diff options
author | Rong Yan <rongyan236@gmail.com> | 2014-11-27 05:45:58 +0000 |
---|---|---|
committer | Michael Niedermayer <michaelni@gmx.at> | 2014-11-27 12:07:19 +0100 |
commit | 22e557917d08ed93e9c657f13c9669482341eb02 (patch) | |
tree | 80b608c358bf19d16addc2d693bc9dd225ca4b7b | |
parent | 4ae1d6021be96dcded093e51d04294075b14e511 (diff) | |
download | ffmpeg-streaming-22e557917d08ed93e9c657f13c9669482341eb02.zip ffmpeg-streaming-22e557917d08ed93e9c657f13c9669482341eb02.tar.gz |
libavutil/ppc/util_altivec.h : fix load_with_perm_vec() add marcos vcswapi2s() vcswapc() VEC_SPLAT16() VEC_SLD16() for POWER LE
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
-rw-r--r-- | libavutil/ppc/util_altivec.h | 38 |
1 files changed, 31 insertions, 7 deletions
diff --git a/libavutil/ppc/util_altivec.h b/libavutil/ppc/util_altivec.h index 2f0c6eb..c0b71c8 100644 --- a/libavutil/ppc/util_altivec.h +++ b/libavutil/ppc/util_altivec.h @@ -46,9 +46,21 @@ #define WORD_s1 0x14,0x15,0x16,0x17 #define WORD_s2 0x18,0x19,0x1a,0x1b #define WORD_s3 0x1c,0x1d,0x1e,0x1f - #define vcprm(a,b,c,d) (const vector unsigned char){WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d} +#define SWP_W2S0 0x02,0x03,0x00,0x01 +#define SWP_W2S1 0x06,0x07,0x04,0x05 +#define SWP_W2S2 0x0a,0x0b,0x08,0x09 +#define SWP_W2S3 0x0e,0x0f,0x0c,0x0d +#define SWP_W2Ss0 0x12,0x13,0x10,0x11 +#define SWP_W2Ss1 0x16,0x17,0x14,0x15 +#define SWP_W2Ss2 0x1a,0x1b,0x18,0x19 +#define SWP_W2Ss3 0x1e,0x1f,0x1c,0x1d +#define vcswapi2s(a,b,c,d) (const vector unsigned char){SWP_W2S ## a, SWP_W2S ## b, SWP_W2S ## c, SWP_W2S ## d} + +#define vcswapc() \ + (const vector unsigned char){0x0f,0x0e,0x0d,0x0c,0x0b,0x0a,0x09,0x08,0x07,0x06,0x05,0x04,0x03,0x02,0x01,0x00} + // Transpose 8x8 matrix of 16-bit elements (in-place) #define TRANSPOSE8(a,b,c,d,e,f,g,h) \ @@ -103,8 +115,15 @@ static inline vector unsigned char unaligned_load(int offset, const uint8_t *src register vector unsigned char mask = vec_lvsl(offset, src); return vec_perm(first, second, mask); } +static inline vec_u8 load_with_perm_vec(int offset, const uint8_t *src, vec_u8 perm_vec) +{ + vec_u8 a = vec_ld(offset, src); + vec_u8 b = vec_ld(offset+15, src); + return vec_perm(a, b, perm_vec); +} #else #define unaligned_load(a,b) VEC_LD(a,b) +#define load_with_perm_vec(a,b,c) VEC_LD(a,b) #endif @@ -112,12 +131,6 @@ static inline vector unsigned char unaligned_load(int offset, const uint8_t *src * loads vector known misalignment * @param perm_vec the align permute vector to combine the two loads from lvsl */ -static inline vec_u8 load_with_perm_vec(int offset, const uint8_t *src, vec_u8 perm_vec) -{ - vec_u8 a = vec_ld(offset, src); - vec_u8 b = vec_ld(offset+15, src); - return vec_perm(a, b, perm_vec); -} #define vec_unaligned_load(b) VEC_LD(0, b) @@ -135,6 +148,17 @@ static inline vec_u8 load_with_perm_vec(int offset, const uint8_t *src, vec_u8 p #define VEC_ST(a,b,c) vec_vsx_st(a,b,c) #endif +#if HAVE_BIGENDIAN +#define VEC_SPLAT16(a,b) vec_splat((vec_s16)a, b) +#else +#define VEC_SPLAT16(a,b) vec_splat((vec_s16)(vec_perm(a, a, vcswapi2s(0,1,2,3))), b) +#endif + +#if HAVE_BIGENDIAN +#define VEC_SLD16(a,b,c) vec_sld(a, b, c) +#else +#define VEC_SLD16(a,b,c) vec_sld(b, a, c) +#endif #endif /* HAVE_ALTIVEC */ |