cannam@154: /* Copyright (c) 2003-2008 Jean-Marc Valin
cannam@154:    Copyright (c) 2007-2008 CSIRO
cannam@154:    Copyright (c) 2007-2009 Xiph.Org Foundation
cannam@154:    Written by Jean-Marc Valin */
cannam@154: /**
cannam@154:    @file arch.h
cannam@154:    @brief Various architecture definitions for CELT
cannam@154: */
cannam@154: /*
cannam@154:    Redistribution and use in source and binary forms, with or without
cannam@154:    modification, are permitted provided that the following conditions
cannam@154:    are met:
cannam@154: 
cannam@154:    - Redistributions of source code must retain the above copyright
cannam@154:    notice, this list of conditions and the following disclaimer.
cannam@154: 
cannam@154:    - Redistributions in binary form must reproduce the above copyright
cannam@154:    notice, this list of conditions and the following disclaimer in the
cannam@154:    documentation and/or other materials provided with the distribution.
cannam@154: 
cannam@154:    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
cannam@154:    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
cannam@154:    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
cannam@154:    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
cannam@154:    OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
cannam@154:    EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
cannam@154:    PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
cannam@154:    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
cannam@154:    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
cannam@154:    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
cannam@154:    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
cannam@154: */
cannam@154: 
cannam@154: #ifndef ARCH_H
cannam@154: #define ARCH_H
cannam@154: 
cannam@154: #include "opus_types.h"
cannam@154: #include "opus_defines.h"
cannam@154: 
cannam@154: # if !defined(__GNUC_PREREQ)
cannam@154: #  if defined(__GNUC__)&&defined(__GNUC_MINOR__)
cannam@154: #   define __GNUC_PREREQ(_maj,_min) \
cannam@154:  ((__GNUC__<<16)+__GNUC_MINOR__>=((_maj)<<16)+(_min))
cannam@154: #  else
cannam@154: #   define __GNUC_PREREQ(_maj,_min) 0
cannam@154: #  endif
cannam@154: # endif
cannam@154: 
cannam@154: #if OPUS_GNUC_PREREQ(3, 0)
cannam@154: #define opus_likely(x)       (__builtin_expect(!!(x), 1))
cannam@154: #define opus_unlikely(x)     (__builtin_expect(!!(x), 0))
cannam@154: #else
cannam@154: #define opus_likely(x)       (!!(x))
cannam@154: #define opus_unlikely(x)     (!!(x))
cannam@154: #endif
cannam@154: 
cannam@154: #define CELT_SIG_SCALE 32768.f
cannam@154: 
cannam@154: #define CELT_FATAL(str) celt_fatal(str, __FILE__, __LINE__);
cannam@154: 
cannam@154: #if defined(ENABLE_ASSERTIONS) || defined(ENABLE_HARDENING)
cannam@154: #ifdef __GNUC__
cannam@154: __attribute__((noreturn))
cannam@154: #endif
cannam@154: void celt_fatal(const char *str, const char *file, int line);
cannam@154: 
cannam@154: #if defined(CELT_C) && !defined(OVERRIDE_celt_fatal)
cannam@154: #include <stdio.h>
cannam@154: #include <stdlib.h>
cannam@154: #ifdef __GNUC__
cannam@154: __attribute__((noreturn))
cannam@154: #endif
cannam@154: void celt_fatal(const char *str, const char *file, int line)
cannam@154: {
cannam@154:    fprintf (stderr, "Fatal (internal) error in %s, line %d: %s\n", file, line, str);
cannam@154:    abort();
cannam@154: }
cannam@154: #endif
cannam@154: 
cannam@154: #define celt_assert(cond) {if (!(cond)) {CELT_FATAL("assertion failed: " #cond);}}
cannam@154: #define celt_assert2(cond, message) {if (!(cond)) {CELT_FATAL("assertion failed: " #cond "\n" message);}}
cannam@154: #define MUST_SUCCEED(call) celt_assert((call) == OPUS_OK)
cannam@154: #else
cannam@154: #define celt_assert(cond)
cannam@154: #define celt_assert2(cond, message)
cannam@154: #define MUST_SUCCEED(call) do {if((call) != OPUS_OK) {RESTORE_STACK; return OPUS_INTERNAL_ERROR;} } while (0)
cannam@154: #endif
cannam@154: 
cannam@154: #if defined(ENABLE_ASSERTIONS)
cannam@154: #define celt_sig_assert(cond) {if (!(cond)) {CELT_FATAL("signal assertion failed: " #cond);}}
cannam@154: #else
cannam@154: #define celt_sig_assert(cond)
cannam@154: #endif
cannam@154: 
cannam@154: #define IMUL32(a,b) ((a)*(b))
cannam@154: 
cannam@154: #define MIN16(a,b) ((a) < (b) ? (a) : (b))   /**< Minimum 16-bit value.   */
cannam@154: #define MAX16(a,b) ((a) > (b) ? (a) : (b))   /**< Maximum 16-bit value.   */
cannam@154: #define MIN32(a,b) ((a) < (b) ? (a) : (b))   /**< Minimum 32-bit value.   */
cannam@154: #define MAX32(a,b) ((a) > (b) ? (a) : (b))   /**< Maximum 32-bit value.   */
cannam@154: #define IMIN(a,b) ((a) < (b) ? (a) : (b))   /**< Minimum int value.   */
cannam@154: #define IMAX(a,b) ((a) > (b) ? (a) : (b))   /**< Maximum int value.   */
cannam@154: #define UADD32(a,b) ((a)+(b))
cannam@154: #define USUB32(a,b) ((a)-(b))
cannam@154: 
cannam@154: /* Set this if opus_int64 is a native type of the CPU. */
cannam@154: /* Assume that all LP64 architectures have fast 64-bit types; also x86_64
cannam@154:    (which can be ILP32 for x32) and Win64 (which is LLP64). */
cannam@154: #if defined(__x86_64__) || defined(__LP64__) || defined(_WIN64)
cannam@154: #define OPUS_FAST_INT64 1
cannam@154: #else
cannam@154: #define OPUS_FAST_INT64 0
cannam@154: #endif
cannam@154: 
cannam@154: #define PRINT_MIPS(file)
cannam@154: 
cannam@154: #ifdef FIXED_POINT
cannam@154: 
cannam@154: typedef opus_int16 opus_val16;
cannam@154: typedef opus_int32 opus_val32;
cannam@154: typedef opus_int64 opus_val64;
cannam@154: 
cannam@154: typedef opus_val32 celt_sig;
cannam@154: typedef opus_val16 celt_norm;
cannam@154: typedef opus_val32 celt_ener;
cannam@154: 
cannam@154: #define celt_isnan(x) 0
cannam@154: 
cannam@154: #define Q15ONE 32767
cannam@154: 
cannam@154: #define SIG_SHIFT 12
cannam@154: /* Safe saturation value for 32-bit signals. Should be less than
cannam@154:    2^31*(1-0.85) to avoid blowing up on DC at deemphasis.*/
cannam@154: #define SIG_SAT (300000000)
cannam@154: 
cannam@154: #define NORM_SCALING 16384
cannam@154: 
cannam@154: #define DB_SHIFT 10
cannam@154: 
cannam@154: #define EPSILON 1
cannam@154: #define VERY_SMALL 0
cannam@154: #define VERY_LARGE16 ((opus_val16)32767)
cannam@154: #define Q15_ONE ((opus_val16)32767)
cannam@154: 
cannam@154: #define SCALEIN(a)      (a)
cannam@154: #define SCALEOUT(a)     (a)
cannam@154: 
cannam@154: #define ABS16(x) ((x) < 0 ? (-(x)) : (x))
cannam@154: #define ABS32(x) ((x) < 0 ? (-(x)) : (x))
cannam@154: 
cannam@154: static OPUS_INLINE opus_int16 SAT16(opus_int32 x) {
cannam@154:    return x > 32767 ? 32767 : x < -32768 ? -32768 : (opus_int16)x;
cannam@154: }
cannam@154: 
cannam@154: #ifdef FIXED_DEBUG
cannam@154: #include "fixed_debug.h"
cannam@154: #else
cannam@154: 
cannam@154: #include "fixed_generic.h"
cannam@154: 
cannam@154: #ifdef OPUS_ARM_PRESUME_AARCH64_NEON_INTR
cannam@154: #include "arm/fixed_arm64.h"
cannam@154: #elif OPUS_ARM_INLINE_EDSP
cannam@154: #include "arm/fixed_armv5e.h"
cannam@154: #elif defined (OPUS_ARM_INLINE_ASM)
cannam@154: #include "arm/fixed_armv4.h"
cannam@154: #elif defined (BFIN_ASM)
cannam@154: #include "fixed_bfin.h"
cannam@154: #elif defined (TI_C5X_ASM)
cannam@154: #include "fixed_c5x.h"
cannam@154: #elif defined (TI_C6X_ASM)
cannam@154: #include "fixed_c6x.h"
cannam@154: #endif
cannam@154: 
cannam@154: #endif
cannam@154: 
cannam@154: #else /* FIXED_POINT */
cannam@154: 
cannam@154: typedef float opus_val16;
cannam@154: typedef float opus_val32;
cannam@154: typedef float opus_val64;
cannam@154: 
cannam@154: typedef float celt_sig;
cannam@154: typedef float celt_norm;
cannam@154: typedef float celt_ener;
cannam@154: 
cannam@154: #ifdef FLOAT_APPROX
cannam@154: /* This code should reliably detect NaN/inf even when -ffast-math is used.
cannam@154:    Assumes IEEE 754 format. */
cannam@154: static OPUS_INLINE int celt_isnan(float x)
cannam@154: {
cannam@154:    union {float f; opus_uint32 i;} in;
cannam@154:    in.f = x;
cannam@154:    return ((in.i>>23)&0xFF)==0xFF && (in.i&0x007FFFFF)!=0;
cannam@154: }
cannam@154: #else
cannam@154: #ifdef __FAST_MATH__
cannam@154: #error Cannot build libopus with -ffast-math unless FLOAT_APPROX is defined. This could result in crashes on extreme (e.g. NaN) input
cannam@154: #endif
cannam@154: #define celt_isnan(x) ((x)!=(x))
cannam@154: #endif
cannam@154: 
cannam@154: #define Q15ONE 1.0f
cannam@154: 
cannam@154: #define NORM_SCALING 1.f
cannam@154: 
cannam@154: #define EPSILON 1e-15f
cannam@154: #define VERY_SMALL 1e-30f
cannam@154: #define VERY_LARGE16 1e15f
cannam@154: #define Q15_ONE ((opus_val16)1.f)
cannam@154: 
cannam@154: /* This appears to be the same speed as C99's fabsf() but it's more portable. */
cannam@154: #define ABS16(x) ((float)fabs(x))
cannam@154: #define ABS32(x) ((float)fabs(x))
cannam@154: 
cannam@154: #define QCONST16(x,bits) (x)
cannam@154: #define QCONST32(x,bits) (x)
cannam@154: 
cannam@154: #define NEG16(x) (-(x))
cannam@154: #define NEG32(x) (-(x))
cannam@154: #define NEG32_ovflw(x) (-(x))
cannam@154: #define EXTRACT16(x) (x)
cannam@154: #define EXTEND32(x) (x)
cannam@154: #define SHR16(a,shift) (a)
cannam@154: #define SHL16(a,shift) (a)
cannam@154: #define SHR32(a,shift) (a)
cannam@154: #define SHL32(a,shift) (a)
cannam@154: #define PSHR32(a,shift) (a)
cannam@154: #define VSHR32(a,shift) (a)
cannam@154: 
cannam@154: #define PSHR(a,shift)   (a)
cannam@154: #define SHR(a,shift)    (a)
cannam@154: #define SHL(a,shift)    (a)
cannam@154: #define SATURATE(x,a)   (x)
cannam@154: #define SATURATE16(x)   (x)
cannam@154: 
cannam@154: #define ROUND16(a,shift)  (a)
cannam@154: #define SROUND16(a,shift) (a)
cannam@154: #define HALF16(x)       (.5f*(x))
cannam@154: #define HALF32(x)       (.5f*(x))
cannam@154: 
cannam@154: #define ADD16(a,b) ((a)+(b))
cannam@154: #define SUB16(a,b) ((a)-(b))
cannam@154: #define ADD32(a,b) ((a)+(b))
cannam@154: #define SUB32(a,b) ((a)-(b))
cannam@154: #define ADD32_ovflw(a,b) ((a)+(b))
cannam@154: #define SUB32_ovflw(a,b) ((a)-(b))
cannam@154: #define MULT16_16_16(a,b)     ((a)*(b))
cannam@154: #define MULT16_16(a,b)     ((opus_val32)(a)*(opus_val32)(b))
cannam@154: #define MAC16_16(c,a,b)     ((c)+(opus_val32)(a)*(opus_val32)(b))
cannam@154: 
cannam@154: #define MULT16_32_Q15(a,b)     ((a)*(b))
cannam@154: #define MULT16_32_Q16(a,b)     ((a)*(b))
cannam@154: 
cannam@154: #define MULT32_32_Q31(a,b)     ((a)*(b))
cannam@154: 
cannam@154: #define MAC16_32_Q15(c,a,b)     ((c)+(a)*(b))
cannam@154: #define MAC16_32_Q16(c,a,b)     ((c)+(a)*(b))
cannam@154: 
cannam@154: #define MULT16_16_Q11_32(a,b)     ((a)*(b))
cannam@154: #define MULT16_16_Q11(a,b)     ((a)*(b))
cannam@154: #define MULT16_16_Q13(a,b)     ((a)*(b))
cannam@154: #define MULT16_16_Q14(a,b)     ((a)*(b))
cannam@154: #define MULT16_16_Q15(a,b)     ((a)*(b))
cannam@154: #define MULT16_16_P15(a,b)     ((a)*(b))
cannam@154: #define MULT16_16_P13(a,b)     ((a)*(b))
cannam@154: #define MULT16_16_P14(a,b)     ((a)*(b))
cannam@154: #define MULT16_32_P16(a,b)     ((a)*(b))
cannam@154: 
cannam@154: #define DIV32_16(a,b)     (((opus_val32)(a))/(opus_val16)(b))
cannam@154: #define DIV32(a,b)     (((opus_val32)(a))/(opus_val32)(b))
cannam@154: 
cannam@154: #define SCALEIN(a)      ((a)*CELT_SIG_SCALE)
cannam@154: #define SCALEOUT(a)     ((a)*(1/CELT_SIG_SCALE))
cannam@154: 
cannam@154: #define SIG2WORD16(x) (x)
cannam@154: 
cannam@154: #endif /* !FIXED_POINT */
cannam@154: 
cannam@154: #ifndef GLOBAL_STACK_SIZE
cannam@154: #ifdef FIXED_POINT
cannam@154: #define GLOBAL_STACK_SIZE 120000
cannam@154: #else
cannam@154: #define GLOBAL_STACK_SIZE 120000
cannam@154: #endif
cannam@154: #endif
cannam@154: 
cannam@154: #endif /* ARCH_H */