(void); +/*! + * @brief Frees an @ref XXH32_state_t. + * + * @param statePtr A pointer to an @ref XXH32_state_t allocated with @ref XXH32_createState(). + * + * @return @ref XXH_OK. + * + * @note @p statePtr must be allocated with XXH32_createState(). + * + * @see @ref streaming_example "Streaming Example" + * + */ +XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr); +/*! + * @brief Copies one @ref XXH32_state_t to another. + * + * @param dst_state The state to copy to. + * @param src_state The state to copy from. + * @pre + * @p dst_state and @p src_state must not be `NULL` and must not overlap. + */ +XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dst_state, const XXH32_state_t* src_state); + +/*! + * @brief Resets an @ref XXH32_state_t to begin a new hash. + * + * @param statePtr The state struct to reset. + * @param seed The 32-bit seed to alter the hash result predictably. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note This function resets and seeds a state. Call it before @ref XXH32_update(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH32_reset (XXH32_state_t* statePtr, XXH32_hash_t seed); + +/*! + * @brief Consumes a block of @p input to an @ref XXH32_state_t. + * + * @param statePtr The state struct to update. + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * + * @pre + * @p statePtr must not be `NULL`. + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note Call this to incrementally consume blocks of data. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length); + +/*! + * @brief Returns the calculated hash value from an @ref XXH32_state_t. + * + * @param statePtr The state struct to calculate the hash from. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return The calculated 32-bit xxHash32 value from that state. + * + * @note + * Calling XXH32_digest() will not affect @p statePtr, so you can update, + * digest, and update again. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32_digest (const XXH32_state_t* statePtr); +#endif /* !XXH_NO_STREAM */ + +/******* Canonical representation *******/ + +/*! + * @brief Canonical (big endian) representation of @ref XXH32_hash_t. + */ +typedef struct { + unsigned char digest[4]; /*!< Hash bytes, big endian */ +} XXH32_canonical_t; + +/*! + * @brief Converts an @ref XXH32_hash_t to a big endian @ref XXH32_canonical_t. + * + * @param dst The @ref XXH32_canonical_t pointer to be stored to. + * @param hash The @ref XXH32_hash_t to be converted. + * + * @pre + * @p dst must not be `NULL`. + * + * @see @ref canonical_representation_example "Canonical Representation Example" + */ +XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash); + +/*! + * @brief Converts an @ref XXH32_canonical_t to a native @ref XXH32_hash_t. + * + * @param src The @ref XXH32_canonical_t to convert. + * + * @pre + * @p src must not be `NULL`. + * + * @return The converted hash. + * + * @see @ref canonical_representation_example "Canonical Representation Example" + */ +XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src); + + +/*! @cond Doxygen ignores this part */ +#ifdef __has_attribute +# define XXH_HAS_ATTRIBUTE(x) __has_attribute(x) +#else +# define XXH_HAS_ATTRIBUTE(x) 0 +#endif +/*! @endcond */ + +/*! @cond Doxygen ignores this part */ +/* + * C23 __STDC_VERSION__ number hasn't been specified yet. For now + * leave as `201711L` (C17 + 1). + * TODO: Update to correct value when its been specified. + */ +#define XXH_C23_VN 201711L +/*! @endcond */ + +/*! @cond Doxygen ignores this part */ +/* C-language Attributes are added in C23. */ +#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= XXH_C23_VN) && defined(__has_c_attribute) +# define XXH_HAS_C_ATTRIBUTE(x) __has_c_attribute(x) +#else +# define XXH_HAS_C_ATTRIBUTE(x) 0 +#endif +/*! @endcond */ + +/*! @cond Doxygen ignores this part */ +#if defined(__cplusplus) && defined(__has_cpp_attribute) +# define XXH_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x) +#else +# define XXH_HAS_CPP_ATTRIBUTE(x) 0 +#endif +/*! @endcond */ + +/*! @cond Doxygen ignores this part */ +/* + * Define XXH_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute + * introduced in CPP17 and C23. + * CPP17 : https://en.cppreference.com/w/cpp/language/attributes/fallthrough + * C23 : https://en.cppreference.com/w/c/language/attributes/fallthrough + */ +#if XXH_HAS_C_ATTRIBUTE(fallthrough) || XXH_HAS_CPP_ATTRIBUTE(fallthrough) +# define XXH_FALLTHROUGH [[fallthrough]] +#elif XXH_HAS_ATTRIBUTE(__fallthrough__) +# define XXH_FALLTHROUGH __attribute__ ((__fallthrough__)) +#else +# define XXH_FALLTHROUGH /* fallthrough */ +#endif +/*! @endcond */ + +/*! @cond Doxygen ignores this part */ +/* + * Define XXH_NOESCAPE for annotated pointers in public API. + * https://clang.llvm.org/docs/AttributeReference.html#noescape + * As of writing this, only supported by clang. + */ +#if XXH_HAS_ATTRIBUTE(noescape) +# define XXH_NOESCAPE __attribute__((__noescape__)) +#else +# define XXH_NOESCAPE +#endif +/*! @endcond */ + + +/*! + * @} + * @ingroup public + * @{ + */ + +#ifndef XXH_NO_LONG_LONG +/*-********************************************************************** +* 64-bit hash +************************************************************************/ +#if defined(XXH_DOXYGEN) /* don't include */ +/*! + * @brief An unsigned 64-bit integer. + * + * Not necessarily defined to `uint64_t` but functionally equivalent. + */ +typedef uint64_t XXH64_hash_t; +#elif !defined (__VMS) \ + && (defined (__cplusplus) \ + || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) +# ifdef _AIX +# include +# else +# include +# endif + typedef uint64_t XXH64_hash_t; +#else +# include +# if defined(__LP64__) && ULONG_MAX == 0xFFFFFFFFFFFFFFFFULL + /* LP64 ABI says uint64_t is unsigned long */ + typedef unsigned long XXH64_hash_t; +# else + /* the following type must have a width of 64-bit */ + typedef unsigned long long XXH64_hash_t; +# endif +#endif + +/*! + * @} + * + * @defgroup XXH64_family XXH64 family + * @ingroup public + * @{ + * Contains functions used in the classic 64-bit xxHash algorithm. + * + * @note + * XXH3 provides competitive speed for both 32-bit and 64-bit systems, + * and offers true 64/128 bit hash results. + * It provides better speed for systems with vector processing capabilities. + */ + +/*! + * @brief Calculates the 64-bit hash of @p input using xxHash64. + * + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * @param seed The 64-bit seed to alter the hash's output predictably. + * + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return The calculated 64-bit xxHash64 value. + * + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed); + +/******* Streaming *******/ +#ifndef XXH_NO_STREAM +/*! + * @brief The opaque state struct for the XXH64 streaming API. + * + * @see XXH64_state_s for details. + * @see @ref streaming_example "Streaming Example" + */ +typedef struct XXH64_state_s XXH64_state_t; /* incomplete type */ + +/*! + * @brief Allocates an @ref XXH64_state_t. + * + * @return An allocated pointer of @ref XXH64_state_t on success. + * @return `NULL` on failure. + * + * @note Must be freed with XXH64_freeState(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_MALLOCF XXH64_state_t* XXH64_createState(void); + +/*! + * @brief Frees an @ref XXH64_state_t. + * + * @param statePtr A pointer to an @ref XXH64_state_t allocated with @ref XXH64_createState(). + * + * @return @ref XXH_OK. + * + * @note @p statePtr must be allocated with XXH64_createState(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr); + +/*! + * @brief Copies one @ref XXH64_state_t to another. + * + * @param dst_state The state to copy to. + * @param src_state The state to copy from. + * @pre + * @p dst_state and @p src_state must not be `NULL` and must not overlap. + */ +XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t* dst_state, const XXH64_state_t* src_state); + +/*! + * @brief Resets an @ref XXH64_state_t to begin a new hash. + * + * @param statePtr The state struct to reset. + * @param seed The 64-bit seed to alter the hash result predictably. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note This function resets and seeds a state. Call it before @ref XXH64_update(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH64_reset (XXH_NOESCAPE XXH64_state_t* statePtr, XXH64_hash_t seed); + +/*! + * @brief Consumes a block of @p input to an @ref XXH64_state_t. + * + * @param statePtr The state struct to update. + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * + * @pre + * @p statePtr must not be `NULL`. + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note Call this to incrementally consume blocks of data. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH_NOESCAPE XXH64_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length); + +/*! + * @brief Returns the calculated hash value from an @ref XXH64_state_t. + * + * @param statePtr The state struct to calculate the hash from. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return The calculated 64-bit xxHash64 value from that state. + * + * @note + * Calling XXH64_digest() will not affect @p statePtr, so you can update, + * digest, and update again. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_digest (XXH_NOESCAPE const XXH64_state_t* statePtr); +#endif /* !XXH_NO_STREAM */ +/******* Canonical representation *******/ + +/*! + * @brief Canonical (big endian) representation of @ref XXH64_hash_t. + */ +typedef struct { unsigned char digest[sizeof(XXH64_hash_t)]; } XXH64_canonical_t; + +/*! + * @brief Converts an @ref XXH64_hash_t to a big endian @ref XXH64_canonical_t. + * + * @param dst The @ref XXH64_canonical_t pointer to be stored to. + * @param hash The @ref XXH64_hash_t to be converted. + * + * @pre + * @p dst must not be `NULL`. + * + * @see @ref canonical_representation_example "Canonical Representation Example" + */ +XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t* dst, XXH64_hash_t hash); + +/*! + * @brief Converts an @ref XXH64_canonical_t to a native @ref XXH64_hash_t. + * + * @param src The @ref XXH64_canonical_t to convert. + * + * @pre + * @p src must not be `NULL`. + * + * @return The converted hash. + * + * @see @ref canonical_representation_example "Canonical Representation Example" + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t* src); + +#ifndef XXH_NO_XXH3 + +/*! + * @} + * ************************************************************************ + * @defgroup XXH3_family XXH3 family + * @ingroup public + * @{ + * + * XXH3 is a more recent hash algorithm featuring: + * - Improved speed for both small and large inputs + * - True 64-bit and 128-bit outputs + * - SIMD acceleration + * - Improved 32-bit viability + * + * Speed analysis methodology is explained here: + * + * https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html + * + * Compared to XXH64, expect XXH3 to run approximately + * ~2x faster on large inputs and >3x faster on small ones, + * exact differences vary depending on platform. + * + * XXH3's speed benefits greatly from SIMD and 64-bit arithmetic, + * but does not require it. + * Most 32-bit and 64-bit targets that can run XXH32 smoothly can run XXH3 + * at competitive speeds, even without vector support. Further details are + * explained in the implementation. + * + * XXH3 has a fast scalar implementation, but it also includes accelerated SIMD + * implementations for many common platforms: + * - AVX512 + * - AVX2 + * - SSE2 + * - ARM NEON + * - WebAssembly SIMD128 + * - POWER8 VSX + * - s390x ZVector + * This can be controlled via the @ref XXH_VECTOR macro, but it automatically + * selects the best version according to predefined macros. For the x86 family, an + * automatic runtime dispatcher is included separately in @ref xxh_x86dispatch.c. + * + * XXH3 implementation is portable: + * it has a generic C90 formulation that can be compiled on any platform, + * all implementations generate exactly the same hash value on all platforms. + * Starting from v0.8.0, it's also labelled "stable", meaning that + * any future version will also generate the same hash value. + * + * XXH3 offers 2 variants, _64bits and _128bits. + * + * When only 64 bits are needed, prefer invoking the _64bits variant, as it + * reduces the amount of mixing, resulting in faster speed on small inputs. + * It's also generally simpler to manipulate a scalar return type than a struct. + * + * The API supports one-shot hashing, streaming mode, and custom secrets. + */ + +/*! + * @ingroup tuning + * @brief Possible values for @ref XXH_VECTOR. + * + * Unless set explicitly, determined automatically. + */ +# define XXH_SCALAR 0 /*!< Portable scalar version */ +# define XXH_SSE2 1 /*!< SSE2 for Pentium 4, Opteron, all x86_64. */ +# define XXH_AVX2 2 /*!< AVX2 for Haswell and Bulldozer */ +# define XXH_AVX512 3 /*!< AVX512 for Skylake and Icelake */ +# define XXH_NEON 4 /*!< NEON for most ARMv7-A, all AArch64, and WASM SIMD128 */ +# define XXH_VSX 5 /*!< VSX and ZVector for POWER8/z13 (64-bit) */ +# define XXH_SVE 6 /*!< SVE for some ARMv8-A and ARMv9-A */ +# define XXH_LSX 7 /*!< LSX (128-bit SIMD) for LoongArch64 */ + + +/*-********************************************************************** +* XXH3 64-bit variant +************************************************************************/ + +/*! + * @brief Calculates 64-bit unseeded variant of XXH3 hash of @p input. + * + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return The calculated 64-bit XXH3 hash value. + * + * @note + * This is equivalent to @ref XXH3_64bits_withSeed() with a seed of `0`, however + * it may have slightly better performance due to constant propagation of the + * defaults. + * + * @see + * XXH3_64bits_withSeed(), XXH3_64bits_withSecret(): other seeding variants + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void* input, size_t length); + +/*! + * @brief Calculates 64-bit seeded variant of XXH3 hash of @p input. + * + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * @param seed The 64-bit seed to alter the hash result predictably. + * + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return The calculated 64-bit XXH3 hash value. + * + * @note + * seed == 0 produces the same results as @ref XXH3_64bits(). + * + * This variant generates a custom secret on the fly based on default secret + * altered using the @p seed value. + * + * While this operation is decently fast, note that it's not completely free. + * + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSeed(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed); + +/*! + * The bare minimum size for a custom secret. + * + * @see + * XXH3_64bits_withSecret(), XXH3_64bits_reset_withSecret(), + * XXH3_128bits_withSecret(), XXH3_128bits_reset_withSecret(). + */ +#define XXH3_SECRET_SIZE_MIN 136 + +/*! + * @brief Calculates 64-bit variant of XXH3 with a custom "secret". + * + * @param data The block of data to be hashed, at least @p len bytes in size. + * @param len The length of @p data, in bytes. + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * + * @return The calculated 64-bit XXH3 hash value. + * + * @pre + * The memory between @p data and @p data + @p len must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p data may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * It's possible to provide any blob of bytes as a "secret" to generate the hash. + * This makes it more difficult for an external actor to prepare an intentional collision. + * The main condition is that @p secretSize *must* be large enough (>= @ref XXH3_SECRET_SIZE_MIN). + * However, the quality of the secret impacts the dispersion of the hash algorithm. + * Therefore, the secret _must_ look like a bunch of random bytes. + * Avoid "trivial" or structured data such as repeated sequences or a text document. + * Whenever in doubt about the "randomness" of the blob of bytes, + * consider employing @ref XXH3_generateSecret() instead (see below). + * It will generate a proper high entropy secret derived from the blob of bytes. + * Another advantage of using XXH3_generateSecret() is that + * it guarantees that all bits within the initial blob of bytes + * will impact every bit of the output. + * This is not necessarily the case when using the blob of bytes directly + * because, when hashing _small_ inputs, only a portion of the secret is employed. + * + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSecret(XXH_NOESCAPE const void* data, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize); + + +/******* Streaming *******/ +#ifndef XXH_NO_STREAM +/* + * Streaming requires state maintenance. + * This operation costs memory and CPU. + * As a consequence, streaming is slower than one-shot hashing. + * For better performance, prefer one-shot functions whenever applicable. + */ + +/*! + * @brief The opaque state struct for the XXH3 streaming API. + * + * @see XXH3_state_s for details. + * @see @ref streaming_example "Streaming Example" + */ +typedef struct XXH3_state_s XXH3_state_t; +XXH_PUBLIC_API XXH_MALLOCF XXH3_state_t* XXH3_createState(void); +XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr); + +/*! + * @brief Copies one @ref XXH3_state_t to another. + * + * @param dst_state The state to copy to. + * @param src_state The state to copy from. + * @pre + * @p dst_state and @p src_state must not be `NULL` and must not overlap. + */ +XXH_PUBLIC_API void XXH3_copyState(XXH_NOESCAPE XXH3_state_t* dst_state, XXH_NOESCAPE const XXH3_state_t* src_state); + +/*! + * @brief Resets an @ref XXH3_state_t to begin a new hash. + * + * @param statePtr The state struct to reset. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note + * - This function resets `statePtr` and generate a secret with default parameters. + * - Call this function before @ref XXH3_64bits_update(). + * - Digest will be equivalent to `XXH3_64bits()`. + * + * @see @ref streaming_example "Streaming Example" + * + */ +XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr); + +/*! + * @brief Resets an @ref XXH3_state_t with 64-bit seed to begin a new hash. + * + * @param statePtr The state struct to reset. + * @param seed The 64-bit seed to alter the hash result predictably. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note + * - This function resets `statePtr` and generate a secret from `seed`. + * - Call this function before @ref XXH3_64bits_update(). + * - Digest will be equivalent to `XXH3_64bits_withSeed()`. + * + * @see @ref streaming_example "Streaming Example" + * + */ +XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed); + +/*! + * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash. + * + * @param statePtr The state struct to reset. + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note + * `secret` is referenced, it _must outlive_ the hash streaming session. + * + * Similar to one-shot API, `secretSize` must be >= @ref XXH3_SECRET_SIZE_MIN, + * and the quality of produced hash values depends on secret's entropy + * (secret's content should look like a bunch of random bytes). + * When in doubt about the randomness of a candidate `secret`, + * consider employing `XXH3_generateSecret()` instead (see below). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize); + +/*! + * @brief Consumes a block of @p input to an @ref XXH3_state_t. + * + * @param statePtr The state struct to update. + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * + * @pre + * @p statePtr must not be `NULL`. + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note Call this to incrementally consume blocks of data. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update (XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length); + +/*! + * @brief Returns the calculated XXH3 64-bit hash value from an @ref XXH3_state_t. + * + * @param statePtr The state struct to calculate the hash from. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return The calculated XXH3 64-bit hash value from that state. + * + * @note + * Calling XXH3_64bits_digest() will not affect @p statePtr, so you can update, + * digest, and update again. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_digest (XXH_NOESCAPE const XXH3_state_t* statePtr); +#endif /* !XXH_NO_STREAM */ + +/* note : canonical representation of XXH3 is the same as XXH64 + * since they both produce XXH64_hash_t values */ + + +/*-********************************************************************** +* XXH3 128-bit variant +************************************************************************/ + +/*! + * @brief The return value from 128-bit hashes. + * + * Stored in little endian order, although the fields themselves are in native + * endianness. + */ +typedef struct { + XXH64_hash_t low64; /*!< `value & 0xFFFFFFFFFFFFFFFF` */ + XXH64_hash_t high64; /*!< `value >> 64` */ +} XXH128_hash_t; + +/*! + * @brief Calculates 128-bit unseeded variant of XXH3 of @p data. + * + * @param data The block of data to be hashed, at least @p length bytes in size. + * @param len The length of @p data, in bytes. + * + * @return The calculated 128-bit variant of XXH3 value. + * + * The 128-bit variant of XXH3 has more strength, but it has a bit of overhead + * for shorter inputs. + * + * This is equivalent to @ref XXH3_128bits_withSeed() with a seed of `0`, however + * it may have slightly better performance due to constant propagation of the + * defaults. + * + * @see XXH3_128bits_withSeed(), XXH3_128bits_withSecret(): other seeding variants + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void* data, size_t len); +/*! @brief Calculates 128-bit seeded variant of XXH3 hash of @p data. + * + * @param data The block of data to be hashed, at least @p length bytes in size. + * @param len The length of @p data, in bytes. + * @param seed The 64-bit seed to alter the hash result predictably. + * + * @return The calculated 128-bit variant of XXH3 value. + * + * @note + * seed == 0 produces the same results as @ref XXH3_64bits(). + * + * This variant generates a custom secret on the fly based on default secret + * altered using the @p seed value. + * + * While this operation is decently fast, note that it's not completely free. + * + * @see XXH3_128bits(), XXH3_128bits_withSecret(): other seeding variants + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSeed(XXH_NOESCAPE const void* data, size_t len, XXH64_hash_t seed); +/*! + * @brief Calculates 128-bit variant of XXH3 with a custom "secret". + * + * @param data The block of data to be hashed, at least @p len bytes in size. + * @param len The length of @p data, in bytes. + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * + * @return The calculated 128-bit variant of XXH3 value. + * + * It's possible to provide any blob of bytes as a "secret" to generate the hash. + * This makes it more difficult for an external actor to prepare an intentional collision. + * The main condition is that @p secretSize *must* be large enough (>= @ref XXH3_SECRET_SIZE_MIN). + * However, the quality of the secret impacts the dispersion of the hash algorithm. + * Therefore, the secret _must_ look like a bunch of random bytes. + * Avoid "trivial" or structured data such as repeated sequences or a text document. + * Whenever in doubt about the "randomness" of the blob of bytes, + * consider employing @ref XXH3_generateSecret() instead (see below). + * It will generate a proper high entropy secret derived from the blob of bytes. + * Another advantage of using XXH3_generateSecret() is that + * it guarantees that all bits within the initial blob of bytes + * will impact every bit of the output. + * This is not necessarily the case when using the blob of bytes directly + * because, when hashing _small_ inputs, only a portion of the secret is employed. + * + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSecret(XXH_NOESCAPE const void* data, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize); + +/******* Streaming *******/ +#ifndef XXH_NO_STREAM +/* + * Streaming requires state maintenance. + * This operation costs memory and CPU. + * As a consequence, streaming is slower than one-shot hashing. + * For better performance, prefer one-shot functions whenever applicable. + * + * XXH3_128bits uses the same XXH3_state_t as XXH3_64bits(). + * Use already declared XXH3_createState() and XXH3_freeState(). + * + * All reset and streaming functions have same meaning as their 64-bit counterpart. + */ + +/*! + * @brief Resets an @ref XXH3_state_t to begin a new hash. + * + * @param statePtr The state struct to reset. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note + * - This function resets `statePtr` and generate a secret with default parameters. + * - Call it before @ref XXH3_128bits_update(). + * - Digest will be equivalent to `XXH3_128bits()`. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr); + +/*! + * @brief Resets an @ref XXH3_state_t with 64-bit seed to begin a new hash. + * + * @param statePtr The state struct to reset. + * @param seed The 64-bit seed to alter the hash result predictably. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note + * - This function resets `statePtr` and generate a secret from `seed`. + * - Call it before @ref XXH3_128bits_update(). + * - Digest will be equivalent to `XXH3_128bits_withSeed()`. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed); +/*! + * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash. + * + * @param statePtr The state struct to reset. + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * `secret` is referenced, it _must outlive_ the hash streaming session. + * Similar to one-shot API, `secretSize` must be >= @ref XXH3_SECRET_SIZE_MIN, + * and the quality of produced hash values depends on secret's entropy + * (secret's content should look like a bunch of random bytes). + * When in doubt about the randomness of a candidate `secret`, + * consider employing `XXH3_generateSecret()` instead (see below). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize); + +/*! + * @brief Consumes a block of @p input to an @ref XXH3_state_t. + * + * Call this to incrementally consume blocks of data. + * + * @param statePtr The state struct to update. + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + */ +XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update (XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length); + +/*! + * @brief Returns the calculated XXH3 128-bit hash value from an @ref XXH3_state_t. + * + * @param statePtr The state struct to calculate the hash from. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return The calculated XXH3 128-bit hash value from that state. + * + * @note + * Calling XXH3_128bits_digest() will not affect @p statePtr, so you can update, + * digest, and update again. + * + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_digest (XXH_NOESCAPE const XXH3_state_t* statePtr); +#endif /* !XXH_NO_STREAM */ + +/* Following helper functions make it possible to compare XXH128_hast_t values. + * Since XXH128_hash_t is a structure, this capability is not offered by the language. + * Note: For better performance, these functions can be inlined using XXH_INLINE_ALL */ + +/*! + * @brief Check equality of two XXH128_hash_t values + * + * @param h1 The 128-bit hash value. + * @param h2 Another 128-bit hash value. + * + * @return `1` if `h1` and `h2` are equal. + * @return `0` if they are not. + */ +XXH_PUBLIC_API XXH_PUREF int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2); + +/*! + * @brief Compares two @ref XXH128_hash_t + * + * This comparator is compatible with stdlib's `qsort()`/`bsearch()`. + * + * @param h128_1 Left-hand side value + * @param h128_2 Right-hand side value + * + * @return >0 if @p h128_1 > @p h128_2 + * @return =0 if @p h128_1 == @p h128_2 + * @return <0 if @p h128_1 < @p h128_2 + */ +XXH_PUBLIC_API XXH_PUREF int XXH128_cmp(XXH_NOESCAPE const void* h128_1, XXH_NOESCAPE const void* h128_2); + + +/******* Canonical representation *******/ +typedef struct { unsigned char digest[sizeof(XXH128_hash_t)]; } XXH128_canonical_t; + + +/*! + * @brief Converts an @ref XXH128_hash_t to a big endian @ref XXH128_canonical_t. + * + * @param dst The @ref XXH128_canonical_t pointer to be stored to. + * @param hash The @ref XXH128_hash_t to be converted. + * + * @pre + * @p dst must not be `NULL`. + * @see @ref canonical_representation_example "Canonical Representation Example" + */ +XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH_NOESCAPE XXH128_canonical_t* dst, XXH128_hash_t hash); + +/*! + * @brief Converts an @ref XXH128_canonical_t to a native @ref XXH128_hash_t. + * + * @param src The @ref XXH128_canonical_t to convert. + * + * @pre + * @p src must not be `NULL`. + * + * @return The converted hash. + * @see @ref canonical_representation_example "Canonical Representation Example" + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t* src); + + +#endif /* !XXH_NO_XXH3 */ +#endif /* XXH_NO_LONG_LONG */ + +/*! + * @} + */ +#endif /* XXHASH_H_5627135585666179 */ + + + +#if defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742) +#define XXHASH_H_STATIC_13879238742 +/* **************************************************************************** + * This section contains declarations which are not guaranteed to remain stable. + * They may change in future versions, becoming incompatible with a different + * version of the library. + * These declarations should only be used with static linking. + * Never use them in association with dynamic linking! + ***************************************************************************** */ + +/* + * These definitions are only present to allow static allocation + * of XXH states, on stack or in a struct, for example. + * Never **ever** access their members directly. + */ + +/*! + * @internal + * @brief Structure for XXH32 streaming API. + * + * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY, + * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is + * an opaque type. This allows fields to safely be changed. + * + * Typedef'd to @ref XXH32_state_t. + * Do not access the members of this struct directly. + * @see XXH64_state_s, XXH3_state_s + */ +struct XXH32_state_s { + XXH32_hash_t total_len_32; /*!< Total length hashed, modulo 2^32 */ + XXH32_hash_t large_len; /*!< Whether the hash is >= 16 (handles @ref total_len_32 overflow) */ + XXH32_hash_t acc[4]; /*!< Accumulator lanes */ + unsigned char buffer[16]; /*!< Internal buffer for partial reads. */ + XXH32_hash_t bufferedSize; /*!< Amount of data in @ref buffer */ + XXH32_hash_t reserved; /*!< Reserved field. Do not read nor write to it. */ +}; /* typedef'd to XXH32_state_t */ + + +#ifndef XXH_NO_LONG_LONG /* defined when there is no 64-bit support */ + +/*! + * @internal + * @brief Structure for XXH64 streaming API. + * + * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY, + * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is + * an opaque type. This allows fields to safely be changed. + * + * Typedef'd to @ref XXH64_state_t. + * Do not access the members of this struct directly. + * @see XXH32_state_s, XXH3_state_s + */ +struct XXH64_state_s { + XXH64_hash_t total_len; /*!< Total length hashed. This is always 64-bit. */ + XXH64_hash_t acc[4]; /*!< Accumulator lanes */ + unsigned char buffer[32]; /*!< Internal buffer for partial reads.. */ + XXH32_hash_t bufferedSize; /*!< Amount of data in @ref buffer */ + XXH32_hash_t reserved32; /*!< Reserved field, needed for padding anyways*/ + XXH64_hash_t reserved64; /*!< Reserved field. Do not read or write to it. */ +}; /* typedef'd to XXH64_state_t */ + +#ifndef XXH_NO_XXH3 + +#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* >= C11 */ +# define XXH_ALIGN(n) _Alignas(n) +#elif defined(__cplusplus) && (__cplusplus >= 201103L) /* >= C++11 */ +/* In C++ alignas() is a keyword */ +# define XXH_ALIGN(n) alignas(n) +#elif defined(__GNUC__) +# define XXH_ALIGN(n) __attribute__ ((aligned(n))) +#elif defined(_MSC_VER) +# define XXH_ALIGN(n) __declspec(align(n)) +#else +# define XXH_ALIGN(n) /* disabled */ +#endif + +/* Old GCC versions only accept the attribute after the type in structures. */ +#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)) /* C11+ */ \ + && ! (defined(__cplusplus) && (__cplusplus >= 201103L)) /* >= C++11 */ \ + && defined(__GNUC__) +# define XXH_ALIGN_MEMBER(align, type) type XXH_ALIGN(align) +#else +# define XXH_ALIGN_MEMBER(align, type) XXH_ALIGN(align) type +#endif + +/*! + * @brief The size of the internal XXH3 buffer. + * + * This is the optimal update size for incremental hashing. + * + * @see XXH3_64b_update(), XXH3_128b_update(). + */ +#define XXH3_INTERNALBUFFER_SIZE 256 + +/*! + * @internal + * @brief Default size of the secret buffer (and @ref XXH3_kSecret). + * + * This is the size used in @ref XXH3_kSecret and the seeded functions. + * + * Not to be confused with @ref XXH3_SECRET_SIZE_MIN. + */ +#define XXH3_SECRET_DEFAULT_SIZE 192 + +/*! + * @internal + * @brief Structure for XXH3 streaming API. + * + * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY, + * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. + * Otherwise it is an opaque type. + * Never use this definition in combination with dynamic library. + * This allows fields to safely be changed in the future. + * + * @note ** This structure has a strict alignment requirement of 64 bytes!! ** + * Do not allocate this with `malloc()` or `new`, + * it will not be sufficiently aligned. + * Use @ref XXH3_createState() and @ref XXH3_freeState(), or stack allocation. + * + * Typedef'd to @ref XXH3_state_t. + * Do never access the members of this struct directly. + * + * @see XXH3_INITSTATE() for stack initialization. + * @see XXH3_createState(), XXH3_freeState(). + * @see XXH32_state_s, XXH64_state_s + */ +struct XXH3_state_s { + XXH_ALIGN_MEMBER(64, XXH64_hash_t acc[8]); + /*!< The 8 accumulators. See @ref XXH32_state_s::v and @ref XXH64_state_s::v */ + XXH_ALIGN_MEMBER(64, unsigned char customSecret[XXH3_SECRET_DEFAULT_SIZE]); + /*!< Used to store a custom secret generated from a seed. */ + XXH_ALIGN_MEMBER(64, unsigned char buffer[XXH3_INTERNALBUFFER_SIZE]); + /*!< The internal buffer. @see XXH32_state_s::mem32 */ + XXH32_hash_t bufferedSize; + /*!< The amount of memory in @ref buffer, @see XXH32_state_s::memsize */ + XXH32_hash_t useSeed; + /*!< Reserved field. Needed for padding on 64-bit. */ + size_t nbStripesSoFar; + /*!< Number or stripes processed. */ + XXH64_hash_t totalLen; + /*!< Total length hashed. 64-bit even on 32-bit targets. */ + size_t nbStripesPerBlock; + /*!< Number of stripes per block. */ + size_t secretLimit; + /*!< Size of @ref customSecret or @ref extSecret */ + XXH64_hash_t seed; + /*!< Seed for _withSeed variants. Must be zero otherwise, @see XXH3_INITSTATE() */ + XXH64_hash_t reserved64; + /*!< Reserved field. */ + const unsigned char* extSecret; + /*!< Reference to an external secret for the _withSecret variants, NULL + * for other variants. */ + /* note: there may be some padding at the end due to alignment on 64 bytes */ +}; /* typedef'd to XXH3_state_t */ + +#undef XXH_ALIGN_MEMBER + +/*! + * @brief Initializes a stack-allocated `XXH3_state_s`. + * + * When the @ref XXH3_state_t structure is merely emplaced on stack, + * it should be initialized with XXH3_INITSTATE() or a memset() + * in case its first reset uses XXH3_NNbits_reset_withSeed(). + * This init can be omitted if the first reset uses default or _withSecret mode. + * This operation isn't necessary when the state is created with XXH3_createState(). + * Note that this doesn't prepare the state for a streaming operation, + * it's still necessary to use XXH3_NNbits_reset*() afterwards. + */ +#define XXH3_INITSTATE(XXH3_state_ptr) \ + do { \ + XXH3_state_t* tmp_xxh3_state_ptr = (XXH3_state_ptr); \ + tmp_xxh3_state_ptr->seed = 0; \ + tmp_xxh3_state_ptr->extSecret = NULL; \ + } while(0) + + +/*! + * @brief Calculates the 128-bit hash of @p data using XXH3. + * + * @param data The block of data to be hashed, at least @p len bytes in size. + * @param len The length of @p data, in bytes. + * @param seed The 64-bit seed to alter the hash's output predictably. + * + * @pre + * The memory between @p data and @p data + @p len must be valid, + * readable, contiguous memory. However, if @p len is `0`, @p data may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return The calculated 128-bit XXH3 value. + * + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128(XXH_NOESCAPE const void* data, size_t len, XXH64_hash_t seed); + + +/* === Experimental API === */ +/* Symbols defined below must be considered tied to a specific library version. */ + +/*! + * @brief Derive a high-entropy secret from any user-defined content, named customSeed. + * + * @param secretBuffer A writable buffer for derived high-entropy secret data. + * @param secretSize Size of secretBuffer, in bytes. Must be >= XXH3_SECRET_SIZE_MIN. + * @param customSeed A user-defined content. + * @param customSeedSize Size of customSeed, in bytes. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * The generated secret can be used in combination with `*_withSecret()` functions. + * The `_withSecret()` variants are useful to provide a higher level of protection + * than 64-bit seed, as it becomes much more difficult for an external actor to + * guess how to impact the calculation logic. + * + * The function accepts as input a custom seed of any length and any content, + * and derives from it a high-entropy secret of length @p secretSize into an + * already allocated buffer @p secretBuffer. + * + * The generated secret can then be used with any `*_withSecret()` variant. + * The functions @ref XXH3_128bits_withSecret(), @ref XXH3_64bits_withSecret(), + * @ref XXH3_128bits_reset_withSecret() and @ref XXH3_64bits_reset_withSecret() + * are part of this list. They all accept a `secret` parameter + * which must be large enough for implementation reasons (>= @ref XXH3_SECRET_SIZE_MIN) + * _and_ feature very high entropy (consist of random-looking bytes). + * These conditions can be a high bar to meet, so @ref XXH3_generateSecret() can + * be employed to ensure proper quality. + * + * @p customSeed can be anything. It can have any size, even small ones, + * and its content can be anything, even "poor entropy" sources such as a bunch + * of zeroes. The resulting `secret` will nonetheless provide all required qualities. + * + * @pre + * - @p secretSize must be >= @ref XXH3_SECRET_SIZE_MIN + * - When @p customSeedSize > 0, supplying NULL as customSeed is undefined behavior. + * + * Example code: + * @code{.c} + * #include + * #include + * #include + * #define XXH_STATIC_LINKING_ONLY // expose unstable API + * #include "xxhash.h" + * // Hashes argv[2] using the entropy from argv[1]. + * int main(int argc, char* argv[]) + * { + * char secret[XXH3_SECRET_SIZE_MIN]; + * if (argv != 3) { return 1; } + * XXH3_generateSecret(secret, sizeof(secret), argv[1], strlen(argv[1])); + * XXH64_hash_t h = XXH3_64bits_withSecret( + * argv[2], strlen(argv[2]), + * secret, sizeof(secret) + * ); + * printf("%016llx\n", (unsigned long long) h); + * } + * @endcode + */ +XXH_PUBLIC_API XXH_errorcode XXH3_generateSecret(XXH_NOESCAPE void* secretBuffer, size_t secretSize, XXH_NOESCAPE const void* customSeed, size_t customSeedSize); + +/*! + * @brief Generate the same secret as the _withSeed() variants. + * + * @param secretBuffer A writable buffer of @ref XXH3_SECRET_DEFAULT_SIZE bytes + * @param seed The 64-bit seed to alter the hash result predictably. + * + * The generated secret can be used in combination with + *`*_withSecret()` and `_withSecretandSeed()` variants. + * + * Example C++ `std::string` hash class: + * @code{.cpp} + * #include + * #define XXH_STATIC_LINKING_ONLY // expose unstable API + * #include "xxhash.h" + * // Slow, seeds each time + * class HashSlow { + * XXH64_hash_t seed; + * public: + * HashSlow(XXH64_hash_t s) : seed{s} {} + * size_t operator()(const std::string& x) const { + * return size_t{XXH3_64bits_withSeed(x.c_str(), x.length(), seed)}; + * } + * }; + * // Fast, caches the seeded secret for future uses. + * class HashFast { + * unsigned char secret[XXH3_SECRET_DEFAULT_SIZE]; + * public: + * HashFast(XXH64_hash_t s) { + * XXH3_generateSecret_fromSeed(secret, seed); + * } + * size_t operator()(const std::string& x) const { + * return size_t{ + * XXH3_64bits_withSecret(x.c_str(), x.length(), secret, sizeof(secret)) + * }; + * } + * }; + * @endcode + */ +XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(XXH_NOESCAPE void* secretBuffer, XXH64_hash_t seed); + +/*! + * @brief Maximum size of "short" key in bytes. + */ +#define XXH3_MIDSIZE_MAX 240 + +/*! + * @brief Calculates 64/128-bit seeded variant of XXH3 hash of @p data. + * + * @param data The block of data to be hashed, at least @p len bytes in size. + * @param len The length of @p data, in bytes. + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * @param seed The 64-bit seed to alter the hash result predictably. + * + * These variants generate hash values using either: + * - @p seed for "short" keys (< @ref XXH3_MIDSIZE_MAX = 240 bytes) + * - @p secret for "large" keys (>= @ref XXH3_MIDSIZE_MAX). + * + * This generally benefits speed, compared to `_withSeed()` or `_withSecret()`. + * `_withSeed()` has to generate the secret on the fly for "large" keys. + * It's fast, but can be perceptible for "not so large" keys (< 1 KB). + * `_withSecret()` has to generate the masks on the fly for "small" keys, + * which requires more instructions than _withSeed() variants. + * Therefore, _withSecretandSeed variant combines the best of both worlds. + * + * When @p secret has been generated by XXH3_generateSecret_fromSeed(), + * this variant produces *exactly* the same results as `_withSeed()` variant, + * hence offering only a pure speed benefit on "large" input, + * by skipping the need to regenerate the secret for every large input. + * + * Another usage scenario is to hash the secret to a 64-bit hash value, + * for example with XXH3_64bits(), which then becomes the seed, + * and then employ both the seed and the secret in _withSecretandSeed(). + * On top of speed, an added benefit is that each bit in the secret + * has a 50% chance to swap each bit in the output, via its impact to the seed. + * + * This is not guaranteed when using the secret directly in "small data" scenarios, + * because only portions of the secret are employed for small data. + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t +XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* data, size_t len, + XXH_NOESCAPE const void* secret, size_t secretSize, + XXH64_hash_t seed); + +/*! + * @brief Calculates 128-bit seeded variant of XXH3 hash of @p data. + * + * @param data The memory segment to be hashed, at least @p len bytes in size. + * @param length The length of @p data, in bytes. + * @param secret The secret used to alter hash result predictably. + * @param secretSize The length of @p secret, in bytes (must be >= XXH3_SECRET_SIZE_MIN) + * @param seed64 The 64-bit seed to alter the hash result predictably. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @see XXH3_64bits_withSecretandSeed(): contract is the same. + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t +XXH3_128bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length, + XXH_NOESCAPE const void* secret, size_t secretSize, + XXH64_hash_t seed64); + +#ifndef XXH_NO_STREAM +/*! + * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash. + * + * @param statePtr A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState(). + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * @param seed64 The 64-bit seed to alter the hash result predictably. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @see XXH3_64bits_withSecretandSeed(). Contract is identical. + */ +XXH_PUBLIC_API XXH_errorcode +XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, + XXH_NOESCAPE const void* secret, size_t secretSize, + XXH64_hash_t seed64); + +/*! + * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash. + * + * @param statePtr A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState(). + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * @param seed64 The 64-bit seed to alter the hash result predictably. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @see XXH3_64bits_withSecretandSeed(). Contract is identical. + * + * Note: there was a bug in an earlier version of this function (<= v0.8.2) + * that would make it generate an incorrect hash value + * when @p seed == 0 and @p length < XXH3_MIDSIZE_MAX + * and @p secret is different from XXH3_generateSecret_fromSeed(). + * As stated in the contract, the correct hash result must be + * the same as XXH3_128bits_withSeed() when @p length <= XXH3_MIDSIZE_MAX. + * Results generated by this older version are wrong, hence not comparable. + */ +XXH_PUBLIC_API XXH_errorcode +XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, + XXH_NOESCAPE const void* secret, size_t secretSize, + XXH64_hash_t seed64); + +#endif /* !XXH_NO_STREAM */ + +#endif /* !XXH_NO_XXH3 */ +#endif /* XXH_NO_LONG_LONG */ +#if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) +# define XXH_IMPLEMENTATION +#endif + +#endif /* defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742) */ + + +/* ======================================================================== */ +/* ======================================================================== */ +/* ======================================================================== */ + + +/*-********************************************************************** + * xxHash implementation + *-********************************************************************** + * xxHash's implementation used to be hosted inside xxhash.c. + * + * However, inlining requires implementation to be visible to the compiler, + * hence be included alongside the header. + * Previously, implementation was hosted inside xxhash.c, + * which was then #included when inlining was activated. + * This construction created issues with a few build and install systems, + * as it required xxhash.c to be stored in /include directory. + * + * xxHash implementation is now directly integrated within xxhash.h. + * As a consequence, xxhash.c is no longer needed in /include. + * + * xxhash.c is still available and is still useful. + * In a "normal" setup, when xxhash is not inlined, + * xxhash.h only exposes the prototypes and public symbols, + * while xxhash.c can be built into an object file xxhash.o + * which can then be linked into the final binary. + ************************************************************************/ + +#if ( defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) \ + || defined(XXH_IMPLEMENTATION) ) && !defined(XXH_IMPLEM_13a8737387) +# define XXH_IMPLEM_13a8737387 + +/* ************************************* +* Tuning parameters +***************************************/ + +/*! + * @defgroup tuning Tuning parameters + * @{ + * + * Various macros to control xxHash's behavior. + */ +#ifdef XXH_DOXYGEN +/*! + * @brief Define this to disable 64-bit code. + * + * Useful if only using the @ref XXH32_family and you have a strict C90 compiler. + */ +# define XXH_NO_LONG_LONG +# undef XXH_NO_LONG_LONG /* don't actually */ +/*! + * @brief Controls how unaligned memory is accessed. + * + * By default, access to unaligned memory is controlled by `memcpy()`, which is + * safe and portable. + * + * Unfortunately, on some target/compiler combinations, the generated assembly + * is sub-optimal. + * + * The below switch allow selection of a different access method + * in the search for improved performance. + * + * @par Possible options: + * + * - `XXH_FORCE_MEMORY_ACCESS=0` (default): `memcpy` + * @par + * Use `memcpy()`. Safe and portable. Note that most modern compilers will + * eliminate the function call and treat it as an unaligned access. + * + * - `XXH_FORCE_MEMORY_ACCESS=1`: `__attribute__((aligned(1)))` + * @par + * Depends on compiler extensions and is therefore not portable. + * This method is safe _if_ your compiler supports it, + * and *generally* as fast or faster than `memcpy`. + * + * - `XXH_FORCE_MEMORY_ACCESS=2`: Direct cast + * @par + * Casts directly and dereferences. This method doesn't depend on the + * compiler, but it violates the C standard as it directly dereferences an + * unaligned pointer. It can generate buggy code on targets which do not + * support unaligned memory accesses, but in some circumstances, it's the + * only known way to get the most performance. + * + * - `XXH_FORCE_MEMORY_ACCESS=3`: Byteshift + * @par + * Also portable. This can generate the best code on old compilers which don't + * inline small `memcpy()` calls, and it might also be faster on big-endian + * systems which lack a native byteswap instruction. However, some compilers + * will emit literal byteshifts even if the target supports unaligned access. + * + * + * @warning + * Methods 1 and 2 rely on implementation-defined behavior. Use these with + * care, as what works on one compiler/platform/optimization level may cause + * another to read garbage data or even crash. + * + * See https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html for details. + * + * Prefer these methods in priority order (0 > 3 > 1 > 2) + */ +# define XXH_FORCE_MEMORY_ACCESS 0 + +/*! + * @def XXH_SIZE_OPT + * @brief Controls how much xxHash optimizes for size. + * + * xxHash, when compiled, tends to result in a rather large binary size. This + * is mostly due to heavy usage to forced inlining and constant folding of the + * @ref XXH3_family to increase performance. + * + * However, some developers prefer size over speed. This option can + * significantly reduce the size of the generated code. When using the `-Os` + * or `-Oz` options on GCC or Clang, this is defined to 1 by default, + * otherwise it is defined to 0. + * + * Most of these size optimizations can be controlled manually. + * + * This is a number from 0-2. + * - `XXH_SIZE_OPT` == 0: Default. xxHash makes no size optimizations. Speed + * comes first. + * - `XXH_SIZE_OPT` == 1: Default for `-Os` and `-Oz`. xxHash is more + * conservative and disables hacks that increase code size. It implies the + * options @ref XXH_NO_INLINE_HINTS == 1, @ref XXH_FORCE_ALIGN_CHECK == 0, + * and @ref XXH3_NEON_LANES == 8 if they are not already defined. + * - `XXH_SIZE_OPT` == 2: xxHash tries to make itself as small as possible. + * Performance may cry. For example, the single shot functions just use the + * streaming API. + */ +# define XXH_SIZE_OPT 0 + +/*! + * @def XXH_FORCE_ALIGN_CHECK + * @brief If defined to non-zero, adds a special path for aligned inputs (XXH32() + * and XXH64() only). + * + * This is an important performance trick for architectures without decent + * unaligned memory access performance. + * + * It checks for input alignment, and when conditions are met, uses a "fast + * path" employing direct 32-bit/64-bit reads, resulting in _dramatically + * faster_ read speed. + * + * The check costs one initial branch per hash, which is generally negligible, + * but not zero. + * + * Moreover, it's not useful to generate an additional code path if memory + * access uses the same instruction for both aligned and unaligned + * addresses (e.g. x86 and aarch64). + * + * In these cases, the alignment check can be removed by setting this macro to 0. + * Then the code will always use unaligned memory access. + * Align check is automatically disabled on x86, x64, ARM64, and some ARM chips + * which are platforms known to offer good unaligned memory accesses performance. + * + * It is also disabled by default when @ref XXH_SIZE_OPT >= 1. + * + * This option does not affect XXH3 (only XXH32 and XXH64). + */ +# define XXH_FORCE_ALIGN_CHECK 0 + +/*! + * @def XXH_NO_INLINE_HINTS + * @brief When non-zero, sets all functions to `static`. + * + * By default, xxHash tries to force the compiler to inline almost all internal + * functions. + * + * This can usually improve performance due to reduced jumping and improved + * constant folding, but significantly increases the size of the binary which + * might not be favorable. + * + * Additionally, sometimes the forced inlining can be detrimental to performance, + * depending on the architecture. + * + * XXH_NO_INLINE_HINTS marks all internal functions as static, giving the + * compiler full control on whether to inline or not. + * + * When not optimizing (-O0), using `-fno-inline` with GCC or Clang, or if + * @ref XXH_SIZE_OPT >= 1, this will automatically be defined. + */ +# define XXH_NO_INLINE_HINTS 0 + +/*! + * @def XXH3_INLINE_SECRET + * @brief Determines whether to inline the XXH3 withSecret code. + * + * When the secret size is known, the compiler can improve the performance + * of XXH3_64bits_withSecret() and XXH3_128bits_withSecret(). + * + * However, if the secret size is not known, it doesn't have any benefit. This + * happens when xxHash is compiled into a global symbol. Therefore, if + * @ref XXH_INLINE_ALL is *not* defined, this will be defined to 0. + * + * Additionally, this defaults to 0 on GCC 12+, which has an issue with function pointers + * that are *sometimes* force inline on -Og, and it is impossible to automatically + * detect this optimization level. + */ +# define XXH3_INLINE_SECRET 0 + +/*! + * @def XXH32_ENDJMP + * @brief Whether to use a jump for `XXH32_finalize`. + * + * For performance, `XXH32_finalize` uses multiple branches in the finalizer. + * This is generally preferable for performance, + * but depending on exact architecture, a jmp may be preferable. + * + * This setting is only possibly making a difference for very small inputs. + */ +# define XXH32_ENDJMP 0 + +/*! + * @internal + * @brief Redefines old internal names. + * + * For compatibility with code that uses xxHash's internals before the names + * were changed to improve namespacing. There is no other reason to use this. + */ +# define XXH_OLD_NAMES +# undef XXH_OLD_NAMES /* don't actually use, it is ugly. */ + +/*! + * @def XXH_NO_STREAM + * @brief Disables the streaming API. + * + * When xxHash is not inlined and the streaming functions are not used, disabling + * the streaming functions can improve code size significantly, especially with + * the @ref XXH3_family which tends to make constant folded copies of itself. + */ +# define XXH_NO_STREAM +# undef XXH_NO_STREAM /* don't actually */ +#endif /* XXH_DOXYGEN */ +/*! + * @} + */ + +#ifndef XXH_FORCE_MEMORY_ACCESS /* can be defined externally, on command line for example */ + /* prefer __packed__ structures (method 1) for GCC + * < ARMv7 with unaligned access (e.g. Raspbian armhf) still uses byte shifting, so we use memcpy + * which for some reason does unaligned loads. */ +# if defined(__GNUC__) && !(defined(__ARM_ARCH) && __ARM_ARCH < 7 && defined(__ARM_FEATURE_UNALIGNED)) +# define XXH_FORCE_MEMORY_ACCESS 1 +# endif +#endif + +#ifndef XXH_SIZE_OPT + /* default to 1 for -Os or -Oz */ +# if (defined(__GNUC__) || defined(__clang__)) && defined(__OPTIMIZE_SIZE__) +# define XXH_SIZE_OPT 1 +# else +# define XXH_SIZE_OPT 0 +# endif +#endif + +#ifndef XXH_FORCE_ALIGN_CHECK /* can be defined externally */ + /* don't check on sizeopt, x86, aarch64, or arm when unaligned access is available */ +# if XXH_SIZE_OPT >= 1 || \ + defined(__i386) || defined(__x86_64__) || defined(__aarch64__) || defined(__ARM_FEATURE_UNALIGNED) \ + || defined(_M_IX86) || defined(_M_X64) || defined(_M_ARM64) || defined(_M_ARM) /* visual */ +# define XXH_FORCE_ALIGN_CHECK 0 +# else +# define XXH_FORCE_ALIGN_CHECK 1 +# endif +#endif + +#ifndef XXH_NO_INLINE_HINTS +# if XXH_SIZE_OPT >= 1 || defined(__NO_INLINE__) /* -O0, -fno-inline */ +# define XXH_NO_INLINE_HINTS 1 +# else +# define XXH_NO_INLINE_HINTS 0 +# endif +#endif + +#ifndef XXH3_INLINE_SECRET +# if (defined(__GNUC__) && !defined(__clang__) && __GNUC__ >= 12) \ + || !defined(XXH_INLINE_ALL) +# define XXH3_INLINE_SECRET 0 +# else +# define XXH3_INLINE_SECRET 1 +# endif +#endif + +#ifndef XXH32_ENDJMP +/* generally preferable for performance */ +# define XXH32_ENDJMP 0 +#endif + +/*! + * @defgroup impl Implementation + * @{ + */ + + +/* ************************************* +* Includes & Memory related functions +***************************************/ +#if defined(XXH_NO_STREAM) +/* nothing */ +#elif defined(XXH_NO_STDLIB) + +/* When requesting to disable any mention of stdlib, + * the library loses the ability to invoked malloc / free. + * In practice, it means that functions like `XXH*_createState()` + * will always fail, and return NULL. + * This flag is useful in situations where + * xxhash.h is integrated into some kernel, embedded or limited environment + * without access to dynamic allocation. + */ + +static XXH_CONSTF void* XXH_malloc(size_t s) { (void)s; return NULL; } +static void XXH_free(void* p) { (void)p; } + +#else + +/* + * Modify the local functions below should you wish to use + * different memory routines for malloc() and free() + */ +#include + +/*! + * @internal + * @brief Modify this function to use a different routine than malloc(). + */ +static XXH_MALLOCF void* XXH_malloc(size_t s) { return malloc(s); } + +/*! + * @internal + * @brief Modify this function to use a different routine than free(). + */ +static void XXH_free(void* p) { free(p); } + +#endif /* XXH_NO_STDLIB */ + +#include + +/*! + * @internal + * @brief Modify this function to use a different routine than memcpy(). + */ +static void* XXH_memcpy(void* dest, const void* src, size_t size) +{ + return memcpy(dest,src,size); +} + +#include /* ULLONG_MAX */ + + +/* ************************************* +* Compiler Specific Options +***************************************/ +#ifdef _MSC_VER /* Visual Studio warning fix */ +# pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */ +#endif + +#if XXH_NO_INLINE_HINTS /* disable inlining hints */ +# if defined(__GNUC__) || defined(__clang__) +# define XXH_FORCE_INLINE static __attribute__((__unused__)) +# else +# define XXH_FORCE_INLINE static +# endif +# define XXH_NO_INLINE static +/* enable inlining hints */ +#elif defined(__GNUC__) || defined(__clang__) +# define XXH_FORCE_INLINE static __inline__ __attribute__((__always_inline__, __unused__)) +# define XXH_NO_INLINE static __attribute__((__noinline__)) +#elif defined(_MSC_VER) /* Visual Studio */ +# define XXH_FORCE_INLINE static __forceinline +# define XXH_NO_INLINE static __declspec(noinline) +#elif defined (__cplusplus) \ + || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L)) /* C99 */ +# define XXH_FORCE_INLINE static inline +# define XXH_NO_INLINE static +#else +# define XXH_FORCE_INLINE static +# define XXH_NO_INLINE static +#endif + +#if defined(XXH_INLINE_ALL) +# define XXH_STATIC XXH_FORCE_INLINE +#else +# define XXH_STATIC static +#endif + +#if XXH3_INLINE_SECRET +# define XXH3_WITH_SECRET_INLINE XXH_FORCE_INLINE +#else +# define XXH3_WITH_SECRET_INLINE XXH_NO_INLINE +#endif + +#if ((defined(sun) || defined(__sun)) && __cplusplus) /* Solaris includes __STDC_VERSION__ with C++. Tested with GCC 5.5 */ +# define XXH_RESTRICT /* disable */ +#elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* >= C99 */ +# define XXH_RESTRICT restrict +#elif (defined (__GNUC__) && ((__GNUC__ > 3) || (__GNUC__ == 3 && __GNUC_MINOR__ >= 1))) \ + || (defined (__clang__)) \ + || (defined (_MSC_VER) && (_MSC_VER >= 1400)) \ + || (defined (__INTEL_COMPILER) && (__INTEL_COMPILER >= 1300)) +/* + * There are a LOT more compilers that recognize __restrict but this + * covers the major ones. + */ +# define XXH_RESTRICT __restrict +#else +# define XXH_RESTRICT /* disable */ +#endif + +/* ************************************* +* Debug +***************************************/ +/*! + * @ingroup tuning + * @def XXH_DEBUGLEVEL + * @brief Sets the debugging level. + * + * XXH_DEBUGLEVEL is expected to be defined externally, typically via the + * compiler's command line options. The value must be a number. + */ +#ifndef XXH_DEBUGLEVEL +# ifdef DEBUGLEVEL /* backwards compat */ +# define XXH_DEBUGLEVEL DEBUGLEVEL +# else +# define XXH_DEBUGLEVEL 0 +# endif +#endif + +#if (XXH_DEBUGLEVEL>=1) +# include /* note: can still be disabled with NDEBUG */ +# define XXH_ASSERT(c) assert(c) +#else +# if defined(__INTEL_COMPILER) +# define XXH_ASSERT(c) XXH_ASSUME((unsigned char) (c)) +# else +# define XXH_ASSERT(c) XXH_ASSUME(c) +# endif +#endif + +/* note: use after variable declarations */ +#ifndef XXH_STATIC_ASSERT +# if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* C11 */ +# define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { _Static_assert((c),m); } while(0) +# elif defined(__cplusplus) && (__cplusplus >= 201103L) /* C++11 */ +# define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { static_assert((c),m); } while(0) +# else +# define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { struct xxh_sa { char x[(c) ? 1 : -1]; }; } while(0) +# endif +# define XXH_STATIC_ASSERT(c) XXH_STATIC_ASSERT_WITH_MESSAGE((c),#c) +#endif + +/*! + * @internal + * @def XXH_COMPILER_GUARD(var) + * @brief Used to prevent unwanted optimizations for @p var. + * + * It uses an empty GCC inline assembly statement with a register constraint + * which forces @p var into a general purpose register (eg eax, ebx, ecx + * on x86) and marks it as modified. + * + * This is used in a few places to avoid unwanted autovectorization (e.g. + * XXH32_round()). All vectorization we want is explicit via intrinsics, + * and _usually_ isn't wanted elsewhere. + * + * We also use it to prevent unwanted constant folding for AArch64 in + * XXH3_initCustomSecret_scalar(). + */ +#if defined(__GNUC__) || defined(__clang__) +# define XXH_COMPILER_GUARD(var) __asm__("" : "+r" (var)) +#else +# define XXH_COMPILER_GUARD(var) ((void)0) +#endif + +/* Specifically for NEON vectors which use the "w" constraint, on + * Clang. */ +#if defined(__clang__) && defined(__ARM_ARCH) && !defined(__wasm__) +# define XXH_COMPILER_GUARD_CLANG_NEON(var) __asm__("" : "+w" (var)) +#else +# define XXH_COMPILER_GUARD_CLANG_NEON(var) ((void)0) +#endif + +/* ************************************* +* Basic Types +***************************************/ +#if !defined (__VMS) \ + && (defined (__cplusplus) \ + || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) +# ifdef _AIX +# include +# else +# include +# endif + typedef uint8_t xxh_u8; +#else + typedef unsigned char xxh_u8; +#endif +typedef XXH32_hash_t xxh_u32; + +#ifdef XXH_OLD_NAMES +# warning "XXH_OLD_NAMES is planned to be removed starting v0.9. If the program depends on it, consider moving away from it by employing newer type names directly" +# define BYTE xxh_u8 +# define U8 xxh_u8 +# define U32 xxh_u32 +#endif + +/* *** Memory access *** */ + +/*! + * @internal + * @fn xxh_u32 XXH_read32(const void* ptr) + * @brief Reads an unaligned 32-bit integer from @p ptr in native endianness. + * + * Affected by @ref XXH_FORCE_MEMORY_ACCESS. + * + * @param ptr The pointer to read from. + * @return The 32-bit native endian integer from the bytes at @p ptr. + */ + +/*! + * @internal + * @fn xxh_u32 XXH_readLE32(const void* ptr) + * @brief Reads an unaligned 32-bit little endian integer from @p ptr. + * + * Affected by @ref XXH_FORCE_MEMORY_ACCESS. + * + * @param ptr The pointer to read from. + * @return The 32-bit little endian integer from the bytes at @p ptr. + */ + +/*! + * @internal + * @fn xxh_u32 XXH_readBE32(const void* ptr) + * @brief Reads an unaligned 32-bit big endian integer from @p ptr. + * + * Affected by @ref XXH_FORCE_MEMORY_ACCESS. + * + * @param ptr The pointer to read from. + * @return The 32-bit big endian integer from the bytes at @p ptr. + */ + +/*! + * @internal + * @fn xxh_u32 XXH_readLE32_align(const void* ptr, XXH_alignment align) + * @brief Like @ref XXH_readLE32(), but has an option for aligned reads. + * + * Affected by @ref XXH_FORCE_MEMORY_ACCESS. + * Note that when @ref XXH_FORCE_ALIGN_CHECK == 0, the @p align parameter is + * always @ref XXH_alignment::XXH_unaligned. + * + * @param ptr The pointer to read from. + * @param align Whether @p ptr is aligned. + * @pre + * If @p align == @ref XXH_alignment::XXH_aligned, @p ptr must be 4 byte + * aligned. + * @return The 32-bit little endian integer from the bytes at @p ptr. + */ + +#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3)) +/* + * Manual byteshift. Best for old compilers which don't inline memcpy. + * We actually directly use XXH_readLE32 and XXH_readBE32. + */ +#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2)) + +/* + * Force direct memory access. Only works on CPU which support unaligned memory + * access in hardware. + */ +static xxh_u32 XXH_read32(const void* memPtr) { return *(const xxh_u32*) memPtr; } + +#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1)) + +/* + * __attribute__((aligned(1))) is supported by gcc and clang. Originally the + * documentation claimed that it only increased the alignment, but actually it + * can decrease it on gcc, clang, and icc: + * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502, + * https://gcc.godbolt.org/z/xYez1j67Y. + */ +#ifdef XXH_OLD_NAMES +typedef union { xxh_u32 u32; } __attribute__((__packed__)) unalign; +#endif +static xxh_u32 XXH_read32(const void* ptr) +{ + typedef __attribute__((__aligned__(1))) xxh_u32 xxh_unalign32; + return *((const xxh_unalign32*)ptr); +} + +#else + +/* + * Portable and safe solution. Generally efficient. + * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html + */ +static xxh_u32 XXH_read32(const void* memPtr) +{ + xxh_u32 val; + XXH_memcpy(&val, memPtr, sizeof(val)); + return val; +} + +#endif /* XXH_FORCE_DIRECT_MEMORY_ACCESS */ + + +/* *** Endianness *** */ + +/*! + * @ingroup tuning + * @def XXH_CPU_LITTLE_ENDIAN + * @brief Whether the target is little endian. + * + * Defined to 1 if the target is little endian, or 0 if it is big endian. + * It can be defined externally, for example on the compiler command line. + * + * If it is not defined, + * a runtime check (which is usually constant folded) is used instead. + * + * @note + * This is not necessarily defined to an integer constant. + * + * @see XXH_isLittleEndian() for the runtime check. + */ +#ifndef XXH_CPU_LITTLE_ENDIAN +/* + * Try to detect endianness automatically, to avoid the nonstandard behavior + * in `XXH_isLittleEndian()` + */ +# if defined(_WIN32) /* Windows is always little endian */ \ + || defined(__LITTLE_ENDIAN__) \ + || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) +# define XXH_CPU_LITTLE_ENDIAN 1 +# elif defined(__BIG_ENDIAN__) \ + || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) +# define XXH_CPU_LITTLE_ENDIAN 0 +# else +/*! + * @internal + * @brief Runtime check for @ref XXH_CPU_LITTLE_ENDIAN. + * + * Most compilers will constant fold this. + */ +static int XXH_isLittleEndian(void) +{ + /* + * Portable and well-defined behavior. + * Don't use static: it is detrimental to performance. + */ + const union { xxh_u32 u; xxh_u8 c[4]; } one = { 1 }; + return one.c[0]; +} +# define XXH_CPU_LITTLE_ENDIAN XXH_isLittleEndian() +# endif +#endif + + + + +/* **************************************** +* Compiler-specific Functions and Macros +******************************************/ +#define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__) + +#ifdef __has_builtin +# define XXH_HAS_BUILTIN(x) __has_builtin(x) +#else +# define XXH_HAS_BUILTIN(x) 0 +#endif + + + +/* + * C23 and future versions have standard "unreachable()". + * Once it has been implemented reliably we can add it as an + * additional case: + * + * ``` + * #if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= XXH_C23_VN) + * # include + * # ifdef unreachable + * # define XXH_UNREACHABLE() unreachable() + * # endif + * #endif + * ``` + * + * Note C++23 also has std::unreachable() which can be detected + * as follows: + * ``` + * #if defined(__cpp_lib_unreachable) && (__cpp_lib_unreachable >= 202202L) + * # include + * # define XXH_UNREACHABLE() std::unreachable() + * #endif + * ``` + * NB: `__cpp_lib_unreachable` is defined in the `` header. + * We don't use that as including `` in `extern "C"` blocks + * doesn't work on GCC12 + */ + +#if XXH_HAS_BUILTIN(__builtin_unreachable) +# define XXH_UNREACHABLE() __builtin_unreachable() + +#elif defined(_MSC_VER) +# define XXH_UNREACHABLE() __assume(0) + +#else +# define XXH_UNREACHABLE() +#endif + +#if XXH_HAS_BUILTIN(__builtin_assume) +# define XXH_ASSUME(c) __builtin_assume(c) +#else +# define XXH_ASSUME(c) if (!(c)) { XXH_UNREACHABLE(); } +#endif + +/*! + * @internal + * @def XXH_rotl32(x,r) + * @brief 32-bit rotate left. + * + * @param x The 32-bit integer to be rotated. + * @param r The number of bits to rotate. + * @pre + * @p r > 0 && @p r < 32 + * @note + * @p x and @p r may be evaluated multiple times. + * @return The rotated result. + */ +#if !defined(NO_CLANG_BUILTIN) && XXH_HAS_BUILTIN(__builtin_rotateleft32) \ + && XXH_HAS_BUILTIN(__builtin_rotateleft64) +# define XXH_rotl32 __builtin_rotateleft32 +# define XXH_rotl64 __builtin_rotateleft64 +#elif XXH_HAS_BUILTIN(__builtin_stdc_rotate_left) +# define XXH_rotl32 __builtin_stdc_rotate_left +# define XXH_rotl64 __builtin_stdc_rotate_left +/* Note: although _rotl exists for minGW (GCC under windows), performance seems poor */ +#elif defined(_MSC_VER) +# define XXH_rotl32(x,r) _rotl(x,r) +# define XXH_rotl64(x,r) _rotl64(x,r) +#else +# define XXH_rotl32(x,r) (((x) << (r)) | ((x) >> (32 - (r)))) +# define XXH_rotl64(x,r) (((x) << (r)) | ((x) >> (64 - (r)))) +#endif + +/*! + * @internal + * @fn xxh_u32 XXH_swap32(xxh_u32 x) + * @brief A 32-bit byteswap. + * + * @param x The 32-bit integer to byteswap. + * @return @p x, byteswapped. + */ +#if defined(_MSC_VER) /* Visual Studio */ +# define XXH_swap32 _byteswap_ulong +#elif XXH_GCC_VERSION >= 403 +# define XXH_swap32 __builtin_bswap32 +#else +static xxh_u32 XXH_swap32 (xxh_u32 x) +{ + return ((x << 24) & 0xff000000 ) | + ((x << 8) & 0x00ff0000 ) | + ((x >> 8) & 0x0000ff00 ) | + ((x >> 24) & 0x000000ff ); +} +#endif + + +/* *************************** +* Memory reads +*****************************/ + +/*! + * @internal + * @brief Enum to indicate whether a pointer is aligned. + */ +typedef enum { + XXH_aligned, /*!< Aligned */ + XXH_unaligned /*!< Possibly unaligned */ +} XXH_alignment; + +/* + * XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load. + * + * This is ideal for older compilers which don't inline memcpy. + */ +#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3)) + +XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* memPtr) +{ + const xxh_u8* bytePtr = (const xxh_u8 *)memPtr; + return bytePtr[0] + | ((xxh_u32)bytePtr[1] << 8) + | ((xxh_u32)bytePtr[2] << 16) + | ((xxh_u32)bytePtr[3] << 24); +} + +XXH_FORCE_INLINE xxh_u32 XXH_readBE32(const void* memPtr) +{ + const xxh_u8* bytePtr = (const xxh_u8 *)memPtr; + return bytePtr[3] + | ((xxh_u32)bytePtr[2] << 8) + | ((xxh_u32)bytePtr[1] << 16) + | ((xxh_u32)bytePtr[0] << 24); +} + +#else +XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* ptr) +{ + return XXH_CPU_LITTLE_ENDIAN ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr)); +} + +static xxh_u32 XXH_readBE32(const void* ptr) +{ + return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr); +} +#endif + +XXH_FORCE_INLINE xxh_u32 +XXH_readLE32_align(const void* ptr, XXH_alignment align) +{ + if (align==XXH_unaligned) { + return XXH_readLE32(ptr); + } else { + return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u32*)ptr : XXH_swap32(*(const xxh_u32*)ptr); + } +} + + +/* ************************************* +* Misc +***************************************/ +/*! @ingroup public */ +XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; } + + +/* ******************************************************************* +* 32-bit hash functions +*********************************************************************/ +/*! + * @} + * @defgroup XXH32_impl XXH32 implementation + * @ingroup impl + * + * Details on the XXH32 implementation. + * @{ + */ + /* #define instead of static const, to be used as initializers */ +#define XXH_PRIME32_1 0x9E3779B1U /*!< 0b10011110001101110111100110110001 */ +#define XXH_PRIME32_2 0x85EBCA77U /*!< 0b10000101111010111100101001110111 */ +#define XXH_PRIME32_3 0xC2B2AE3DU /*!< 0b11000010101100101010111000111101 */ +#define XXH_PRIME32_4 0x27D4EB2FU /*!< 0b00100111110101001110101100101111 */ +#define XXH_PRIME32_5 0x165667B1U /*!< 0b00010110010101100110011110110001 */ + +#ifdef XXH_OLD_NAMES +# define PRIME32_1 XXH_PRIME32_1 +# define PRIME32_2 XXH_PRIME32_2 +# define PRIME32_3 XXH_PRIME32_3 +# define PRIME32_4 XXH_PRIME32_4 +# define PRIME32_5 XXH_PRIME32_5 +#endif + +/*! + * @internal + * @brief Normal stripe processing routine. + * + * This shuffles the bits so that any bit from @p input impacts several bits in + * @p acc. + * + * @param acc The accumulator lane. + * @param input The stripe of input to mix. + * @return The mixed accumulator lane. + */ +static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input) +{ + acc += input * XXH_PRIME32_2; + acc = XXH_rotl32(acc, 13); + acc *= XXH_PRIME32_1; +#if (defined(__SSE4_1__) || defined(__aarch64__) || defined(__wasm_simd128__)) && !defined(XXH_ENABLE_AUTOVECTORIZE) + /* + * UGLY HACK: + * A compiler fence is used to prevent GCC and Clang from + * autovectorizing the XXH32 loop (pragmas and attributes don't work for some + * reason) without globally disabling SSE4.1. + * + * The reason we want to avoid vectorization is because despite working on + * 4 integers at a time, there are multiple factors slowing XXH32 down on + * SSE4: + * - There's a ridiculous amount of lag from pmulld (10 cycles of latency on + * newer chips!) making it slightly slower to multiply four integers at + * once compared to four integers independently. Even when pmulld was + * fastest, Sandy/Ivy Bridge, it is still not worth it to go into SSE + * just to multiply unless doing a long operation. + * + * - Four instructions are required to rotate, + * movqda tmp, v // not required with VEX encoding + * pslld tmp, 13 // tmp <<= 13 + * psrld v, 19 // x >>= 19 + * por v, tmp // x |= tmp + * compared to one for scalar: + * roll v, 13 // reliably fast across the board + * shldl v, v, 13 // Sandy Bridge and later prefer this for some reason + * + * - Instruction level parallelism is actually more beneficial here because + * the SIMD actually serializes this operation: While v1 is rotating, v2 + * can load data, while v3 can multiply. SSE forces them to operate + * together. + * + * This is also enabled on AArch64, as Clang is *very aggressive* in vectorizing + * the loop. NEON is only faster on the A53, and with the newer cores, it is less + * than half the speed. + * + * Additionally, this is used on WASM SIMD128 because it JITs to the same + * SIMD instructions and has the same issue. + */ + XXH_COMPILER_GUARD(acc); +#endif + return acc; +} + +/*! + * @internal + * @brief Mixes all bits to finalize the hash. + * + * The final mix ensures that all input bits have a chance to impact any bit in + * the output digest, resulting in an unbiased distribution. + * + * @param hash The hash to avalanche. + * @return The avalanched hash. + */ +static xxh_u32 XXH32_avalanche(xxh_u32 hash) +{ + hash ^= hash >> 15; + hash *= XXH_PRIME32_2; + hash ^= hash >> 13; + hash *= XXH_PRIME32_3; + hash ^= hash >> 16; + return hash; +} + +#define XXH_get32bits(p) XXH_readLE32_align(p, align) + +/*! + * @internal + * @brief Sets up the initial accumulator state for XXH32(). + */ +XXH_FORCE_INLINE void +XXH32_initAccs(xxh_u32 *acc, xxh_u32 seed) +{ + XXH_ASSERT(acc != NULL); + acc[0] = seed + XXH_PRIME32_1 + XXH_PRIME32_2; + acc[1] = seed + XXH_PRIME32_2; + acc[2] = seed + 0; + acc[3] = seed - XXH_PRIME32_1; +} + +/*! + * @internal + * @brief Consumes a block of data for XXH32(). + * + * @return the end input pointer. + */ +XXH_FORCE_INLINE const xxh_u8 * +XXH32_consumeLong( + xxh_u32 *XXH_RESTRICT acc, + xxh_u8 const *XXH_RESTRICT input, + size_t len, + XXH_alignment align +) +{ + const xxh_u8* const bEnd = input + len; + const xxh_u8* const limit = bEnd - 15; + XXH_ASSERT(acc != NULL); + XXH_ASSERT(input != NULL); + XXH_ASSERT(len >= 16); + do { + acc[0] = XXH32_round(acc[0], XXH_get32bits(input)); input += 4; + acc[1] = XXH32_round(acc[1], XXH_get32bits(input)); input += 4; + acc[2] = XXH32_round(acc[2], XXH_get32bits(input)); input += 4; + acc[3] = XXH32_round(acc[3], XXH_get32bits(input)); input += 4; + } while (input < limit); + + return input; +} + +/*! + * @internal + * @brief Merges the accumulator lanes together for XXH32() + */ +XXH_FORCE_INLINE XXH_PUREF xxh_u32 +XXH32_mergeAccs(const xxh_u32 *acc) +{ + XXH_ASSERT(acc != NULL); + return XXH_rotl32(acc[0], 1) + XXH_rotl32(acc[1], 7) + + XXH_rotl32(acc[2], 12) + XXH_rotl32(acc[3], 18); +} + +/*! + * @internal + * @brief Processes the last 0-15 bytes of @p ptr. + * + * There may be up to 15 bytes remaining to consume from the input. + * This final stage will digest them to ensure that all input bytes are present + * in the final mix. + * + * @param hash The hash to finalize. + * @param ptr The pointer to the remaining input. + * @param len The remaining length, modulo 16. + * @param align Whether @p ptr is aligned. + * @return The finalized hash. + * @see XXH64_finalize(). + */ +static XXH_PUREF xxh_u32 +XXH32_finalize(xxh_u32 hash, const xxh_u8* ptr, size_t len, XXH_alignment align) +{ +#define XXH_PROCESS1 do { \ + hash += (*ptr++) * XXH_PRIME32_5; \ + hash = XXH_rotl32(hash, 11) * XXH_PRIME32_1; \ +} while (0) + +#define XXH_PROCESS4 do { \ + hash += XXH_get32bits(ptr) * XXH_PRIME32_3; \ + ptr += 4; \ + hash = XXH_rotl32(hash, 17) * XXH_PRIME32_4; \ +} while (0) + + if (ptr==NULL) XXH_ASSERT(len == 0); + + /* Compact rerolled version; generally faster */ + if (!XXH32_ENDJMP) { + len &= 15; + while (len >= 4) { + XXH_PROCESS4; + len -= 4; + } + while (len > 0) { + XXH_PROCESS1; + --len; + } + return XXH32_avalanche(hash); + } else { + switch(len&15) /* or switch(bEnd - p) */ { + case 12: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 8: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 4: XXH_PROCESS4; + return XXH32_avalanche(hash); + + case 13: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 9: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 5: XXH_PROCESS4; + XXH_PROCESS1; + return XXH32_avalanche(hash); + + case 14: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 10: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 6: XXH_PROCESS4; + XXH_PROCESS1; + XXH_PROCESS1; + return XXH32_avalanche(hash); + + case 15: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 11: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 7: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 3: XXH_PROCESS1; + XXH_FALLTHROUGH; /* fallthrough */ + case 2: XXH_PROCESS1; + XXH_FALLTHROUGH; /* fallthrough */ + case 1: XXH_PROCESS1; + XXH_FALLTHROUGH; /* fallthrough */ + case 0: return XXH32_avalanche(hash); + } + XXH_ASSERT(0); + return hash; /* reaching this point is deemed impossible */ + } +} + +#ifdef XXH_OLD_NAMES +# define PROCESS1 XXH_PROCESS1 +# define PROCESS4 XXH_PROCESS4 +#else +# undef XXH_PROCESS1 +# undef XXH_PROCESS4 +#endif + +/*! + * @internal + * @brief The implementation for @ref XXH32(). + * + * @param input , len , seed Directly passed from @ref XXH32(). + * @param align Whether @p input is aligned. + * @return The calculated hash. + */ +XXH_FORCE_INLINE XXH_PUREF xxh_u32 +XXH32_endian_align(const xxh_u8* input, size_t len, xxh_u32 seed, XXH_alignment align) +{ + xxh_u32 h32; + + if (input==NULL) XXH_ASSERT(len == 0); + + if (len>=16) { + xxh_u32 acc[4]; + XXH32_initAccs(acc, seed); + + input = XXH32_consumeLong(acc, input, len, align); + + h32 = XXH32_mergeAccs(acc); + } else { + h32 = seed + XXH_PRIME32_5; + } + + h32 += (xxh_u32)len; + + return XXH32_finalize(h32, input, len&15, align); +} + +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t len, XXH32_hash_t seed) +{ +#if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2 + /* Simple version, good for code maintenance, but unfortunately slow for small inputs */ + XXH32_state_t state; + XXH32_reset(&state, seed); + XXH32_update(&state, (const xxh_u8*)input, len); + return XXH32_digest(&state); +#else + if (XXH_FORCE_ALIGN_CHECK) { + if ((((size_t)input) & 3) == 0) { /* Input is 4-bytes aligned, leverage the speed benefit */ + return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_aligned); + } } + + return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned); +#endif +} + + + +/******* Hash streaming *******/ +#ifndef XXH_NO_STREAM +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void) +{ + return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t)); +} +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr) +{ + XXH_free(statePtr); + return XXH_OK; +} + +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState) +{ + XXH_memcpy(dstState, srcState, sizeof(*dstState)); +} + +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, XXH32_hash_t seed) +{ + XXH_ASSERT(statePtr != NULL); + memset(statePtr, 0, sizeof(*statePtr)); + XXH32_initAccs(statePtr->acc, seed); + return XXH_OK; +} + + +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH_errorcode +XXH32_update(XXH32_state_t* state, const void* input, size_t len) +{ + if (input==NULL) { + XXH_ASSERT(len == 0); + return XXH_OK; + } + + state->total_len_32 += (XXH32_hash_t)len; + state->large_len |= (XXH32_hash_t)((len>=16) | (state->total_len_32>=16)); + + XXH_ASSERT(state->bufferedSize < sizeof(state->buffer)); + if (len < sizeof(state->buffer) - state->bufferedSize) { /* fill in tmp buffer */ + XXH_memcpy(state->buffer + state->bufferedSize, input, len); + state->bufferedSize += (XXH32_hash_t)len; + return XXH_OK; + } + + { const xxh_u8* xinput = (const xxh_u8*)input; + const xxh_u8* const bEnd = xinput + len; + + if (state->bufferedSize) { /* non-empty buffer: complete first */ + XXH_memcpy(state->buffer + state->bufferedSize, xinput, sizeof(state->buffer) - state->bufferedSize); + xinput += sizeof(state->buffer) - state->bufferedSize; + /* then process one round */ + (void)XXH32_consumeLong(state->acc, state->buffer, sizeof(state->buffer), XXH_aligned); + state->bufferedSize = 0; + } + + XXH_ASSERT(xinput <= bEnd); + if ((size_t)(bEnd - xinput) >= sizeof(state->buffer)) { + /* Process the remaining data */ + xinput = XXH32_consumeLong(state->acc, xinput, (size_t)(bEnd - xinput), XXH_unaligned); + } + + if (xinput < bEnd) { + /* Copy the leftover to the tmp buffer */ + XXH_memcpy(state->buffer, xinput, (size_t)(bEnd-xinput)); + state->bufferedSize = (unsigned)(bEnd-xinput); + } + } + + return XXH_OK; +} + + +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t* state) +{ + xxh_u32 h32; + + if (state->large_len) { + h32 = XXH32_mergeAccs(state->acc); + } else { + h32 = state->acc[2] /* == seed */ + XXH_PRIME32_5; + } + + h32 += state->total_len_32; + + return XXH32_finalize(h32, state->buffer, state->bufferedSize, XXH_aligned); +} +#endif /* !XXH_NO_STREAM */ + +/******* Canonical representation *******/ + +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash) +{ + XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t)); + if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash); + XXH_memcpy(dst, &hash, sizeof(*dst)); +} +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src) +{ + return XXH_readBE32(src); +} + + +#ifndef XXH_NO_LONG_LONG + +/* ******************************************************************* +* 64-bit hash functions +*********************************************************************/ +/*! + * @} + * @ingroup impl + * @{ + */ +/******* Memory access *******/ + +typedef XXH64_hash_t xxh_u64; + +#ifdef XXH_OLD_NAMES +# define U64 xxh_u64 +#endif + +#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3)) +/* + * Manual byteshift. Best for old compilers which don't inline memcpy. + * We actually directly use XXH_readLE64 and XXH_readBE64. + */ +#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2)) + +/* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */ +static xxh_u64 XXH_read64(const void* memPtr) +{ + return *(const xxh_u64*) memPtr; +} + +#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1)) + +/* + * __attribute__((aligned(1))) is supported by gcc and clang. Originally the + * documentation claimed that it only increased the alignment, but actually it + * can decrease it on gcc, clang, and icc: + * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502, + * https://gcc.godbolt.org/z/xYez1j67Y. + */ +#ifdef XXH_OLD_NAMES +typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((__packed__)) unalign64; +#endif +static xxh_u64 XXH_read64(const void* ptr) +{ + typedef __attribute__((__aligned__(1))) xxh_u64 xxh_unalign64; + return *((const xxh_unalign64*)ptr); +} + +#else + +/* + * Portable and safe solution. Generally efficient. + * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html + */ +static xxh_u64 XXH_read64(const void* memPtr) +{ + xxh_u64 val; + XXH_memcpy(&val, memPtr, sizeof(val)); + return val; +} + +#endif /* XXH_FORCE_DIRECT_MEMORY_ACCESS */ + +#if defined(_MSC_VER) /* Visual Studio */ +# define XXH_swap64 _byteswap_uint64 +#elif XXH_GCC_VERSION >= 403 +# define XXH_swap64 __builtin_bswap64 +#else +static xxh_u64 XXH_swap64(xxh_u64 x) +{ + return ((x << 56) & 0xff00000000000000ULL) | + ((x << 40) & 0x00ff000000000000ULL) | + ((x << 24) & 0x0000ff0000000000ULL) | + ((x << 8) & 0x000000ff00000000ULL) | + ((x >> 8) & 0x00000000ff000000ULL) | + ((x >> 24) & 0x0000000000ff0000ULL) | + ((x >> 40) & 0x000000000000ff00ULL) | + ((x >> 56) & 0x00000000000000ffULL); +} +#endif + + +/* XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load. */ +#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3)) + +XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* memPtr) +{ + const xxh_u8* bytePtr = (const xxh_u8 *)memPtr; + return bytePtr[0] + | ((xxh_u64)bytePtr[1] << 8) + | ((xxh_u64)bytePtr[2] << 16) + | ((xxh_u64)bytePtr[3] << 24) + | ((xxh_u64)bytePtr[4] << 32) + | ((xxh_u64)bytePtr[5] << 40) + | ((xxh_u64)bytePtr[6] << 48) + | ((xxh_u64)bytePtr[7] << 56); +} + +XXH_FORCE_INLINE xxh_u64 XXH_readBE64(const void* memPtr) +{ + const xxh_u8* bytePtr = (const xxh_u8 *)memPtr; + return bytePtr[7] + | ((xxh_u64)bytePtr[6] << 8) + | ((xxh_u64)bytePtr[5] << 16) + | ((xxh_u64)bytePtr[4] << 24) + | ((xxh_u64)bytePtr[3] << 32) + | ((xxh_u64)bytePtr[2] << 40) + | ((xxh_u64)bytePtr[1] << 48) + | ((xxh_u64)bytePtr[0] << 56); +} + +#else +XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* ptr) +{ + return XXH_CPU_LITTLE_ENDIAN ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr)); +} + +static xxh_u64 XXH_readBE64(const void* ptr) +{ + return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr); +} +#endif + +XXH_FORCE_INLINE xxh_u64 +XXH_readLE64_align(const void* ptr, XXH_alignment align) +{ + if (align==XXH_unaligned) + return XXH_readLE64(ptr); + else + return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u64*)ptr : XXH_swap64(*(const xxh_u64*)ptr); +} + + +/******* xxh64 *******/ +/*! + * @} + * @defgroup XXH64_impl XXH64 implementation + * @ingroup impl + * + * Details on the XXH64 implementation. + * @{ + */ +/* #define rather that static const, to be used as initializers */ +#define XXH_PRIME64_1 0x9E3779B185EBCA87ULL /*!< 0b1001111000110111011110011011000110000101111010111100101010000111 */ +#define XXH_PRIME64_2 0xC2B2AE3D27D4EB4FULL /*!< 0b1100001010110010101011100011110100100111110101001110101101001111 */ +#define XXH_PRIME64_3 0x165667B19E3779F9ULL /*!< 0b0001011001010110011001111011000110011110001101110111100111111001 */ +#define XXH_PRIME64_4 0x85EBCA77C2B2AE63ULL /*!< 0b1000010111101011110010100111011111000010101100101010111001100011 */ +#define XXH_PRIME64_5 0x27D4EB2F165667C5ULL /*!< 0b0010011111010100111010110010111100010110010101100110011111000101 */ + +#ifdef XXH_OLD_NAMES +# define PRIME64_1 XXH_PRIME64_1 +# define PRIME64_2 XXH_PRIME64_2 +# define PRIME64_3 XXH_PRIME64_3 +# define PRIME64_4 XXH_PRIME64_4 +# define PRIME64_5 XXH_PRIME64_5 +#endif + +/*! @copydoc XXH32_round */ +static xxh_u64 XXH64_round(xxh_u64 acc, xxh_u64 input) +{ + acc += input * XXH_PRIME64_2; + acc = XXH_rotl64(acc, 31); + acc *= XXH_PRIME64_1; +#if (defined(__AVX512F__)) && !defined(XXH_ENABLE_AUTOVECTORIZE) + /* + * DISABLE AUTOVECTORIZATION: + * A compiler fence is used to prevent GCC and Clang from + * autovectorizing the XXH64 loop (pragmas and attributes don't work for some + * reason) without globally disabling AVX512. + * + * Autovectorization of XXH64 tends to be detrimental, + * though the exact outcome may change depending on exact cpu and compiler version. + * For information, it has been reported as detrimental for Skylake-X, + * but possibly beneficial for Zen4. + * + * The default is to disable auto-vectorization, + * but you can select to enable it instead using `XXH_ENABLE_AUTOVECTORIZE` build variable. + */ + XXH_COMPILER_GUARD(acc); +#endif + return acc; +} + +static xxh_u64 XXH64_mergeRound(xxh_u64 acc, xxh_u64 val) +{ + val = XXH64_round(0, val); + acc ^= val; + acc = acc * XXH_PRIME64_1 + XXH_PRIME64_4; + return acc; +} + +/*! @copydoc XXH32_avalanche */ +static xxh_u64 XXH64_avalanche(xxh_u64 hash) +{ + hash ^= hash >> 33; + hash *= XXH_PRIME64_2; + hash ^= hash >> 29; + hash *= XXH_PRIME64_3; + hash ^= hash >> 32; + return hash; +} + + +#define XXH_get64bits(p) XXH_readLE64_align(p, align) + +/*! + * @internal + * @brief Sets up the initial accumulator state for XXH64(). + */ +XXH_FORCE_INLINE void +XXH64_initAccs(xxh_u64 *acc, xxh_u64 seed) +{ + XXH_ASSERT(acc != NULL); + acc[0] = seed + XXH_PRIME64_1 + XXH_PRIME64_2; + acc[1] = seed + XXH_PRIME64_2; + acc[2] = seed + 0; + acc[3] = seed - XXH_PRIME64_1; +} + +/*! + * @internal + * @brief Consumes a block of data for XXH64(). + * + * @return the end input pointer. + */ +XXH_FORCE_INLINE const xxh_u8 * +XXH64_consumeLong( + xxh_u64 *XXH_RESTRICT acc, + xxh_u8 const *XXH_RESTRICT input, + size_t len, + XXH_alignment align +) +{ + const xxh_u8* const bEnd = input + len; + const xxh_u8* const limit = bEnd - 31; + XXH_ASSERT(acc != NULL); + XXH_ASSERT(input != NULL); + XXH_ASSERT(len >= 32); + do { + /* reroll on 32-bit */ + if (sizeof(void *) < sizeof(xxh_u64)) { + size_t i; + for (i = 0; i < 4; i++) { + acc[i] = XXH64_round(acc[i], XXH_get64bits(input)); + input += 8; + } + } else { + acc[0] = XXH64_round(acc[0], XXH_get64bits(input)); input += 8; + acc[1] = XXH64_round(acc[1], XXH_get64bits(input)); input += 8; + acc[2] = XXH64_round(acc[2], XXH_get64bits(input)); input += 8; + acc[3] = XXH64_round(acc[3], XXH_get64bits(input)); input += 8; + } + } while (input < limit); + + return input; +} + +/*! + * @internal + * @brief Merges the accumulator lanes together for XXH64() + */ +XXH_FORCE_INLINE XXH_PUREF xxh_u64 +XXH64_mergeAccs(const xxh_u64 *acc) +{ + XXH_ASSERT(acc != NULL); + { + xxh_u64 h64 = XXH_rotl64(acc[0], 1) + XXH_rotl64(acc[1], 7) + + XXH_rotl64(acc[2], 12) + XXH_rotl64(acc[3], 18); + /* reroll on 32-bit */ + if (sizeof(void *) < sizeof(xxh_u64)) { + size_t i; + for (i = 0; i < 4; i++) { + h64 = XXH64_mergeRound(h64, acc[i]); + } + } else { + h64 = XXH64_mergeRound(h64, acc[0]); + h64 = XXH64_mergeRound(h64, acc[1]); + h64 = XXH64_mergeRound(h64, acc[2]); + h64 = XXH64_mergeRound(h64, acc[3]); + } + return h64; + } +} + +/*! + * @internal + * @brief Processes the last 0-31 bytes of @p ptr. + * + * There may be up to 31 bytes remaining to consume from the input. + * This final stage will digest them to ensure that all input bytes are present + * in the final mix. + * + * @param hash The hash to finalize. + * @param ptr The pointer to the remaining input. + * @param len The remaining length, modulo 32. + * @param align Whether @p ptr is aligned. + * @return The finalized hash + * @see XXH32_finalize(). + */ +XXH_STATIC XXH_PUREF xxh_u64 +XXH64_finalize(xxh_u64 hash, const xxh_u8* ptr, size_t len, XXH_alignment align) +{ + if (ptr==NULL) XXH_ASSERT(len == 0); + len &= 31; + while (len >= 8) { + xxh_u64 const k1 = XXH64_round(0, XXH_get64bits(ptr)); + ptr += 8; + hash ^= k1; + hash = XXH_rotl64(hash,27) * XXH_PRIME64_1 + XXH_PRIME64_4; + len -= 8; + } + if (len >= 4) { + hash ^= (xxh_u64)(XXH_get32bits(ptr)) * XXH_PRIME64_1; + ptr += 4; + hash = XXH_rotl64(hash, 23) * XXH_PRIME64_2 + XXH_PRIME64_3; + len -= 4; + } + while (len > 0) { + hash ^= (*ptr++) * XXH_PRIME64_5; + hash = XXH_rotl64(hash, 11) * XXH_PRIME64_1; + --len; + } + return XXH64_avalanche(hash); +} + +#ifdef XXH_OLD_NAMES +# define PROCESS1_64 XXH_PROCESS1_64 +# define PROCESS4_64 XXH_PROCESS4_64 +# define PROCESS8_64 XXH_PROCESS8_64 +#else +# undef XXH_PROCESS1_64 +# undef XXH_PROCESS4_64 +# undef XXH_PROCESS8_64 +#endif + +/*! + * @internal + * @brief The implementation for @ref XXH64(). + * + * @param input , len , seed Directly passed from @ref XXH64(). + * @param align Whether @p input is aligned. + * @return The calculated hash. + */ +XXH_FORCE_INLINE XXH_PUREF xxh_u64 +XXH64_endian_align(const xxh_u8* input, size_t len, xxh_u64 seed, XXH_alignment align) +{ + xxh_u64 h64; + if (input==NULL) XXH_ASSERT(len == 0); + + if (len>=32) { /* Process a large block of data */ + xxh_u64 acc[4]; + XXH64_initAccs(acc, seed); + + input = XXH64_consumeLong(acc, input, len, align); + + h64 = XXH64_mergeAccs(acc); + } else { + h64 = seed + XXH_PRIME64_5; + } + + h64 += (xxh_u64) len; + + return XXH64_finalize(h64, input, len, align); +} + + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API XXH64_hash_t XXH64 (XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed) +{ +#if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2 + /* Simple version, good for code maintenance, but unfortunately slow for small inputs */ + XXH64_state_t state; + XXH64_reset(&state, seed); + XXH64_update(&state, (const xxh_u8*)input, len); + return XXH64_digest(&state); +#else + if (XXH_FORCE_ALIGN_CHECK) { + if ((((size_t)input) & 7)==0) { /* Input is aligned, let's leverage the speed advantage */ + return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_aligned); + } } + + return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned); + +#endif +} + +/******* Hash Streaming *******/ +#ifndef XXH_NO_STREAM +/*! @ingroup XXH64_family*/ +XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void) +{ + return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t)); +} +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr) +{ + XXH_free(statePtr); + return XXH_OK; +} + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t* dstState, const XXH64_state_t* srcState) +{ + XXH_memcpy(dstState, srcState, sizeof(*dstState)); +} + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH_NOESCAPE XXH64_state_t* statePtr, XXH64_hash_t seed) +{ + XXH_ASSERT(statePtr != NULL); + memset(statePtr, 0, sizeof(*statePtr)); + XXH64_initAccs(statePtr->acc, seed); + return XXH_OK; +} + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API XXH_errorcode +XXH64_update (XXH_NOESCAPE XXH64_state_t* state, XXH_NOESCAPE const void* input, size_t len) +{ + if (input==NULL) { + XXH_ASSERT(len == 0); + return XXH_OK; + } + + state->total_len += len; + + XXH_ASSERT(state->bufferedSize <= sizeof(state->buffer)); + if (len < sizeof(state->buffer) - state->bufferedSize) { /* fill in tmp buffer */ + XXH_memcpy(state->buffer + state->bufferedSize, input, len); + state->bufferedSize += (XXH32_hash_t)len; + return XXH_OK; + } + + { const xxh_u8* xinput = (const xxh_u8*)input; + const xxh_u8* const bEnd = xinput + len; + + if (state->bufferedSize) { /* non-empty buffer => complete first */ + XXH_memcpy(state->buffer + state->bufferedSize, xinput, sizeof(state->buffer) - state->bufferedSize); + xinput += sizeof(state->buffer) - state->bufferedSize; + /* and process one round */ + (void)XXH64_consumeLong(state->acc, state->buffer, sizeof(state->buffer), XXH_aligned); + state->bufferedSize = 0; + } + + XXH_ASSERT(xinput <= bEnd); + if ((size_t)(bEnd - xinput) >= sizeof(state->buffer)) { + /* Process the remaining data */ + xinput = XXH64_consumeLong(state->acc, xinput, (size_t)(bEnd - xinput), XXH_unaligned); + } + + if (xinput < bEnd) { + /* Copy the leftover to the tmp buffer */ + XXH_memcpy(state->buffer, xinput, (size_t)(bEnd-xinput)); + state->bufferedSize = (unsigned)(bEnd-xinput); + } + } + + return XXH_OK; +} + + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API XXH64_hash_t XXH64_digest(XXH_NOESCAPE const XXH64_state_t* state) +{ + xxh_u64 h64; + + if (state->total_len >= 32) { + h64 = XXH64_mergeAccs(state->acc); + } else { + h64 = state->acc[2] /*seed*/ + XXH_PRIME64_5; + } + + h64 += (xxh_u64) state->total_len; + + return XXH64_finalize(h64, state->buffer, (size_t)state->total_len, XXH_aligned); +} +#endif /* !XXH_NO_STREAM */ + +/******* Canonical representation *******/ + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t* dst, XXH64_hash_t hash) +{ + XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t)); + if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash); + XXH_memcpy(dst, &hash, sizeof(*dst)); +} + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t* src) +{ + return XXH_readBE64(src); +} + +#ifndef XXH_NO_XXH3 + +/* ********************************************************************* +* XXH3 +* New generation hash designed for speed on small keys and vectorization +************************************************************************ */ +/*! + * @} + * @defgroup XXH3_impl XXH3 implementation + * @ingroup impl + * @{ + */ + +/* === Compiler specifics === */ + + +#if (defined(__GNUC__) && (__GNUC__ >= 3)) \ + || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) \ + || defined(__clang__) +# define XXH_likely(x) __builtin_expect(x, 1) +# define XXH_unlikely(x) __builtin_expect(x, 0) +#else +# define XXH_likely(x) (x) +# define XXH_unlikely(x) (x) +#endif + +#ifndef XXH_HAS_INCLUDE +# ifdef __has_include +/* + * Not defined as XXH_HAS_INCLUDE(x) (function-like) because + * this causes segfaults in Apple Clang 4.2 (on Mac OS X 10.7 Lion) + */ +# define XXH_HAS_INCLUDE __has_include +# else +# define XXH_HAS_INCLUDE(x) 0 +# endif +#endif + +#if defined(__GNUC__) || defined(__clang__) +# if defined(__ARM_FEATURE_SVE) +# include +# endif +# if defined(__ARM_NEON__) || defined(__ARM_NEON) \ + || (defined(_M_ARM) && _M_ARM >= 7) \ + || defined(_M_ARM64) || defined(_M_ARM64EC) \ + || (defined(__wasm_simd128__) && XXH_HAS_INCLUDE()) /* WASM SIMD128 via SIMDe */ +# define inline __inline__ /* circumvent a clang bug */ +# include +# undef inline +# elif defined(__AVX2__) +# include +# elif defined(__SSE2__) +# include +# elif defined(__loongarch_sx) +# include +# endif +#endif + +#if defined(_MSC_VER) +# include +#endif + +/* + * One goal of XXH3 is to make it fast on both 32-bit and 64-bit, while + * remaining a true 64-bit/128-bit hash function. + * + * This is done by prioritizing a subset of 64-bit operations that can be + * emulated without too many steps on the average 32-bit machine. + * + * For example, these two lines seem similar, and run equally fast on 64-bit: + * + * xxh_u64 x; + * x ^= (x >> 47); // good + * x ^= (x >> 13); // bad + * + * However, to a 32-bit machine, there is a major difference. + * + * x ^= (x >> 47) looks like this: + * + * x.lo ^= (x.hi >> (47 - 32)); + * + * while x ^= (x >> 13) looks like this: + * + * // note: funnel shifts are not usually cheap. + * x.lo ^= (x.lo >> 13) | (x.hi << (32 - 13)); + * x.hi ^= (x.hi >> 13); + * + * The first one is significantly faster than the second, simply because the + * shift is larger than 32. This means: + * - All the bits we need are in the upper 32 bits, so we can ignore the lower + * 32 bits in the shift. + * - The shift result will always fit in the lower 32 bits, and therefore, + * we can ignore the upper 32 bits in the xor. + * + * Thanks to this optimization, XXH3 only requires these features to be efficient: + * + * - Usable unaligned access + * - A 32-bit or 64-bit ALU + * - If 32-bit, a decent ADC instruction + * - A 32 or 64-bit multiply with a 64-bit result + * - For the 128-bit variant, a decent byteswap helps short inputs. + * + * The first two are already required by XXH32, and almost all 32-bit and 64-bit + * platforms which can run XXH32 can run XXH3 efficiently. + * + * Thumb-1, the classic 16-bit only subset of ARM's instruction set, is one + * notable exception. + * + * First of all, Thumb-1 lacks support for the UMULL instruction which + * performs the important long multiply. This means numerous __aeabi_lmul + * calls. + * + * Second of all, the 8 functional registers are just not enough. + * Setup for __aeabi_lmul, byteshift loads, pointers, and all arithmetic need + * Lo registers, and this shuffling results in thousands more MOVs than A32. + * + * A32 and T32 don't have this limitation. They can access all 14 registers, + * do a 32->64 multiply with UMULL, and the flexible operand allowing free + * shifts is helpful, too. + * + * Therefore, we do a quick sanity check. + * + * If compiling Thumb-1 for a target which supports ARM instructions, we will + * emit a warning, as it is not a "sane" platform to compile for. + * + * Usually, if this happens, it is because of an accident and you probably need + * to specify -march, as you likely meant to compile for a newer architecture. + * + * Credit: large sections of the vectorial and asm source code paths + * have been contributed by @easyaspi314 + */ +#if defined(__thumb__) && !defined(__thumb2__) && defined(__ARM_ARCH_ISA_ARM) +# warning "XXH3 is highly inefficient without ARM or Thumb-2." +#endif + +/* ========================================== + * Vectorization detection + * ========================================== */ + +#ifdef XXH_DOXYGEN +/*! + * @ingroup tuning + * @brief Overrides the vectorization implementation chosen for XXH3. + * + * Can be defined to 0 to disable SIMD or any of the values mentioned in + * @ref XXH_VECTOR_TYPE. + * + * If this is not defined, it uses predefined macros to determine the best + * implementation. + */ +# define XXH_VECTOR XXH_SCALAR +/*! + * @ingroup tuning + * @brief Selects the minimum alignment for XXH3's accumulators. + * + * When using SIMD, this should match the alignment required for said vector + * type, so, for example, 32 for AVX2. + * + * Default: Auto detected. + */ +# define XXH_ACC_ALIGN 8 +#endif + +/* Actual definition */ +#ifndef XXH_DOXYGEN +#endif + +#ifndef XXH_VECTOR /* can be defined on command line */ +# if defined(__ARM_FEATURE_SVE) +# define XXH_VECTOR XXH_SVE +# elif ( \ + defined(__ARM_NEON__) || defined(__ARM_NEON) /* gcc */ \ + || defined(_M_ARM) || defined(_M_ARM64) || defined(_M_ARM64EC) /* msvc */ \ + || (defined(__wasm_simd128__) && XXH_HAS_INCLUDE()) /* wasm simd128 via SIMDe */ \ + ) && ( \ + defined(_WIN32) || defined(__LITTLE_ENDIAN__) /* little endian only */ \ + || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ + ) +# define XXH_VECTOR XXH_NEON +# elif defined(__AVX512F__) +# define XXH_VECTOR XXH_AVX512 +# elif defined(__AVX2__) +# define XXH_VECTOR XXH_AVX2 +# elif defined(__SSE2__) || defined(_M_AMD64) || defined(_M_X64) || (defined(_M_IX86_FP) && (_M_IX86_FP == 2)) +# define XXH_VECTOR XXH_SSE2 +# elif (defined(__PPC64__) && defined(__POWER8_VECTOR__)) \ + || (defined(__s390x__) && defined(__VEC__)) \ + && defined(__GNUC__) /* TODO: IBM XL */ +# define XXH_VECTOR XXH_VSX +# elif defined(__loongarch_sx) +# define XXH_VECTOR XXH_LSX +# else +# define XXH_VECTOR XXH_SCALAR +# endif +#endif + +/* __ARM_FEATURE_SVE is only supported by GCC & Clang. */ +#if (XXH_VECTOR == XXH_SVE) && !defined(__ARM_FEATURE_SVE) +# ifdef _MSC_VER +# pragma warning(once : 4606) +# else +# warning "__ARM_FEATURE_SVE isn't supported. Use SCALAR instead." +# endif +# undef XXH_VECTOR +# define XXH_VECTOR XXH_SCALAR +#endif + +/* + * Controls the alignment of the accumulator, + * for compatibility with aligned vector loads, which are usually faster. + */ +#ifndef XXH_ACC_ALIGN +# if defined(XXH_X86DISPATCH) +# define XXH_ACC_ALIGN 64 /* for compatibility with avx512 */ +# elif XXH_VECTOR == XXH_SCALAR /* scalar */ +# define XXH_ACC_ALIGN 8 +# elif XXH_VECTOR == XXH_SSE2 /* sse2 */ +# define XXH_ACC_ALIGN 16 +# elif XXH_VECTOR == XXH_AVX2 /* avx2 */ +# define XXH_ACC_ALIGN 32 +# elif XXH_VECTOR == XXH_NEON /* neon */ +# define XXH_ACC_ALIGN 16 +# elif XXH_VECTOR == XXH_VSX /* vsx */ +# define XXH_ACC_ALIGN 16 +# elif XXH_VECTOR == XXH_AVX512 /* avx512 */ +# define XXH_ACC_ALIGN 64 +# elif XXH_VECTOR == XXH_SVE /* sve */ +# define XXH_ACC_ALIGN 64 +# elif XXH_VECTOR == XXH_LSX /* lsx */ +# define XXH_ACC_ALIGN 64 +# endif +#endif + +#if defined(XXH_X86DISPATCH) || XXH_VECTOR == XXH_SSE2 \ + || XXH_VECTOR == XXH_AVX2 || XXH_VECTOR == XXH_AVX512 +# define XXH_SEC_ALIGN XXH_ACC_ALIGN +#elif XXH_VECTOR == XXH_SVE +# define XXH_SEC_ALIGN XXH_ACC_ALIGN +#else +# define XXH_SEC_ALIGN 8 +#endif + +#if defined(__GNUC__) || defined(__clang__) +# define XXH_ALIASING __attribute__((__may_alias__)) +#else +# define XXH_ALIASING /* nothing */ +#endif + +/* + * UGLY HACK: + * GCC usually generates the best code with -O3 for xxHash. + * + * However, when targeting AVX2, it is overzealous in its unrolling resulting + * in code roughly 3/4 the speed of Clang. + * + * There are other issues, such as GCC splitting _mm256_loadu_si256 into + * _mm_loadu_si128 + _mm256_inserti128_si256. This is an optimization which + * only applies to Sandy and Ivy Bridge... which don't even support AVX2. + * + * That is why when compiling the AVX2 version, it is recommended to use either + * -O2 -mavx2 -march=haswell + * or + * -O2 -mavx2 -mno-avx256-split-unaligned-load + * for decent performance, or to use Clang instead. + * + * Fortunately, we can control the first one with a pragma that forces GCC into + * -O2, but the other one we can't control without "failed to inline always + * inline function due to target mismatch" warnings. + */ +#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \ + && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \ + && defined(__OPTIMIZE__) && XXH_SIZE_OPT <= 0 /* respect -O0 and -Os */ +# pragma GCC push_options +# pragma GCC optimize("-O2") +#endif + +#if XXH_VECTOR == XXH_NEON + +/* + * UGLY HACK: While AArch64 GCC on Linux does not seem to care, on macOS, GCC -O3 + * optimizes out the entire hashLong loop because of the aliasing violation. + * + * However, GCC is also inefficient at load-store optimization with vld1q/vst1q, + * so the only option is to mark it as aliasing. + */ +typedef uint64x2_t xxh_aliasing_uint64x2_t XXH_ALIASING; + +/*! + * @internal + * @brief `vld1q_u64` but faster and alignment-safe. + * + * On AArch64, unaligned access is always safe, but on ARMv7-a, it is only + * *conditionally* safe (`vld1` has an alignment bit like `movdq[ua]` in x86). + * + * GCC for AArch64 sees `vld1q_u8` as an intrinsic instead of a load, so it + * prohibits load-store optimizations. Therefore, a direct dereference is used. + * + * Otherwise, `vld1q_u8` is used with `vreinterpretq_u8_u64` to do a safe + * unaligned load. + */ +#if defined(__aarch64__) && defined(__GNUC__) && !defined(__clang__) +XXH_FORCE_INLINE uint64x2_t XXH_vld1q_u64(void const* ptr) /* silence -Wcast-align */ +{ + return *(xxh_aliasing_uint64x2_t const *)ptr; +} +#else +XXH_FORCE_INLINE uint64x2_t XXH_vld1q_u64(void const* ptr) +{ + return vreinterpretq_u64_u8(vld1q_u8((uint8_t const*)ptr)); +} +#endif + +/*! + * @internal + * @brief `vmlal_u32` on low and high halves of a vector. + * + * This is a workaround for AArch64 GCC < 11 which implemented arm_neon.h with + * inline assembly and were therefore incapable of merging the `vget_{low, high}_u32` + * with `vmlal_u32`. + */ +#if defined(__aarch64__) && defined(__GNUC__) && !defined(__clang__) && __GNUC__ < 11 +XXH_FORCE_INLINE uint64x2_t +XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs) +{ + /* Inline assembly is the only way */ + __asm__("umlal %0.2d, %1.2s, %2.2s" : "+w" (acc) : "w" (lhs), "w" (rhs)); + return acc; +} +XXH_FORCE_INLINE uint64x2_t +XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs) +{ + /* This intrinsic works as expected */ + return vmlal_high_u32(acc, lhs, rhs); +} +#else +/* Portable intrinsic versions */ +XXH_FORCE_INLINE uint64x2_t +XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs) +{ + return vmlal_u32(acc, vget_low_u32(lhs), vget_low_u32(rhs)); +} +/*! @copydoc XXH_vmlal_low_u32 + * Assume the compiler converts this to vmlal_high_u32 on aarch64 */ +XXH_FORCE_INLINE uint64x2_t +XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs) +{ + return vmlal_u32(acc, vget_high_u32(lhs), vget_high_u32(rhs)); +} +#endif + +/*! + * @ingroup tuning + * @brief Controls the NEON to scalar ratio for XXH3 + * + * This can be set to 2, 4, 6, or 8. + * + * ARM Cortex CPUs are _very_ sensitive to how their pipelines are used. + * + * For example, the Cortex-A73 can dispatch 3 micro-ops per cycle, but only 2 of those + * can be NEON. If you are only using NEON instructions, you are only using 2/3 of the CPU + * bandwidth. + * + * This is even more noticeable on the more advanced cores like the Cortex-A76 which + * can dispatch 8 micro-ops per cycle, but still only 2 NEON micro-ops at once. + * + * Therefore, to make the most out of the pipeline, it is beneficial to run 6 NEON lanes + * and 2 scalar lanes, which is chosen by default. + * + * This does not apply to Apple processors or 32-bit processors, which run better with + * full NEON. These will default to 8. Additionally, size-optimized builds run 8 lanes. + * + * This change benefits CPUs with large micro-op buffers without negatively affecting + * most other CPUs: + * + * | Chipset | Dispatch type | NEON only | 6:2 hybrid | Diff. | + * |:----------------------|:--------------------|----------:|-----------:|------:| + * | Snapdragon 730 (A76) | 2 NEON/8 micro-ops | 8.8 GB/s | 10.1 GB/s | ~16% | + * | Snapdragon 835 (A73) | 2 NEON/3 micro-ops | 5.1 GB/s | 5.3 GB/s | ~5% | + * | Marvell PXA1928 (A53) | In-order dual-issue | 1.9 GB/s | 1.9 GB/s | 0% | + * | Apple M1 | 4 NEON/8 micro-ops | 37.3 GB/s | 36.1 GB/s | ~-3% | + * + * It also seems to fix some bad codegen on GCC, making it almost as fast as clang. + * + * When using WASM SIMD128, if this is 2 or 6, SIMDe will scalarize 2 of the lanes meaning + * it effectively becomes worse 4. + * + * @see XXH3_accumulate_512_neon() + */ +# ifndef XXH3_NEON_LANES +# if (defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64) || defined(_M_ARM64EC)) \ + && !defined(__APPLE__) && XXH_SIZE_OPT <= 0 +# define XXH3_NEON_LANES 6 +# else +# define XXH3_NEON_LANES XXH_ACC_NB +# endif +# endif +#endif /* XXH_VECTOR == XXH_NEON */ + +/* + * VSX and Z Vector helpers. + * + * This is very messy, and any pull requests to clean this up are welcome. + * + * There are a lot of problems with supporting VSX and s390x, due to + * inconsistent intrinsics, spotty coverage, and multiple endiannesses. + */ +#if XXH_VECTOR == XXH_VSX +/* Annoyingly, these headers _may_ define three macros: `bool`, `vector`, + * and `pixel`. This is a problem for obvious reasons. + * + * These keywords are unnecessary; the spec literally says they are + * equivalent to `__bool`, `__vector`, and `__pixel` and may be undef'd + * after including the header. + * + * We use pragma push_macro/pop_macro to keep the namespace clean. */ +# pragma push_macro("bool") +# pragma push_macro("vector") +# pragma push_macro("pixel") +/* silence potential macro redefined warnings */ +# undef bool +# undef vector +# undef pixel + +# if defined(__s390x__) +# include +# else +# include +# endif + +/* Restore the original macro values, if applicable. */ +# pragma pop_macro("pixel") +# pragma pop_macro("vector") +# pragma pop_macro("bool") + +typedef __vector unsigned long long xxh_u64x2; +typedef __vector unsigned char xxh_u8x16; +typedef __vector unsigned xxh_u32x4; + +/* + * UGLY HACK: Similar to aarch64 macOS GCC, s390x GCC has the same aliasing issue. + */ +typedef xxh_u64x2 xxh_aliasing_u64x2 XXH_ALIASING; + +# ifndef XXH_VSX_BE +# if defined(__BIG_ENDIAN__) \ + || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) +# define XXH_VSX_BE 1 +# elif defined(__VEC_ELEMENT_REG_ORDER__) && __VEC_ELEMENT_REG_ORDER__ == __ORDER_BIG_ENDIAN__ +# warning "-maltivec=be is not recommended. Please use native endianness." +# define XXH_VSX_BE 1 +# else +# define XXH_VSX_BE 0 +# endif +# endif /* !defined(XXH_VSX_BE) */ + +# if XXH_VSX_BE +# if defined(__POWER9_VECTOR__) || (defined(__clang__) && defined(__s390x__)) +# define XXH_vec_revb vec_revb +# else +/*! + * A polyfill for POWER9's vec_revb(). + */ +XXH_FORCE_INLINE xxh_u64x2 XXH_vec_revb(xxh_u64x2 val) +{ + xxh_u8x16 const vByteSwap = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00, + 0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 }; + return vec_perm(val, val, vByteSwap); +} +# endif +# endif /* XXH_VSX_BE */ + +/*! + * Performs an unaligned vector load and byte swaps it on big endian. + */ +XXH_FORCE_INLINE xxh_u64x2 XXH_vec_loadu(const void *ptr) +{ + xxh_u64x2 ret; + XXH_memcpy(&ret, ptr, sizeof(xxh_u64x2)); +# if XXH_VSX_BE + ret = XXH_vec_revb(ret); +# endif + return ret; +} + +/* + * vec_mulo and vec_mule are very problematic intrinsics on PowerPC + * + * These intrinsics weren't added until GCC 8, despite existing for a while, + * and they are endian dependent. Also, their meaning swap depending on version. + * */ +# if defined(__s390x__) + /* s390x is always big endian, no issue on this platform */ +# define XXH_vec_mulo vec_mulo +# define XXH_vec_mule vec_mule +# elif defined(__clang__) && XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw) && !defined(__ibmxl__) +/* Clang has a better way to control this, we can just use the builtin which doesn't swap. */ + /* The IBM XL Compiler (which defined __clang__) only implements the vec_* operations */ +# define XXH_vec_mulo __builtin_altivec_vmulouw +# define XXH_vec_mule __builtin_altivec_vmuleuw +# else +/* gcc needs inline assembly */ +/* Adapted from https://github.com/google/highwayhash/blob/master/highwayhash/hh_vsx.h. */ +XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mulo(xxh_u32x4 a, xxh_u32x4 b) +{ + xxh_u64x2 result; + __asm__("vmulouw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b)); + return result; +} +XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mule(xxh_u32x4 a, xxh_u32x4 b) +{ + xxh_u64x2 result; + __asm__("vmuleuw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b)); + return result; +} +# endif /* XXH_vec_mulo, XXH_vec_mule */ +#endif /* XXH_VECTOR == XXH_VSX */ + +#if XXH_VECTOR == XXH_SVE +#define ACCRND(acc, offset) \ +do { \ + svuint64_t input_vec = svld1_u64(mask, xinput + offset); \ + svuint64_t secret_vec = svld1_u64(mask, xsecret + offset); \ + svuint64_t mixed = sveor_u64_x(mask, secret_vec, input_vec); \ + svuint64_t swapped = svtbl_u64(input_vec, kSwap); \ + svuint64_t mixed_lo = svextw_u64_x(mask, mixed); \ + svuint64_t mixed_hi = svlsr_n_u64_x(mask, mixed, 32); \ + svuint64_t mul = svmad_u64_x(mask, mixed_lo, mixed_hi, swapped); \ + acc = svadd_u64_x(mask, acc, mul); \ +} while (0) +#endif /* XXH_VECTOR == XXH_SVE */ + +/* prefetch + * can be disabled, by declaring XXH_NO_PREFETCH build macro */ +#if defined(XXH_NO_PREFETCH) +# define XXH_PREFETCH(ptr) (void)(ptr) /* disabled */ +#else +# if XXH_SIZE_OPT >= 1 +# define XXH_PREFETCH(ptr) (void)(ptr) +# elif defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86)) /* _mm_prefetch() not defined outside of x86/x64 */ +# include /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */ +# define XXH_PREFETCH(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T0) +# elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) ) +# define XXH_PREFETCH(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */) +# else +# define XXH_PREFETCH(ptr) (void)(ptr) /* disabled */ +# endif +#endif /* XXH_NO_PREFETCH */ + + +/* ========================================== + * XXH3 default settings + * ========================================== */ + +#define XXH_SECRET_DEFAULT_SIZE 192 /* minimum XXH3_SECRET_SIZE_MIN */ + +#if (XXH_SECRET_DEFAULT_SIZE < XXH3_SECRET_SIZE_MIN) +# error "default keyset is not large enough" +#endif + +/*! Pseudorandom secret taken directly from FARSH. */ +XXH_ALIGN(64) static const xxh_u8 XXH3_kSecret[XXH_SECRET_DEFAULT_SIZE] = { + 0xb8, 0xfe, 0x6c, 0x39, 0x23, 0xa4, 0x4b, 0xbe, 0x7c, 0x01, 0x81, 0x2c, 0xf7, 0x21, 0xad, 0x1c, + 0xde, 0xd4, 0x6d, 0xe9, 0x83, 0x90, 0x97, 0xdb, 0x72, 0x40, 0xa4, 0xa4, 0xb7, 0xb3, 0x67, 0x1f, + 0xcb, 0x79, 0xe6, 0x4e, 0xcc, 0xc0, 0xe5, 0x78, 0x82, 0x5a, 0xd0, 0x7d, 0xcc, 0xff, 0x72, 0x21, + 0xb8, 0x08, 0x46, 0x74, 0xf7, 0x43, 0x24, 0x8e, 0xe0, 0x35, 0x90, 0xe6, 0x81, 0x3a, 0x26, 0x4c, + 0x3c, 0x28, 0x52, 0xbb, 0x91, 0xc3, 0x00, 0xcb, 0x88, 0xd0, 0x65, 0x8b, 0x1b, 0x53, 0x2e, 0xa3, + 0x71, 0x64, 0x48, 0x97, 0xa2, 0x0d, 0xf9, 0x4e, 0x38, 0x19, 0xef, 0x46, 0xa9, 0xde, 0xac, 0xd8, + 0xa8, 0xfa, 0x76, 0x3f, 0xe3, 0x9c, 0x34, 0x3f, 0xf9, 0xdc, 0xbb, 0xc7, 0xc7, 0x0b, 0x4f, 0x1d, + 0x8a, 0x51, 0xe0, 0x4b, 0xcd, 0xb4, 0x59, 0x31, 0xc8, 0x9f, 0x7e, 0xc9, 0xd9, 0x78, 0x73, 0x64, + 0xea, 0xc5, 0xac, 0x83, 0x34, 0xd3, 0xeb, 0xc3, 0xc5, 0x81, 0xa0, 0xff, 0xfa, 0x13, 0x63, 0xeb, + 0x17, 0x0d, 0xdd, 0x51, 0xb7, 0xf0, 0xda, 0x49, 0xd3, 0x16, 0x55, 0x26, 0x29, 0xd4, 0x68, 0x9e, + 0x2b, 0x16, 0xbe, 0x58, 0x7d, 0x47, 0xa1, 0xfc, 0x8f, 0xf8, 0xb8, 0xd1, 0x7a, 0xd0, 0x31, 0xce, + 0x45, 0xcb, 0x3a, 0x8f, 0x95, 0x16, 0x04, 0x28, 0xaf, 0xd7, 0xfb, 0xca, 0xbb, 0x4b, 0x40, 0x7e, +}; + +static const xxh_u64 PRIME_MX1 = 0x165667919E3779F9ULL; /*!< 0b0001011001010110011001111001000110011110001101110111100111111001 */ +static const xxh_u64 PRIME_MX2 = 0x9FB21C651E98DF25ULL; /*!< 0b1001111110110010000111000110010100011110100110001101111100100101 */ + +#ifdef XXH_OLD_NAMES +# define kSecret XXH3_kSecret +#endif + +#ifdef XXH_DOXYGEN +/*! + * @brief Calculates a 32-bit to 64-bit long multiply. + * + * Implemented as a macro. + * + * Wraps `__emulu` on MSVC x86 because it tends to call `__allmul` when it doesn't + * need to (but it shouldn't need to anyways, it is about 7 instructions to do + * a 64x64 multiply...). Since we know that this will _always_ emit `MULL`, we + * use that instead of the normal method. + * + * If you are compiling for platforms like Thumb-1 and don't have a better option, + * you may also want to write your own long multiply routine here. + * + * @param x, y Numbers to be multiplied + * @return 64-bit product of the low 32 bits of @p x and @p y. + */ +XXH_FORCE_INLINE xxh_u64 +XXH_mult32to64(xxh_u64 x, xxh_u64 y) +{ + return (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF); +} +#elif defined(_MSC_VER) && defined(_M_IX86) +# define XXH_mult32to64(x, y) __emulu((unsigned)(x), (unsigned)(y)) +#else +/* + * Downcast + upcast is usually better than masking on older compilers like + * GCC 4.2 (especially 32-bit ones), all without affecting newer compilers. + * + * The other method, (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF), will AND both operands + * and perform a full 64x64 multiply -- entirely redundant on 32-bit. + */ +# define XXH_mult32to64(x, y) ((xxh_u64)(xxh_u32)(x) * (xxh_u64)(xxh_u32)(y)) +#endif + +/*! + * @brief Calculates a 64->128-bit long multiply. + * + * Uses `__uint128_t` and `_umul128` if available, otherwise uses a scalar + * version. + * + * @param lhs , rhs The 64-bit integers to be multiplied + * @return The 128-bit result represented in an @ref XXH128_hash_t. + */ +static XXH128_hash_t +XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs) +{ + /* + * GCC/Clang __uint128_t method. + * + * On most 64-bit targets, GCC and Clang define a __uint128_t type. + * This is usually the best way as it usually uses a native long 64-bit + * multiply, such as MULQ on x86_64 or MUL + UMULH on aarch64. + * + * Usually. + * + * Despite being a 32-bit platform, Clang (and emscripten) define this type + * despite not having the arithmetic for it. This results in a laggy + * compiler builtin call which calculates a full 128-bit multiply. + * In that case it is best to use the portable one. + * https://github.com/Cyan4973/xxHash/issues/211#issuecomment-515575677 + */ +#if (defined(__GNUC__) || defined(__clang__)) && !defined(__wasm__) \ + && defined(__SIZEOF_INT128__) \ + || (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128) + + __uint128_t const product = (__uint128_t)lhs * (__uint128_t)rhs; + XXH128_hash_t r128; + r128.low64 = (xxh_u64)(product); + r128.high64 = (xxh_u64)(product >> 64); + return r128; + + /* + * MSVC for x64's _umul128 method. + * + * xxh_u64 _umul128(xxh_u64 Multiplier, xxh_u64 Multiplicand, xxh_u64 *HighProduct); + * + * This compiles to single operand MUL on x64. + */ +#elif (defined(_M_X64) || defined(_M_IA64)) && !defined(_M_ARM64EC) + +#ifndef _MSC_VER +# pragma intrinsic(_umul128) +#endif + xxh_u64 product_high; + xxh_u64 const product_low = _umul128(lhs, rhs, &product_high); + XXH128_hash_t r128; + r128.low64 = product_low; + r128.high64 = product_high; + return r128; + + /* + * MSVC for ARM64's __umulh method. + * + * This compiles to the same MUL + UMULH as GCC/Clang's __uint128_t method. + */ +#elif defined(_M_ARM64) || defined(_M_ARM64EC) + +#ifndef _MSC_VER +# pragma intrinsic(__umulh) +#endif + XXH128_hash_t r128; + r128.low64 = lhs * rhs; + r128.high64 = __umulh(lhs, rhs); + return r128; + +#else + /* + * Portable scalar method. Optimized for 32-bit and 64-bit ALUs. + * + * This is a fast and simple grade school multiply, which is shown below + * with base 10 arithmetic instead of base 0x100000000. + * + * 9 3 // D2 lhs = 93 + * x 7 5 // D2 rhs = 75 + * ---------- + * 1 5 // D2 lo_lo = (93 % 10) * (75 % 10) = 15 + * 4 5 | // D2 hi_lo = (93 / 10) * (75 % 10) = 45 + * 2 1 | // D2 lo_hi = (93 % 10) * (75 / 10) = 21 + * + 6 3 | | // D2 hi_hi = (93 / 10) * (75 / 10) = 63 + * --------- + * 2 7 | // D2 cross = (15 / 10) + (45 % 10) + 21 = 27 + * + 6 7 | | // D2 upper = (27 / 10) + (45 / 10) + 63 = 67 + * --------- + * 6 9 7 5 // D4 res = (27 * 10) + (15 % 10) + (67 * 100) = 6975 + * + * The reasons for adding the products like this are: + * 1. It avoids manual carry tracking. Just like how + * (9 * 9) + 9 + 9 = 99, the same applies with this for UINT64_MAX. + * This avoids a lot of complexity. + * + * 2. It hints for, and on Clang, compiles to, the powerful UMAAL + * instruction available in ARM's Digital Signal Processing extension + * in 32-bit ARMv6 and later, which is shown below: + * + * void UMAAL(xxh_u32 *RdLo, xxh_u32 *RdHi, xxh_u32 Rn, xxh_u32 Rm) + * { + * xxh_u64 product = (xxh_u64)*RdLo * (xxh_u64)*RdHi + Rn + Rm; + * *RdLo = (xxh_u32)(product & 0xFFFFFFFF); + * *RdHi = (xxh_u32)(product >> 32); + * } + * + * This instruction was designed for efficient long multiplication, and + * allows this to be calculated in only 4 instructions at speeds + * comparable to some 64-bit ALUs. + * + * 3. It isn't terrible on other platforms. Usually this will be a couple + * of 32-bit ADD/ADCs. + */ + + /* First calculate all of the cross products. */ + xxh_u64 const lo_lo = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs & 0xFFFFFFFF); + xxh_u64 const hi_lo = XXH_mult32to64(lhs >> 32, rhs & 0xFFFFFFFF); + xxh_u64 const lo_hi = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs >> 32); + xxh_u64 const hi_hi = XXH_mult32to64(lhs >> 32, rhs >> 32); + + /* Now add the products together. These will never overflow. */ + xxh_u64 const cross = (lo_lo >> 32) + (hi_lo & 0xFFFFFFFF) + lo_hi; + xxh_u64 const upper = (hi_lo >> 32) + (cross >> 32) + hi_hi; + xxh_u64 const lower = (cross << 32) | (lo_lo & 0xFFFFFFFF); + + XXH128_hash_t r128; + r128.low64 = lower; + r128.high64 = upper; + return r128; +#endif +} + +/*! + * @brief Calculates a 64-bit to 128-bit multiply, then XOR folds it. + * + * The reason for the separate function is to prevent passing too many structs + * around by value. This will hopefully inline the multiply, but we don't force it. + * + * @param lhs , rhs The 64-bit integers to multiply + * @return The low 64 bits of the product XOR'd by the high 64 bits. + * @see XXH_mult64to128() + */ +static xxh_u64 +XXH3_mul128_fold64(xxh_u64 lhs, xxh_u64 rhs) +{ + XXH128_hash_t product = XXH_mult64to128(lhs, rhs); + return product.low64 ^ product.high64; +} + +/*! Seems to produce slightly better code on GCC for some reason. */ +XXH_FORCE_INLINE XXH_CONSTF xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift) +{ + XXH_ASSERT(0 <= shift && shift < 64); + return v64 ^ (v64 >> shift); +} + +/* + * This is a fast avalanche stage, + * suitable when input bits are already partially mixed + */ +static XXH64_hash_t XXH3_avalanche(xxh_u64 h64) +{ + h64 = XXH_xorshift64(h64, 37); + h64 *= PRIME_MX1; + h64 = XXH_xorshift64(h64, 32); + return h64; +} + +/* + * This is a stronger avalanche, + * inspired by Pelle Evensen's rrmxmx + * preferable when input has not been previously mixed + */ +static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len) +{ + /* this mix is inspired by Pelle Evensen's rrmxmx */ + h64 ^= XXH_rotl64(h64, 49) ^ XXH_rotl64(h64, 24); + h64 *= PRIME_MX2; + h64 ^= (h64 >> 35) + len ; + h64 *= PRIME_MX2; + return XXH_xorshift64(h64, 28); +} + + +/* ========================================== + * Short keys + * ========================================== + * One of the shortcomings of XXH32 and XXH64 was that their performance was + * sub-optimal on short lengths. It used an iterative algorithm which strongly + * favored lengths that were a multiple of 4 or 8. + * + * Instead of iterating over individual inputs, we use a set of single shot + * functions which piece together a range of lengths and operate in constant time. + * + * Additionally, the number of multiplies has been significantly reduced. This + * reduces latency, especially when emulating 64-bit multiplies on 32-bit. + * + * Depending on the platform, this may or may not be faster than XXH32, but it + * is almost guaranteed to be faster than XXH64. + */ + +/* + * At very short lengths, there isn't enough input to fully hide secrets, or use + * the entire secret. + * + * There is also only a limited amount of mixing we can do before significantly + * impacting performance. + * + * Therefore, we use different sections of the secret and always mix two secret + * samples with an XOR. This should have no effect on performance on the + * seedless or withSeed variants because everything _should_ be constant folded + * by modern compilers. + * + * The XOR mixing hides individual parts of the secret and increases entropy. + * + * This adds an extra layer of strength for custom secrets. + */ +XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t +XXH3_len_1to3_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(input != NULL); + XXH_ASSERT(1 <= len && len <= 3); + XXH_ASSERT(secret != NULL); + /* + * len = 1: combined = { input[0], 0x01, input[0], input[0] } + * len = 2: combined = { input[1], 0x02, input[0], input[1] } + * len = 3: combined = { input[2], 0x03, input[0], input[1] } + */ + { xxh_u8 const c1 = input[0]; + xxh_u8 const c2 = input[len >> 1]; + xxh_u8 const c3 = input[len - 1]; + xxh_u32 const combined = ((xxh_u32)c1 << 16) | ((xxh_u32)c2 << 24) + | ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8); + xxh_u64 const bitflip = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed; + xxh_u64 const keyed = (xxh_u64)combined ^ bitflip; + return XXH64_avalanche(keyed); + } +} + +XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t +XXH3_len_4to8_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(input != NULL); + XXH_ASSERT(secret != NULL); + XXH_ASSERT(4 <= len && len <= 8); + seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32; + { xxh_u32 const input1 = XXH_readLE32(input); + xxh_u32 const input2 = XXH_readLE32(input + len - 4); + xxh_u64 const bitflip = (XXH_readLE64(secret+8) ^ XXH_readLE64(secret+16)) - seed; + xxh_u64 const input64 = input2 + (((xxh_u64)input1) << 32); + xxh_u64 const keyed = input64 ^ bitflip; + return XXH3_rrmxmx(keyed, len); + } +} + +XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t +XXH3_len_9to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(input != NULL); + XXH_ASSERT(secret != NULL); + XXH_ASSERT(9 <= len && len <= 16); + { xxh_u64 const bitflip1 = (XXH_readLE64(secret+24) ^ XXH_readLE64(secret+32)) + seed; + xxh_u64 const bitflip2 = (XXH_readLE64(secret+40) ^ XXH_readLE64(secret+48)) - seed; + xxh_u64 const input_lo = XXH_readLE64(input) ^ bitflip1; + xxh_u64 const input_hi = XXH_readLE64(input + len - 8) ^ bitflip2; + xxh_u64 const acc = len + + XXH_swap64(input_lo) + input_hi + + XXH3_mul128_fold64(input_lo, input_hi); + return XXH3_avalanche(acc); + } +} + +XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t +XXH3_len_0to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(len <= 16); + { if (XXH_likely(len > 8)) return XXH3_len_9to16_64b(input, len, secret, seed); + if (XXH_likely(len >= 4)) return XXH3_len_4to8_64b(input, len, secret, seed); + if (len) return XXH3_len_1to3_64b(input, len, secret, seed); + return XXH64_avalanche(seed ^ (XXH_readLE64(secret+56) ^ XXH_readLE64(secret+64))); + } +} + +/* + * DISCLAIMER: There are known *seed-dependent* multicollisions here due to + * multiplication by zero, affecting hashes of lengths 17 to 240. + * + * However, they are very unlikely. + * + * Keep this in mind when using the unseeded XXH3_64bits() variant: As with all + * unseeded non-cryptographic hashes, it does not attempt to defend itself + * against specially crafted inputs, only random inputs. + * + * Compared to classic UMAC where a 1 in 2^31 chance of 4 consecutive bytes + * cancelling out the secret is taken an arbitrary number of times (addressed + * in XXH3_accumulate_512), this collision is very unlikely with random inputs + * and/or proper seeding: + * + * This only has a 1 in 2^63 chance of 8 consecutive bytes cancelling out, in a + * function that is only called up to 16 times per hash with up to 240 bytes of + * input. + * + * This is not too bad for a non-cryptographic hash function, especially with + * only 64 bit outputs. + * + * The 128-bit variant (which trades some speed for strength) is NOT affected + * by this, although it is always a good idea to use a proper seed if you care + * about strength. + */ +XXH_FORCE_INLINE xxh_u64 XXH3_mix16B(const xxh_u8* XXH_RESTRICT input, + const xxh_u8* XXH_RESTRICT secret, xxh_u64 seed64) +{ +#if defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \ + && defined(__i386__) && defined(__SSE2__) /* x86 + SSE2 */ \ + && !defined(XXH_ENABLE_AUTOVECTORIZE) /* Define to disable like XXH32 hack */ + /* + * UGLY HACK: + * GCC for x86 tends to autovectorize the 128-bit multiply, resulting in + * slower code. + * + * By forcing seed64 into a register, we disrupt the cost model and + * cause it to scalarize. See `XXH32_round()` + * + * FIXME: Clang's output is still _much_ faster -- On an AMD Ryzen 3600, + * XXH3_64bits @ len=240 runs at 4.6 GB/s with Clang 9, but 3.3 GB/s on + * GCC 9.2, despite both emitting scalar code. + * + * GCC generates much better scalar code than Clang for the rest of XXH3, + * which is why finding a more optimal codepath is an interest. + */ + XXH_COMPILER_GUARD(seed64); +#endif + { xxh_u64 const input_lo = XXH_readLE64(input); + xxh_u64 const input_hi = XXH_readLE64(input+8); + return XXH3_mul128_fold64( + input_lo ^ (XXH_readLE64(secret) + seed64), + input_hi ^ (XXH_readLE64(secret+8) - seed64) + ); + } +} + +/* For mid range keys, XXH3 uses a Mum-hash variant. */ +XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t +XXH3_len_17to128_64b(const xxh_u8* XXH_RESTRICT input, size_t len, + const xxh_u8* XXH_RESTRICT secret, size_t secretSize, + XXH64_hash_t seed) +{ + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize; + XXH_ASSERT(16 < len && len <= 128); + + { xxh_u64 acc = len * XXH_PRIME64_1; +#if XXH_SIZE_OPT >= 1 + /* Smaller and cleaner, but slightly slower. */ + unsigned int i = (unsigned int)(len - 1) / 32; + do { + acc += XXH3_mix16B(input+16 * i, secret+32*i, seed); + acc += XXH3_mix16B(input+len-16*(i+1), secret+32*i+16, seed); + } while (i-- != 0); +#else + if (len > 32) { + if (len > 64) { + if (len > 96) { + acc += XXH3_mix16B(input+48, secret+96, seed); + acc += XXH3_mix16B(input+len-64, secret+112, seed); + } + acc += XXH3_mix16B(input+32, secret+64, seed); + acc += XXH3_mix16B(input+len-48, secret+80, seed); + } + acc += XXH3_mix16B(input+16, secret+32, seed); + acc += XXH3_mix16B(input+len-32, secret+48, seed); + } + acc += XXH3_mix16B(input+0, secret+0, seed); + acc += XXH3_mix16B(input+len-16, secret+16, seed); +#endif + return XXH3_avalanche(acc); + } +} + +XXH_NO_INLINE XXH_PUREF XXH64_hash_t +XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len, + const xxh_u8* XXH_RESTRICT secret, size_t secretSize, + XXH64_hash_t seed) +{ + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize; + XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX); + + #define XXH3_MIDSIZE_STARTOFFSET 3 + #define XXH3_MIDSIZE_LASTOFFSET 17 + + { xxh_u64 acc = len * XXH_PRIME64_1; + xxh_u64 acc_end; + unsigned int const nbRounds = (unsigned int)len / 16; + unsigned int i; + XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX); + for (i=0; i<8; i++) { + acc += XXH3_mix16B(input+(16*i), secret+(16*i), seed); + } + /* last bytes */ + acc_end = XXH3_mix16B(input + len - 16, secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET, seed); + XXH_ASSERT(nbRounds >= 8); + acc = XXH3_avalanche(acc); +#if defined(__clang__) /* Clang */ \ + && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */ \ + && !defined(XXH_ENABLE_AUTOVECTORIZE) /* Define to disable */ + /* + * UGLY HACK: + * Clang for ARMv7-A tries to vectorize this loop, similar to GCC x86. + * In everywhere else, it uses scalar code. + * + * For 64->128-bit multiplies, even if the NEON was 100% optimal, it + * would still be slower than UMAAL (see XXH_mult64to128). + * + * Unfortunately, Clang doesn't handle the long multiplies properly and + * converts them to the nonexistent "vmulq_u64" intrinsic, which is then + * scalarized into an ugly mess of VMOV.32 instructions. + * + * This mess is difficult to avoid without turning autovectorization + * off completely, but they are usually relatively minor and/or not + * worth it to fix. + * + * This loop is the easiest to fix, as unlike XXH32, this pragma + * _actually works_ because it is a loop vectorization instead of an + * SLP vectorization. + */ + #pragma clang loop vectorize(disable) +#endif + for (i=8 ; i < nbRounds; i++) { + /* + * Prevents clang for unrolling the acc loop and interleaving with this one. + */ + XXH_COMPILER_GUARD(acc); + acc_end += XXH3_mix16B(input+(16*i), secret+(16*(i-8)) + XXH3_MIDSIZE_STARTOFFSET, seed); + } + return XXH3_avalanche(acc + acc_end); + } +} + + +/* ======= Long Keys ======= */ + +#define XXH_STRIPE_LEN 64 +#define XXH_SECRET_CONSUME_RATE 8 /* nb of secret bytes consumed at each accumulation */ +#define XXH_ACC_NB (XXH_STRIPE_LEN / sizeof(xxh_u64)) + +#ifdef XXH_OLD_NAMES +# define STRIPE_LEN XXH_STRIPE_LEN +# define ACC_NB XXH_ACC_NB +#endif + +#ifndef XXH_PREFETCH_DIST +# ifdef __clang__ +# define XXH_PREFETCH_DIST 320 +# else +# if (XXH_VECTOR == XXH_AVX512) +# define XXH_PREFETCH_DIST 512 +# else +# define XXH_PREFETCH_DIST 384 +# endif +# endif /* __clang__ */ +#endif /* XXH_PREFETCH_DIST */ + +/* + * These macros are to generate an XXH3_accumulate() function. + * The two arguments select the name suffix and target attribute. + * + * The name of this symbol is XXH3_accumulate_() and it calls + * XXH3_accumulate_512_(). + * + * It may be useful to hand implement this function if the compiler fails to + * optimize the inline function. + */ +#define XXH3_ACCUMULATE_TEMPLATE(name) \ +void \ +XXH3_accumulate_##name(xxh_u64* XXH_RESTRICT acc, \ + const xxh_u8* XXH_RESTRICT input, \ + const xxh_u8* XXH_RESTRICT secret, \ + size_t nbStripes) \ +{ \ + size_t n; \ + for (n = 0; n < nbStripes; n++ ) { \ + const xxh_u8* const in = input + n*XXH_STRIPE_LEN; \ + XXH_PREFETCH(in + XXH_PREFETCH_DIST); \ + XXH3_accumulate_512_##name( \ + acc, \ + in, \ + secret + n*XXH_SECRET_CONSUME_RATE); \ + } \ +} + + +XXH_FORCE_INLINE void XXH_writeLE64(void* dst, xxh_u64 v64) +{ + if (!XXH_CPU_LITTLE_ENDIAN) v64 = XXH_swap64(v64); + XXH_memcpy(dst, &v64, sizeof(v64)); +} + +/* Several intrinsic functions below are supposed to accept __int64 as argument, + * as documented in https://software.intel.com/sites/landingpage/IntrinsicsGuide/ . + * However, several environments do not define __int64 type, + * requiring a workaround. + */ +#if !defined (__VMS) \ + && (defined (__cplusplus) \ + || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) + typedef int64_t xxh_i64; +#else + /* the following type must have a width of 64-bit */ + typedef long long xxh_i64; +#endif + + +/* + * XXH3_accumulate_512 is the tightest loop for long inputs, and it is the most optimized. + * + * It is a hardened version of UMAC, based off of FARSH's implementation. + * + * This was chosen because it adapts quite well to 32-bit, 64-bit, and SIMD + * implementations, and it is ridiculously fast. + * + * We harden it by mixing the original input to the accumulators as well as the product. + * + * This means that in the (relatively likely) case of a multiply by zero, the + * original input is preserved. + * + * On 128-bit inputs, we swap 64-bit pairs when we add the input to improve + * cross-pollination, as otherwise the upper and lower halves would be + * essentially independent. + * + * This doesn't matter on 64-bit hashes since they all get merged together in + * the end, so we skip the extra step. + * + * Both XXH3_64bits and XXH3_128bits use this subroutine. + */ + +#if (XXH_VECTOR == XXH_AVX512) \ + || (defined(XXH_DISPATCH_AVX512) && XXH_DISPATCH_AVX512 != 0) + +#ifndef XXH_TARGET_AVX512 +# define XXH_TARGET_AVX512 /* disable attribute target */ +#endif + +XXH_FORCE_INLINE XXH_TARGET_AVX512 void +XXH3_accumulate_512_avx512(void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + __m512i* const xacc = (__m512i *) acc; + XXH_ASSERT((((size_t)acc) & 63) == 0); + XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i)); + + { + /* data_vec = input[0]; */ + __m512i const data_vec = _mm512_loadu_si512 (input); + /* key_vec = secret[0]; */ + __m512i const key_vec = _mm512_loadu_si512 (secret); + /* data_key = data_vec ^ key_vec; */ + __m512i const data_key = _mm512_xor_si512 (data_vec, key_vec); + /* data_key_lo = data_key >> 32; */ + __m512i const data_key_lo = _mm512_srli_epi64 (data_key, 32); + /* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */ + __m512i const product = _mm512_mul_epu32 (data_key, data_key_lo); + /* xacc[0] += swap(data_vec); */ + __m512i const data_swap = _mm512_shuffle_epi32(data_vec, (_MM_PERM_ENUM)_MM_SHUFFLE(1, 0, 3, 2)); + __m512i const sum = _mm512_add_epi64(*xacc, data_swap); + /* xacc[0] += product; */ + *xacc = _mm512_add_epi64(product, sum); + } +} +XXH_FORCE_INLINE XXH_TARGET_AVX512 XXH3_ACCUMULATE_TEMPLATE(avx512) + +/* + * XXH3_scrambleAcc: Scrambles the accumulators to improve mixing. + * + * Multiplication isn't perfect, as explained by Google in HighwayHash: + * + * // Multiplication mixes/scrambles bytes 0-7 of the 64-bit result to + * // varying degrees. In descending order of goodness, bytes + * // 3 4 2 5 1 6 0 7 have quality 228 224 164 160 100 96 36 32. + * // As expected, the upper and lower bytes are much worse. + * + * Source: https://github.com/google/highwayhash/blob/0aaf66b/highwayhash/hh_avx2.h#L291 + * + * Since our algorithm uses a pseudorandom secret to add some variance into the + * mix, we don't need to (or want to) mix as often or as much as HighwayHash does. + * + * This isn't as tight as XXH3_accumulate, but still written in SIMD to avoid + * extraction. + * + * Both XXH3_64bits and XXH3_128bits use this subroutine. + */ + +XXH_FORCE_INLINE XXH_TARGET_AVX512 void +XXH3_scrambleAcc_avx512(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 63) == 0); + XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i)); + { __m512i* const xacc = (__m512i*) acc; + const __m512i prime32 = _mm512_set1_epi32((int)XXH_PRIME32_1); + + /* xacc[0] ^= (xacc[0] >> 47) */ + __m512i const acc_vec = *xacc; + __m512i const shifted = _mm512_srli_epi64 (acc_vec, 47); + /* xacc[0] ^= secret; */ + __m512i const key_vec = _mm512_loadu_si512 (secret); + __m512i const data_key = _mm512_ternarylogic_epi32(key_vec, acc_vec, shifted, 0x96 /* key_vec ^ acc_vec ^ shifted */); + + /* xacc[0] *= XXH_PRIME32_1; */ + __m512i const data_key_hi = _mm512_srli_epi64 (data_key, 32); + __m512i const prod_lo = _mm512_mul_epu32 (data_key, prime32); + __m512i const prod_hi = _mm512_mul_epu32 (data_key_hi, prime32); + *xacc = _mm512_add_epi64(prod_lo, _mm512_slli_epi64(prod_hi, 32)); + } +} + +XXH_FORCE_INLINE XXH_TARGET_AVX512 void +XXH3_initCustomSecret_avx512(void* XXH_RESTRICT customSecret, xxh_u64 seed64) +{ + XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 63) == 0); + XXH_STATIC_ASSERT(XXH_SEC_ALIGN == 64); + XXH_ASSERT(((size_t)customSecret & 63) == 0); + (void)(&XXH_writeLE64); + { int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m512i); + __m512i const seed_pos = _mm512_set1_epi64((xxh_i64)seed64); + __m512i const seed = _mm512_mask_sub_epi64(seed_pos, 0xAA, _mm512_set1_epi8(0), seed_pos); + + const __m512i* const src = (const __m512i*) ((const void*) XXH3_kSecret); + __m512i* const dest = ( __m512i*) customSecret; + int i; + XXH_ASSERT(((size_t)src & 63) == 0); /* control alignment */ + XXH_ASSERT(((size_t)dest & 63) == 0); + for (i=0; i < nbRounds; ++i) { + dest[i] = _mm512_add_epi64(_mm512_load_si512(src + i), seed); + } } +} + +#endif + +#if (XXH_VECTOR == XXH_AVX2) \ + || (defined(XXH_DISPATCH_AVX2) && XXH_DISPATCH_AVX2 != 0) + +#ifndef XXH_TARGET_AVX2 +# define XXH_TARGET_AVX2 /* disable attribute target */ +#endif + +XXH_FORCE_INLINE XXH_TARGET_AVX2 void +XXH3_accumulate_512_avx2( void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 31) == 0); + { __m256i* const xacc = (__m256i *) acc; + /* Unaligned. This is mainly for pointer arithmetic, and because + * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */ + const __m256i* const xinput = (const __m256i *) input; + /* Unaligned. This is mainly for pointer arithmetic, and because + * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */ + const __m256i* const xsecret = (const __m256i *) secret; + + size_t i; + for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) { + /* data_vec = xinput[i]; */ + __m256i const data_vec = _mm256_loadu_si256 (xinput+i); + /* key_vec = xsecret[i]; */ + __m256i const key_vec = _mm256_loadu_si256 (xsecret+i); + /* data_key = data_vec ^ key_vec; */ + __m256i const data_key = _mm256_xor_si256 (data_vec, key_vec); + /* data_key_lo = data_key >> 32; */ + __m256i const data_key_lo = _mm256_srli_epi64 (data_key, 32); + /* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */ + __m256i const product = _mm256_mul_epu32 (data_key, data_key_lo); + /* xacc[i] += swap(data_vec); */ + __m256i const data_swap = _mm256_shuffle_epi32(data_vec, _MM_SHUFFLE(1, 0, 3, 2)); + __m256i const sum = _mm256_add_epi64(xacc[i], data_swap); + /* xacc[i] += product; */ + xacc[i] = _mm256_add_epi64(product, sum); + } } +} +XXH_FORCE_INLINE XXH_TARGET_AVX2 XXH3_ACCUMULATE_TEMPLATE(avx2) + +XXH_FORCE_INLINE XXH_TARGET_AVX2 void +XXH3_scrambleAcc_avx2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 31) == 0); + { __m256i* const xacc = (__m256i*) acc; + /* Unaligned. This is mainly for pointer arithmetic, and because + * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */ + const __m256i* const xsecret = (const __m256i *) secret; + const __m256i prime32 = _mm256_set1_epi32((int)XXH_PRIME32_1); + + size_t i; + for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) { + /* xacc[i] ^= (xacc[i] >> 47) */ + __m256i const acc_vec = xacc[i]; + __m256i const shifted = _mm256_srli_epi64 (acc_vec, 47); + __m256i const data_vec = _mm256_xor_si256 (acc_vec, shifted); + /* xacc[i] ^= xsecret; */ + __m256i const key_vec = _mm256_loadu_si256 (xsecret+i); + __m256i const data_key = _mm256_xor_si256 (data_vec, key_vec); + + /* xacc[i] *= XXH_PRIME32_1; */ + __m256i const data_key_hi = _mm256_srli_epi64 (data_key, 32); + __m256i const prod_lo = _mm256_mul_epu32 (data_key, prime32); + __m256i const prod_hi = _mm256_mul_epu32 (data_key_hi, prime32); + xacc[i] = _mm256_add_epi64(prod_lo, _mm256_slli_epi64(prod_hi, 32)); + } + } +} + +XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_initCustomSecret_avx2(void* XXH_RESTRICT customSecret, xxh_u64 seed64) +{ + XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 31) == 0); + XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE / sizeof(__m256i)) == 6); + XXH_STATIC_ASSERT(XXH_SEC_ALIGN <= 64); + (void)(&XXH_writeLE64); + XXH_PREFETCH(customSecret); + { __m256i const seed = _mm256_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64, (xxh_i64)(0U - seed64), (xxh_i64)seed64); + + const __m256i* const src = (const __m256i*) ((const void*) XXH3_kSecret); + __m256i* dest = ( __m256i*) customSecret; + +# if defined(__GNUC__) || defined(__clang__) + /* + * On GCC & Clang, marking 'dest' as modified will cause the compiler: + * - do not extract the secret from sse registers in the internal loop + * - use less common registers, and avoid pushing these reg into stack + */ + XXH_COMPILER_GUARD(dest); +# endif + XXH_ASSERT(((size_t)src & 31) == 0); /* control alignment */ + XXH_ASSERT(((size_t)dest & 31) == 0); + + /* GCC -O2 need unroll loop manually */ + dest[0] = _mm256_add_epi64(_mm256_load_si256(src+0), seed); + dest[1] = _mm256_add_epi64(_mm256_load_si256(src+1), seed); + dest[2] = _mm256_add_epi64(_mm256_load_si256(src+2), seed); + dest[3] = _mm256_add_epi64(_mm256_load_si256(src+3), seed); + dest[4] = _mm256_add_epi64(_mm256_load_si256(src+4), seed); + dest[5] = _mm256_add_epi64(_mm256_load_si256(src+5), seed); + } +} + +#endif + +/* x86dispatch always generates SSE2 */ +#if (XXH_VECTOR == XXH_SSE2) || defined(XXH_X86DISPATCH) + +#ifndef XXH_TARGET_SSE2 +# define XXH_TARGET_SSE2 /* disable attribute target */ +#endif + +XXH_FORCE_INLINE XXH_TARGET_SSE2 void +XXH3_accumulate_512_sse2( void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + /* SSE2 is just a half-scale version of the AVX2 version. */ + XXH_ASSERT((((size_t)acc) & 15) == 0); + { __m128i* const xacc = (__m128i *) acc; + /* Unaligned. This is mainly for pointer arithmetic, and because + * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */ + const __m128i* const xinput = (const __m128i *) input; + /* Unaligned. This is mainly for pointer arithmetic, and because + * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */ + const __m128i* const xsecret = (const __m128i *) secret; + + size_t i; + for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) { + /* data_vec = xinput[i]; */ + __m128i const data_vec = _mm_loadu_si128 (xinput+i); + /* key_vec = xsecret[i]; */ + __m128i const key_vec = _mm_loadu_si128 (xsecret+i); + /* data_key = data_vec ^ key_vec; */ + __m128i const data_key = _mm_xor_si128 (data_vec, key_vec); + /* data_key_lo = data_key >> 32; */ + __m128i const data_key_lo = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1)); + /* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */ + __m128i const product = _mm_mul_epu32 (data_key, data_key_lo); + /* xacc[i] += swap(data_vec); */ + __m128i const data_swap = _mm_shuffle_epi32(data_vec, _MM_SHUFFLE(1,0,3,2)); + __m128i const sum = _mm_add_epi64(xacc[i], data_swap); + /* xacc[i] += product; */ + xacc[i] = _mm_add_epi64(product, sum); + } } +} +XXH_FORCE_INLINE XXH_TARGET_SSE2 XXH3_ACCUMULATE_TEMPLATE(sse2) + +XXH_FORCE_INLINE XXH_TARGET_SSE2 void +XXH3_scrambleAcc_sse2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 15) == 0); + { __m128i* const xacc = (__m128i*) acc; + /* Unaligned. This is mainly for pointer arithmetic, and because + * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */ + const __m128i* const xsecret = (const __m128i *) secret; + const __m128i prime32 = _mm_set1_epi32((int)XXH_PRIME32_1); + + size_t i; + for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) { + /* xacc[i] ^= (xacc[i] >> 47) */ + __m128i const acc_vec = xacc[i]; + __m128i const shifted = _mm_srli_epi64 (acc_vec, 47); + __m128i const data_vec = _mm_xor_si128 (acc_vec, shifted); + /* xacc[i] ^= xsecret[i]; */ + __m128i const key_vec = _mm_loadu_si128 (xsecret+i); + __m128i const data_key = _mm_xor_si128 (data_vec, key_vec); + + /* xacc[i] *= XXH_PRIME32_1; */ + __m128i const data_key_hi = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1)); + __m128i const prod_lo = _mm_mul_epu32 (data_key, prime32); + __m128i const prod_hi = _mm_mul_epu32 (data_key_hi, prime32); + xacc[i] = _mm_add_epi64(prod_lo, _mm_slli_epi64(prod_hi, 32)); + } + } +} + +XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_initCustomSecret_sse2(void* XXH_RESTRICT customSecret, xxh_u64 seed64) +{ + XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0); + (void)(&XXH_writeLE64); + { int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m128i); + +# if defined(_MSC_VER) && defined(_M_IX86) && _MSC_VER < 1900 + /* MSVC 32bit mode does not support _mm_set_epi64x before 2015 */ + XXH_ALIGN(16) const xxh_i64 seed64x2[2] = { (xxh_i64)seed64, (xxh_i64)(0U - seed64) }; + __m128i const seed = _mm_load_si128((__m128i const*)seed64x2); +# else + __m128i const seed = _mm_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64); +# endif + int i; + + const void* const src16 = XXH3_kSecret; + __m128i* dst16 = (__m128i*) customSecret; +# if defined(__GNUC__) || defined(__clang__) + /* + * On GCC & Clang, marking 'dest' as modified will cause the compiler: + * - do not extract the secret from sse registers in the internal loop + * - use less common registers, and avoid pushing these reg into stack + */ + XXH_COMPILER_GUARD(dst16); +# endif + XXH_ASSERT(((size_t)src16 & 15) == 0); /* control alignment */ + XXH_ASSERT(((size_t)dst16 & 15) == 0); + + for (i=0; i < nbRounds; ++i) { + dst16[i] = _mm_add_epi64(_mm_load_si128((const __m128i *)src16+i), seed); + } } +} + +#endif + +#if (XXH_VECTOR == XXH_NEON) + +/* forward declarations for the scalar routines */ +XXH_FORCE_INLINE void +XXH3_scalarRound(void* XXH_RESTRICT acc, void const* XXH_RESTRICT input, + void const* XXH_RESTRICT secret, size_t lane); + +XXH_FORCE_INLINE void +XXH3_scalarScrambleRound(void* XXH_RESTRICT acc, + void const* XXH_RESTRICT secret, size_t lane); + +/*! + * @internal + * @brief The bulk processing loop for NEON and WASM SIMD128. + * + * The NEON code path is actually partially scalar when running on AArch64. This + * is to optimize the pipelining and can have up to 15% speedup depending on the + * CPU, and it also mitigates some GCC codegen issues. + * + * @see XXH3_NEON_LANES for configuring this and details about this optimization. + * + * NEON's 32-bit to 64-bit long multiply takes a half vector of 32-bit + * integers instead of the other platforms which mask full 64-bit vectors, + * so the setup is more complicated than just shifting right. + * + * Additionally, there is an optimization for 4 lanes at once noted below. + * + * Since, as stated, the most optimal amount of lanes for Cortexes is 6, + * there needs to be *three* versions of the accumulate operation used + * for the remaining 2 lanes. + * + * WASM's SIMD128 uses SIMDe's arm_neon.h polyfill because the intrinsics overlap + * nearly perfectly. + */ + +XXH_FORCE_INLINE void +XXH3_accumulate_512_neon( void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 15) == 0); + XXH_STATIC_ASSERT(XXH3_NEON_LANES > 0 && XXH3_NEON_LANES <= XXH_ACC_NB && XXH3_NEON_LANES % 2 == 0); + { /* GCC for darwin arm64 does not like aliasing here */ + xxh_aliasing_uint64x2_t* const xacc = (xxh_aliasing_uint64x2_t*) acc; + /* We don't use a uint32x4_t pointer because it causes bus errors on ARMv7. */ + uint8_t const* xinput = (const uint8_t *) input; + uint8_t const* xsecret = (const uint8_t *) secret; + + size_t i; +#ifdef __wasm_simd128__ + /* + * On WASM SIMD128, Clang emits direct address loads when XXH3_kSecret + * is constant propagated, which results in it converting it to this + * inside the loop: + * + * a = v128.load(XXH3_kSecret + 0 + $secret_offset, offset = 0) + * b = v128.load(XXH3_kSecret + 16 + $secret_offset, offset = 0) + * ... + * + * This requires a full 32-bit address immediate (and therefore a 6 byte + * instruction) as well as an add for each offset. + * + * Putting an asm guard prevents it from folding (at the cost of losing + * the alignment hint), and uses the free offset in `v128.load` instead + * of adding secret_offset each time which overall reduces code size by + * about a kilobyte and improves performance. + */ + XXH_COMPILER_GUARD(xsecret); +#endif + /* Scalar lanes use the normal scalarRound routine */ + for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) { + XXH3_scalarRound(acc, input, secret, i); + } + i = 0; + /* 4 NEON lanes at a time. */ + for (; i+1 < XXH3_NEON_LANES / 2; i+=2) { + /* data_vec = xinput[i]; */ + uint64x2_t data_vec_1 = XXH_vld1q_u64(xinput + (i * 16)); + uint64x2_t data_vec_2 = XXH_vld1q_u64(xinput + ((i+1) * 16)); + /* key_vec = xsecret[i]; */ + uint64x2_t key_vec_1 = XXH_vld1q_u64(xsecret + (i * 16)); + uint64x2_t key_vec_2 = XXH_vld1q_u64(xsecret + ((i+1) * 16)); + /* data_swap = swap(data_vec) */ + uint64x2_t data_swap_1 = vextq_u64(data_vec_1, data_vec_1, 1); + uint64x2_t data_swap_2 = vextq_u64(data_vec_2, data_vec_2, 1); + /* data_key = data_vec ^ key_vec; */ + uint64x2_t data_key_1 = veorq_u64(data_vec_1, key_vec_1); + uint64x2_t data_key_2 = veorq_u64(data_vec_2, key_vec_2); + + /* + * If we reinterpret the 64x2 vectors as 32x4 vectors, we can use a + * de-interleave operation for 4 lanes in 1 step with `vuzpq_u32` to + * get one vector with the low 32 bits of each lane, and one vector + * with the high 32 bits of each lane. + * + * The intrinsic returns a double vector because the original ARMv7-a + * instruction modified both arguments in place. AArch64 and SIMD128 emit + * two instructions from this intrinsic. + * + * [ dk11L | dk11H | dk12L | dk12H ] -> [ dk11L | dk12L | dk21L | dk22L ] + * [ dk21L | dk21H | dk22L | dk22H ] -> [ dk11H | dk12H | dk21H | dk22H ] + */ + uint32x4x2_t unzipped = vuzpq_u32( + vreinterpretq_u32_u64(data_key_1), + vreinterpretq_u32_u64(data_key_2) + ); + /* data_key_lo = data_key & 0xFFFFFFFF */ + uint32x4_t data_key_lo = unzipped.val[0]; + /* data_key_hi = data_key >> 32 */ + uint32x4_t data_key_hi = unzipped.val[1]; + /* + * Then, we can split the vectors horizontally and multiply which, as for most + * widening intrinsics, have a variant that works on both high half vectors + * for free on AArch64. A similar instruction is available on SIMD128. + * + * sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi + */ + uint64x2_t sum_1 = XXH_vmlal_low_u32(data_swap_1, data_key_lo, data_key_hi); + uint64x2_t sum_2 = XXH_vmlal_high_u32(data_swap_2, data_key_lo, data_key_hi); + /* + * Clang reorders + * a += b * c; // umlal swap.2d, dkl.2s, dkh.2s + * c += a; // add acc.2d, acc.2d, swap.2d + * to + * c += a; // add acc.2d, acc.2d, swap.2d + * c += b * c; // umlal acc.2d, dkl.2s, dkh.2s + * + * While it would make sense in theory since the addition is faster, + * for reasons likely related to umlal being limited to certain NEON + * pipelines, this is worse. A compiler guard fixes this. + */ + XXH_COMPILER_GUARD_CLANG_NEON(sum_1); + XXH_COMPILER_GUARD_CLANG_NEON(sum_2); + /* xacc[i] = acc_vec + sum; */ + xacc[i] = vaddq_u64(xacc[i], sum_1); + xacc[i+1] = vaddq_u64(xacc[i+1], sum_2); + } + /* Operate on the remaining NEON lanes 2 at a time. */ + for (; i < XXH3_NEON_LANES / 2; i++) { + /* data_vec = xinput[i]; */ + uint64x2_t data_vec = XXH_vld1q_u64(xinput + (i * 16)); + /* key_vec = xsecret[i]; */ + uint64x2_t key_vec = XXH_vld1q_u64(xsecret + (i * 16)); + /* acc_vec_2 = swap(data_vec) */ + uint64x2_t data_swap = vextq_u64(data_vec, data_vec, 1); + /* data_key = data_vec ^ key_vec; */ + uint64x2_t data_key = veorq_u64(data_vec, key_vec); + /* For two lanes, just use VMOVN and VSHRN. */ + /* data_key_lo = data_key & 0xFFFFFFFF; */ + uint32x2_t data_key_lo = vmovn_u64(data_key); + /* data_key_hi = data_key >> 32; */ + uint32x2_t data_key_hi = vshrn_n_u64(data_key, 32); + /* sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi; */ + uint64x2_t sum = vmlal_u32(data_swap, data_key_lo, data_key_hi); + /* Same Clang workaround as before */ + XXH_COMPILER_GUARD_CLANG_NEON(sum); + /* xacc[i] = acc_vec + sum; */ + xacc[i] = vaddq_u64 (xacc[i], sum); + } + } +} +XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(neon) + +XXH_FORCE_INLINE void +XXH3_scrambleAcc_neon(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 15) == 0); + + { xxh_aliasing_uint64x2_t* xacc = (xxh_aliasing_uint64x2_t*) acc; + uint8_t const* xsecret = (uint8_t const*) secret; + + size_t i; + /* WASM uses operator overloads and doesn't need these. */ +#ifndef __wasm_simd128__ + /* { prime32_1, prime32_1 } */ + uint32x2_t const kPrimeLo = vdup_n_u32(XXH_PRIME32_1); + /* { 0, prime32_1, 0, prime32_1 } */ + uint32x4_t const kPrimeHi = vreinterpretq_u32_u64(vdupq_n_u64((xxh_u64)XXH_PRIME32_1 << 32)); +#endif + + /* AArch64 uses both scalar and neon at the same time */ + for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) { + XXH3_scalarScrambleRound(acc, secret, i); + } + for (i=0; i < XXH3_NEON_LANES / 2; i++) { + /* xacc[i] ^= (xacc[i] >> 47); */ + uint64x2_t acc_vec = xacc[i]; + uint64x2_t shifted = vshrq_n_u64(acc_vec, 47); + uint64x2_t data_vec = veorq_u64(acc_vec, shifted); + + /* xacc[i] ^= xsecret[i]; */ + uint64x2_t key_vec = XXH_vld1q_u64(xsecret + (i * 16)); + uint64x2_t data_key = veorq_u64(data_vec, key_vec); + /* xacc[i] *= XXH_PRIME32_1 */ +#ifdef __wasm_simd128__ + /* SIMD128 has multiply by u64x2, use it instead of expanding and scalarizing */ + xacc[i] = data_key * XXH_PRIME32_1; +#else + /* + * Expanded version with portable NEON intrinsics + * + * lo(x) * lo(y) + (hi(x) * lo(y) << 32) + * + * prod_hi = hi(data_key) * lo(prime) << 32 + * + * Since we only need 32 bits of this multiply a trick can be used, reinterpreting the vector + * as a uint32x4_t and multiplying by { 0, prime, 0, prime } to cancel out the unwanted bits + * and avoid the shift. + */ + uint32x4_t prod_hi = vmulq_u32 (vreinterpretq_u32_u64(data_key), kPrimeHi); + /* Extract low bits for vmlal_u32 */ + uint32x2_t data_key_lo = vmovn_u64(data_key); + /* xacc[i] = prod_hi + lo(data_key) * XXH_PRIME32_1; */ + xacc[i] = vmlal_u32(vreinterpretq_u64_u32(prod_hi), data_key_lo, kPrimeLo); +#endif + } + } +} +#endif + +#if (XXH_VECTOR == XXH_VSX) + +XXH_FORCE_INLINE void +XXH3_accumulate_512_vsx( void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + /* presumed aligned */ + xxh_aliasing_u64x2* const xacc = (xxh_aliasing_u64x2*) acc; + xxh_u8 const* const xinput = (xxh_u8 const*) input; /* no alignment restriction */ + xxh_u8 const* const xsecret = (xxh_u8 const*) secret; /* no alignment restriction */ + xxh_u64x2 const v32 = { 32, 32 }; + size_t i; + for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) { + /* data_vec = xinput[i]; */ + xxh_u64x2 const data_vec = XXH_vec_loadu(xinput + 16*i); + /* key_vec = xsecret[i]; */ + xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + 16*i); + xxh_u64x2 const data_key = data_vec ^ key_vec; + /* shuffled = (data_key << 32) | (data_key >> 32); */ + xxh_u32x4 const shuffled = (xxh_u32x4)vec_rl(data_key, v32); + /* product = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)shuffled & 0xFFFFFFFF); */ + xxh_u64x2 const product = XXH_vec_mulo((xxh_u32x4)data_key, shuffled); + /* acc_vec = xacc[i]; */ + xxh_u64x2 acc_vec = xacc[i]; + acc_vec += product; + + /* swap high and low halves */ +#ifdef __s390x__ + acc_vec += vec_permi(data_vec, data_vec, 2); +#else + acc_vec += vec_xxpermdi(data_vec, data_vec, 2); +#endif + xacc[i] = acc_vec; + } +} +XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(vsx) + +XXH_FORCE_INLINE void +XXH3_scrambleAcc_vsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 15) == 0); + + { xxh_aliasing_u64x2* const xacc = (xxh_aliasing_u64x2*) acc; + const xxh_u8* const xsecret = (const xxh_u8*) secret; + /* constants */ + xxh_u64x2 const v32 = { 32, 32 }; + xxh_u64x2 const v47 = { 47, 47 }; + xxh_u32x4 const prime = { XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1 }; + size_t i; + for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) { + /* xacc[i] ^= (xacc[i] >> 47); */ + xxh_u64x2 const acc_vec = xacc[i]; + xxh_u64x2 const data_vec = acc_vec ^ (acc_vec >> v47); + + /* xacc[i] ^= xsecret[i]; */ + xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + 16*i); + xxh_u64x2 const data_key = data_vec ^ key_vec; + + /* xacc[i] *= XXH_PRIME32_1 */ + /* prod_lo = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)prime & 0xFFFFFFFF); */ + xxh_u64x2 const prod_even = XXH_vec_mule((xxh_u32x4)data_key, prime); + /* prod_hi = ((xxh_u64x2)data_key >> 32) * ((xxh_u64x2)prime >> 32); */ + xxh_u64x2 const prod_odd = XXH_vec_mulo((xxh_u32x4)data_key, prime); + xacc[i] = prod_odd + (prod_even << v32); + } } +} + +#endif + +#if (XXH_VECTOR == XXH_SVE) + +XXH_FORCE_INLINE void +XXH3_accumulate_512_sve( void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + uint64_t *xacc = (uint64_t *)acc; + const uint64_t *xinput = (const uint64_t *)(const void *)input; + const uint64_t *xsecret = (const uint64_t *)(const void *)secret; + svuint64_t kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1); + uint64_t element_count = svcntd(); + if (element_count >= 8) { + svbool_t mask = svptrue_pat_b64(SV_VL8); + svuint64_t vacc = svld1_u64(mask, xacc); + ACCRND(vacc, 0); + svst1_u64(mask, xacc, vacc); + } else if (element_count == 2) { /* sve128 */ + svbool_t mask = svptrue_pat_b64(SV_VL2); + svuint64_t acc0 = svld1_u64(mask, xacc + 0); + svuint64_t acc1 = svld1_u64(mask, xacc + 2); + svuint64_t acc2 = svld1_u64(mask, xacc + 4); + svuint64_t acc3 = svld1_u64(mask, xacc + 6); + ACCRND(acc0, 0); + ACCRND(acc1, 2); + ACCRND(acc2, 4); + ACCRND(acc3, 6); + svst1_u64(mask, xacc + 0, acc0); + svst1_u64(mask, xacc + 2, acc1); + svst1_u64(mask, xacc + 4, acc2); + svst1_u64(mask, xacc + 6, acc3); + } else { + svbool_t mask = svptrue_pat_b64(SV_VL4); + svuint64_t acc0 = svld1_u64(mask, xacc + 0); + svuint64_t acc1 = svld1_u64(mask, xacc + 4); + ACCRND(acc0, 0); + ACCRND(acc1, 4); + svst1_u64(mask, xacc + 0, acc0); + svst1_u64(mask, xacc + 4, acc1); + } +} + +XXH_FORCE_INLINE void +XXH3_accumulate_sve(xxh_u64* XXH_RESTRICT acc, + const xxh_u8* XXH_RESTRICT input, + const xxh_u8* XXH_RESTRICT secret, + size_t nbStripes) +{ + if (nbStripes != 0) { + uint64_t *xacc = (uint64_t *)acc; + const uint64_t *xinput = (const uint64_t *)(const void *)input; + const uint64_t *xsecret = (const uint64_t *)(const void *)secret; + svuint64_t kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1); + uint64_t element_count = svcntd(); + if (element_count >= 8) { + svbool_t mask = svptrue_pat_b64(SV_VL8); + svuint64_t vacc = svld1_u64(mask, xacc + 0); + do { + /* svprfd(svbool_t, void *, enum svfprop); */ + svprfd(mask, xinput + 128, SV_PLDL1STRM); + ACCRND(vacc, 0); + xinput += 8; + xsecret += 1; + nbStripes--; + } while (nbStripes != 0); + + svst1_u64(mask, xacc + 0, vacc); + } else if (element_count == 2) { /* sve128 */ + svbool_t mask = svptrue_pat_b64(SV_VL2); + svuint64_t acc0 = svld1_u64(mask, xacc + 0); + svuint64_t acc1 = svld1_u64(mask, xacc + 2); + svuint64_t acc2 = svld1_u64(mask, xacc + 4); + svuint64_t acc3 = svld1_u64(mask, xacc + 6); + do { + svprfd(mask, xinput + 128, SV_PLDL1STRM); + ACCRND(acc0, 0); + ACCRND(acc1, 2); + ACCRND(acc2, 4); + ACCRND(acc3, 6); + xinput += 8; + xsecret += 1; + nbStripes--; + } while (nbStripes != 0); + + svst1_u64(mask, xacc + 0, acc0); + svst1_u64(mask, xacc + 2, acc1); + svst1_u64(mask, xacc + 4, acc2); + svst1_u64(mask, xacc + 6, acc3); + } else { + svbool_t mask = svptrue_pat_b64(SV_VL4); + svuint64_t acc0 = svld1_u64(mask, xacc + 0); + svuint64_t acc1 = svld1_u64(mask, xacc + 4); + do { + svprfd(mask, xinput + 128, SV_PLDL1STRM); + ACCRND(acc0, 0); + ACCRND(acc1, 4); + xinput += 8; + xsecret += 1; + nbStripes--; + } while (nbStripes != 0); + + svst1_u64(mask, xacc + 0, acc0); + svst1_u64(mask, xacc + 4, acc1); + } + } +} + +#endif + +#if (XXH_VECTOR == XXH_LSX) +#define _LSX_SHUFFLE(z, y, x, w) (((z) << 6) | ((y) << 4) | ((x) << 2) | (w)) + +XXH_FORCE_INLINE void +XXH3_accumulate_512_lsx( void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 15) == 0); + { + __m128i* const xacc = (__m128i *) acc; + const __m128i* const xinput = (const __m128i *) input; + const __m128i* const xsecret = (const __m128i *) secret; + + for (size_t i = 0; i < XXH_STRIPE_LEN / sizeof(__m128i); i++) { + /* data_vec = xinput[i]; */ + __m128i const data_vec = __lsx_vld(xinput + i, 0); + /* key_vec = xsecret[i]; */ + __m128i const key_vec = __lsx_vld(xsecret + i, 0); + /* data_key = data_vec ^ key_vec; */ + __m128i const data_key = __lsx_vxor_v(data_vec, key_vec); + /* data_key_lo = data_key >> 32; */ + __m128i const data_key_lo = __lsx_vsrli_d(data_key, 32); + // __m128i const data_key_lo = __lsx_vsrli_d(data_key, 32); + /* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */ + __m128i const product = __lsx_vmulwev_d_wu(data_key, data_key_lo); + /* xacc[i] += swap(data_vec); */ + __m128i const data_swap = __lsx_vshuf4i_w(data_vec, _LSX_SHUFFLE(1, 0, 3, 2)); + __m128i const sum = __lsx_vadd_d(xacc[i], data_swap); + /* xacc[i] += product; */ + xacc[i] = __lsx_vadd_d(product, sum); + } + } +} +XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(lsx) + +XXH_FORCE_INLINE void +XXH3_scrambleAcc_lsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 15) == 0); + { + __m128i* const xacc = (__m128i*) acc; + const __m128i* const xsecret = (const __m128i *) secret; + const __m128i prime32 = __lsx_vreplgr2vr_w((int)XXH_PRIME32_1); + + for (size_t i = 0; i < XXH_STRIPE_LEN / sizeof(__m128i); i++) { + /* xacc[i] ^= (xacc[i] >> 47) */ + __m128i const acc_vec = xacc[i]; + __m128i const shifted = __lsx_vsrli_d(acc_vec, 47); + __m128i const data_vec = __lsx_vxor_v(acc_vec, shifted); + /* xacc[i] ^= xsecret[i]; */ + __m128i const key_vec = __lsx_vld(xsecret + i, 0); + __m128i const data_key = __lsx_vxor_v(data_vec, key_vec); + + /* xacc[i] *= XXH_PRIME32_1; */ + __m128i const data_key_hi = __lsx_vsrli_d(data_key, 32); + __m128i const prod_lo = __lsx_vmulwev_d_wu(data_key, prime32); + __m128i const prod_hi = __lsx_vmulwev_d_wu(data_key_hi, prime32); + xacc[i] = __lsx_vadd_d(prod_lo, __lsx_vslli_d(prod_hi, 32)); + } + } +} + +#endif + +/* scalar variants - universal */ + +#if defined(__aarch64__) && (defined(__GNUC__) || defined(__clang__)) +/* + * In XXH3_scalarRound(), GCC and Clang have a similar codegen issue, where they + * emit an excess mask and a full 64-bit multiply-add (MADD X-form). + * + * While this might not seem like much, as AArch64 is a 64-bit architecture, only + * big Cortex designs have a full 64-bit multiplier. + * + * On the little cores, the smaller 32-bit multiplier is used, and full 64-bit + * multiplies expand to 2-3 multiplies in microcode. This has a major penalty + * of up to 4 latency cycles and 2 stall cycles in the multiply pipeline. + * + * Thankfully, AArch64 still provides the 32-bit long multiply-add (UMADDL) which does + * not have this penalty and does the mask automatically. + */ +XXH_FORCE_INLINE xxh_u64 +XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs, xxh_u64 acc) +{ + xxh_u64 ret; + /* note: %x = 64-bit register, %w = 32-bit register */ + __asm__("umaddl %x0, %w1, %w2, %x3" : "=r" (ret) : "r" (lhs), "r" (rhs), "r" (acc)); + return ret; +} +#else +XXH_FORCE_INLINE xxh_u64 +XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs, xxh_u64 acc) +{ + return XXH_mult32to64((xxh_u32)lhs, (xxh_u32)rhs) + acc; +} +#endif + +/*! + * @internal + * @brief Scalar round for @ref XXH3_accumulate_512_scalar(). + * + * This is extracted to its own function because the NEON path uses a combination + * of NEON and scalar. + */ +XXH_FORCE_INLINE void +XXH3_scalarRound(void* XXH_RESTRICT acc, + void const* XXH_RESTRICT input, + void const* XXH_RESTRICT secret, + size_t lane) +{ + xxh_u64* xacc = (xxh_u64*) acc; + xxh_u8 const* xinput = (xxh_u8 const*) input; + xxh_u8 const* xsecret = (xxh_u8 const*) secret; + XXH_ASSERT(lane < XXH_ACC_NB); + XXH_ASSERT(((size_t)acc & (XXH_ACC_ALIGN-1)) == 0); + { + xxh_u64 const data_val = XXH_readLE64(xinput + lane * 8); + xxh_u64 const data_key = data_val ^ XXH_readLE64(xsecret + lane * 8); + xacc[lane ^ 1] += data_val; /* swap adjacent lanes */ + xacc[lane] = XXH_mult32to64_add64(data_key /* & 0xFFFFFFFF */, data_key >> 32, xacc[lane]); + } +} + +/*! + * @internal + * @brief Processes a 64 byte block of data using the scalar path. + */ +XXH_FORCE_INLINE void +XXH3_accumulate_512_scalar(void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + size_t i; + /* ARM GCC refuses to unroll this loop, resulting in a 24% slowdown on ARMv6. */ +#if defined(__GNUC__) && !defined(__clang__) \ + && (defined(__arm__) || defined(__thumb2__)) \ + && defined(__ARM_FEATURE_UNALIGNED) /* no unaligned access just wastes bytes */ \ + && XXH_SIZE_OPT <= 0 +# pragma GCC unroll 8 +#endif + for (i=0; i < XXH_ACC_NB; i++) { + XXH3_scalarRound(acc, input, secret, i); + } +} +XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(scalar) + +/*! + * @internal + * @brief Scalar scramble step for @ref XXH3_scrambleAcc_scalar(). + * + * This is extracted to its own function because the NEON path uses a combination + * of NEON and scalar. + */ +XXH_FORCE_INLINE void +XXH3_scalarScrambleRound(void* XXH_RESTRICT acc, + void const* XXH_RESTRICT secret, + size_t lane) +{ + xxh_u64* const xacc = (xxh_u64*) acc; /* presumed aligned */ + const xxh_u8* const xsecret = (const xxh_u8*) secret; /* no alignment restriction */ + XXH_ASSERT((((size_t)acc) & (XXH_ACC_ALIGN-1)) == 0); + XXH_ASSERT(lane < XXH_ACC_NB); + { + xxh_u64 const key64 = XXH_readLE64(xsecret + lane * 8); + xxh_u64 acc64 = xacc[lane]; + acc64 = XXH_xorshift64(acc64, 47); + acc64 ^= key64; + acc64 *= XXH_PRIME32_1; + xacc[lane] = acc64; + } +} + +/*! + * @internal + * @brief Scrambles the accumulators after a large chunk has been read + */ +XXH_FORCE_INLINE void +XXH3_scrambleAcc_scalar(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + size_t i; + for (i=0; i < XXH_ACC_NB; i++) { + XXH3_scalarScrambleRound(acc, secret, i); + } +} + +XXH_FORCE_INLINE void +XXH3_initCustomSecret_scalar(void* XXH_RESTRICT customSecret, xxh_u64 seed64) +{ + /* + * We need a separate pointer for the hack below, + * which requires a non-const pointer. + * Any decent compiler will optimize this out otherwise. + */ + const xxh_u8* kSecretPtr = XXH3_kSecret; + XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0); + +#if defined(__GNUC__) && defined(__aarch64__) + /* + * UGLY HACK: + * GCC and Clang generate a bunch of MOV/MOVK pairs for aarch64, and they are + * placed sequentially, in order, at the top of the unrolled loop. + * + * While MOVK is great for generating constants (2 cycles for a 64-bit + * constant compared to 4 cycles for LDR), it fights for bandwidth with + * the arithmetic instructions. + * + * I L S + * MOVK + * MOVK + * MOVK + * MOVK + * ADD + * SUB STR + * STR + * By forcing loads from memory (as the asm line causes the compiler to assume + * that XXH3_kSecretPtr has been changed), the pipelines are used more + * efficiently: + * I L S + * LDR + * ADD LDR + * SUB STR + * STR + * + * See XXH3_NEON_LANES for details on the pipsline. + * + * XXH3_64bits_withSeed, len == 256, Snapdragon 835 + * without hack: 2654.4 MB/s + * with hack: 3202.9 MB/s + */ + XXH_COMPILER_GUARD(kSecretPtr); +#endif + { int const nbRounds = XXH_SECRET_DEFAULT_SIZE / 16; + int i; + for (i=0; i < nbRounds; i++) { + /* + * The asm hack causes the compiler to assume that kSecretPtr aliases with + * customSecret, and on aarch64, this prevented LDP from merging two + * loads together for free. Putting the loads together before the stores + * properly generates LDP. + */ + xxh_u64 lo = XXH_readLE64(kSecretPtr + 16*i) + seed64; + xxh_u64 hi = XXH_readLE64(kSecretPtr + 16*i + 8) - seed64; + XXH_writeLE64((xxh_u8*)customSecret + 16*i, lo); + XXH_writeLE64((xxh_u8*)customSecret + 16*i + 8, hi); + } } +} + + +typedef void (*XXH3_f_accumulate)(xxh_u64* XXH_RESTRICT, const xxh_u8* XXH_RESTRICT, const xxh_u8* XXH_RESTRICT, size_t); +typedef void (*XXH3_f_scrambleAcc)(void* XXH_RESTRICT, const void*); +typedef void (*XXH3_f_initCustomSecret)(void* XXH_RESTRICT, xxh_u64); + + +#if (XXH_VECTOR == XXH_AVX512) + +#define XXH3_accumulate_512 XXH3_accumulate_512_avx512 +#define XXH3_accumulate XXH3_accumulate_avx512 +#define XXH3_scrambleAcc XXH3_scrambleAcc_avx512 +#define XXH3_initCustomSecret XXH3_initCustomSecret_avx512 + +#elif (XXH_VECTOR == XXH_AVX2) + +#define XXH3_accumulate_512 XXH3_accumulate_512_avx2 +#define XXH3_accumulate XXH3_accumulate_avx2 +#define XXH3_scrambleAcc XXH3_scrambleAcc_avx2 +#define XXH3_initCustomSecret XXH3_initCustomSecret_avx2 + +#elif (XXH_VECTOR == XXH_SSE2) + +#define XXH3_accumulate_512 XXH3_accumulate_512_sse2 +#define XXH3_accumulate XXH3_accumulate_sse2 +#define XXH3_scrambleAcc XXH3_scrambleAcc_sse2 +#define XXH3_initCustomSecret XXH3_initCustomSecret_sse2 + +#elif (XXH_VECTOR == XXH_NEON) + +#define XXH3_accumulate_512 XXH3_accumulate_512_neon +#define XXH3_accumulate XXH3_accumulate_neon +#define XXH3_scrambleAcc XXH3_scrambleAcc_neon +#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar + +#elif (XXH_VECTOR == XXH_VSX) + +#define XXH3_accumulate_512 XXH3_accumulate_512_vsx +#define XXH3_accumulate XXH3_accumulate_vsx +#define XXH3_scrambleAcc XXH3_scrambleAcc_vsx +#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar + +#elif (XXH_VECTOR == XXH_SVE) +#define XXH3_accumulate_512 XXH3_accumulate_512_sve +#define XXH3_accumulate XXH3_accumulate_sve +#define XXH3_scrambleAcc XXH3_scrambleAcc_scalar +#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar + +#elif (XXH_VECTOR == XXH_LSX) +#define XXH3_accumulate_512 XXH3_accumulate_512_lsx +#define XXH3_accumulate XXH3_accumulate_lsx +#define XXH3_scrambleAcc XXH3_scrambleAcc_lsx +#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar + +#else /* scalar */ + +#define XXH3_accumulate_512 XXH3_accumulate_512_scalar +#define XXH3_accumulate XXH3_accumulate_scalar +#define XXH3_scrambleAcc XXH3_scrambleAcc_scalar +#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar + +#endif + +#if XXH_SIZE_OPT >= 1 /* don't do SIMD for initialization */ +# undef XXH3_initCustomSecret +# define XXH3_initCustomSecret XXH3_initCustomSecret_scalar +#endif + +XXH_FORCE_INLINE void +XXH3_hashLong_internal_loop(xxh_u64* XXH_RESTRICT acc, + const xxh_u8* XXH_RESTRICT input, size_t len, + const xxh_u8* XXH_RESTRICT secret, size_t secretSize, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble) +{ + size_t const nbStripesPerBlock = (secretSize - XXH_STRIPE_LEN) / XXH_SECRET_CONSUME_RATE; + size_t const block_len = XXH_STRIPE_LEN * nbStripesPerBlock; + size_t const nb_blocks = (len - 1) / block_len; + + size_t n; + + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); + + for (n = 0; n < nb_blocks; n++) { + f_acc(acc, input + n*block_len, secret, nbStripesPerBlock); + f_scramble(acc, secret + secretSize - XXH_STRIPE_LEN); + } + + /* last partial block */ + XXH_ASSERT(len > XXH_STRIPE_LEN); + { size_t const nbStripes = ((len - 1) - (block_len * nb_blocks)) / XXH_STRIPE_LEN; + XXH_ASSERT(nbStripes <= (secretSize / XXH_SECRET_CONSUME_RATE)); + f_acc(acc, input + nb_blocks*block_len, secret, nbStripes); + + /* last stripe */ + { const xxh_u8* const p = input + len - XXH_STRIPE_LEN; +#define XXH_SECRET_LASTACC_START 7 /* not aligned on 8, last secret is different from acc & scrambler */ + XXH3_accumulate_512(acc, p, secret + secretSize - XXH_STRIPE_LEN - XXH_SECRET_LASTACC_START); + } } +} + +XXH_FORCE_INLINE xxh_u64 +XXH3_mix2Accs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret) +{ + return XXH3_mul128_fold64( + acc[0] ^ XXH_readLE64(secret), + acc[1] ^ XXH_readLE64(secret+8) ); +} + +static XXH_PUREF XXH64_hash_t +XXH3_mergeAccs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret, xxh_u64 start) +{ + xxh_u64 result64 = start; + size_t i = 0; + + for (i = 0; i < 4; i++) { + result64 += XXH3_mix2Accs(acc+2*i, secret + 16*i); +#if defined(__clang__) /* Clang */ \ + && (defined(__arm__) || defined(__thumb__)) /* ARMv7 */ \ + && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */ \ + && !defined(XXH_ENABLE_AUTOVECTORIZE) /* Define to disable */ + /* + * UGLY HACK: + * Prevent autovectorization on Clang ARMv7-a. Exact same problem as + * the one in XXH3_len_129to240_64b. Speeds up shorter keys > 240b. + * XXH3_64bits, len == 256, Snapdragon 835: + * without hack: 2063.7 MB/s + * with hack: 2560.7 MB/s + */ + XXH_COMPILER_GUARD(result64); +#endif + } + + return XXH3_avalanche(result64); +} + +/* do not align on 8, so that the secret is different from the accumulator */ +#define XXH_SECRET_MERGEACCS_START 11 + +static XXH_PUREF XXH64_hash_t +XXH3_finalizeLong_64b(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret, xxh_u64 len) +{ + return XXH3_mergeAccs(acc, secret + XXH_SECRET_MERGEACCS_START, len * XXH_PRIME64_1); +} + +#define XXH3_INIT_ACC { XXH_PRIME32_3, XXH_PRIME64_1, XXH_PRIME64_2, XXH_PRIME64_3, \ + XXH_PRIME64_4, XXH_PRIME32_2, XXH_PRIME64_5, XXH_PRIME32_1 } + +XXH_FORCE_INLINE XXH64_hash_t +XXH3_hashLong_64b_internal(const void* XXH_RESTRICT input, size_t len, + const void* XXH_RESTRICT secret, size_t secretSize, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble) +{ + XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC; + + XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, (const xxh_u8*)secret, secretSize, f_acc, f_scramble); + + /* converge into final hash */ + XXH_STATIC_ASSERT(sizeof(acc) == 64); + XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START); + return XXH3_finalizeLong_64b(acc, (const xxh_u8*)secret, (xxh_u64)len); +} + +/* + * It's important for performance to transmit secret's size (when it's static) + * so that the compiler can properly optimize the vectorized loop. + * This makes a big performance difference for "medium" keys (<1 KB) when using AVX instruction set. + * When the secret size is unknown, or on GCC 12 where the mix of NO_INLINE and FORCE_INLINE + * breaks -Og, this is XXH_NO_INLINE. + */ +XXH3_WITH_SECRET_INLINE XXH64_hash_t +XXH3_hashLong_64b_withSecret(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen) +{ + (void)seed64; + return XXH3_hashLong_64b_internal(input, len, secret, secretLen, XXH3_accumulate, XXH3_scrambleAcc); +} + +/* + * It's preferable for performance that XXH3_hashLong is not inlined, + * as it results in a smaller function for small data, easier to the instruction cache. + * Note that inside this no_inline function, we do inline the internal loop, + * and provide a statically defined secret size to allow optimization of vector loop. + */ +XXH_NO_INLINE XXH_PUREF XXH64_hash_t +XXH3_hashLong_64b_default(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen) +{ + (void)seed64; (void)secret; (void)secretLen; + return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_accumulate, XXH3_scrambleAcc); +} + +/* + * XXH3_hashLong_64b_withSeed(): + * Generate a custom key based on alteration of default XXH3_kSecret with the seed, + * and then use this key for long mode hashing. + * + * This operation is decently fast but nonetheless costs a little bit of time. + * Try to avoid it whenever possible (typically when seed==0). + * + * It's important for performance that XXH3_hashLong is not inlined. Not sure + * why (uop cache maybe?), but the difference is large and easily measurable. + */ +XXH_FORCE_INLINE XXH64_hash_t +XXH3_hashLong_64b_withSeed_internal(const void* input, size_t len, + XXH64_hash_t seed, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble, + XXH3_f_initCustomSecret f_initSec) +{ +#if XXH_SIZE_OPT <= 0 + if (seed == 0) + return XXH3_hashLong_64b_internal(input, len, + XXH3_kSecret, sizeof(XXH3_kSecret), + f_acc, f_scramble); +#endif + { XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE]; + f_initSec(secret, seed); + return XXH3_hashLong_64b_internal(input, len, secret, sizeof(secret), + f_acc, f_scramble); + } +} + +/* + * It's important for performance that XXH3_hashLong is not inlined. + */ +XXH_NO_INLINE XXH64_hash_t +XXH3_hashLong_64b_withSeed(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed, const xxh_u8* XXH_RESTRICT secret, size_t secretLen) +{ + (void)secret; (void)secretLen; + return XXH3_hashLong_64b_withSeed_internal(input, len, seed, + XXH3_accumulate, XXH3_scrambleAcc, XXH3_initCustomSecret); +} + + +typedef XXH64_hash_t (*XXH3_hashLong64_f)(const void* XXH_RESTRICT, size_t, + XXH64_hash_t, const xxh_u8* XXH_RESTRICT, size_t); + +XXH_FORCE_INLINE XXH64_hash_t +XXH3_64bits_internal(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen, + XXH3_hashLong64_f f_hashLong) +{ + XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN); + /* + * If an action is to be taken if `secretLen` condition is not respected, + * it should be done here. + * For now, it's a contract pre-condition. + * Adding a check and a branch here would cost performance at every hash. + * Also, note that function signature doesn't offer room to return an error. + */ + if (len <= 16) + return XXH3_len_0to16_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64); + if (len <= 128) + return XXH3_len_17to128_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64); + if (len <= XXH3_MIDSIZE_MAX) + return XXH3_len_129to240_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64); + return f_hashLong(input, len, seed64, (const xxh_u8*)secret, secretLen); +} + + +/* === Public entry point === */ + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void* input, size_t length) +{ + return XXH3_64bits_internal(input, length, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_default); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH64_hash_t +XXH3_64bits_withSecret(XXH_NOESCAPE const void* input, size_t length, XXH_NOESCAPE const void* secret, size_t secretSize) +{ + return XXH3_64bits_internal(input, length, 0, secret, secretSize, XXH3_hashLong_64b_withSecret); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH64_hash_t +XXH3_64bits_withSeed(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed) +{ + return XXH3_64bits_internal(input, length, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed); +} + +XXH_PUBLIC_API XXH64_hash_t +XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed) +{ + if (length <= XXH3_MIDSIZE_MAX) + return XXH3_64bits_internal(input, length, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL); + return XXH3_hashLong_64b_withSecret(input, length, seed, (const xxh_u8*)secret, secretSize); +} + + +/* === XXH3 streaming === */ +#ifndef XXH_NO_STREAM +/* + * Malloc's a pointer that is always aligned to @align. + * + * This must be freed with `XXH_alignedFree()`. + * + * malloc typically guarantees 16 byte alignment on 64-bit systems and 8 byte + * alignment on 32-bit. This isn't enough for the 32 byte aligned loads in AVX2 + * or on 32-bit, the 16 byte aligned loads in SSE2 and NEON. + * + * This underalignment previously caused a rather obvious crash which went + * completely unnoticed due to XXH3_createState() not actually being tested. + * Credit to RedSpah for noticing this bug. + * + * The alignment is done manually: Functions like posix_memalign or _mm_malloc + * are avoided: To maintain portability, we would have to write a fallback + * like this anyways, and besides, testing for the existence of library + * functions without relying on external build tools is impossible. + * + * The method is simple: Overallocate, manually align, and store the offset + * to the original behind the returned pointer. + * + * Align must be a power of 2 and 8 <= align <= 128. + */ +static XXH_MALLOCF void* XXH_alignedMalloc(size_t s, size_t align) +{ + XXH_ASSERT(align <= 128 && align >= 8); /* range check */ + XXH_ASSERT((align & (align-1)) == 0); /* power of 2 */ + XXH_ASSERT(s != 0 && s < (s + align)); /* empty/overflow */ + { /* Overallocate to make room for manual realignment and an offset byte */ + xxh_u8* base = (xxh_u8*)XXH_malloc(s + align); + if (base != NULL) { + /* + * Get the offset needed to align this pointer. + * + * Even if the returned pointer is aligned, there will always be + * at least one byte to store the offset to the original pointer. + */ + size_t offset = align - ((size_t)base & (align - 1)); /* base % align */ + /* Add the offset for the now-aligned pointer */ + xxh_u8* ptr = base + offset; + + XXH_ASSERT((size_t)ptr % align == 0); + + /* Store the offset immediately before the returned pointer. */ + ptr[-1] = (xxh_u8)offset; + return ptr; + } + return NULL; + } +} +/* + * Frees an aligned pointer allocated by XXH_alignedMalloc(). Don't pass + * normal malloc'd pointers, XXH_alignedMalloc has a specific data layout. + */ +static void XXH_alignedFree(void* p) +{ + if (p != NULL) { + xxh_u8* ptr = (xxh_u8*)p; + /* Get the offset byte we added in XXH_malloc. */ + xxh_u8 offset = ptr[-1]; + /* Free the original malloc'd pointer */ + xxh_u8* base = ptr - offset; + XXH_free(base); + } +} +/*! @ingroup XXH3_family */ +/*! + * @brief Allocate an @ref XXH3_state_t. + * + * @return An allocated pointer of @ref XXH3_state_t on success. + * @return `NULL` on failure. + * + * @note Must be freed with XXH3_freeState(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void) +{ + XXH3_state_t* const state = (XXH3_state_t*)XXH_alignedMalloc(sizeof(XXH3_state_t), 64); + if (state==NULL) return NULL; + XXH3_INITSTATE(state); + return state; +} + +/*! @ingroup XXH3_family */ +/*! + * @brief Frees an @ref XXH3_state_t. + * + * @param statePtr A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState(). + * + * @return @ref XXH_OK. + * + * @note Must be allocated with XXH3_createState(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr) +{ + XXH_alignedFree(statePtr); + return XXH_OK; +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API void +XXH3_copyState(XXH_NOESCAPE XXH3_state_t* dst_state, XXH_NOESCAPE const XXH3_state_t* src_state) +{ + XXH_memcpy(dst_state, src_state, sizeof(*dst_state)); +} + +static void +XXH3_reset_internal(XXH3_state_t* statePtr, + XXH64_hash_t seed, + const void* secret, size_t secretSize) +{ + size_t const initStart = offsetof(XXH3_state_t, bufferedSize); + size_t const initLength = offsetof(XXH3_state_t, nbStripesPerBlock) - initStart; + XXH_ASSERT(offsetof(XXH3_state_t, nbStripesPerBlock) > initStart); + XXH_ASSERT(statePtr != NULL); + /* set members from bufferedSize to nbStripesPerBlock (excluded) to 0 */ + memset((char*)statePtr + initStart, 0, initLength); + statePtr->acc[0] = XXH_PRIME32_3; + statePtr->acc[1] = XXH_PRIME64_1; + statePtr->acc[2] = XXH_PRIME64_2; + statePtr->acc[3] = XXH_PRIME64_3; + statePtr->acc[4] = XXH_PRIME64_4; + statePtr->acc[5] = XXH_PRIME32_2; + statePtr->acc[6] = XXH_PRIME64_5; + statePtr->acc[7] = XXH_PRIME32_1; + statePtr->seed = seed; + statePtr->useSeed = (seed != 0); + statePtr->extSecret = (const unsigned char*)secret; + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); + statePtr->secretLimit = secretSize - XXH_STRIPE_LEN; + statePtr->nbStripesPerBlock = statePtr->secretLimit / XXH_SECRET_CONSUME_RATE; +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr) +{ + if (statePtr == NULL) return XXH_ERROR; + XXH3_reset_internal(statePtr, 0, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE); + return XXH_OK; +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_64bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize) +{ + if (statePtr == NULL) return XXH_ERROR; + XXH3_reset_internal(statePtr, 0, secret, secretSize); + if (secret == NULL) return XXH_ERROR; + if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR; + return XXH_OK; +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_64bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed) +{ + if (statePtr == NULL) return XXH_ERROR; + if (seed==0) return XXH3_64bits_reset(statePtr); + if ((seed != statePtr->seed) || (statePtr->extSecret != NULL)) + XXH3_initCustomSecret(statePtr->customSecret, seed); + XXH3_reset_internal(statePtr, seed, NULL, XXH_SECRET_DEFAULT_SIZE); + return XXH_OK; +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed64) +{ + if (statePtr == NULL) return XXH_ERROR; + if (secret == NULL) return XXH_ERROR; + if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR; + XXH3_reset_internal(statePtr, seed64, secret, secretSize); + statePtr->useSeed = 1; /* always, even if seed64==0 */ + return XXH_OK; +} + +/*! + * @internal + * @brief Processes a large input for XXH3_update() and XXH3_digest_long(). + * + * Unlike XXH3_hashLong_internal_loop(), this can process data that overlaps a block. + * + * @param acc Pointer to the 8 accumulator lanes + * @param nbStripesSoFarPtr In/out pointer to the number of leftover stripes in the block* + * @param nbStripesPerBlock Number of stripes in a block + * @param input Input pointer + * @param nbStripes Number of stripes to process + * @param secret Secret pointer + * @param secretLimit Offset of the last block in @p secret + * @param f_acc Pointer to an XXH3_accumulate implementation + * @param f_scramble Pointer to an XXH3_scrambleAcc implementation + * @return Pointer past the end of @p input after processing + */ +XXH_FORCE_INLINE const xxh_u8 * +XXH3_consumeStripes(xxh_u64* XXH_RESTRICT acc, + size_t* XXH_RESTRICT nbStripesSoFarPtr, size_t nbStripesPerBlock, + const xxh_u8* XXH_RESTRICT input, size_t nbStripes, + const xxh_u8* XXH_RESTRICT secret, size_t secretLimit, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble) +{ + const xxh_u8* initialSecret = secret + *nbStripesSoFarPtr * XXH_SECRET_CONSUME_RATE; + /* Process full blocks */ + if (nbStripes >= (nbStripesPerBlock - *nbStripesSoFarPtr)) { + /* Process the initial partial block... */ + size_t nbStripesThisIter = nbStripesPerBlock - *nbStripesSoFarPtr; + + do { + /* Accumulate and scramble */ + f_acc(acc, input, initialSecret, nbStripesThisIter); + f_scramble(acc, secret + secretLimit); + input += nbStripesThisIter * XXH_STRIPE_LEN; + nbStripes -= nbStripesThisIter; + /* Then continue the loop with the full block size */ + nbStripesThisIter = nbStripesPerBlock; + initialSecret = secret; + } while (nbStripes >= nbStripesPerBlock); + *nbStripesSoFarPtr = 0; + } + /* Process a partial block */ + if (nbStripes > 0) { + f_acc(acc, input, initialSecret, nbStripes); + input += nbStripes * XXH_STRIPE_LEN; + *nbStripesSoFarPtr += nbStripes; + } + /* Return end pointer */ + return input; +} + +#ifndef XXH3_STREAM_USE_STACK +# if XXH_SIZE_OPT <= 0 && !defined(__clang__) /* clang doesn't need additional stack space */ +# define XXH3_STREAM_USE_STACK 1 +# endif +#endif +/* + * Both XXH3_64bits_update and XXH3_128bits_update use this routine. + */ +XXH_FORCE_INLINE XXH_errorcode +XXH3_update(XXH3_state_t* XXH_RESTRICT const state, + const xxh_u8* XXH_RESTRICT input, size_t len, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble) +{ + if (input==NULL) { + XXH_ASSERT(len == 0); + return XXH_OK; + } + + XXH_ASSERT(state != NULL); + { const xxh_u8* const bEnd = input + len; + const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret; +#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1 + /* For some reason, gcc and MSVC seem to suffer greatly + * when operating accumulators directly into state. + * Operating into stack space seems to enable proper optimization. + * clang, on the other hand, doesn't seem to need this trick */ + XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[8]; + XXH_memcpy(acc, state->acc, sizeof(acc)); +#else + xxh_u64* XXH_RESTRICT const acc = state->acc; +#endif + state->totalLen += len; + XXH_ASSERT(state->bufferedSize <= XXH3_INTERNALBUFFER_SIZE); + + /* small input : just fill in tmp buffer */ + if (len <= XXH3_INTERNALBUFFER_SIZE - state->bufferedSize) { + XXH_memcpy(state->buffer + state->bufferedSize, input, len); + state->bufferedSize += (XXH32_hash_t)len; + return XXH_OK; + } + + /* total input is now > XXH3_INTERNALBUFFER_SIZE */ + #define XXH3_INTERNALBUFFER_STRIPES (XXH3_INTERNALBUFFER_SIZE / XXH_STRIPE_LEN) + XXH_STATIC_ASSERT(XXH3_INTERNALBUFFER_SIZE % XXH_STRIPE_LEN == 0); /* clean multiple */ + + /* + * Internal buffer is partially filled (always, except at beginning) + * Complete it, then consume it. + */ + if (state->bufferedSize) { + size_t const loadSize = XXH3_INTERNALBUFFER_SIZE - state->bufferedSize; + XXH_memcpy(state->buffer + state->bufferedSize, input, loadSize); + input += loadSize; + XXH3_consumeStripes(acc, + &state->nbStripesSoFar, state->nbStripesPerBlock, + state->buffer, XXH3_INTERNALBUFFER_STRIPES, + secret, state->secretLimit, + f_acc, f_scramble); + state->bufferedSize = 0; + } + XXH_ASSERT(input < bEnd); + if (bEnd - input > XXH3_INTERNALBUFFER_SIZE) { + size_t nbStripes = (size_t)(bEnd - 1 - input) / XXH_STRIPE_LEN; + input = XXH3_consumeStripes(acc, + &state->nbStripesSoFar, state->nbStripesPerBlock, + input, nbStripes, + secret, state->secretLimit, + f_acc, f_scramble); + XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN); + + } + /* Some remaining input (always) : buffer it */ + XXH_ASSERT(input < bEnd); + XXH_ASSERT(bEnd - input <= XXH3_INTERNALBUFFER_SIZE); + XXH_ASSERT(state->bufferedSize == 0); + XXH_memcpy(state->buffer, input, (size_t)(bEnd-input)); + state->bufferedSize = (XXH32_hash_t)(bEnd-input); +#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1 + /* save stack accumulators into state */ + XXH_memcpy(state->acc, acc, sizeof(acc)); +#endif + } + + return XXH_OK; +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_64bits_update(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len) +{ + return XXH3_update(state, (const xxh_u8*)input, len, + XXH3_accumulate, XXH3_scrambleAcc); +} + + +XXH_FORCE_INLINE void +XXH3_digest_long (XXH64_hash_t* acc, + const XXH3_state_t* state, + const unsigned char* secret) +{ + xxh_u8 lastStripe[XXH_STRIPE_LEN]; + const xxh_u8* lastStripePtr; + + /* + * Digest on a local copy. This way, the state remains unaltered, and it can + * continue ingesting more input afterwards. + */ + XXH_memcpy(acc, state->acc, sizeof(state->acc)); + if (state->bufferedSize >= XXH_STRIPE_LEN) { + /* Consume remaining stripes then point to remaining data in buffer */ + size_t const nbStripes = (state->bufferedSize - 1) / XXH_STRIPE_LEN; + size_t nbStripesSoFar = state->nbStripesSoFar; + XXH3_consumeStripes(acc, + &nbStripesSoFar, state->nbStripesPerBlock, + state->buffer, nbStripes, + secret, state->secretLimit, + XXH3_accumulate, XXH3_scrambleAcc); + lastStripePtr = state->buffer + state->bufferedSize - XXH_STRIPE_LEN; + } else { /* bufferedSize < XXH_STRIPE_LEN */ + /* Copy to temp buffer */ + size_t const catchupSize = XXH_STRIPE_LEN - state->bufferedSize; + XXH_ASSERT(state->bufferedSize > 0); /* there is always some input buffered */ + XXH_memcpy(lastStripe, state->buffer + sizeof(state->buffer) - catchupSize, catchupSize); + XXH_memcpy(lastStripe + catchupSize, state->buffer, state->bufferedSize); + lastStripePtr = lastStripe; + } + /* Last stripe */ + XXH3_accumulate_512(acc, + lastStripePtr, + secret + state->secretLimit - XXH_SECRET_LASTACC_START); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (XXH_NOESCAPE const XXH3_state_t* state) +{ + const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret; + if (state->totalLen > XXH3_MIDSIZE_MAX) { + XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB]; + XXH3_digest_long(acc, state, secret); + return XXH3_finalizeLong_64b(acc, secret, (xxh_u64)state->totalLen); + } + /* totalLen <= XXH3_MIDSIZE_MAX: digesting a short input */ + if (state->useSeed) + return XXH3_64bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed); + return XXH3_64bits_withSecret(state->buffer, (size_t)(state->totalLen), + secret, state->secretLimit + XXH_STRIPE_LEN); +} +#endif /* !XXH_NO_STREAM */ + + +/* ========================================== + * XXH3 128 bits (a.k.a XXH128) + * ========================================== + * XXH3's 128-bit variant has better mixing and strength than the 64-bit variant, + * even without counting the significantly larger output size. + * + * For example, extra steps are taken to avoid the seed-dependent collisions + * in 17-240 byte inputs (See XXH3_mix16B and XXH128_mix32B). + * + * This strength naturally comes at the cost of some speed, especially on short + * lengths. Note that longer hashes are about as fast as the 64-bit version + * due to it using only a slight modification of the 64-bit loop. + * + * XXH128 is also more oriented towards 64-bit machines. It is still extremely + * fast for a _128-bit_ hash on 32-bit (it usually clears XXH64). + */ + +XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t +XXH3_len_1to3_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + /* A doubled version of 1to3_64b with different constants. */ + XXH_ASSERT(input != NULL); + XXH_ASSERT(1 <= len && len <= 3); + XXH_ASSERT(secret != NULL); + /* + * len = 1: combinedl = { input[0], 0x01, input[0], input[0] } + * len = 2: combinedl = { input[1], 0x02, input[0], input[1] } + * len = 3: combinedl = { input[2], 0x03, input[0], input[1] } + */ + { xxh_u8 const c1 = input[0]; + xxh_u8 const c2 = input[len >> 1]; + xxh_u8 const c3 = input[len - 1]; + xxh_u32 const combinedl = ((xxh_u32)c1 <<16) | ((xxh_u32)c2 << 24) + | ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8); + xxh_u32 const combinedh = XXH_rotl32(XXH_swap32(combinedl), 13); + xxh_u64 const bitflipl = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed; + xxh_u64 const bitfliph = (XXH_readLE32(secret+8) ^ XXH_readLE32(secret+12)) - seed; + xxh_u64 const keyed_lo = (xxh_u64)combinedl ^ bitflipl; + xxh_u64 const keyed_hi = (xxh_u64)combinedh ^ bitfliph; + XXH128_hash_t h128; + h128.low64 = XXH64_avalanche(keyed_lo); + h128.high64 = XXH64_avalanche(keyed_hi); + return h128; + } +} + +XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t +XXH3_len_4to8_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(input != NULL); + XXH_ASSERT(secret != NULL); + XXH_ASSERT(4 <= len && len <= 8); + seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32; + { xxh_u32 const input_lo = XXH_readLE32(input); + xxh_u32 const input_hi = XXH_readLE32(input + len - 4); + xxh_u64 const input_64 = input_lo + ((xxh_u64)input_hi << 32); + xxh_u64 const bitflip = (XXH_readLE64(secret+16) ^ XXH_readLE64(secret+24)) + seed; + xxh_u64 const keyed = input_64 ^ bitflip; + + /* Shift len to the left to ensure it is even, this avoids even multiplies. */ + XXH128_hash_t m128 = XXH_mult64to128(keyed, XXH_PRIME64_1 + (len << 2)); + + m128.high64 += (m128.low64 << 1); + m128.low64 ^= (m128.high64 >> 3); + + m128.low64 = XXH_xorshift64(m128.low64, 35); + m128.low64 *= PRIME_MX2; + m128.low64 = XXH_xorshift64(m128.low64, 28); + m128.high64 = XXH3_avalanche(m128.high64); + return m128; + } +} + +XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t +XXH3_len_9to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(input != NULL); + XXH_ASSERT(secret != NULL); + XXH_ASSERT(9 <= len && len <= 16); + { xxh_u64 const bitflipl = (XXH_readLE64(secret+32) ^ XXH_readLE64(secret+40)) - seed; + xxh_u64 const bitfliph = (XXH_readLE64(secret+48) ^ XXH_readLE64(secret+56)) + seed; + xxh_u64 const input_lo = XXH_readLE64(input); + xxh_u64 input_hi = XXH_readLE64(input + len - 8); + XXH128_hash_t m128 = XXH_mult64to128(input_lo ^ input_hi ^ bitflipl, XXH_PRIME64_1); + /* + * Put len in the middle of m128 to ensure that the length gets mixed to + * both the low and high bits in the 128x64 multiply below. + */ + m128.low64 += (xxh_u64)(len - 1) << 54; + input_hi ^= bitfliph; + /* + * Add the high 32 bits of input_hi to the high 32 bits of m128, then + * add the long product of the low 32 bits of input_hi and XXH_PRIME32_2 to + * the high 64 bits of m128. + * + * The best approach to this operation is different on 32-bit and 64-bit. + */ + if (sizeof(void *) < sizeof(xxh_u64)) { /* 32-bit */ + /* + * 32-bit optimized version, which is more readable. + * + * On 32-bit, it removes an ADC and delays a dependency between the two + * halves of m128.high64, but it generates an extra mask on 64-bit. + */ + m128.high64 += (input_hi & 0xFFFFFFFF00000000ULL) + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2); + } else { + /* + * 64-bit optimized (albeit more confusing) version. + * + * Uses some properties of addition and multiplication to remove the mask: + * + * Let: + * a = input_hi.lo = (input_hi & 0x00000000FFFFFFFF) + * b = input_hi.hi = (input_hi & 0xFFFFFFFF00000000) + * c = XXH_PRIME32_2 + * + * a + (b * c) + * Inverse Property: x + y - x == y + * a + (b * (1 + c - 1)) + * Distributive Property: x * (y + z) == (x * y) + (x * z) + * a + (b * 1) + (b * (c - 1)) + * Identity Property: x * 1 == x + * a + b + (b * (c - 1)) + * + * Substitute a, b, and c: + * input_hi.hi + input_hi.lo + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1)) + * + * Since input_hi.hi + input_hi.lo == input_hi, we get this: + * input_hi + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1)) + */ + m128.high64 += input_hi + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2 - 1); + } + /* m128 ^= XXH_swap64(m128 >> 64); */ + m128.low64 ^= XXH_swap64(m128.high64); + + { /* 128x64 multiply: h128 = m128 * XXH_PRIME64_2; */ + XXH128_hash_t h128 = XXH_mult64to128(m128.low64, XXH_PRIME64_2); + h128.high64 += m128.high64 * XXH_PRIME64_2; + + h128.low64 = XXH3_avalanche(h128.low64); + h128.high64 = XXH3_avalanche(h128.high64); + return h128; + } } +} + +/* + * Assumption: `secret` size is >= XXH3_SECRET_SIZE_MIN + */ +XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t +XXH3_len_0to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(len <= 16); + { if (len > 8) return XXH3_len_9to16_128b(input, len, secret, seed); + if (len >= 4) return XXH3_len_4to8_128b(input, len, secret, seed); + if (len) return XXH3_len_1to3_128b(input, len, secret, seed); + { XXH128_hash_t h128; + xxh_u64 const bitflipl = XXH_readLE64(secret+64) ^ XXH_readLE64(secret+72); + xxh_u64 const bitfliph = XXH_readLE64(secret+80) ^ XXH_readLE64(secret+88); + h128.low64 = XXH64_avalanche(seed ^ bitflipl); + h128.high64 = XXH64_avalanche( seed ^ bitfliph); + return h128; + } } +} + +/* + * A bit slower than XXH3_mix16B, but handles multiply by zero better. + */ +XXH_FORCE_INLINE XXH128_hash_t +XXH128_mix32B(XXH128_hash_t acc, const xxh_u8* input_1, const xxh_u8* input_2, + const xxh_u8* secret, XXH64_hash_t seed) +{ + acc.low64 += XXH3_mix16B (input_1, secret+0, seed); + acc.low64 ^= XXH_readLE64(input_2) + XXH_readLE64(input_2 + 8); + acc.high64 += XXH3_mix16B (input_2, secret+16, seed); + acc.high64 ^= XXH_readLE64(input_1) + XXH_readLE64(input_1 + 8); + return acc; +} + + +XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t +XXH3_len_17to128_128b(const xxh_u8* XXH_RESTRICT input, size_t len, + const xxh_u8* XXH_RESTRICT secret, size_t secretSize, + XXH64_hash_t seed) +{ + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize; + XXH_ASSERT(16 < len && len <= 128); + + { XXH128_hash_t acc; + acc.low64 = len * XXH_PRIME64_1; + acc.high64 = 0; + +#if XXH_SIZE_OPT >= 1 + { + /* Smaller, but slightly slower. */ + unsigned int i = (unsigned int)(len - 1) / 32; + do { + acc = XXH128_mix32B(acc, input+16*i, input+len-16*(i+1), secret+32*i, seed); + } while (i-- != 0); + } +#else + if (len > 32) { + if (len > 64) { + if (len > 96) { + acc = XXH128_mix32B(acc, input+48, input+len-64, secret+96, seed); + } + acc = XXH128_mix32B(acc, input+32, input+len-48, secret+64, seed); + } + acc = XXH128_mix32B(acc, input+16, input+len-32, secret+32, seed); + } + acc = XXH128_mix32B(acc, input, input+len-16, secret, seed); +#endif + { XXH128_hash_t h128; + h128.low64 = acc.low64 + acc.high64; + h128.high64 = (acc.low64 * XXH_PRIME64_1) + + (acc.high64 * XXH_PRIME64_4) + + ((len - seed) * XXH_PRIME64_2); + h128.low64 = XXH3_avalanche(h128.low64); + h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64); + return h128; + } + } +} + +XXH_NO_INLINE XXH_PUREF XXH128_hash_t +XXH3_len_129to240_128b(const xxh_u8* XXH_RESTRICT input, size_t len, + const xxh_u8* XXH_RESTRICT secret, size_t secretSize, + XXH64_hash_t seed) +{ + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize; + XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX); + + { XXH128_hash_t acc; + unsigned i; + acc.low64 = len * XXH_PRIME64_1; + acc.high64 = 0; + /* + * We set as `i` as offset + 32. We do this so that unchanged + * `len` can be used as upper bound. This reaches a sweet spot + * where both x86 and aarch64 get simple agen and good codegen + * for the loop. + */ + for (i = 32; i < 160; i += 32) { + acc = XXH128_mix32B(acc, + input + i - 32, + input + i - 16, + secret + i - 32, + seed); + } + acc.low64 = XXH3_avalanche(acc.low64); + acc.high64 = XXH3_avalanche(acc.high64); + /* + * NB: `i <= len` will duplicate the last 32-bytes if + * len % 32 was zero. This is an unfortunate necessity to keep + * the hash result stable. + */ + for (i=160; i <= len; i += 32) { + acc = XXH128_mix32B(acc, + input + i - 32, + input + i - 16, + secret + XXH3_MIDSIZE_STARTOFFSET + i - 160, + seed); + } + /* last bytes */ + acc = XXH128_mix32B(acc, + input + len - 16, + input + len - 32, + secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET - 16, + (XXH64_hash_t)0 - seed); + + { XXH128_hash_t h128; + h128.low64 = acc.low64 + acc.high64; + h128.high64 = (acc.low64 * XXH_PRIME64_1) + + (acc.high64 * XXH_PRIME64_4) + + ((len - seed) * XXH_PRIME64_2); + h128.low64 = XXH3_avalanche(h128.low64); + h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64); + return h128; + } + } +} + +static XXH_PUREF XXH128_hash_t +XXH3_finalizeLong_128b(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret, size_t secretSize, xxh_u64 len) +{ + XXH128_hash_t h128; + h128.low64 = XXH3_finalizeLong_64b(acc, secret, len); + h128.high64 = XXH3_mergeAccs(acc, secret + secretSize + - XXH_STRIPE_LEN - XXH_SECRET_MERGEACCS_START, + ~(len * XXH_PRIME64_2)); + return h128; +} + +XXH_FORCE_INLINE XXH128_hash_t +XXH3_hashLong_128b_internal(const void* XXH_RESTRICT input, size_t len, + const xxh_u8* XXH_RESTRICT secret, size_t secretSize, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble) +{ + XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC; + + XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, secret, secretSize, f_acc, f_scramble); + + /* converge into final hash */ + XXH_STATIC_ASSERT(sizeof(acc) == 64); + XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START); + return XXH3_finalizeLong_128b(acc, secret, secretSize, (xxh_u64)len); +} + +/* + * It's important for performance that XXH3_hashLong() is not inlined. + */ +XXH_NO_INLINE XXH_PUREF XXH128_hash_t +XXH3_hashLong_128b_default(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, + const void* XXH_RESTRICT secret, size_t secretLen) +{ + (void)seed64; (void)secret; (void)secretLen; + return XXH3_hashLong_128b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret), + XXH3_accumulate, XXH3_scrambleAcc); +} + +/* + * It's important for performance to pass @p secretLen (when it's static) + * to the compiler, so that it can properly optimize the vectorized loop. + * + * When the secret size is unknown, or on GCC 12 where the mix of NO_INLINE and FORCE_INLINE + * breaks -Og, this is XXH_NO_INLINE. + */ +XXH3_WITH_SECRET_INLINE XXH128_hash_t +XXH3_hashLong_128b_withSecret(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, + const void* XXH_RESTRICT secret, size_t secretLen) +{ + (void)seed64; + return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, secretLen, + XXH3_accumulate, XXH3_scrambleAcc); +} + +XXH_FORCE_INLINE XXH128_hash_t +XXH3_hashLong_128b_withSeed_internal(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble, + XXH3_f_initCustomSecret f_initSec) +{ + if (seed64 == 0) + return XXH3_hashLong_128b_internal(input, len, + XXH3_kSecret, sizeof(XXH3_kSecret), + f_acc, f_scramble); + { XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE]; + f_initSec(secret, seed64); + return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, sizeof(secret), + f_acc, f_scramble); + } +} + +/* + * It's important for performance that XXH3_hashLong is not inlined. + */ +XXH_NO_INLINE XXH128_hash_t +XXH3_hashLong_128b_withSeed(const void* input, size_t len, + XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen) +{ + (void)secret; (void)secretLen; + return XXH3_hashLong_128b_withSeed_internal(input, len, seed64, + XXH3_accumulate, XXH3_scrambleAcc, XXH3_initCustomSecret); +} + +typedef XXH128_hash_t (*XXH3_hashLong128_f)(const void* XXH_RESTRICT, size_t, + XXH64_hash_t, const void* XXH_RESTRICT, size_t); + +XXH_FORCE_INLINE XXH128_hash_t +XXH3_128bits_internal(const void* input, size_t len, + XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen, + XXH3_hashLong128_f f_hl128) +{ + XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN); + /* + * If an action is to be taken if `secret` conditions are not respected, + * it should be done here. + * For now, it's a contract pre-condition. + * Adding a check and a branch here would cost performance at every hash. + */ + if (len <= 16) + return XXH3_len_0to16_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64); + if (len <= 128) + return XXH3_len_17to128_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64); + if (len <= XXH3_MIDSIZE_MAX) + return XXH3_len_129to240_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64); + return f_hl128(input, len, seed64, secret, secretLen); +} + + +/* === Public XXH128 API === */ + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void* input, size_t len) +{ + return XXH3_128bits_internal(input, len, 0, + XXH3_kSecret, sizeof(XXH3_kSecret), + XXH3_hashLong_128b_default); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t +XXH3_128bits_withSecret(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize) +{ + return XXH3_128bits_internal(input, len, 0, + (const xxh_u8*)secret, secretSize, + XXH3_hashLong_128b_withSecret); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t +XXH3_128bits_withSeed(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed) +{ + return XXH3_128bits_internal(input, len, seed, + XXH3_kSecret, sizeof(XXH3_kSecret), + XXH3_hashLong_128b_withSeed); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t +XXH3_128bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed) +{ + if (len <= XXH3_MIDSIZE_MAX) + return XXH3_128bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL); + return XXH3_hashLong_128b_withSecret(input, len, seed, secret, secretSize); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t +XXH128(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed) +{ + return XXH3_128bits_withSeed(input, len, seed); +} + + +/* === XXH3 128-bit streaming === */ +#ifndef XXH_NO_STREAM +/* + * All initialization and update functions are identical to 64-bit streaming variant. + * The only difference is the finalization routine. + */ + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr) +{ + return XXH3_64bits_reset(statePtr); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize) +{ + return XXH3_64bits_reset_withSecret(statePtr, secret, secretSize); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_128bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed) +{ + return XXH3_64bits_reset_withSeed(statePtr, seed); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed) +{ + return XXH3_64bits_reset_withSecretandSeed(statePtr, secret, secretSize, seed); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_128bits_update(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len) +{ + return XXH3_64bits_update(state, input, len); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (XXH_NOESCAPE const XXH3_state_t* state) +{ + const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret; + if (state->totalLen > XXH3_MIDSIZE_MAX) { + XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB]; + XXH3_digest_long(acc, state, secret); + XXH_ASSERT(state->secretLimit + XXH_STRIPE_LEN >= sizeof(acc) + XXH_SECRET_MERGEACCS_START); + return XXH3_finalizeLong_128b(acc, secret, state->secretLimit + XXH_STRIPE_LEN, (xxh_u64)state->totalLen); + } + /* len <= XXH3_MIDSIZE_MAX : short code */ + if (state->useSeed) + return XXH3_128bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed); + return XXH3_128bits_withSecret(state->buffer, (size_t)(state->totalLen), + secret, state->secretLimit + XXH_STRIPE_LEN); +} +#endif /* !XXH_NO_STREAM */ +/* 128-bit utility functions */ + +#include /* memcmp, memcpy */ + +/* return : 1 is equal, 0 if different */ +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2) +{ + /* note : XXH128_hash_t is compact, it has no padding byte */ + return !(memcmp(&h1, &h2, sizeof(h1))); +} + +/* This prototype is compatible with stdlib's qsort(). + * @return : >0 if *h128_1 > *h128_2 + * <0 if *h128_1 < *h128_2 + * =0 if *h128_1 == *h128_2 */ +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API int XXH128_cmp(XXH_NOESCAPE const void* h128_1, XXH_NOESCAPE const void* h128_2) +{ + XXH128_hash_t const h1 = *(const XXH128_hash_t*)h128_1; + XXH128_hash_t const h2 = *(const XXH128_hash_t*)h128_2; + int const hcmp = (h1.high64 > h2.high64) - (h2.high64 > h1.high64); + /* note : bets that, in most cases, hash values are different */ + if (hcmp) return hcmp; + return (h1.low64 > h2.low64) - (h2.low64 > h1.low64); +} + + +/*====== Canonical representation ======*/ +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API void +XXH128_canonicalFromHash(XXH_NOESCAPE XXH128_canonical_t* dst, XXH128_hash_t hash) +{ + XXH_STATIC_ASSERT(sizeof(XXH128_canonical_t) == sizeof(XXH128_hash_t)); + if (XXH_CPU_LITTLE_ENDIAN) { + hash.high64 = XXH_swap64(hash.high64); + hash.low64 = XXH_swap64(hash.low64); + } + XXH_memcpy(dst, &hash.high64, sizeof(hash.high64)); + XXH_memcpy((char*)dst + sizeof(hash.high64), &hash.low64, sizeof(hash.low64)); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t +XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t* src) +{ + XXH128_hash_t h; + h.high64 = XXH_readBE64(src); + h.low64 = XXH_readBE64(src->digest + 8); + return h; +} + + + +/* ========================================== + * Secret generators + * ========================================== + */ +#define XXH_MIN(x, y) (((x) > (y)) ? (y) : (x)) + +XXH_FORCE_INLINE void XXH3_combine16(void* dst, XXH128_hash_t h128) +{ + XXH_writeLE64( dst, XXH_readLE64(dst) ^ h128.low64 ); + XXH_writeLE64( (char*)dst+8, XXH_readLE64((char*)dst+8) ^ h128.high64 ); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_generateSecret(XXH_NOESCAPE void* secretBuffer, size_t secretSize, XXH_NOESCAPE const void* customSeed, size_t customSeedSize) +{ +#if (XXH_DEBUGLEVEL >= 1) + XXH_ASSERT(secretBuffer != NULL); + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); +#else + /* production mode, assert() are disabled */ + if (secretBuffer == NULL) return XXH_ERROR; + if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR; +#endif + + if (customSeedSize == 0) { + customSeed = XXH3_kSecret; + customSeedSize = XXH_SECRET_DEFAULT_SIZE; + } +#if (XXH_DEBUGLEVEL >= 1) + XXH_ASSERT(customSeed != NULL); +#else + if (customSeed == NULL) return XXH_ERROR; +#endif + + /* Fill secretBuffer with a copy of customSeed - repeat as needed */ + { size_t pos = 0; + while (pos < secretSize) { + size_t const toCopy = XXH_MIN((secretSize - pos), customSeedSize); + memcpy((char*)secretBuffer + pos, customSeed, toCopy); + pos += toCopy; + } } + + { size_t const nbSeg16 = secretSize / 16; + size_t n; + XXH128_canonical_t scrambler; + XXH128_canonicalFromHash(&scrambler, XXH128(customSeed, customSeedSize, 0)); + for (n=0; n