diff options
| author | xiangmingzhu <xiangzhu@cisco.com> | 2014-04-30 11:48:07 +0400 |
|---|---|---|
| committer | Jean-Marc Valin <jmvalin@jmvalin.ca> | 2014-10-04 05:16:00 +0400 |
| commit | c95c9a048f3283afb2e412b10085d4f7c19e1412 (patch) | |
| tree | ed8873af6559d7a98922e0fed85be47c826ef521 /silk/macros.h | |
| parent | 80460334b77d70e665a390503cd8992cdad06c10 (diff) | |
Cisco optimization for x86 & fixed point
1. Only for fixed point on x86 platform (32bit and 64bit, uses SIMD
intrinsics up to SSE4.2)
2. Use "configure --enable-fixed-point --enable-intrinsics" to enable
optimization, default is disabled.
3. Official test cases are verified and passed.
Signed-off-by: Timothy B. Terriberry <tterribe@xiph.org>
Diffstat (limited to 'silk/macros.h')
| -rw-r--r-- | silk/macros.h | 28 |
1 files changed, 28 insertions, 0 deletions
diff --git a/silk/macros.h b/silk/macros.h index 612c2cc5..2f249509 100644 --- a/silk/macros.h +++ b/silk/macros.h @@ -35,19 +35,39 @@ POSSIBILITY OF SUCH DAMAGE. #include "opus_types.h" #include "opus_defines.h" +#if OPUS_GNUC_PREREQ(3, 0) +#define opus_likely(x) (__builtin_expect(!!(x), 1)) +#define opus_unlikely(x) (__builtin_expect(!!(x), 0)) +#else +#define opus_likely(x) (!!(x)) +#define opus_unlikely(x) (!!(x)) +#endif + /* This is an OPUS_INLINE header file for general platform. */ /* (a32 * (opus_int32)((opus_int16)(b32))) >> 16 output have to be 32bit int */ +#if defined(__x86_64__) || defined(__LP64__) || defined(_WIN64) +#define silk_SMULWB(a32, b32) (((a32) * (opus_int64)((opus_int16)(b32))) >> 16) +#else #define silk_SMULWB(a32, b32) ((((a32) >> 16) * (opus_int32)((opus_int16)(b32))) + ((((a32) & 0x0000FFFF) * (opus_int32)((opus_int16)(b32))) >> 16)) +#endif /* a32 + (b32 * (opus_int32)((opus_int16)(c32))) >> 16 output have to be 32bit int */ +#if defined(__x86_64__) || defined(__LP64__) || defined(_WIN64) +#define silk_SMLAWB(a32, b32, c32) ((a32) + (((b32) * (opus_int64)((opus_int16)(c32))) >> 16)) +#else #define silk_SMLAWB(a32, b32, c32) ((a32) + ((((b32) >> 16) * (opus_int32)((opus_int16)(c32))) + ((((b32) & 0x0000FFFF) * (opus_int32)((opus_int16)(c32))) >> 16))) +#endif /* (a32 * (b32 >> 16)) >> 16 */ #define silk_SMULWT(a32, b32) (((a32) >> 16) * ((b32) >> 16) + ((((a32) & 0x0000FFFF) * ((b32) >> 16)) >> 16)) /* a32 + (b32 * (c32 >> 16)) >> 16 */ +#if defined(__x86_64__) || defined(__LP64__) || defined(_WIN64) +#define silk_SMLAWT(a32, b32, c32) ((a32) + (((b32) * ((opus_int64)(c32) >> 16)) >> 16)) +#else #define silk_SMLAWT(a32, b32, c32) ((a32) + (((b32) >> 16) * ((c32) >> 16)) + ((((b32) & 0x0000FFFF) * ((c32) >> 16)) >> 16)) +#endif /* (opus_int32)((opus_int16)(a3))) * (opus_int32)((opus_int16)(b32)) output have to be 32bit int */ #define silk_SMULBB(a32, b32) ((opus_int32)((opus_int16)(a32)) * (opus_int32)((opus_int16)(b32))) @@ -65,10 +85,18 @@ POSSIBILITY OF SUCH DAMAGE. #define silk_SMLAL(a64, b32, c32) (silk_ADD64((a64), ((opus_int64)(b32) * (opus_int64)(c32)))) /* (a32 * b32) >> 16 */ +#if defined(__x86_64__) || defined(__LP64__) || defined(_WIN64) +#define silk_SMULWW(a32, b32) (((opus_int64)(a32) * (b32)) >> 16) +#else #define silk_SMULWW(a32, b32) silk_MLA(silk_SMULWB((a32), (b32)), (a32), silk_RSHIFT_ROUND((b32), 16)) +#endif /* a32 + ((b32 * c32) >> 16) */ +#if defined(__x86_64__) || defined(__LP64__) || defined(_WIN64) +#define silk_SMLAWW(a32, b32, c32) ((a32) + (((opus_int64)(b32) * (c32)) >> 16)) +#else #define silk_SMLAWW(a32, b32, c32) silk_MLA(silk_SMLAWB((a32), (b32), (c32)), (b32), silk_RSHIFT_ROUND((c32), 16)) +#endif /* add/subtract with output saturated */ #define silk_ADD_SAT32(a, b) ((((opus_uint32)(a) + (opus_uint32)(b)) & 0x80000000) == 0 ? \ |