/* $Id: sph_types.h 260 2011-07-21 01:02:38Z tp $ */
/**
* Basic type definitions.
*
* This header file defines the generic integer types that will be used
* for the implementation of hash functions; it also contains helper
* functions which encode and decode multi-byte integer values, using
* either little-endian or big-endian conventions.
*
* This file contains a compile-time test on the size of a byte
* (the unsigned char C type). If bytes are not octets,
* i.e. if they do not have a size of exactly 8 bits, then compilation
* is aborted. Architectures where bytes are not octets are relatively
* rare, even in the embedded devices market. We forbid non-octet bytes
* because there is no clear convention on how octet streams are encoded
* on such systems.
*
* ==========================(LICENSE BEGIN)============================
*
* Copyright (c) 2007-2010 Projet RNRT SAPHIR
*
* Permission is hereby granted, free of charge, to any person obtaining
* a copy of this software and associated documentation files (the
* "Software"), to deal in the Software without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
* distribute, sublicense, and/or sell copies of the Software, and to
* permit persons to whom the Software is furnished to do so, subject to
* the following conditions:
*
* The above copyright notice and this permission notice shall be
* included in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*
* ===========================(LICENSE END)=============================
*
* @file sph_types.h
* @author Thomas Pornin
*/
#ifndef SPH_TYPES_H__
#define SPH_TYPES_H__
#include
/*
* All our I/O functions are defined over octet streams. We do not know
* how to handle input data if bytes are not octets.
*/
#if CHAR_BIT != 8
#error This code requires 8-bit bytes
#endif
/* ============= BEGIN documentation block for Doxygen ============ */
#ifdef DOXYGEN_IGNORE
/** @mainpage sphlib C code documentation
*
* @section overview Overview
*
* sphlib is a library which contains implementations of
* various cryptographic hash functions. These pages have been generated
* with doxygen and
* document the API for the C implementations.
*
* The API is described in appropriate header files, which are available
* in the "Files" section. Each hash function family has its own header,
* whose name begins with "sph_" and contains the family
* name. For instance, the API for the RIPEMD hash functions is available
* in the header file sph_ripemd.h.
*
* @section principles API structure and conventions
*
* @subsection io Input/output conventions
*
* In all generality, hash functions operate over strings of bits.
* Individual bits are rarely encountered in C programming or actual
* communication protocols; most protocols converge on the ubiquitous
* "octet" which is a group of eight bits. Data is thus expressed as a
* stream of octets. The C programming language contains the notion of a
* "byte", which is a data unit managed under the type "unsigned
* char". The C standard prescribes that a byte should hold at
* least eight bits, but possibly more. Most modern architectures, even
* in the embedded world, feature eight-bit bytes, i.e. map bytes to
* octets.
*
* Nevertheless, for some of the implemented hash functions, an extra
* API has been added, which allows the input of arbitrary sequences of
* bits: when the computation is about to be closed, 1 to 7 extra bits
* can be added. The functions for which this API is implemented include
* the SHA-2 functions and all SHA-3 candidates.
*
* sphlib defines hash function which may hash octet streams,
* i.e. streams of bits where the number of bits is a multiple of eight.
* The data input functions in the sphlib API expect data
* as anonymous pointers ("const void *") with a length
* (of type "size_t") which gives the input data chunk length
* in bytes. A byte is assumed to be an octet; the sph_types.h
* header contains a compile-time test which prevents compilation on
* architectures where this property is not met.
*
* The hash function output is also converted into bytes. All currently
* implemented hash functions have an output width which is a multiple of
* eight, and this is likely to remain true for new designs.
*
* Most hash functions internally convert input data into 32-bit of 64-bit
* words, using either little-endian or big-endian conversion. The hash
* output also often consists of such words, which are encoded into output
* bytes with a similar endianness convention. Some hash functions have
* been only loosely specified on that subject; when necessary,
* sphlib has been tested against published "reference"
* implementations in order to use the same conventions.
*
* @subsection shortname Function short name
*
* Each implemented hash function has a "short name" which is used
* internally to derive the identifiers for the functions and context
* structures which the function uses. For instance, MD5 has the short
* name "md5". Short names are listed in the next section,
* for the implemented hash functions. In subsequent sections, the
* short name will be assumed to be "XXX": replace with the
* actual hash function name to get the C identifier.
*
* Note: some functions within the same family share the same core
* elements, such as update function or context structure. Correspondingly,
* some of the defined types or functions may actually be macros which
* transparently evaluate to another type or function name.
*
* @subsection context Context structure
*
* Each implemented hash fonction has its own context structure, available
* under the type name "sph_XXX_context" for the hash function
* with short name "XXX". This structure holds all needed
* state for a running hash computation.
*
* The contents of these structures are meant to be opaque, and private
* to the implementation. However, these contents are specified in the
* header files so that application code which uses sphlib
* may access the size of those structures.
*
* The caller is responsible for allocating the context structure,
* whether by dynamic allocation (malloc() or equivalent),
* static allocation (a global permanent variable), as an automatic
* variable ("on the stack"), or by any other mean which ensures proper
* structure alignment. sphlib code performs no dynamic
* allocation by itself.
*
* The context must be initialized before use, using the
* sph_XXX_init() function. This function sets the context
* state to proper initial values for hashing.
*
* Since all state data is contained within the context structure,
* sphlib is thread-safe and reentrant: several hash
* computations may be performed in parallel, provided that they do not
* operate on the same context. Moreover, a running computation can be
* cloned by copying the context (with a simple memcpy()):
* the context and its clone are then independant and may be updated
* with new data and/or closed without interfering with each other.
* Similarly, a context structure can be moved in memory at will:
* context structures contain no pointer, in particular no pointer to
* themselves.
*
* @subsection dataio Data input
*
* Hashed data is input with the sph_XXX() fonction, which
* takes as parameters a pointer to the context, a pointer to the data
* to hash, and the number of data bytes to hash. The context is updated
* with the new data.
*
* Data can be input in one or several calls, with arbitrary input lengths.
* However, it is best, performance wise, to input data by relatively big
* chunks (say a few kilobytes), because this allows sphlib to
* optimize things and avoid internal copying.
*
* When all data has been input, the context can be closed with
* sph_XXX_close(). The hash output is computed and written
* into the provided buffer. The caller must take care to provide a
* buffer of appropriate length; e.g., when using SHA-1, the output is
* a 20-byte word, therefore the output buffer must be at least 20-byte
* long.
*
* For some hash functions, the sph_XXX_addbits_and_close()
* function can be used instead of sph_XXX_close(). This
* function can take a few extra bits to be added at
* the end of the input message. This allows hashing messages with a
* bit length which is not a multiple of 8. The extra bits are provided
* as an unsigned integer value, and a bit count. The bit count must be
* between 0 and 7, inclusive. The extra bits are provided as bits 7 to
* 0 (bits of numerical value 128, 64, 32... downto 0), in that order.
* For instance, to add three bits of value 1, 1 and 0, the unsigned
* integer will have value 192 (1*128 + 1*64 + 0*32) and the bit count
* will be 3.
*
* The SPH_SIZE_XXX macro is defined for each hash function;
* it evaluates to the function output size, expressed in bits. For instance,
* SPH_SIZE_sha1 evaluates to 160.
*
* When closed, the context is automatically reinitialized and can be
* immediately used for another computation. It is not necessary to call
* sph_XXX_init() after a close. Note that
* sph_XXX_init() can still be called to "reset" a context,
* i.e. forget previously input data, and get back to the initial state.
*
* @subsection alignment Data alignment
*
* "Alignment" is a property of data, which is said to be "properly
* aligned" when its emplacement in memory is such that the data can
* be optimally read by full words. This depends on the type of access;
* basically, some hash functions will read data by 32-bit or 64-bit
* words. sphlib does not mandate such alignment for input
* data, but using aligned data can substantially improve performance.
*
* As a rule, it is best to input data by chunks whose length (in bytes)
* is a multiple of eight, and which begins at "generally aligned"
* addresses, such as the base address returned by a call to
* malloc().
*
* @section functions Implemented functions
*
* We give here the list of implemented functions. They are grouped by
* family; to each family corresponds a specific header file. Each
* individual function has its associated "short name". Please refer to
* the documentation for that header file to get details on the hash
* function denomination and provenance.
*
* Note: the functions marked with a '(64)' in the list below are
* available only if the C compiler provides an integer type of length
* 64 bits or more. Such a type is mandatory in the latest C standard
* (ISO 9899:1999, aka "C99") and is present in several older compilers
* as well, so chances are that such a type is available.
*
* - HAVAL family: file sph_haval.h
* - HAVAL-128/3 (128-bit, 3 passes): short name: haval128_3
* - HAVAL-128/4 (128-bit, 4 passes): short name: haval128_4
* - HAVAL-128/5 (128-bit, 5 passes): short name: haval128_5
* - HAVAL-160/3 (160-bit, 3 passes): short name: haval160_3
* - HAVAL-160/4 (160-bit, 4 passes): short name: haval160_4
* - HAVAL-160/5 (160-bit, 5 passes): short name: haval160_5
* - HAVAL-192/3 (192-bit, 3 passes): short name: haval192_3
* - HAVAL-192/4 (192-bit, 4 passes): short name: haval192_4
* - HAVAL-192/5 (192-bit, 5 passes): short name: haval192_5
* - HAVAL-224/3 (224-bit, 3 passes): short name: haval224_3
* - HAVAL-224/4 (224-bit, 4 passes): short name: haval224_4
* - HAVAL-224/5 (224-bit, 5 passes): short name: haval224_5
* - HAVAL-256/3 (256-bit, 3 passes): short name: haval256_3
* - HAVAL-256/4 (256-bit, 4 passes): short name: haval256_4
* - HAVAL-256/5 (256-bit, 5 passes): short name: haval256_5
* - MD2: file sph_md2.h, short name: md2
* - MD4: file sph_md4.h, short name: md4
* - MD5: file sph_md5.h, short name: md5
* - PANAMA: file sph_panama.h, short name: panama
* - RadioGatun family: file sph_radiogatun.h
* - RadioGatun[32]: short name: radiogatun32
* - RadioGatun[64]: short name: radiogatun64 (64)
* - RIPEMD family: file sph_ripemd.h
* - RIPEMD: short name: ripemd
* - RIPEMD-128: short name: ripemd128
* - RIPEMD-160: short name: ripemd160
* - SHA-0: file sph_sha0.h, short name: sha0
* - SHA-1: file sph_sha1.h, short name: sha1
* - SHA-2 family, 32-bit hashes: file sph_sha2.h
* - SHA-224: short name: sha224
* - SHA-256: short name: sha256
* - SHA-384: short name: sha384 (64)
* - SHA-512: short name: sha512 (64)
* - Tiger family: file sph_tiger.h
* - Tiger: short name: tiger (64)
* - Tiger2: short name: tiger2 (64)
* - WHIRLPOOL family: file sph_whirlpool.h
* - WHIRLPOOL-0: short name: whirlpool0 (64)
* - WHIRLPOOL-1: short name: whirlpool1 (64)
* - WHIRLPOOL: short name: whirlpool (64)
*
* The fourteen second-round SHA-3 candidates are also implemented;
* when applicable, the implementations follow the "final" specifications
* as published for the third round of the SHA-3 competition (BLAKE,
* Groestl, JH, Keccak and Skein have been tweaked for third round).
*
* - BLAKE family: file sph_blake.h
* - BLAKE-224: short name: blake224
* - BLAKE-256: short name: blake256
* - BLAKE-384: short name: blake384
* - BLAKE-512: short name: blake512
* - BMW (Blue Midnight Wish) family: file sph_bmw.h
* - BMW-224: short name: bmw224
* - BMW-256: short name: bmw256
* - BMW-384: short name: bmw384 (64)
* - BMW-512: short name: bmw512 (64)
* - CubeHash family: file sph_cubehash.h (specified as
* CubeHash16/32 in the CubeHash specification)
* - CubeHash-224: short name: cubehash224
* - CubeHash-256: short name: cubehash256
* - CubeHash-384: short name: cubehash384
* - CubeHash-512: short name: cubehash512
* - ECHO family: file sph_echo.h
* - ECHO-224: short name: echo224
* - ECHO-256: short name: echo256
* - ECHO-384: short name: echo384
* - ECHO-512: short name: echo512
* - Fugue family: file sph_fugue.h
* - Fugue-224: short name: fugue224
* - Fugue-256: short name: fugue256
* - Fugue-384: short name: fugue384
* - Fugue-512: short name: fugue512
* - Groestl family: file sph_groestl.h
* - Groestl-224: short name: groestl224
* - Groestl-256: short name: groestl256
* - Groestl-384: short name: groestl384
* - Groestl-512: short name: groestl512
* - Hamsi family: file sph_hamsi.h
* - Hamsi-224: short name: hamsi224
* - Hamsi-256: short name: hamsi256
* - Hamsi-384: short name: hamsi384
* - Hamsi-512: short name: hamsi512
* - JH family: file sph_jh.h
* - JH-224: short name: jh224
* - JH-256: short name: jh256
* - JH-384: short name: jh384
* - JH-512: short name: jh512
* - Keccak family: file sph_keccak.h
* - Keccak-224: short name: keccak224
* - Keccak-256: short name: keccak256
* - Keccak-384: short name: keccak384
* - Keccak-512: short name: keccak512
* - Luffa family: file sph_luffa.h
* - Luffa-224: short name: luffa224
* - Luffa-256: short name: luffa256
* - Luffa-384: short name: luffa384
* - Luffa-512: short name: luffa512
* - Shabal family: file sph_shabal.h
* - Shabal-192: short name: shabal192
* - Shabal-224: short name: shabal224
* - Shabal-256: short name: shabal256
* - Shabal-384: short name: shabal384
* - Shabal-512: short name: shabal512
* - SHAvite-3 family: file sph_shavite.h
* - SHAvite-224 (nominally "SHAvite-3 with 224-bit output"):
* short name: shabal224
* - SHAvite-256 (nominally "SHAvite-3 with 256-bit output"):
* short name: shabal256
* - SHAvite-384 (nominally "SHAvite-3 with 384-bit output"):
* short name: shabal384
* - SHAvite-512 (nominally "SHAvite-3 with 512-bit output"):
* short name: shabal512
* - SIMD family: file sph_simd.h
* - SIMD-224: short name: simd224
* - SIMD-256: short name: simd256
* - SIMD-384: short name: simd384
* - SIMD-512: short name: simd512
* - Skein family: file sph_skein.h
* - Skein-224 (nominally specified as Skein-512-224): short name:
* skein224 (64)
* - Skein-256 (nominally specified as Skein-512-256): short name:
* skein256 (64)
* - Skein-384 (nominally specified as Skein-512-384): short name:
* skein384 (64)
* - Skein-512 (nominally specified as Skein-512-512): short name:
* skein512 (64)
*
* For the second-round SHA-3 candidates, the functions are as specified
* for round 2, i.e. with the "tweaks" that some candidates added
* between round 1 and round 2. Also, some of the submitted packages for
* round 2 contained errors, in the specification, reference code, or
* both. sphlib implements the corrected versions.
*/
/** @hideinitializer
* Unsigned integer type whose length is at least 32 bits; on most
* architectures, it will have a width of exactly 32 bits. Unsigned C
* types implement arithmetics modulo a power of 2; use the
* SPH_T32() macro to ensure that the value is truncated
* to exactly 32 bits. Unless otherwise specified, all macros and
* functions which accept sph_u32 values assume that these
* values fit on 32 bits, i.e. do not exceed 2^32-1, even on architectures
* where sph_u32 is larger than that.
*/
typedef __arch_dependant__ sph_u32;
/** @hideinitializer
* Signed integer type corresponding to sph_u32; it has
* width 32 bits or more.
*/
typedef __arch_dependant__ sph_s32;
/** @hideinitializer
* Unsigned integer type whose length is at least 64 bits; on most
* architectures which feature such a type, it will have a width of
* exactly 64 bits. C99-compliant platform will have this type; it
* is also defined when the GNU compiler (gcc) is used, and on
* platforms where unsigned long is large enough. If this
* type is not available, then some hash functions which depends on
* a 64-bit type will not be available (most notably SHA-384, SHA-512,
* Tiger and WHIRLPOOL).
*/
typedef __arch_dependant__ sph_u64;
/** @hideinitializer
* Signed integer type corresponding to sph_u64; it has
* width 64 bits or more.
*/
typedef __arch_dependant__ sph_s64;
/**
* This macro expands the token x into a suitable
* constant expression of type sph_u32. Depending on
* how this type is defined, a suffix such as UL may
* be appended to the argument.
*
* @param x the token to expand into a suitable constant expression
*/
#define SPH_C32(x)
/**
* Truncate a 32-bit value to exactly 32 bits. On most systems, this is
* a no-op, recognized as such by the compiler.
*
* @param x the value to truncate (of type sph_u32)
*/
#define SPH_T32(x)
/**
* Rotate a 32-bit value by a number of bits to the left. The rotate
* count must reside between 1 and 31. This macro assumes that its
* first argument fits in 32 bits (no extra bit allowed on machines where
* sph_u32 is wider); both arguments may be evaluated
* several times.
*
* @param x the value to rotate (of type sph_u32)
* @param n the rotation count (between 1 and 31, inclusive)
*/
#define SPH_ROTL32(x, n)
/**
* Rotate a 32-bit value by a number of bits to the left. The rotate
* count must reside between 1 and 31. This macro assumes that its
* first argument fits in 32 bits (no extra bit allowed on machines where
* sph_u32 is wider); both arguments may be evaluated
* several times.
*
* @param x the value to rotate (of type sph_u32)
* @param n the rotation count (between 1 and 31, inclusive)
*/
#define SPH_ROTR32(x, n)
/**
* This macro is defined on systems for which a 64-bit type has been
* detected, and is used for sph_u64.
*/
#define SPH_64
/**
* This macro is defined on systems for the "native" integer size is
* 64 bits (64-bit values fit in one register).
*/
#define SPH_64_TRUE
/**
* This macro expands the token x into a suitable
* constant expression of type sph_u64. Depending on
* how this type is defined, a suffix such as ULL may
* be appended to the argument. This macro is defined only if a
* 64-bit type was detected and used for sph_u64.
*
* @param x the token to expand into a suitable constant expression
*/
#define SPH_C64(x)
/**
* Truncate a 64-bit value to exactly 64 bits. On most systems, this is
* a no-op, recognized as such by the compiler. This macro is defined only
* if a 64-bit type was detected and used for sph_u64.
*
* @param x the value to truncate (of type sph_u64)
*/
#define SPH_T64(x)
/**
* Rotate a 64-bit value by a number of bits to the left. The rotate
* count must reside between 1 and 63. This macro assumes that its
* first argument fits in 64 bits (no extra bit allowed on machines where
* sph_u64 is wider); both arguments may be evaluated
* several times. This macro is defined only if a 64-bit type was detected
* and used for sph_u64.
*
* @param x the value to rotate (of type sph_u64)
* @param n the rotation count (between 1 and 63, inclusive)
*/
#define SPH_ROTL64(x, n)
/**
* Rotate a 64-bit value by a number of bits to the left. The rotate
* count must reside between 1 and 63. This macro assumes that its
* first argument fits in 64 bits (no extra bit allowed on machines where
* sph_u64 is wider); both arguments may be evaluated
* several times. This macro is defined only if a 64-bit type was detected
* and used for sph_u64.
*
* @param x the value to rotate (of type sph_u64)
* @param n the rotation count (between 1 and 63, inclusive)
*/
#define SPH_ROTR64(x, n)
/**
* This macro evaluates to inline or an equivalent construction,
* if available on the compilation platform, or to nothing otherwise. This
* is used to declare inline functions, for which the compiler should
* endeavour to include the code directly in the caller. Inline functions
* are typically defined in header files as replacement for macros.
*/
#define SPH_INLINE
/**
* This macro is defined if the platform has been detected as using
* little-endian convention. This implies that the sph_u32
* type (and the sph_u64 type also, if it is defined) has
* an exact width (i.e. exactly 32-bit, respectively 64-bit).
*/
#define SPH_LITTLE_ENDIAN
/**
* This macro is defined if the platform has been detected as using
* big-endian convention. This implies that the sph_u32
* type (and the sph_u64 type also, if it is defined) has
* an exact width (i.e. exactly 32-bit, respectively 64-bit).
*/
#define SPH_BIG_ENDIAN
/**
* This macro is defined if 32-bit words (and 64-bit words, if defined)
* can be read from and written to memory efficiently in little-endian
* convention. This is the case for little-endian platforms, and also
* for the big-endian platforms which have special little-endian access
* opcodes (e.g. Ultrasparc).
*/
#define SPH_LITTLE_FAST
/**
* This macro is defined if 32-bit words (and 64-bit words, if defined)
* can be read from and written to memory efficiently in big-endian
* convention. This is the case for little-endian platforms, and also
* for the little-endian platforms which have special big-endian access
* opcodes.
*/
#define SPH_BIG_FAST
/**
* On some platforms, this macro is defined to an unsigned integer type
* into which pointer values may be cast. The resulting value can then
* be tested for being a multiple of 2, 4 or 8, indicating an aligned
* pointer for, respectively, 16-bit, 32-bit or 64-bit memory accesses.
*/
#define SPH_UPTR
/**
* When defined, this macro indicates that unaligned memory accesses
* are possible with only a minor penalty, and thus should be prefered
* over strategies which first copy data to an aligned buffer.
*/
#define SPH_UNALIGNED
/**
* Byte-swap a 32-bit word (i.e. 0x12345678 becomes
* 0x78563412). This is an inline function which resorts
* to inline assembly on some platforms, for better performance.
*
* @param x the 32-bit value to byte-swap
* @return the byte-swapped value
*/
static inline sph_u32 sph_bswap32(sph_u32 x);
/**
* Byte-swap a 64-bit word. This is an inline function which resorts
* to inline assembly on some platforms, for better performance. This
* function is defined only if a suitable 64-bit type was found for
* sph_u64
*
* @param x the 64-bit value to byte-swap
* @return the byte-swapped value
*/
static inline sph_u64 sph_bswap64(sph_u64 x);
/**
* Decode a 16-bit unsigned value from memory, in little-endian convention
* (least significant byte comes first).
*
* @param src the source address
* @return the decoded value
*/
static inline unsigned sph_dec16le(const void *src);
/**
* Encode a 16-bit unsigned value into memory, in little-endian convention
* (least significant byte comes first).
*
* @param dst the destination buffer
* @param val the value to encode
*/
static inline void sph_enc16le(void *dst, unsigned val);
/**
* Decode a 16-bit unsigned value from memory, in big-endian convention
* (most significant byte comes first).
*
* @param src the source address
* @return the decoded value
*/
static inline unsigned sph_dec16be(const void *src);
/**
* Encode a 16-bit unsigned value into memory, in big-endian convention
* (most significant byte comes first).
*
* @param dst the destination buffer
* @param val the value to encode
*/
static inline void sph_enc16be(void *dst, unsigned val);
/**
* Decode a 32-bit unsigned value from memory, in little-endian convention
* (least significant byte comes first).
*
* @param src the source address
* @return the decoded value
*/
static inline sph_u32 sph_dec32le(const void *src);
/**
* Decode a 32-bit unsigned value from memory, in little-endian convention
* (least significant byte comes first). This function assumes that the
* source address is suitably aligned for a direct access, if the platform
* supports such things; it can thus be marginally faster than the generic
* sph_dec32le() function.
*
* @param src the source address
* @return the decoded value
*/
static inline sph_u32 sph_dec32le_aligned(const void *src);
/**
* Encode a 32-bit unsigned value into memory, in little-endian convention
* (least significant byte comes first).
*
* @param dst the destination buffer
* @param val the value to encode
*/
static inline void sph_enc32le(void *dst, sph_u32 val);
/**
* Encode a 32-bit unsigned value into memory, in little-endian convention
* (least significant byte comes first). This function assumes that the
* destination address is suitably aligned for a direct access, if the
* platform supports such things; it can thus be marginally faster than
* the generic sph_enc32le() function.
*
* @param dst the destination buffer
* @param val the value to encode
*/
static inline void sph_enc32le_aligned(void *dst, sph_u32 val);
/**
* Decode a 32-bit unsigned value from memory, in big-endian convention
* (most significant byte comes first).
*
* @param src the source address
* @return the decoded value
*/
static inline sph_u32 sph_dec32be(const void *src);
/**
* Decode a 32-bit unsigned value from memory, in big-endian convention
* (most significant byte comes first). This function assumes that the
* source address is suitably aligned for a direct access, if the platform
* supports such things; it can thus be marginally faster than the generic
* sph_dec32be() function.
*
* @param src the source address
* @return the decoded value
*/
static inline sph_u32 sph_dec32be_aligned(const void *src);
/**
* Encode a 32-bit unsigned value into memory, in big-endian convention
* (most significant byte comes first).
*
* @param dst the destination buffer
* @param val the value to encode
*/
static inline void sph_enc32be(void *dst, sph_u32 val);
/**
* Encode a 32-bit unsigned value into memory, in big-endian convention
* (most significant byte comes first). This function assumes that the
* destination address is suitably aligned for a direct access, if the
* platform supports such things; it can thus be marginally faster than
* the generic sph_enc32be() function.
*
* @param dst the destination buffer
* @param val the value to encode
*/
static inline void sph_enc32be_aligned(void *dst, sph_u32 val);
/**
* Decode a 64-bit unsigned value from memory, in little-endian convention
* (least significant byte comes first). This function is defined only
* if a suitable 64-bit type was detected and used for sph_u64.
*
* @param src the source address
* @return the decoded value
*/
static inline sph_u64 sph_dec64le(const void *src);
/**
* Decode a 64-bit unsigned value from memory, in little-endian convention
* (least significant byte comes first). This function assumes that the
* source address is suitably aligned for a direct access, if the platform
* supports such things; it can thus be marginally faster than the generic
* sph_dec64le() function. This function is defined only
* if a suitable 64-bit type was detected and used for sph_u64.
*
* @param src the source address
* @return the decoded value
*/
static inline sph_u64 sph_dec64le_aligned(const void *src);
/**
* Encode a 64-bit unsigned value into memory, in little-endian convention
* (least significant byte comes first). This function is defined only
* if a suitable 64-bit type was detected and used for sph_u64.
*
* @param dst the destination buffer
* @param val the value to encode
*/
static inline void sph_enc64le(void *dst, sph_u64 val);
/**
* Encode a 64-bit unsigned value into memory, in little-endian convention
* (least significant byte comes first). This function assumes that the
* destination address is suitably aligned for a direct access, if the
* platform supports such things; it can thus be marginally faster than
* the generic sph_enc64le() function. This function is defined
* only if a suitable 64-bit type was detected and used for
* sph_u64.
*
* @param dst the destination buffer
* @param val the value to encode
*/
static inline void sph_enc64le_aligned(void *dst, sph_u64 val);
/**
* Decode a 64-bit unsigned value from memory, in big-endian convention
* (most significant byte comes first). This function is defined only
* if a suitable 64-bit type was detected and used for sph_u64.
*
* @param src the source address
* @return the decoded value
*/
static inline sph_u64 sph_dec64be(const void *src);
/**
* Decode a 64-bit unsigned value from memory, in big-endian convention
* (most significant byte comes first). This function assumes that the
* source address is suitably aligned for a direct access, if the platform
* supports such things; it can thus be marginally faster than the generic
* sph_dec64be() function. This function is defined only
* if a suitable 64-bit type was detected and used for sph_u64.
*
* @param src the source address
* @return the decoded value
*/
static inline sph_u64 sph_dec64be_aligned(const void *src);
/**
* Encode a 64-bit unsigned value into memory, in big-endian convention
* (most significant byte comes first). This function is defined only
* if a suitable 64-bit type was detected and used for sph_u64.
*
* @param dst the destination buffer
* @param val the value to encode
*/
static inline void sph_enc64be(void *dst, sph_u64 val);
/**
* Encode a 64-bit unsigned value into memory, in big-endian convention
* (most significant byte comes first). This function assumes that the
* destination address is suitably aligned for a direct access, if the
* platform supports such things; it can thus be marginally faster than
* the generic sph_enc64be() function. This function is defined
* only if a suitable 64-bit type was detected and used for
* sph_u64.
*
* @param dst the destination buffer
* @param val the value to encode
*/
static inline void sph_enc64be_aligned(void *dst, sph_u64 val);
#endif
/* ============== END documentation block for Doxygen ============= */
#ifndef DOXYGEN_IGNORE
/*
* We want to define the types "sph_u32" and "sph_u64" which hold
* unsigned values of at least, respectively, 32 and 64 bits. These
* tests should select appropriate types for most platforms. The
* macro "SPH_64" is defined if the 64-bit is supported.
*/
#undef SPH_64
#undef SPH_64_TRUE
#if defined __STDC__ && __STDC_VERSION__ >= 199901L
/*
* On C99 implementations, we can use to get an exact 64-bit
* type, if any, or otherwise use a wider type (which must exist, for
* C99 conformance).
*/
#include
#ifdef UINT32_MAX
typedef uint32_t sph_u32;
typedef int32_t sph_s32;
#else
typedef uint_fast32_t sph_u32;
typedef int_fast32_t sph_s32;
#endif
#if !SPH_NO_64
#ifdef UINT64_MAX
typedef uint64_t sph_u64;
typedef int64_t sph_s64;
#else
typedef uint_fast64_t sph_u64;
typedef int_fast64_t sph_s64;
#endif
#endif
#define SPH_C32(x) ((sph_u32)(x))
#if !SPH_NO_64
#define SPH_C64(x) ((sph_u64)(x))
#define SPH_64 1
#endif
#else
/*
* On non-C99 systems, we use "unsigned int" if it is wide enough,
* "unsigned long" otherwise. This supports all "reasonable" architectures.
* We have to be cautious: pre-C99 preprocessors handle constants
* differently in '#if' expressions. Hence the shifts to test UINT_MAX.
*/
#if ((UINT_MAX >> 11) >> 11) >= 0x3FF
typedef unsigned int sph_u32;
typedef int sph_s32;
#define SPH_C32(x) ((sph_u32)(x ## U))
#else
typedef unsigned long sph_u32;
typedef long sph_s32;
#define SPH_C32(x) ((sph_u32)(x ## UL))
#endif
#if !SPH_NO_64
/*
* We want a 64-bit type. We use "unsigned long" if it is wide enough (as
* is common on 64-bit architectures such as AMD64, Alpha or Sparcv9),
* "unsigned long long" otherwise, if available. We use ULLONG_MAX to
* test whether "unsigned long long" is available; we also know that
* gcc features this type, even if the libc header do not know it.
*/
#if ((ULONG_MAX >> 31) >> 31) >= 3
typedef unsigned long sph_u64;
typedef long sph_s64;
#define SPH_C64(x) ((sph_u64)(x ## UL))
#define SPH_64 1
#elif ((ULLONG_MAX >> 31) >> 31) >= 3 || defined __GNUC__
typedef unsigned long long sph_u64;
typedef long long sph_s64;
#define SPH_C64(x) ((sph_u64)(x ## ULL))
#define SPH_64 1
#else
/*
* No 64-bit type...
*/
#endif
#endif
#endif
/*
* If the "unsigned long" type has length 64 bits or more, then this is
* a "true" 64-bit architectures. This is also true with Visual C on
* amd64, even though the "long" type is limited to 32 bits.
*/
#if SPH_64 && (((ULONG_MAX >> 31) >> 31) >= 3 || defined _M_X64)
#define SPH_64_TRUE 1
#endif
/*
* Implementation note: some processors have specific opcodes to perform
* a rotation. Recent versions of gcc recognize the expression above and
* use the relevant opcodes, when appropriate.
*/
#define SPH_T32(x) ((x) & SPH_C32(0xFFFFFFFF))
#define SPH_ROTL32(x, n) SPH_T32(((x) << (n)) | ((x) >> (32 - (n))))
#define SPH_ROTR32(x, n) SPH_ROTL32(x, (32 - (n)))
#if SPH_64
#define SPH_T64(x) ((x) & SPH_C64(0xFFFFFFFFFFFFFFFF))
#define SPH_ROTL64(x, n) SPH_T64(((x) << (n)) | ((x) >> (64 - (n))))
#define SPH_ROTR64(x, n) SPH_ROTL64(x, (64 - (n)))
#endif
#ifndef DOXYGEN_IGNORE
/*
* Define SPH_INLINE to be an "inline" qualifier, if available. We define
* some small macro-like functions which benefit greatly from being inlined.
*/
#if (defined __STDC__ && __STDC_VERSION__ >= 199901L) || defined __GNUC__
#define SPH_INLINE inline
#elif defined _MSC_VER
#define SPH_INLINE __inline
#else
#define SPH_INLINE
#endif
#endif
/*
* We define some macros which qualify the architecture. These macros
* may be explicit set externally (e.g. as compiler parameters). The
* code below sets those macros if they are not already defined.
*
* Most macros are boolean, thus evaluate to either zero or non-zero.
* The SPH_UPTR macro is special, in that it evaluates to a C type,
* or is not defined.
*
* SPH_UPTR if defined: unsigned type to cast pointers into
*
* SPH_UNALIGNED non-zero if unaligned accesses are efficient
* SPH_LITTLE_ENDIAN non-zero if architecture is known to be little-endian
* SPH_BIG_ENDIAN non-zero if architecture is known to be big-endian
* SPH_LITTLE_FAST non-zero if little-endian decoding is fast
* SPH_BIG_FAST non-zero if big-endian decoding is fast
*
* If SPH_UPTR is defined, then encoding and decoding of 32-bit and 64-bit
* values will try to be "smart". Either SPH_LITTLE_ENDIAN or SPH_BIG_ENDIAN
* _must_ be non-zero in those situations. The 32-bit and 64-bit types
* _must_ also have an exact width.
*
* SPH_SPARCV9_GCC_32 UltraSPARC-compatible with gcc, 32-bit mode
* SPH_SPARCV9_GCC_64 UltraSPARC-compatible with gcc, 64-bit mode
* SPH_SPARCV9_GCC UltraSPARC-compatible with gcc
* SPH_I386_GCC x86-compatible (32-bit) with gcc
* SPH_I386_MSVC x86-compatible (32-bit) with Microsoft Visual C
* SPH_AMD64_GCC x86-compatible (64-bit) with gcc
* SPH_AMD64_MSVC x86-compatible (64-bit) with Microsoft Visual C
* SPH_PPC32_GCC PowerPC, 32-bit, with gcc
* SPH_PPC64_GCC PowerPC, 64-bit, with gcc
*
* TODO: enhance automatic detection, for more architectures and compilers.
* Endianness is the most important. SPH_UNALIGNED and SPH_UPTR help with
* some very fast functions (e.g. MD4) when using unaligned input data.
* The CPU-specific-with-GCC macros are useful only for inline assembly,
* normally restrained to this header file.
*/
/*
* 32-bit x86, aka "i386 compatible".
*/
#if defined __i386__ || defined _M_IX86
#define SPH_DETECT_UNALIGNED 1
#define SPH_DETECT_LITTLE_ENDIAN 1
#define SPH_DETECT_UPTR sph_u32
#ifdef __GNUC__
#define SPH_DETECT_I386_GCC 1
#endif
#ifdef _MSC_VER
#define SPH_DETECT_I386_MSVC 1
#endif
/*
* 64-bit x86, hereafter known as "amd64".
*/
#elif defined __x86_64 || defined _M_X64
#define SPH_DETECT_UNALIGNED 1
#define SPH_DETECT_LITTLE_ENDIAN 1
#define SPH_DETECT_UPTR sph_u64
#ifdef __GNUC__
#define SPH_DETECT_AMD64_GCC 1
#endif
#ifdef _MSC_VER
#define SPH_DETECT_AMD64_MSVC 1
#endif
/*
* 64-bit Sparc architecture (implies v9).
*/
#elif ((defined __sparc__ || defined __sparc) && defined __arch64__) \
|| defined __sparcv9
#define SPH_DETECT_BIG_ENDIAN 1
#define SPH_DETECT_UPTR sph_u64
#ifdef __GNUC__
#define SPH_DETECT_SPARCV9_GCC_64 1
#define SPH_DETECT_LITTLE_FAST 1
#endif
/*
* 32-bit Sparc.
*/
#elif (defined __sparc__ || defined __sparc) \
&& !(defined __sparcv9 || defined __arch64__)
#define SPH_DETECT_BIG_ENDIAN 1
#define SPH_DETECT_UPTR sph_u32
#if defined __GNUC__ && defined __sparc_v9__
#define SPH_DETECT_SPARCV9_GCC_32 1
#define SPH_DETECT_LITTLE_FAST 1
#endif
/*
* ARM, little-endian.
*/
#elif defined __arm__ && __ARMEL__
#define SPH_DETECT_LITTLE_ENDIAN 1
/*
* MIPS, little-endian.
*/
#elif MIPSEL || _MIPSEL || __MIPSEL || __MIPSEL__
#define SPH_DETECT_LITTLE_ENDIAN 1
/*
* MIPS, big-endian.
*/
#elif MIPSEB || _MIPSEB || __MIPSEB || __MIPSEB__
#define SPH_DETECT_BIG_ENDIAN 1
/*
* PowerPC.
*/
#elif defined __powerpc__ || defined __POWERPC__ || defined __ppc__ \
|| defined _ARCH_PPC
/*
* Note: we do not declare cross-endian access to be "fast": even if
* using inline assembly, implementation should still assume that
* keeping the decoded word in a temporary is faster than decoding
* it again.
*/
#if defined __GNUC__
#if SPH_64_TRUE
#define SPH_DETECT_PPC64_GCC 1
#else
#define SPH_DETECT_PPC32_GCC 1
#endif
#endif
#if defined __BIG_ENDIAN__ || defined _BIG_ENDIAN
#define SPH_DETECT_BIG_ENDIAN 1
#elif defined __LITTLE_ENDIAN__ || defined _LITTLE_ENDIAN
#define SPH_DETECT_LITTLE_ENDIAN 1
#endif
/*
* Itanium, 64-bit.
*/
#elif defined __ia64 || defined __ia64__ \
|| defined __itanium__ || defined _M_IA64
#if defined __BIG_ENDIAN__ || defined _BIG_ENDIAN
#define SPH_DETECT_BIG_ENDIAN 1
#else
#define SPH_DETECT_LITTLE_ENDIAN 1
#endif
#if defined __LP64__ || defined _LP64
#define SPH_DETECT_UPTR sph_u64
#else
#define SPH_DETECT_UPTR sph_u32
#endif
#endif
#if defined SPH_DETECT_SPARCV9_GCC_32 || defined SPH_DETECT_SPARCV9_GCC_64
#define SPH_DETECT_SPARCV9_GCC 1
#endif
#if defined SPH_DETECT_UNALIGNED && !defined SPH_UNALIGNED
#define SPH_UNALIGNED SPH_DETECT_UNALIGNED
#endif
#if defined SPH_DETECT_UPTR && !defined SPH_UPTR
#define SPH_UPTR SPH_DETECT_UPTR
#endif
#if defined SPH_DETECT_LITTLE_ENDIAN && !defined SPH_LITTLE_ENDIAN
#define SPH_LITTLE_ENDIAN SPH_DETECT_LITTLE_ENDIAN
#endif
#if defined SPH_DETECT_BIG_ENDIAN && !defined SPH_BIG_ENDIAN
#define SPH_BIG_ENDIAN SPH_DETECT_BIG_ENDIAN
#endif
#if defined SPH_DETECT_LITTLE_FAST && !defined SPH_LITTLE_FAST
#define SPH_LITTLE_FAST SPH_DETECT_LITTLE_FAST
#endif
#if defined SPH_DETECT_BIG_FAST && !defined SPH_BIG_FAST
#define SPH_BIG_FAST SPH_DETECT_BIG_FAST
#endif
#if defined SPH_DETECT_SPARCV9_GCC_32 && !defined SPH_SPARCV9_GCC_32
#define SPH_SPARCV9_GCC_32 SPH_DETECT_SPARCV9_GCC_32
#endif
#if defined SPH_DETECT_SPARCV9_GCC_64 && !defined SPH_SPARCV9_GCC_64
#define SPH_SPARCV9_GCC_64 SPH_DETECT_SPARCV9_GCC_64
#endif
#if defined SPH_DETECT_SPARCV9_GCC && !defined SPH_SPARCV9_GCC
#define SPH_SPARCV9_GCC SPH_DETECT_SPARCV9_GCC
#endif
#if defined SPH_DETECT_I386_GCC && !defined SPH_I386_GCC
#define SPH_I386_GCC SPH_DETECT_I386_GCC
#endif
#if defined SPH_DETECT_I386_MSVC && !defined SPH_I386_MSVC
#define SPH_I386_MSVC SPH_DETECT_I386_MSVC
#endif
#if defined SPH_DETECT_AMD64_GCC && !defined SPH_AMD64_GCC
#define SPH_AMD64_GCC SPH_DETECT_AMD64_GCC
#endif
#if defined SPH_DETECT_AMD64_MSVC && !defined SPH_AMD64_MSVC
#define SPH_AMD64_MSVC SPH_DETECT_AMD64_MSVC
#endif
#if defined SPH_DETECT_PPC32_GCC && !defined SPH_PPC32_GCC
#define SPH_PPC32_GCC SPH_DETECT_PPC32_GCC
#endif
#if defined SPH_DETECT_PPC64_GCC && !defined SPH_PPC64_GCC
#define SPH_PPC64_GCC SPH_DETECT_PPC64_GCC
#endif
#if SPH_LITTLE_ENDIAN && !defined SPH_LITTLE_FAST
#define SPH_LITTLE_FAST 1
#endif
#if SPH_BIG_ENDIAN && !defined SPH_BIG_FAST
#define SPH_BIG_FAST 1
#endif
#if defined SPH_UPTR && !(SPH_LITTLE_ENDIAN || SPH_BIG_ENDIAN)
#error SPH_UPTR defined, but endianness is not known.
#endif
#if SPH_I386_GCC && !SPH_NO_ASM
/*
* On x86 32-bit, with gcc, we use the bswapl opcode to byte-swap 32-bit
* values.
*/
static SPH_INLINE sph_u32
sph_bswap32(sph_u32 x)
{
__asm__ __volatile__ ("bswapl %0" : "=r" (x) : "0" (x));
return x;
}
#if SPH_64
static SPH_INLINE sph_u64
sph_bswap64(sph_u64 x)
{
return ((sph_u64)sph_bswap32((sph_u32)x) << 32)
| (sph_u64)sph_bswap32((sph_u32)(x >> 32));
}
#endif
#elif SPH_AMD64_GCC && !SPH_NO_ASM
/*
* On x86 64-bit, with gcc, we use the bswapl opcode to byte-swap 32-bit
* and 64-bit values.
*/
static SPH_INLINE sph_u32
sph_bswap32(sph_u32 x)
{
__asm__ __volatile__ ("bswapl %0" : "=r" (x) : "0" (x));
return x;
}
#if SPH_64
static SPH_INLINE sph_u64
sph_bswap64(sph_u64 x)
{
__asm__ __volatile__ ("bswapq %0" : "=r" (x) : "0" (x));
return x;
}
#endif
/*
* Disabled code. Apparently, Microsoft Visual C 2005 is smart enough
* to generate proper opcodes for endianness swapping with the pure C
* implementation below.
*
#elif SPH_I386_MSVC && !SPH_NO_ASM
static __inline sph_u32 __declspec(naked) __fastcall
sph_bswap32(sph_u32 x)
{
__asm {
bswap ecx
mov eax,ecx
ret
}
}
#if SPH_64
static SPH_INLINE sph_u64
sph_bswap64(sph_u64 x)
{
return ((sph_u64)sph_bswap32((sph_u32)x) << 32)
| (sph_u64)sph_bswap32((sph_u32)(x >> 32));
}
#endif
*
* [end of disabled code]
*/
#else
static SPH_INLINE sph_u32
sph_bswap32(sph_u32 x)
{
x = SPH_T32((x << 16) | (x >> 16));
x = ((x & SPH_C32(0xFF00FF00)) >> 8)
| ((x & SPH_C32(0x00FF00FF)) << 8);
return x;
}
#if SPH_64
/**
* Byte-swap a 64-bit value.
*
* @param x the input value
* @return the byte-swapped value
*/
static SPH_INLINE sph_u64
sph_bswap64(sph_u64 x)
{
x = SPH_T64((x << 32) | (x >> 32));
x = ((x & SPH_C64(0xFFFF0000FFFF0000)) >> 16)
| ((x & SPH_C64(0x0000FFFF0000FFFF)) << 16);
x = ((x & SPH_C64(0xFF00FF00FF00FF00)) >> 8)
| ((x & SPH_C64(0x00FF00FF00FF00FF)) << 8);
return x;
}
#endif
#endif
#if SPH_SPARCV9_GCC && !SPH_NO_ASM
/*
* On UltraSPARC systems, native ordering is big-endian, but it is
* possible to perform little-endian read accesses by specifying the
* address space 0x88 (ASI_PRIMARY_LITTLE). Basically, either we use
* the opcode "lda [%reg]0x88,%dst", where %reg is the register which
* contains the source address and %dst is the destination register,
* or we use "lda [%reg+imm]%asi,%dst", which uses the %asi register
* to get the address space name. The latter format is better since it
* combines an addition and the actual access in a single opcode; but
* it requires the setting (and subsequent resetting) of %asi, which is
* slow. Some operations (i.e. MD5 compression function) combine many
* successive little-endian read accesses, which may share the same
* %asi setting. The macros below contain the appropriate inline
* assembly.
*/
#define SPH_SPARCV9_SET_ASI \
sph_u32 sph_sparcv9_asi; \
__asm__ __volatile__ ( \
"rd %%asi,%0\n\twr %%g0,0x88,%%asi" : "=r" (sph_sparcv9_asi));
#define SPH_SPARCV9_RESET_ASI \
__asm__ __volatile__ ("wr %%g0,%0,%%asi" : : "r" (sph_sparcv9_asi));
#define SPH_SPARCV9_DEC32LE(base, idx) ({ \
sph_u32 sph_sparcv9_tmp; \
__asm__ __volatile__ ("lda [%1+" #idx "*4]%%asi,%0" \
: "=r" (sph_sparcv9_tmp) : "r" (base)); \
sph_sparcv9_tmp; \
})
#endif
static SPH_INLINE void
sph_enc16be(void *dst, unsigned val)
{
((unsigned char *)dst)[0] = (val >> 8);
((unsigned char *)dst)[1] = val;
}
static SPH_INLINE unsigned
sph_dec16be(const void *src)
{
return ((unsigned)(((const unsigned char *)src)[0]) << 8)
| (unsigned)(((const unsigned char *)src)[1]);
}
static SPH_INLINE void
sph_enc16le(void *dst, unsigned val)
{
((unsigned char *)dst)[0] = val;
((unsigned char *)dst)[1] = val >> 8;
}
static SPH_INLINE unsigned
sph_dec16le(const void *src)
{
return (unsigned)(((const unsigned char *)src)[0])
| ((unsigned)(((const unsigned char *)src)[1]) << 8);
}
/**
* Encode a 32-bit value into the provided buffer (big endian convention).
*
* @param dst the destination buffer
* @param val the 32-bit value to encode
*/
static SPH_INLINE void
sph_enc32be(void *dst, sph_u32 val)
{
#if defined SPH_UPTR
#if SPH_UNALIGNED
#if SPH_LITTLE_ENDIAN
val = sph_bswap32(val);
#endif
*(sph_u32 *)dst = val;
#else
if (((SPH_UPTR)dst & 3) == 0) {
#if SPH_LITTLE_ENDIAN
val = sph_bswap32(val);
#endif
*(sph_u32 *)dst = val;
} else {
((unsigned char *)dst)[0] = (val >> 24);
((unsigned char *)dst)[1] = (val >> 16);
((unsigned char *)dst)[2] = (val >> 8);
((unsigned char *)dst)[3] = val;
}
#endif
#else
((unsigned char *)dst)[0] = (val >> 24);
((unsigned char *)dst)[1] = (val >> 16);
((unsigned char *)dst)[2] = (val >> 8);
((unsigned char *)dst)[3] = val;
#endif
}
/**
* Encode a 32-bit value into the provided buffer (big endian convention).
* The destination buffer must be properly aligned.
*
* @param dst the destination buffer (32-bit aligned)
* @param val the value to encode
*/
static SPH_INLINE void
sph_enc32be_aligned(void *dst, sph_u32 val)
{
#if SPH_LITTLE_ENDIAN
*(sph_u32 *)dst = sph_bswap32(val);
#elif SPH_BIG_ENDIAN
*(sph_u32 *)dst = val;
#else
((unsigned char *)dst)[0] = (val >> 24);
((unsigned char *)dst)[1] = (val >> 16);
((unsigned char *)dst)[2] = (val >> 8);
((unsigned char *)dst)[3] = val;
#endif
}
/**
* Decode a 32-bit value from the provided buffer (big endian convention).
*
* @param src the source buffer
* @return the decoded value
*/
static SPH_INLINE sph_u32
sph_dec32be(const void *src)
{
#if defined SPH_UPTR
#if SPH_UNALIGNED
#if SPH_LITTLE_ENDIAN
return sph_bswap32(*(const sph_u32 *)src);
#else
return *(const sph_u32 *)src;
#endif
#else
if (((SPH_UPTR)src & 3) == 0) {
#if SPH_LITTLE_ENDIAN
return sph_bswap32(*(const sph_u32 *)src);
#else
return *(const sph_u32 *)src;
#endif
} else {
return ((sph_u32)(((const unsigned char *)src)[0]) << 24)
| ((sph_u32)(((const unsigned char *)src)[1]) << 16)
| ((sph_u32)(((const unsigned char *)src)[2]) << 8)
| (sph_u32)(((const unsigned char *)src)[3]);
}
#endif
#else
return ((sph_u32)(((const unsigned char *)src)[0]) << 24)
| ((sph_u32)(((const unsigned char *)src)[1]) << 16)
| ((sph_u32)(((const unsigned char *)src)[2]) << 8)
| (sph_u32)(((const unsigned char *)src)[3]);
#endif
}
/**
* Decode a 32-bit value from the provided buffer (big endian convention).
* The source buffer must be properly aligned.
*
* @param src the source buffer (32-bit aligned)
* @return the decoded value
*/
static SPH_INLINE sph_u32
sph_dec32be_aligned(const void *src)
{
#if SPH_LITTLE_ENDIAN
return sph_bswap32(*(const sph_u32 *)src);
#elif SPH_BIG_ENDIAN
return *(const sph_u32 *)src;
#else
return ((sph_u32)(((const unsigned char *)src)[0]) << 24)
| ((sph_u32)(((const unsigned char *)src)[1]) << 16)
| ((sph_u32)(((const unsigned char *)src)[2]) << 8)
| (sph_u32)(((const unsigned char *)src)[3]);
#endif
}
/**
* Encode a 32-bit value into the provided buffer (little endian convention).
*
* @param dst the destination buffer
* @param val the 32-bit value to encode
*/
static SPH_INLINE void
sph_enc32le(void *dst, sph_u32 val)
{
#if defined SPH_UPTR
#if SPH_UNALIGNED
#if SPH_BIG_ENDIAN
val = sph_bswap32(val);
#endif
*(sph_u32 *)dst = val;
#else
if (((SPH_UPTR)dst & 3) == 0) {
#if SPH_BIG_ENDIAN
val = sph_bswap32(val);
#endif
*(sph_u32 *)dst = val;
} else {
((unsigned char *)dst)[0] = val;
((unsigned char *)dst)[1] = (val >> 8);
((unsigned char *)dst)[2] = (val >> 16);
((unsigned char *)dst)[3] = (val >> 24);
}
#endif
#else
((unsigned char *)dst)[0] = val;
((unsigned char *)dst)[1] = (val >> 8);
((unsigned char *)dst)[2] = (val >> 16);
((unsigned char *)dst)[3] = (val >> 24);
#endif
}
/**
* Encode a 32-bit value into the provided buffer (little endian convention).
* The destination buffer must be properly aligned.
*
* @param dst the destination buffer (32-bit aligned)
* @param val the value to encode
*/
static SPH_INLINE void
sph_enc32le_aligned(void *dst, sph_u32 val)
{
#if SPH_LITTLE_ENDIAN
*(sph_u32 *)dst = val;
#elif SPH_BIG_ENDIAN
*(sph_u32 *)dst = sph_bswap32(val);
#else
((unsigned char *)dst)[0] = val;
((unsigned char *)dst)[1] = (val >> 8);
((unsigned char *)dst)[2] = (val >> 16);
((unsigned char *)dst)[3] = (val >> 24);
#endif
}
/**
* Decode a 32-bit value from the provided buffer (little endian convention).
*
* @param src the source buffer
* @return the decoded value
*/
static SPH_INLINE sph_u32
sph_dec32le(const void *src)
{
#if defined SPH_UPTR
#if SPH_UNALIGNED
#if SPH_BIG_ENDIAN
return sph_bswap32(*(const sph_u32 *)src);
#else
return *(const sph_u32 *)src;
#endif
#else
if (((SPH_UPTR)src & 3) == 0) {
#if SPH_BIG_ENDIAN
#if SPH_SPARCV9_GCC && !SPH_NO_ASM
sph_u32 tmp;
/*
* "__volatile__" is needed here because without it,
* gcc-3.4.3 miscompiles the code and performs the
* access before the test on the address, thus triggering
* a bus error...
*/
__asm__ __volatile__ (
"lda [%1]0x88,%0" : "=r" (tmp) : "r" (src));
return tmp;
/*
* On PowerPC, this turns out not to be worth the effort: the inline
* assembly makes GCC optimizer uncomfortable, which tends to nullify
* the decoding gains.
*
* For most hash functions, using this inline assembly trick changes
* hashing speed by less than 5% and often _reduces_ it. The biggest
* gains are for MD4 (+11%) and CubeHash (+30%). For all others, it is
* less then 10%. The speed gain on CubeHash is probably due to the
* chronic shortage of registers that CubeHash endures; for the other
* functions, the generic code appears to be efficient enough already.
*
#elif (SPH_PPC32_GCC || SPH_PPC64_GCC) && !SPH_NO_ASM
sph_u32 tmp;
__asm__ __volatile__ (
"lwbrx %0,0,%1" : "=r" (tmp) : "r" (src));
return tmp;
*/
#else
return sph_bswap32(*(const sph_u32 *)src);
#endif
#else
return *(const sph_u32 *)src;
#endif
} else {
return (sph_u32)(((const unsigned char *)src)[0])
| ((sph_u32)(((const unsigned char *)src)[1]) << 8)
| ((sph_u32)(((const unsigned char *)src)[2]) << 16)
| ((sph_u32)(((const unsigned char *)src)[3]) << 24);
}
#endif
#else
return (sph_u32)(((const unsigned char *)src)[0])
| ((sph_u32)(((const unsigned char *)src)[1]) << 8)
| ((sph_u32)(((const unsigned char *)src)[2]) << 16)
| ((sph_u32)(((const unsigned char *)src)[3]) << 24);
#endif
}
/**
* Decode a 32-bit value from the provided buffer (little endian convention).
* The source buffer must be properly aligned.
*
* @param src the source buffer (32-bit aligned)
* @return the decoded value
*/
static SPH_INLINE sph_u32
sph_dec32le_aligned(const void *src)
{
#if SPH_LITTLE_ENDIAN
return *(const sph_u32 *)src;
#elif SPH_BIG_ENDIAN
#if SPH_SPARCV9_GCC && !SPH_NO_ASM
sph_u32 tmp;
__asm__ __volatile__ ("lda [%1]0x88,%0" : "=r" (tmp) : "r" (src));
return tmp;
/*
* Not worth it generally.
*
#elif (SPH_PPC32_GCC || SPH_PPC64_GCC) && !SPH_NO_ASM
sph_u32 tmp;
__asm__ __volatile__ ("lwbrx %0,0,%1" : "=r" (tmp) : "r" (src));
return tmp;
*/
#else
return sph_bswap32(*(const sph_u32 *)src);
#endif
#else
return (sph_u32)(((const unsigned char *)src)[0])
| ((sph_u32)(((const unsigned char *)src)[1]) << 8)
| ((sph_u32)(((const unsigned char *)src)[2]) << 16)
| ((sph_u32)(((const unsigned char *)src)[3]) << 24);
#endif
}
#if SPH_64
/**
* Encode a 64-bit value into the provided buffer (big endian convention).
*
* @param dst the destination buffer
* @param val the 64-bit value to encode
*/
static SPH_INLINE void
sph_enc64be(void *dst, sph_u64 val)
{
#if defined SPH_UPTR
#if SPH_UNALIGNED
#if SPH_LITTLE_ENDIAN
val = sph_bswap64(val);
#endif
*(sph_u64 *)dst = val;
#else
if (((SPH_UPTR)dst & 7) == 0) {
#if SPH_LITTLE_ENDIAN
val = sph_bswap64(val);
#endif
*(sph_u64 *)dst = val;
} else {
((unsigned char *)dst)[0] = (val >> 56);
((unsigned char *)dst)[1] = (val >> 48);
((unsigned char *)dst)[2] = (val >> 40);
((unsigned char *)dst)[3] = (val >> 32);
((unsigned char *)dst)[4] = (val >> 24);
((unsigned char *)dst)[5] = (val >> 16);
((unsigned char *)dst)[6] = (val >> 8);
((unsigned char *)dst)[7] = val;
}
#endif
#else
((unsigned char *)dst)[0] = (val >> 56);
((unsigned char *)dst)[1] = (val >> 48);
((unsigned char *)dst)[2] = (val >> 40);
((unsigned char *)dst)[3] = (val >> 32);
((unsigned char *)dst)[4] = (val >> 24);
((unsigned char *)dst)[5] = (val >> 16);
((unsigned char *)dst)[6] = (val >> 8);
((unsigned char *)dst)[7] = val;
#endif
}
/**
* Encode a 64-bit value into the provided buffer (big endian convention).
* The destination buffer must be properly aligned.
*
* @param dst the destination buffer (64-bit aligned)
* @param val the value to encode
*/
static SPH_INLINE void
sph_enc64be_aligned(void *dst, sph_u64 val)
{
#if SPH_LITTLE_ENDIAN
*(sph_u64 *)dst = sph_bswap64(val);
#elif SPH_BIG_ENDIAN
*(sph_u64 *)dst = val;
#else
((unsigned char *)dst)[0] = (val >> 56);
((unsigned char *)dst)[1] = (val >> 48);
((unsigned char *)dst)[2] = (val >> 40);
((unsigned char *)dst)[3] = (val >> 32);
((unsigned char *)dst)[4] = (val >> 24);
((unsigned char *)dst)[5] = (val >> 16);
((unsigned char *)dst)[6] = (val >> 8);
((unsigned char *)dst)[7] = val;
#endif
}
/**
* Decode a 64-bit value from the provided buffer (big endian convention).
*
* @param src the source buffer
* @return the decoded value
*/
static SPH_INLINE sph_u64
sph_dec64be(const void *src)
{
#if defined SPH_UPTR
#if SPH_UNALIGNED
#if SPH_LITTLE_ENDIAN
return sph_bswap64(*(const sph_u64 *)src);
#else
return *(const sph_u64 *)src;
#endif
#else
if (((SPH_UPTR)src & 7) == 0) {
#if SPH_LITTLE_ENDIAN
return sph_bswap64(*(const sph_u64 *)src);
#else
return *(const sph_u64 *)src;
#endif
} else {
return ((sph_u64)(((const unsigned char *)src)[0]) << 56)
| ((sph_u64)(((const unsigned char *)src)[1]) << 48)
| ((sph_u64)(((const unsigned char *)src)[2]) << 40)
| ((sph_u64)(((const unsigned char *)src)[3]) << 32)
| ((sph_u64)(((const unsigned char *)src)[4]) << 24)
| ((sph_u64)(((const unsigned char *)src)[5]) << 16)
| ((sph_u64)(((const unsigned char *)src)[6]) << 8)
| (sph_u64)(((const unsigned char *)src)[7]);
}
#endif
#else
return ((sph_u64)(((const unsigned char *)src)[0]) << 56)
| ((sph_u64)(((const unsigned char *)src)[1]) << 48)
| ((sph_u64)(((const unsigned char *)src)[2]) << 40)
| ((sph_u64)(((const unsigned char *)src)[3]) << 32)
| ((sph_u64)(((const unsigned char *)src)[4]) << 24)
| ((sph_u64)(((const unsigned char *)src)[5]) << 16)
| ((sph_u64)(((const unsigned char *)src)[6]) << 8)
| (sph_u64)(((const unsigned char *)src)[7]);
#endif
}
/**
* Decode a 64-bit value from the provided buffer (big endian convention).
* The source buffer must be properly aligned.
*
* @param src the source buffer (64-bit aligned)
* @return the decoded value
*/
static SPH_INLINE sph_u64
sph_dec64be_aligned(const void *src)
{
#if SPH_LITTLE_ENDIAN
return sph_bswap64(*(const sph_u64 *)src);
#elif SPH_BIG_ENDIAN
return *(const sph_u64 *)src;
#else
return ((sph_u64)(((const unsigned char *)src)[0]) << 56)
| ((sph_u64)(((const unsigned char *)src)[1]) << 48)
| ((sph_u64)(((const unsigned char *)src)[2]) << 40)
| ((sph_u64)(((const unsigned char *)src)[3]) << 32)
| ((sph_u64)(((const unsigned char *)src)[4]) << 24)
| ((sph_u64)(((const unsigned char *)src)[5]) << 16)
| ((sph_u64)(((const unsigned char *)src)[6]) << 8)
| (sph_u64)(((const unsigned char *)src)[7]);
#endif
}
/**
* Encode a 64-bit value into the provided buffer (little endian convention).
*
* @param dst the destination buffer
* @param val the 64-bit value to encode
*/
static SPH_INLINE void
sph_enc64le(void *dst, sph_u64 val)
{
#if defined SPH_UPTR
#if SPH_UNALIGNED
#if SPH_BIG_ENDIAN
val = sph_bswap64(val);
#endif
*(sph_u64 *)dst = val;
#else
if (((SPH_UPTR)dst & 7) == 0) {
#if SPH_BIG_ENDIAN
val = sph_bswap64(val);
#endif
*(sph_u64 *)dst = val;
} else {
((unsigned char *)dst)[0] = val;
((unsigned char *)dst)[1] = (val >> 8);
((unsigned char *)dst)[2] = (val >> 16);
((unsigned char *)dst)[3] = (val >> 24);
((unsigned char *)dst)[4] = (val >> 32);
((unsigned char *)dst)[5] = (val >> 40);
((unsigned char *)dst)[6] = (val >> 48);
((unsigned char *)dst)[7] = (val >> 56);
}
#endif
#else
((unsigned char *)dst)[0] = val;
((unsigned char *)dst)[1] = (val >> 8);
((unsigned char *)dst)[2] = (val >> 16);
((unsigned char *)dst)[3] = (val >> 24);
((unsigned char *)dst)[4] = (val >> 32);
((unsigned char *)dst)[5] = (val >> 40);
((unsigned char *)dst)[6] = (val >> 48);
((unsigned char *)dst)[7] = (val >> 56);
#endif
}
/**
* Encode a 64-bit value into the provided buffer (little endian convention).
* The destination buffer must be properly aligned.
*
* @param dst the destination buffer (64-bit aligned)
* @param val the value to encode
*/
static SPH_INLINE void
sph_enc64le_aligned(void *dst, sph_u64 val)
{
#if SPH_LITTLE_ENDIAN
*(sph_u64 *)dst = val;
#elif SPH_BIG_ENDIAN
*(sph_u64 *)dst = sph_bswap64(val);
#else
((unsigned char *)dst)[0] = val;
((unsigned char *)dst)[1] = (val >> 8);
((unsigned char *)dst)[2] = (val >> 16);
((unsigned char *)dst)[3] = (val >> 24);
((unsigned char *)dst)[4] = (val >> 32);
((unsigned char *)dst)[5] = (val >> 40);
((unsigned char *)dst)[6] = (val >> 48);
((unsigned char *)dst)[7] = (val >> 56);
#endif
}
/**
* Decode a 64-bit value from the provided buffer (little endian convention).
*
* @param src the source buffer
* @return the decoded value
*/
static SPH_INLINE sph_u64
sph_dec64le(const void *src)
{
#if defined SPH_UPTR
#if SPH_UNALIGNED
#if SPH_BIG_ENDIAN
return sph_bswap64(*(const sph_u64 *)src);
#else
return *(const sph_u64 *)src;
#endif
#else
if (((SPH_UPTR)src & 7) == 0) {
#if SPH_BIG_ENDIAN
#if SPH_SPARCV9_GCC_64 && !SPH_NO_ASM
sph_u64 tmp;
__asm__ __volatile__ (
"ldxa [%1]0x88,%0" : "=r" (tmp) : "r" (src));
return tmp;
/*
* Not worth it generally.
*
#elif SPH_PPC32_GCC && !SPH_NO_ASM
return (sph_u64)sph_dec32le_aligned(src)
| ((sph_u64)sph_dec32le_aligned(
(const char *)src + 4) << 32);
#elif SPH_PPC64_GCC && !SPH_NO_ASM
sph_u64 tmp;
__asm__ __volatile__ (
"ldbrx %0,0,%1" : "=r" (tmp) : "r" (src));
return tmp;
*/
#else
return sph_bswap64(*(const sph_u64 *)src);
#endif
#else
return *(const sph_u64 *)src;
#endif
} else {
return (sph_u64)(((const unsigned char *)src)[0])
| ((sph_u64)(((const unsigned char *)src)[1]) << 8)
| ((sph_u64)(((const unsigned char *)src)[2]) << 16)
| ((sph_u64)(((const unsigned char *)src)[3]) << 24)
| ((sph_u64)(((const unsigned char *)src)[4]) << 32)
| ((sph_u64)(((const unsigned char *)src)[5]) << 40)
| ((sph_u64)(((const unsigned char *)src)[6]) << 48)
| ((sph_u64)(((const unsigned char *)src)[7]) << 56);
}
#endif
#else
return (sph_u64)(((const unsigned char *)src)[0])
| ((sph_u64)(((const unsigned char *)src)[1]) << 8)
| ((sph_u64)(((const unsigned char *)src)[2]) << 16)
| ((sph_u64)(((const unsigned char *)src)[3]) << 24)
| ((sph_u64)(((const unsigned char *)src)[4]) << 32)
| ((sph_u64)(((const unsigned char *)src)[5]) << 40)
| ((sph_u64)(((const unsigned char *)src)[6]) << 48)
| ((sph_u64)(((const unsigned char *)src)[7]) << 56);
#endif
}
/**
* Decode a 64-bit value from the provided buffer (little endian convention).
* The source buffer must be properly aligned.
*
* @param src the source buffer (64-bit aligned)
* @return the decoded value
*/
static SPH_INLINE sph_u64
sph_dec64le_aligned(const void *src)
{
#if SPH_LITTLE_ENDIAN
return *(const sph_u64 *)src;
#elif SPH_BIG_ENDIAN
#if SPH_SPARCV9_GCC_64 && !SPH_NO_ASM
sph_u64 tmp;
__asm__ __volatile__ ("ldxa [%1]0x88,%0" : "=r" (tmp) : "r" (src));
return tmp;
/*
* Not worth it generally.
*
#elif SPH_PPC32_GCC && !SPH_NO_ASM
return (sph_u64)sph_dec32le_aligned(src)
| ((sph_u64)sph_dec32le_aligned((const char *)src + 4) << 32);
#elif SPH_PPC64_GCC && !SPH_NO_ASM
sph_u64 tmp;
__asm__ __volatile__ ("ldbrx %0,0,%1" : "=r" (tmp) : "r" (src));
return tmp;
*/
#else
return sph_bswap64(*(const sph_u64 *)src);
#endif
#else
return (sph_u64)(((const unsigned char *)src)[0])
| ((sph_u64)(((const unsigned char *)src)[1]) << 8)
| ((sph_u64)(((const unsigned char *)src)[2]) << 16)
| ((sph_u64)(((const unsigned char *)src)[3]) << 24)
| ((sph_u64)(((const unsigned char *)src)[4]) << 32)
| ((sph_u64)(((const unsigned char *)src)[5]) << 40)
| ((sph_u64)(((const unsigned char *)src)[6]) << 48)
| ((sph_u64)(((const unsigned char *)src)[7]) << 56);
#endif
}
#endif
#endif /* Doxygen excluded block */
#endif