lavr: add x86-optimized functions for mixing 1-to-2 s16p with flt coeffs

libavresample/x86/audio_mix.asm

@@ -184,3 +184,50 @@ MIX_1_TO_2_FLTP_FLT
 INIT_YMM avx
 MIX_1_TO_2_FLTP_FLT
 %endif
+
+;-----------------------------------------------------------------------------
+; void ff_mix_1_to_2_s16p_flt(int16_t **src, float **matrix, int len,
+;                             int out_ch, int in_ch);
+;-----------------------------------------------------------------------------
+
+%macro MIX_1_TO_2_S16P_FLT 0
+cglobal mix_1_to_2_s16p_flt, 3,5,6, src0, matrix0, len, src1, matrix1
+    mov       src1q, [src0q+gprsize]
+    mov       src0q, [src0q]
+    sub       src1q, src0q
+    mov    matrix1q, [matrix0q+gprsize]
+    mov    matrix0q, [matrix0q]
+    VBROADCASTSS m4, [matrix0q]
+    VBROADCASTSS m5, [matrix1q]
+    ALIGN 16
+.loop:
+    mova         m0, [src0q]
+    S16_TO_S32_SX 0, 2
+    cvtdq2ps     m0, m0
+    cvtdq2ps     m2, m2
+    mulps        m1, m0, m5
+    mulps        m0, m0, m4
+    mulps        m3, m2, m5
+    mulps        m2, m2, m4
+    cvtps2dq     m0, m0
+    cvtps2dq     m1, m1
+    cvtps2dq     m2, m2
+    cvtps2dq     m3, m3
+    packssdw     m0, m2
+    packssdw     m1, m3
+    mova  [src0q      ], m0
+    mova  [src0q+src1q], m1
+    add       src0q, mmsize
+    sub        lend, mmsize/2
+    jg .loop
+    REP_RET
+%endmacro
+
+INIT_XMM sse2
+MIX_1_TO_2_S16P_FLT
+INIT_XMM sse4
+MIX_1_TO_2_S16P_FLT
+%if HAVE_AVX
+INIT_XMM avx
+MIX_1_TO_2_S16P_FLT
+%endif

libavresample/x86/audio_mix_init.c

@@ -40,6 +40,13 @@ extern void ff_mix_1_to_2_fltp_flt_sse(float **src, float **matrix, int len,
 extern void ff_mix_1_to_2_fltp_flt_avx(float **src, float **matrix, int len,
                                        int out_ch, int in_ch);
 
+extern void ff_mix_1_to_2_s16p_flt_sse2(int16_t **src, float **matrix, int len,
+                                        int out_ch, int in_ch);
+extern void ff_mix_1_to_2_s16p_flt_sse4(int16_t **src, float **matrix, int len,
+                                        int out_ch, int in_ch);
+extern void ff_mix_1_to_2_s16p_flt_avx (int16_t **src, float **matrix, int len,
+                                        int out_ch, int in_ch);
+
 av_cold void ff_audio_mix_init_x86(AudioMix *am)
 {
 #if HAVE_YASM
@@ -56,16 +63,22 @@ av_cold void ff_audio_mix_init_x86(AudioMix *am)
                               2, 1, 16, 8, "SSE2", ff_mix_2_to_1_s16p_flt_sse2);
         ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_Q8,
                               2, 1, 16, 8, "SSE2", ff_mix_2_to_1_s16p_q8_sse2);
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,
+                              1, 2, 16, 8, "SSE2", ff_mix_1_to_2_s16p_flt_sse2);
     }
     if (mm_flags & AV_CPU_FLAG_SSE4 && HAVE_SSE) {
         ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,
                               2, 1, 16, 8, "SSE4", ff_mix_2_to_1_s16p_flt_sse4);
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,
+                              1, 2, 16, 8, "SSE4", ff_mix_1_to_2_s16p_flt_sse4);
     }
     if (mm_flags & AV_CPU_FLAG_AVX && HAVE_AVX) {
         ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,
                               2, 1, 32, 16, "AVX", ff_mix_2_to_1_fltp_flt_avx);
         ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,
                               1, 2, 32, 8, "AVX", ff_mix_1_to_2_fltp_flt_avx);
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,
+                              1, 2, 16, 8, "AVX", ff_mix_1_to_2_s16p_flt_avx);
     }
 #endif
 }