lavr: x86: optimized 6-channel s16 to s16p conversion

This commit is contained in:
Justin Ruggles 2012-05-02 17:54:56 -04:00
parent 8eeffa8ada
commit 205ace8843
2 changed files with 74 additions and 0 deletions

View File

@ -862,3 +862,64 @@ CONV_S16_TO_S16P_2CH
INIT_XMM avx
CONV_S16_TO_S16P_2CH
%endif
;------------------------------------------------------------------------------
; void ff_conv_s16_to_s16p_6ch(int16_t *const *dst, int16_t *src, int len,
; int channels);
;------------------------------------------------------------------------------
%macro CONV_S16_TO_S16P_6CH 0
%if ARCH_X86_64
cglobal conv_s16_to_s16p_6ch, 3,8,5, dst, src, len, dst1, dst2, dst3, dst4, dst5
%else
cglobal conv_s16_to_s16p_6ch, 2,7,5, dst, src, dst1, dst2, dst3, dst4, dst5
%define lend dword r2m
%endif
mov dst1q, [dstq+ gprsize]
mov dst2q, [dstq+2*gprsize]
mov dst3q, [dstq+3*gprsize]
mov dst4q, [dstq+4*gprsize]
mov dst5q, [dstq+5*gprsize]
mov dstq, [dstq ]
sub dst1q, dstq
sub dst2q, dstq
sub dst3q, dstq
sub dst4q, dstq
sub dst5q, dstq
.loop:
mova m0, [srcq+0*mmsize] ; m0 = 0, 1, 2, 3, 4, 5, 6, 7
mova m3, [srcq+1*mmsize] ; m3 = 8, 9, 10, 11, 12, 13, 14, 15
mova m2, [srcq+2*mmsize] ; m2 = 16, 17, 18, 19, 20, 21, 22, 23
PALIGNR m1, m3, m0, 12, m4 ; m1 = 6, 7, 8, 9, 10, 11, x, x
shufps m3, m2, q1032 ; m3 = 12, 13, 14, 15, 16, 17, 18, 19
psrldq m2, 4 ; m2 = 18, 19, 20, 21, 22, 23, x, x
SBUTTERFLY2 wd, 0, 1, 4 ; m0 = 0, 6, 1, 7, 2, 8, 3, 9
; m1 = 4, 10, 5, 11, x, x, x, x
SBUTTERFLY2 wd, 3, 2, 4 ; m3 = 12, 18, 13, 19, 14, 20, 15, 21
; m2 = 16, 22, 17, 23, x, x, x, x
SBUTTERFLY2 dq, 0, 3, 4 ; m0 = 0, 6, 12, 18, 1, 7, 13, 19
; m3 = 2, 8, 14, 20, 3, 9, 15, 21
punpckldq m1, m2 ; m1 = 4, 10, 16, 22, 5, 11, 17, 23
movq [dstq ], m0
movhps [dstq+dst1q], m0
movq [dstq+dst2q], m3
movhps [dstq+dst3q], m3
movq [dstq+dst4q], m1
movhps [dstq+dst5q], m1
add srcq, mmsize*3
add dstq, mmsize/2
sub lend, mmsize/4
jg .loop
REP_RET
%endmacro
%define PALIGNR PALIGNR_MMX
INIT_XMM sse2
CONV_S16_TO_S16P_6CH
%define PALIGNR PALIGNR_SSSE3
INIT_XMM ssse3
CONV_S16_TO_S16P_6CH
%if HAVE_AVX
INIT_XMM avx
CONV_S16_TO_S16P_6CH
%endif

View File

@ -99,6 +99,13 @@ extern void ff_conv_s16_to_s16p_2ch_ssse3(int16_t *const *dst, int16_t *src,
extern void ff_conv_s16_to_s16p_2ch_avx (int16_t *const *dst, int16_t *src,
int len, int channels);
extern void ff_conv_s16_to_s16p_6ch_sse2 (int16_t *const *dst, int16_t *src,
int len, int channels);
extern void ff_conv_s16_to_s16p_6ch_ssse3(int16_t *const *dst, int16_t *src,
int len, int channels);
extern void ff_conv_s16_to_s16p_6ch_avx (int16_t *const *dst, int16_t *src,
int len, int channels);
av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
{
#if HAVE_YASM
@ -148,6 +155,8 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
2, 16, 4, "SSE2", ff_conv_fltp_to_s16_2ch_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
2, 16, 8, "SSE2", ff_conv_s16_to_s16p_2ch_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
6, 16, 4, "SSE2", ff_conv_s16_to_s16p_6ch_sse2);
}
if (mm_flags & AV_CPU_FLAG_SSSE3 && HAVE_SSE) {
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
@ -156,6 +165,8 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
2, 16, 4, "SSSE3", ff_conv_fltp_to_s16_2ch_ssse3);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
2, 16, 8, "SSSE3", ff_conv_s16_to_s16p_2ch_ssse3);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
6, 16, 4, "SSSE3", ff_conv_s16_to_s16p_6ch_ssse3);
}
if (mm_flags & AV_CPU_FLAG_SSE4 && HAVE_SSE) {
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16,
@ -182,6 +193,8 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
6, 16, 4, "AVX", ff_conv_fltp_to_flt_6ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
2, 16, 8, "AVX", ff_conv_s16_to_s16p_2ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
6, 16, 4, "AVX", ff_conv_s16_to_s16p_6ch_avx);
}
#endif
}