Chris@69: /* Copyright (c) 2003-2008 Jean-Marc Valin Chris@69: Copyright (c) 2007-2008 CSIRO Chris@69: Copyright (c) 2007-2009 Xiph.Org Foundation Chris@69: Written by Jean-Marc Valin */ Chris@69: /** Chris@69: @file arch.h Chris@69: @brief Various architecture definitions for CELT Chris@69: */ 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: #ifndef ARCH_H Chris@69: #define ARCH_H Chris@69: Chris@69: #include "opus_types.h" Chris@69: #include "opus_defines.h" Chris@69: Chris@69: # if !defined(__GNUC_PREREQ) Chris@69: # if defined(__GNUC__)&&defined(__GNUC_MINOR__) Chris@69: # define __GNUC_PREREQ(_maj,_min) \ Chris@69: ((__GNUC__<<16)+__GNUC_MINOR__>=((_maj)<<16)+(_min)) Chris@69: # else Chris@69: # define __GNUC_PREREQ(_maj,_min) 0 Chris@69: # endif Chris@69: # endif Chris@69: Chris@69: #if OPUS_GNUC_PREREQ(3, 0) Chris@69: #define opus_likely(x) (__builtin_expect(!!(x), 1)) Chris@69: #define opus_unlikely(x) (__builtin_expect(!!(x), 0)) Chris@69: #else Chris@69: #define opus_likely(x) (!!(x)) Chris@69: #define opus_unlikely(x) (!!(x)) Chris@69: #endif Chris@69: Chris@69: #define CELT_SIG_SCALE 32768.f Chris@69: Chris@69: #define CELT_FATAL(str) celt_fatal(str, __FILE__, __LINE__); Chris@69: Chris@69: #if defined(ENABLE_ASSERTIONS) || defined(ENABLE_HARDENING) Chris@69: #ifdef __GNUC__ Chris@69: __attribute__((noreturn)) Chris@69: #endif Chris@69: void celt_fatal(const char *str, const char *file, int line); Chris@69: Chris@69: #if defined(CELT_C) && !defined(OVERRIDE_celt_fatal) Chris@69: #include Chris@69: #include Chris@69: #ifdef __GNUC__ Chris@69: __attribute__((noreturn)) Chris@69: #endif Chris@69: void celt_fatal(const char *str, const char *file, int line) Chris@69: { Chris@69: fprintf (stderr, "Fatal (internal) error in %s, line %d: %s\n", file, line, str); Chris@69: abort(); Chris@69: } Chris@69: #endif Chris@69: Chris@69: #define celt_assert(cond) {if (!(cond)) {CELT_FATAL("assertion failed: " #cond);}} Chris@69: #define celt_assert2(cond, message) {if (!(cond)) {CELT_FATAL("assertion failed: " #cond "\n" message);}} Chris@69: #define MUST_SUCCEED(call) celt_assert((call) == OPUS_OK) Chris@69: #else Chris@69: #define celt_assert(cond) Chris@69: #define celt_assert2(cond, message) Chris@69: #define MUST_SUCCEED(call) do {if((call) != OPUS_OK) {RESTORE_STACK; return OPUS_INTERNAL_ERROR;} } while (0) Chris@69: #endif Chris@69: Chris@69: #if defined(ENABLE_ASSERTIONS) Chris@69: #define celt_sig_assert(cond) {if (!(cond)) {CELT_FATAL("signal assertion failed: " #cond);}} Chris@69: #else Chris@69: #define celt_sig_assert(cond) Chris@69: #endif Chris@69: Chris@69: #define IMUL32(a,b) ((a)*(b)) Chris@69: Chris@69: #define MIN16(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum 16-bit value. */ Chris@69: #define MAX16(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 16-bit value. */ Chris@69: #define MIN32(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum 32-bit value. */ Chris@69: #define MAX32(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 32-bit value. */ Chris@69: #define IMIN(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum int value. */ Chris@69: #define IMAX(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum int value. */ Chris@69: #define UADD32(a,b) ((a)+(b)) Chris@69: #define USUB32(a,b) ((a)-(b)) Chris@69: Chris@69: /* Set this if opus_int64 is a native type of the CPU. */ Chris@69: /* Assume that all LP64 architectures have fast 64-bit types; also x86_64 Chris@69: (which can be ILP32 for x32) and Win64 (which is LLP64). */ Chris@69: #if defined(__x86_64__) || defined(__LP64__) || defined(_WIN64) Chris@69: #define OPUS_FAST_INT64 1 Chris@69: #else Chris@69: #define OPUS_FAST_INT64 0 Chris@69: #endif Chris@69: Chris@69: #define PRINT_MIPS(file) Chris@69: Chris@69: #ifdef FIXED_POINT Chris@69: Chris@69: typedef opus_int16 opus_val16; Chris@69: typedef opus_int32 opus_val32; Chris@69: typedef opus_int64 opus_val64; Chris@69: Chris@69: typedef opus_val32 celt_sig; Chris@69: typedef opus_val16 celt_norm; Chris@69: typedef opus_val32 celt_ener; Chris@69: Chris@69: #define celt_isnan(x) 0 Chris@69: Chris@69: #define Q15ONE 32767 Chris@69: Chris@69: #define SIG_SHIFT 12 Chris@69: /* Safe saturation value for 32-bit signals. Should be less than Chris@69: 2^31*(1-0.85) to avoid blowing up on DC at deemphasis.*/ Chris@69: #define SIG_SAT (300000000) Chris@69: Chris@69: #define NORM_SCALING 16384 Chris@69: Chris@69: #define DB_SHIFT 10 Chris@69: Chris@69: #define EPSILON 1 Chris@69: #define VERY_SMALL 0 Chris@69: #define VERY_LARGE16 ((opus_val16)32767) Chris@69: #define Q15_ONE ((opus_val16)32767) Chris@69: Chris@69: #define SCALEIN(a) (a) Chris@69: #define SCALEOUT(a) (a) Chris@69: Chris@69: #define ABS16(x) ((x) < 0 ? (-(x)) : (x)) Chris@69: #define ABS32(x) ((x) < 0 ? (-(x)) : (x)) Chris@69: Chris@69: static OPUS_INLINE opus_int16 SAT16(opus_int32 x) { Chris@69: return x > 32767 ? 32767 : x < -32768 ? -32768 : (opus_int16)x; Chris@69: } Chris@69: Chris@69: #ifdef FIXED_DEBUG Chris@69: #include "fixed_debug.h" Chris@69: #else Chris@69: Chris@69: #include "fixed_generic.h" Chris@69: Chris@69: #ifdef OPUS_ARM_PRESUME_AARCH64_NEON_INTR Chris@69: #include "arm/fixed_arm64.h" Chris@69: #elif OPUS_ARM_INLINE_EDSP Chris@69: #include "arm/fixed_armv5e.h" Chris@69: #elif defined (OPUS_ARM_INLINE_ASM) Chris@69: #include "arm/fixed_armv4.h" Chris@69: #elif defined (BFIN_ASM) Chris@69: #include "fixed_bfin.h" Chris@69: #elif defined (TI_C5X_ASM) Chris@69: #include "fixed_c5x.h" Chris@69: #elif defined (TI_C6X_ASM) Chris@69: #include "fixed_c6x.h" Chris@69: #endif Chris@69: Chris@69: #endif Chris@69: Chris@69: #else /* FIXED_POINT */ Chris@69: Chris@69: typedef float opus_val16; Chris@69: typedef float opus_val32; Chris@69: typedef float opus_val64; Chris@69: Chris@69: typedef float celt_sig; Chris@69: typedef float celt_norm; Chris@69: typedef float celt_ener; Chris@69: Chris@69: #ifdef FLOAT_APPROX Chris@69: /* This code should reliably detect NaN/inf even when -ffast-math is used. Chris@69: Assumes IEEE 754 format. */ Chris@69: static OPUS_INLINE int celt_isnan(float x) Chris@69: { Chris@69: union {float f; opus_uint32 i;} in; Chris@69: in.f = x; Chris@69: return ((in.i>>23)&0xFF)==0xFF && (in.i&0x007FFFFF)!=0; Chris@69: } Chris@69: #else Chris@69: #ifdef __FAST_MATH__ Chris@69: #error Cannot build libopus with -ffast-math unless FLOAT_APPROX is defined. This could result in crashes on extreme (e.g. NaN) input Chris@69: #endif Chris@69: #define celt_isnan(x) ((x)!=(x)) Chris@69: #endif Chris@69: Chris@69: #define Q15ONE 1.0f Chris@69: Chris@69: #define NORM_SCALING 1.f Chris@69: Chris@69: #define EPSILON 1e-15f Chris@69: #define VERY_SMALL 1e-30f Chris@69: #define VERY_LARGE16 1e15f Chris@69: #define Q15_ONE ((opus_val16)1.f) Chris@69: Chris@69: /* This appears to be the same speed as C99's fabsf() but it's more portable. */ Chris@69: #define ABS16(x) ((float)fabs(x)) Chris@69: #define ABS32(x) ((float)fabs(x)) Chris@69: Chris@69: #define QCONST16(x,bits) (x) Chris@69: #define QCONST32(x,bits) (x) Chris@69: Chris@69: #define NEG16(x) (-(x)) Chris@69: #define NEG32(x) (-(x)) Chris@69: #define NEG32_ovflw(x) (-(x)) Chris@69: #define EXTRACT16(x) (x) Chris@69: #define EXTEND32(x) (x) Chris@69: #define SHR16(a,shift) (a) Chris@69: #define SHL16(a,shift) (a) Chris@69: #define SHR32(a,shift) (a) Chris@69: #define SHL32(a,shift) (a) Chris@69: #define PSHR32(a,shift) (a) Chris@69: #define VSHR32(a,shift) (a) Chris@69: Chris@69: #define PSHR(a,shift) (a) Chris@69: #define SHR(a,shift) (a) Chris@69: #define SHL(a,shift) (a) Chris@69: #define SATURATE(x,a) (x) Chris@69: #define SATURATE16(x) (x) Chris@69: Chris@69: #define ROUND16(a,shift) (a) Chris@69: #define SROUND16(a,shift) (a) Chris@69: #define HALF16(x) (.5f*(x)) Chris@69: #define HALF32(x) (.5f*(x)) Chris@69: Chris@69: #define ADD16(a,b) ((a)+(b)) Chris@69: #define SUB16(a,b) ((a)-(b)) Chris@69: #define ADD32(a,b) ((a)+(b)) Chris@69: #define SUB32(a,b) ((a)-(b)) Chris@69: #define ADD32_ovflw(a,b) ((a)+(b)) Chris@69: #define SUB32_ovflw(a,b) ((a)-(b)) Chris@69: #define MULT16_16_16(a,b) ((a)*(b)) Chris@69: #define MULT16_16(a,b) ((opus_val32)(a)*(opus_val32)(b)) Chris@69: #define MAC16_16(c,a,b) ((c)+(opus_val32)(a)*(opus_val32)(b)) Chris@69: Chris@69: #define MULT16_32_Q15(a,b) ((a)*(b)) Chris@69: #define MULT16_32_Q16(a,b) ((a)*(b)) Chris@69: Chris@69: #define MULT32_32_Q31(a,b) ((a)*(b)) Chris@69: Chris@69: #define MAC16_32_Q15(c,a,b) ((c)+(a)*(b)) Chris@69: #define MAC16_32_Q16(c,a,b) ((c)+(a)*(b)) Chris@69: Chris@69: #define MULT16_16_Q11_32(a,b) ((a)*(b)) Chris@69: #define MULT16_16_Q11(a,b) ((a)*(b)) Chris@69: #define MULT16_16_Q13(a,b) ((a)*(b)) Chris@69: #define MULT16_16_Q14(a,b) ((a)*(b)) Chris@69: #define MULT16_16_Q15(a,b) ((a)*(b)) Chris@69: #define MULT16_16_P15(a,b) ((a)*(b)) Chris@69: #define MULT16_16_P13(a,b) ((a)*(b)) Chris@69: #define MULT16_16_P14(a,b) ((a)*(b)) Chris@69: #define MULT16_32_P16(a,b) ((a)*(b)) Chris@69: Chris@69: #define DIV32_16(a,b) (((opus_val32)(a))/(opus_val16)(b)) Chris@69: #define DIV32(a,b) (((opus_val32)(a))/(opus_val32)(b)) Chris@69: Chris@69: #define SCALEIN(a) ((a)*CELT_SIG_SCALE) Chris@69: #define SCALEOUT(a) ((a)*(1/CELT_SIG_SCALE)) Chris@69: Chris@69: #define SIG2WORD16(x) (x) Chris@69: Chris@69: #endif /* !FIXED_POINT */ Chris@69: Chris@69: #ifndef GLOBAL_STACK_SIZE Chris@69: #ifdef FIXED_POINT Chris@69: #define GLOBAL_STACK_SIZE 120000 Chris@69: #else Chris@69: #define GLOBAL_STACK_SIZE 120000 Chris@69: #endif Chris@69: #endif Chris@69: Chris@69: #endif /* ARCH_H */