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

libavresample/x86/audio_mix.asm

@@ -150,3 +150,37 @@ cglobal mix_2_to_1_s16p_q8, 3,4,6, src, matrix, len, src1
     sub        lend, mmsize/2
     jg .loop
     REP_RET
+
+;-----------------------------------------------------------------------------
+; void ff_mix_1_to_2_fltp_flt(float **src, float **matrix, int len,
+;                             int out_ch, int in_ch);
+;-----------------------------------------------------------------------------
+
+%macro MIX_1_TO_2_FLTP_FLT 0
+cglobal mix_1_to_2_fltp_flt, 3,5,4, src0, matrix0, len, src1, matrix1
+    mov       src1q, [src0q+gprsize]
+    mov       src0q, [src0q]
+    sub       src1q, src0q
+    mov    matrix1q, [matrix0q+gprsize]
+    mov    matrix0q, [matrix0q]
+    VBROADCASTSS m2, [matrix0q]
+    VBROADCASTSS m3, [matrix1q]
+    ALIGN 16
+.loop:
+    mova         m0, [src0q]
+    mulps        m1, m0, m3
+    mulps        m0, m0, m2
+    mova  [src0q      ], m0
+    mova  [src0q+src1q], m1
+    add       src0q, mmsize
+    sub        lend, mmsize/4
+    jg .loop
+    REP_RET
+%endmacro
+
+INIT_XMM sse
+MIX_1_TO_2_FLTP_FLT
+%if HAVE_AVX
+INIT_YMM avx
+MIX_1_TO_2_FLTP_FLT
+%endif

libavresample/x86/audio_mix_init.c

@@ -35,6 +35,11 @@ extern void ff_mix_2_to_1_s16p_flt_sse4(int16_t **src, float **matrix, int len,
 extern void ff_mix_2_to_1_s16p_q8_sse2(int16_t **src, int16_t **matrix,
                                        int len, int out_ch, int in_ch);
 
+extern void ff_mix_1_to_2_fltp_flt_sse(float **src, float **matrix, int len,
+                                       int out_ch, int in_ch);
+extern void ff_mix_1_to_2_fltp_flt_avx(float **src, float **matrix, int len,
+                                       int out_ch, int in_ch);
+
 av_cold void ff_audio_mix_init_x86(AudioMix *am)
 {
 #if HAVE_YASM
@@ -43,6 +48,8 @@ av_cold void ff_audio_mix_init_x86(AudioMix *am)
     if (mm_flags & AV_CPU_FLAG_SSE && HAVE_SSE) {
         ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,
                               2, 1, 16, 8, "SSE", ff_mix_2_to_1_fltp_flt_sse);
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,
+                              1, 2, 16, 4, "SSE", ff_mix_1_to_2_fltp_flt_sse);
     }
     if (mm_flags & AV_CPU_FLAG_SSE2 && HAVE_SSE) {
         ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,
@@ -57,6 +64,8 @@ av_cold void ff_audio_mix_init_x86(AudioMix *am)
     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);
     }
 #endif
 }