Chris@69: /* Copyright (c) 2003-2008 Timothy B. Terriberry
Chris@69:    Copyright (c) 2008 Xiph.Org Foundation */
Chris@69: /*
Chris@69:    Redistribution and use in source and binary forms, with or without
Chris@69:    modification, are permitted provided that the following conditions
Chris@69:    are met:
Chris@69: 
Chris@69:    - Redistributions of source code must retain the above copyright
Chris@69:    notice, this list of conditions and the following disclaimer.
Chris@69: 
Chris@69:    - Redistributions in binary form must reproduce the above copyright
Chris@69:    notice, this list of conditions and the following disclaimer in the
Chris@69:    documentation and/or other materials provided with the distribution.
Chris@69: 
Chris@69:    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
Chris@69:    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
Chris@69:    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
Chris@69:    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
Chris@69:    OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
Chris@69:    EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
Chris@69:    PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
Chris@69:    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
Chris@69:    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
Chris@69:    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
Chris@69:    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
Chris@69: */
Chris@69: 
Chris@69: /*Some common macros for potential platform-specific optimization.*/
Chris@69: #include "opus_types.h"
Chris@69: #include <math.h>
Chris@69: #include <limits.h>
Chris@69: #include "arch.h"
Chris@69: #if !defined(_ecintrin_H)
Chris@69: # define _ecintrin_H (1)
Chris@69: 
Chris@69: /*Some specific platforms may have optimized intrinsic or OPUS_INLINE assembly
Chris@69:    versions of these functions which can substantially improve performance.
Chris@69:   We define macros for them to allow easy incorporation of these non-ANSI
Chris@69:    features.*/
Chris@69: 
Chris@69: /*Modern gcc (4.x) can compile the naive versions of min and max with cmov if
Chris@69:    given an appropriate architecture, but the branchless bit-twiddling versions
Chris@69:    are just as fast, and do not require any special target architecture.
Chris@69:   Earlier gcc versions (3.x) compiled both code to the same assembly
Chris@69:    instructions, because of the way they represented ((_b)>(_a)) internally.*/
Chris@69: # define EC_MINI(_a,_b)      ((_a)+(((_b)-(_a))&-((_b)<(_a))))
Chris@69: 
Chris@69: /*Count leading zeros.
Chris@69:   This macro should only be used for implementing ec_ilog(), if it is defined.
Chris@69:   All other code should use EC_ILOG() instead.*/
Chris@69: #if defined(_MSC_VER) && (_MSC_VER >= 1400)
Chris@69: # include <intrin.h>
Chris@69: /*In _DEBUG mode this is not an intrinsic by default.*/
Chris@69: # pragma intrinsic(_BitScanReverse)
Chris@69: 
Chris@69: static __inline int ec_bsr(unsigned long _x){
Chris@69:   unsigned long ret;
Chris@69:   _BitScanReverse(&ret,_x);
Chris@69:   return (int)ret;
Chris@69: }
Chris@69: # define EC_CLZ0    (1)
Chris@69: # define EC_CLZ(_x) (-ec_bsr(_x))
Chris@69: #elif defined(ENABLE_TI_DSPLIB)
Chris@69: # include "dsplib.h"
Chris@69: # define EC_CLZ0    (31)
Chris@69: # define EC_CLZ(_x) (_lnorm(_x))
Chris@69: #elif __GNUC_PREREQ(3,4)
Chris@69: # if INT_MAX>=2147483647
Chris@69: #  define EC_CLZ0    ((int)sizeof(unsigned)*CHAR_BIT)
Chris@69: #  define EC_CLZ(_x) (__builtin_clz(_x))
Chris@69: # elif LONG_MAX>=2147483647L
Chris@69: #  define EC_CLZ0    ((int)sizeof(unsigned long)*CHAR_BIT)
Chris@69: #  define EC_CLZ(_x) (__builtin_clzl(_x))
Chris@69: # endif
Chris@69: #endif
Chris@69: 
Chris@69: #if defined(EC_CLZ)
Chris@69: /*Note that __builtin_clz is not defined when _x==0, according to the gcc
Chris@69:    documentation (and that of the BSR instruction that implements it on x86).
Chris@69:   The majority of the time we can never pass it zero.
Chris@69:   When we need to, it can be special cased.*/
Chris@69: # define EC_ILOG(_x) (EC_CLZ0-EC_CLZ(_x))
Chris@69: #else
Chris@69: int ec_ilog(opus_uint32 _v);
Chris@69: # define EC_ILOG(_x) (ec_ilog(_x))
Chris@69: #endif
Chris@69: #endif