lavr: x86: optimized 2-channel fltp to s16 conversion

diff --git a/libavresample/x86/audio_convert.asm b/libavresample/x86/audio_convert.asm
index 622a84c..15aaa6a 100644 (file)
--- a/libavresample/x86/audio_convert.asm
+++ b/libavresample/x86/audio_convert.asm
@@ -32,6 +32,7 @@ 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
 pb_shuf_unpack_odd:       db -1, -1,  4,  5, -1, -1,  6,  7, -1, -1, 12, 13, -1, -1, 14, 15
+pb_interleave_words: SHUFFLE_MASK_W  0,  4,  1,  5,  2,  6,  3,  7
 
 SECTION_TEXT
 
@@ -538,6 +539,48 @@ INIT_XMM avx
 CONV_S16P_TO_FLT_6CH
 %endif
 
+;------------------------------------------------------------------------------
+; void ff_conv_fltp_to_s16_2ch(int16_t *dst, float *const *src, int len,
+;                              int channels);
+;------------------------------------------------------------------------------
+
+%macro CONV_FLTP_TO_S16_2CH 0
+cglobal conv_fltp_to_s16_2ch, 3,4,3, dst, src0, len, src1
+    lea      lenq, [4*lend]
+    mov     src1q, [src0q+gprsize]
+    mov     src0q, [src0q        ]
+    add      dstq, lenq
+    add     src0q, lenq
+    add     src1q, lenq
+    neg      lenq
+    mova       m2, [pf_s16_scale]
+%if cpuflag(ssse3)
+    mova       m3, [pb_interleave_words]
+%endif
+.loop:
+    mulps      m0, m2, [src0q+lenq] ; m0 =    0,    2,    4,    6
+    mulps      m1, m2, [src1q+lenq] ; m1 =    1,    3,    5,    7
+    cvtps2dq   m0, m0
+    cvtps2dq   m1, m1
+%if cpuflag(ssse3)
+    packssdw   m0, m1               ; m0 = 0, 2, 4, 6, 1, 3, 5, 7
+    pshufb     m0, m3               ; m0 = 0, 1, 2, 3, 4, 5, 6, 7
+%else
+    packssdw   m0, m0               ; m0 = 0, 2, 4, 6, x, x, x, x
+    packssdw   m1, m1               ; m1 = 1, 3, 5, 7, x, x, x, x
+    punpcklwd  m0, m1               ; m0 = 0, 1, 2, 3, 4, 5, 6, 7
+%endif
+    mova  [dstq+lenq], m0
+    add      lenq, mmsize
+    jl .loop
+    REP_RET
+%endmacro
+
+INIT_XMM sse2
+CONV_FLTP_TO_S16_2CH
+INIT_XMM ssse3
+CONV_FLTP_TO_S16_2CH
+
 ;-----------------------------------------------------------------------------
 ; void ff_conv_fltp_to_flt_6ch(float *dst, float *const *src, int len,
 ;                              int channels);

diff --git a/libavresample/x86/audio_convert_init.c b/libavresample/x86/audio_convert_init.c
index 6bcf093..3098658 100644 (file)
--- a/libavresample/x86/audio_convert_init.c
+++ b/libavresample/x86/audio_convert_init.c
@@ -66,6 +66,11 @@ extern void ff_conv_s16p_to_flt_6ch_ssse3(float *dst, int16_t *const *src,
 extern void ff_conv_s16p_to_flt_6ch_avx  (float *dst, int16_t *const *src,
                                           int len, int channels);
 
+extern void ff_conv_fltp_to_s16_2ch_sse2 (int16_t *dst, float *const *src,
+                                          int len, int channels);
+extern void ff_conv_fltp_to_s16_2ch_ssse3(int16_t *dst, float *const *src,
+                                          int len, int channels);
+
 extern void ff_conv_fltp_to_flt_6ch_mmx (float *dst, float *const *src, int len,
                                          int channels);
 extern void ff_conv_fltp_to_flt_6ch_sse4(float *dst, float *const *src, int len,
@@ -110,10 +115,14 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
                                   2, 16, 8, "SSE2", ff_conv_s16p_to_flt_2ch_sse2);
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
                                   6, 16, 4, "SSE2", ff_conv_s16p_to_flt_6ch_sse2);
+        ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP,
+                                  2, 16, 4, "SSE2", ff_conv_fltp_to_s16_2ch_sse2);
     }
     if (mm_flags & AV_CPU_FLAG_SSSE3 && HAVE_SSE) {
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
                                   6, 16, 4, "SSSE3", ff_conv_s16p_to_flt_6ch_ssse3);
+        ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP,
+                                  2, 16, 4, "SSSE3", ff_conv_fltp_to_s16_2ch_ssse3);
     }
     if (mm_flags & AV_CPU_FLAG_SSE4 && HAVE_SSE) {
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16,