lavr: x86: optimized 2-channel flt to s16p conversion

libavresample/x86/audio_convert.asm

@@ -1045,3 +1045,49 @@ CONV_S16_TO_FLTP_6CH
 INIT_XMM avx
 CONV_S16_TO_FLTP_6CH
 %endif
+
+;------------------------------------------------------------------------------
+; void ff_conv_flt_to_s16p_2ch(int16_t *const *dst, float *src, int len,
+;                              int channels);
+;------------------------------------------------------------------------------
+
+%macro CONV_FLT_TO_S16P_2CH 0
+cglobal conv_flt_to_s16p_2ch, 3,4,6, dst0, src, len, dst1
+    lea       lenq, [2*lend]
+    mov      dst1q, [dst0q+gprsize]
+    mov      dst0q, [dst0q        ]
+    lea       srcq, [srcq+4*lenq]
+    add      dst0q, lenq
+    add      dst1q, lenq
+    neg       lenq
+    mova        m5, [pf_s16_scale]
+.loop:
+    mova       m0, [srcq+4*lenq         ]
+    mova       m1, [srcq+4*lenq+  mmsize]
+    mova       m2, [srcq+4*lenq+2*mmsize]
+    mova       m3, [srcq+4*lenq+3*mmsize]
+    DEINT2_PS   0, 1, 4
+    DEINT2_PS   2, 3, 4
+    mulps      m0, m0, m5
+    mulps      m1, m1, m5
+    mulps      m2, m2, m5
+    mulps      m3, m3, m5
+    cvtps2dq   m0, m0
+    cvtps2dq   m1, m1
+    cvtps2dq   m2, m2
+    cvtps2dq   m3, m3
+    packssdw   m0, m2
+    packssdw   m1, m3
+    mova  [dst0q+lenq], m0
+    mova  [dst1q+lenq], m1
+    add      lenq, mmsize
+    jl .loop
+    REP_RET
+%endmacro
+
+INIT_XMM sse2
+CONV_FLT_TO_S16P_2CH
+%if HAVE_AVX
+INIT_XMM avx
+CONV_FLT_TO_S16P_2CH
+%endif

libavresample/x86/audio_convert_init.c

@@ -120,6 +120,11 @@ extern void ff_conv_s16_to_fltp_6ch_sse4 (float *const *dst, int16_t *src,
 extern void ff_conv_s16_to_fltp_6ch_avx  (float *const *dst, int16_t *src,
                                           int len, int channels);
 
+extern void ff_conv_flt_to_s16p_2ch_sse2(int16_t *const *dst, float *src,
+                                         int len, int channels);
+extern void ff_conv_flt_to_s16p_2ch_avx (int16_t *const *dst, float *src,
+                                         int len, int channels);
+
 av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
 {
 #if HAVE_YASM
@@ -175,6 +180,8 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
                                   2, 16, 8, "SSE2", ff_conv_s16_to_fltp_2ch_sse2);
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
                                   6, 16, 4, "SSE2", ff_conv_s16_to_fltp_6ch_sse2);
+        ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_FLT,
+                                  2, 16, 8, "SSE2", ff_conv_flt_to_s16p_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,
@@ -219,6 +226,8 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
                                   2, 16, 8, "AVX", ff_conv_s16_to_fltp_2ch_avx);
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
                                   6, 16, 4, "AVX", ff_conv_s16_to_fltp_6ch_avx);
+        ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_FLT,
+                                  2, 16, 8, "AVX", ff_conv_flt_to_s16p_2ch_avx);
     }
 #endif
 }