diff options
| author | Anton Blanchard <anton@ozlabs.org> | 2018-07-07 14:21:46 +1000 |
|---|---|---|
| committer | Erik de Castro Lopo <erikd@mega-nerd.com> | 2018-08-20 18:16:34 +1000 |
| commit | cdb030cd3749b4547399a049ce3cae5e974ccd48 (patch) | |
| tree | 1ae2e9befc7da38c7c44f27c5ac33ff9afdddb72 /src/libFLAC/include/private/lpc.h | |
| parent | 8e1796b91a1893609b8e2f5813d08f7df3f011cf (diff) | |
| download | flac-cdb030cd3749b4547399a049ce3cae5e974ccd48.tar.gz | |
Add VSX optimised versions of autocorrelation loops
Add a POWER8 and POWER9 version of the autocorrelation functions.
flac --best is about 3.3x faster on POWER9 with this patch.
Signed-off-by: Anton Blanchard <anton@ozlabs.org>
Diffstat (limited to 'src/libFLAC/include/private/lpc.h')
| -rw-r--r-- | src/libFLAC/include/private/lpc.h | 14 |
1 files changed, 14 insertions, 0 deletions
diff --git a/src/libFLAC/include/private/lpc.h b/src/libFLAC/include/private/lpc.h index 63d64324..64dfd1f8 100644 --- a/src/libFLAC/include/private/lpc.h +++ b/src/libFLAC/include/private/lpc.h @@ -91,6 +91,20 @@ void FLAC__lpc_compute_autocorrelation_intrin_sse_lag_12_new(const FLAC__real da void FLAC__lpc_compute_autocorrelation_intrin_sse_lag_16_new(const FLAC__real data[], uint32_t data_len, uint32_t lag, FLAC__real autoc[]); # endif # endif +#if defined(FLAC__CPU_PPC64) && defined(FLAC__USE_VSX) +#ifdef FLAC__HAS_TARGET_POWER9 +void FLAC__lpc_compute_autocorrelation_intrin_power9_vsx_lag_4(const FLAC__real data[], uint32_t data_len, uint32_t lag, FLAC__real autoc[]); +void FLAC__lpc_compute_autocorrelation_intrin_power9_vsx_lag_8(const FLAC__real data[], uint32_t data_len, uint32_t lag, FLAC__real autoc[]); +void FLAC__lpc_compute_autocorrelation_intrin_power9_vsx_lag_12(const FLAC__real data[], uint32_t data_len, uint32_t lag, FLAC__real autoc[]); +void FLAC__lpc_compute_autocorrelation_intrin_power9_vsx_lag_16(const FLAC__real data[], uint32_t data_len, uint32_t lag, FLAC__real autoc[]); +#endif +#ifdef FLAC__HAS_TARGET_POWER8 +void FLAC__lpc_compute_autocorrelation_intrin_power8_vsx_lag_4(const FLAC__real data[], uint32_t data_len, uint32_t lag, FLAC__real autoc[]); +void FLAC__lpc_compute_autocorrelation_intrin_power8_vsx_lag_8(const FLAC__real data[], uint32_t data_len, uint32_t lag, FLAC__real autoc[]); +void FLAC__lpc_compute_autocorrelation_intrin_power8_vsx_lag_12(const FLAC__real data[], uint32_t data_len, uint32_t lag, FLAC__real autoc[]); +void FLAC__lpc_compute_autocorrelation_intrin_power8_vsx_lag_16(const FLAC__real data[], uint32_t data_len, uint32_t lag, FLAC__real autoc[]); +#endif +#endif #endif /* |