diff options
| author | Justin Ruggles | 2012-10-12 15:46:53 -0400 |
|---|---|---|
| committer | Luca Barbato | 2012-10-13 12:34:34 +0200 |
| commit | 10e645e9cb06dc87074232740d4366db18e7ba56 (patch) | |
| tree | bbb634ed7150d0d959c203e1881569f1ea809960 | |
| parent | 92281850a2d878dae1d50e271886ba87013b6ff3 (diff) | |
lavr: handle clipping in the float to s32 conversion
We cannot clip to INT_MAX because that value cannot be exactly
represented by a float value and ends up overflowing during conversion
anyway. We need to use a slightly smaller float value, which ends up
with slightly inaccurate results for samples which clip or nearly clip,
but it is close enough. Using doubles as intermediates in the conversion
would be more accurate, but it takes about twice as much time.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
| -rw-r--r-- | libavresample/x86/audio_convert.asm | 8 |
1 files changed, 7 insertions, 1 deletions
diff --git a/libavresample/x86/audio_convert.asm b/libavresample/x86/audio_convert.asm index 2ebdbc1ec0..25166afff2 100644 --- a/libavresample/x86/audio_convert.asm +++ b/libavresample/x86/audio_convert.asm @@ -28,6 +28,7 @@ SECTION_RODATA 32 pf_s32_inv_scale: times 8 dd 0x30000000 pf_s32_scale: times 8 dd 0x4f000000 +pf_s32_clip: times 8 dd 0x4effffff pf_s16_inv_scale: times 4 dd 0x38000000 pf_s16_scale: times 4 dd 0x47000000 pb_shuf_unpack_even: db -1, -1, 0, 1, -1, -1, 2, 3, -1, -1, 8, 9, -1, -1, 10, 11 @@ -197,17 +198,22 @@ cglobal conv_flt_to_s16, 3,3,5, dst, src, len ;------------------------------------------------------------------------------ %macro CONV_FLT_TO_S32 0 -cglobal conv_flt_to_s32, 3,3,5, dst, src, len +cglobal conv_flt_to_s32, 3,3,6, dst, src, len lea lenq, [lend*4] add srcq, lenq add dstq, lenq neg lenq mova m4, [pf_s32_scale] + mova m5, [pf_s32_clip] .loop: mulps m0, m4, [srcq+lenq ] mulps m1, m4, [srcq+lenq+1*mmsize] mulps m2, m4, [srcq+lenq+2*mmsize] mulps m3, m4, [srcq+lenq+3*mmsize] + minps m0, m0, m5 + minps m1, m1, m5 + minps m2, m2, m5 + minps m3, m3, m5 cvtps2dq m0, m0 cvtps2dq m1, m1 cvtps2dq m2, m2 |