ARM: libswresample: NEON optimised generic fltp to s16 conversion

Adapted to swr by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

2 files changed, 242 insertions, 0 deletions

diff --git a/libswresample/arm/audio_convert_init.c b/libswresample/arm/audio_convert_init.c
index a52ea5a196..9fdb174608 100644
--- a/libswresample/arm/audio_convert_init.c
+++ b/libswresample/arm/audio_convert_init.c
@@ -28,6 +28,7 @@
 
 void swri_oldapi_conv_flt_to_s16_neon(int16_t *dst, const float *src, int len);
 void swri_oldapi_conv_fltp_to_s16_2ch_neon(int16_t *dst, float *const *src, int len, int channels);
+void swri_oldapi_conv_fltp_to_s16_nch_neon(int16_t *dst, float *const *src, int len, int channels);
 
 static void conv_flt_to_s16_neon(uint8_t **dst, const uint8_t **src, int len){
     swri_oldapi_conv_flt_to_s16_neon((int16_t*)*dst, (const float*)*src, len);
@@ -37,6 +38,12 @@ static void conv_fltp_to_s16_2ch_neon(uint8_t **dst, const uint8_t **src, int le
     swri_oldapi_conv_fltp_to_s16_2ch_neon((int16_t*)*dst, (float *const*)src, len, 2);
 }
 
+static void conv_fltp_to_s16_nch_neon(uint8_t **dst, const uint8_t **src, int len){
+    int channels;
+    for(channels=3; channels<SWR_CH_MAX && src[channels]; channels++)
+        ;
+    swri_oldapi_conv_fltp_to_s16_nch_neon((int16_t*)*dst, (float *const*)src, len, channels);
+}
 
 av_cold void swri_audio_convert_init_arm(struct AudioConvert *ac,
                                        enum AVSampleFormat out_fmt,
@@ -52,5 +59,7 @@ av_cold void swri_audio_convert_init_arm(struct AudioConvert *ac,
             ac->simd_f = conv_flt_to_s16_neon;
         if(out_fmt == AV_SAMPLE_FMT_S16 && in_fmt == AV_SAMPLE_FMT_FLTP && channels == 2)
             ac->simd_f = conv_fltp_to_s16_2ch_neon;
+        if(out_fmt == AV_SAMPLE_FMT_S16 && in_fmt == AV_SAMPLE_FMT_FLTP && channels >  2)
+            ac->simd_f = conv_fltp_to_s16_nch_neon;
     }
 }
diff --git a/libswresample/arm/audio_convert_neon.S b/libswresample/arm/audio_convert_neon.S
index f6099dac93..471a2d864f 100644
--- a/libswresample/arm/audio_convert_neon.S
+++ b/libswresample/arm/audio_convert_neon.S
@@ -128,3 +128,236 @@ function swri_oldapi_conv_fltp_to_s16_2ch_neon, export=1
         vst1.16         {q10-q11},[r0,:128]!
         bx              lr
 endfunc
+
+function swri_oldapi_conv_fltp_to_s16_nch_neon, export=1
+        cmp             r3,  #2
+        itt             lt
+        ldrlt           r1,  [r1]
+        blt             swri_oldapi_conv_flt_to_s16_neon
+        beq             swri_oldapi_conv_fltp_to_s16_2ch_neon
+
+        push            {r4-r8, lr}
+        cmp             r3,  #4
+        lsl             r12, r3,  #1
+        blt             4f
+
+        @ 4 channels
+5:      ldm             r1!, {r4-r7}
+        mov             lr,  r2
+        mov             r8,  r0
+        vld1.32         {q8},     [r4,:128]!
+        vcvt.s32.f32    q8,  q8,  #31
+        vld1.32         {q9},     [r5,:128]!
+        vcvt.s32.f32    q9,  q9,  #31
+        vld1.32         {q10},    [r6,:128]!
+        vcvt.s32.f32    q10, q10, #31
+        vld1.32         {q11},    [r7,:128]!
+        vcvt.s32.f32    q11, q11, #31
+6:      subs            lr,  lr,  #8
+        vld1.32         {q0},     [r4,:128]!
+        vcvt.s32.f32    q0,  q0,  #31
+        vsri.32         q9,  q8,  #16
+        vld1.32         {q1},     [r5,:128]!
+        vcvt.s32.f32    q1,  q1,  #31
+        vsri.32         q11, q10, #16
+        vld1.32         {q2},     [r6,:128]!
+        vcvt.s32.f32    q2,  q2,  #31
+        vzip.32         d18, d22
+        vld1.32         {q3},     [r7,:128]!
+        vcvt.s32.f32    q3,  q3,  #31
+        vzip.32         d19, d23
+        vst1.16         {d18},    [r8], r12
+        vsri.32         q1,  q0,  #16
+        vst1.16         {d22},    [r8], r12
+        vsri.32         q3,  q2,  #16
+        vst1.16         {d19},    [r8], r12
+        vzip.32         d2,  d6
+        vst1.16         {d23},    [r8], r12
+        vzip.32         d3,  d7
+        beq             7f
+        vld1.32         {q8},     [r4,:128]!
+        vcvt.s32.f32    q8,  q8,  #31
+        vst1.16         {d2},     [r8], r12
+        vld1.32         {q9},     [r5,:128]!
+        vcvt.s32.f32    q9,  q9,  #31
+        vst1.16         {d6},     [r8], r12
+        vld1.32         {q10},    [r6,:128]!
+        vcvt.s32.f32    q10, q10, #31
+        vst1.16         {d3},     [r8], r12
+        vld1.32         {q11},    [r7,:128]!
+        vcvt.s32.f32    q11, q11, #31
+        vst1.16         {d7},     [r8], r12
+        b               6b
+7:      vst1.16         {d2},     [r8], r12
+        vst1.16         {d6},     [r8], r12
+        vst1.16         {d3},     [r8], r12
+        vst1.16         {d7},     [r8], r12
+        subs            r3,  r3,  #4
+        it              eq
+        popeq           {r4-r8, pc}
+        cmp             r3,  #4
+        add             r0,  r0,  #8
+        bge             5b
+
+        @ 2 channels
+4:      cmp             r3,  #2
+        blt             4f
+        ldm             r1!, {r4-r5}
+        mov             lr,  r2
+        mov             r8,  r0
+        tst             lr,  #8
+        vld1.32         {q8},     [r4,:128]!
+        vcvt.s32.f32    q8,  q8,  #31
+        vld1.32         {q9},     [r5,:128]!
+        vcvt.s32.f32    q9,  q9,  #31
+        vld1.32         {q10},    [r4,:128]!
+        vcvt.s32.f32    q10, q10, #31
+        vld1.32         {q11},    [r5,:128]!
+        vcvt.s32.f32    q11, q11, #31
+        beq             6f
+        subs            lr,  lr,  #8
+        beq             7f
+        vsri.32         d18, d16, #16
+        vsri.32         d19, d17, #16
+        vld1.32         {q8},     [r4,:128]!
+        vcvt.s32.f32    q8,  q8,  #31
+        vst1.32         {d18[0]}, [r8], r12
+        vsri.32         d22, d20, #16
+        vst1.32         {d18[1]}, [r8], r12
+        vsri.32         d23, d21, #16
+        vst1.32         {d19[0]}, [r8], r12
+        vst1.32         {d19[1]}, [r8], r12
+        vld1.32         {q9},     [r5,:128]!
+        vcvt.s32.f32    q9,  q9,  #31
+        vst1.32         {d22[0]}, [r8], r12
+        vst1.32         {d22[1]}, [r8], r12
+        vld1.32         {q10},    [r4,:128]!
+        vcvt.s32.f32    q10, q10, #31
+        vst1.32         {d23[0]}, [r8], r12
+        vst1.32         {d23[1]}, [r8], r12
+        vld1.32         {q11},    [r5,:128]!
+        vcvt.s32.f32    q11, q11, #31
+6:      subs            lr,  lr,  #16
+        vld1.32         {q0},     [r4,:128]!
+        vcvt.s32.f32    q0,  q0,  #31
+        vsri.32         d18, d16, #16
+        vld1.32         {q1},     [r5,:128]!
+        vcvt.s32.f32    q1,  q1,  #31
+        vsri.32         d19, d17, #16
+        vld1.32         {q2},     [r4,:128]!
+        vcvt.s32.f32    q2,  q2,  #31
+        vld1.32         {q3},     [r5,:128]!
+        vcvt.s32.f32    q3,  q3,  #31
+        vst1.32         {d18[0]}, [r8], r12
+        vsri.32         d22, d20, #16
+        vst1.32         {d18[1]}, [r8], r12
+        vsri.32         d23, d21, #16
+        vst1.32         {d19[0]}, [r8], r12
+        vsri.32         d2,  d0,  #16
+        vst1.32         {d19[1]}, [r8], r12
+        vsri.32         d3,  d1,  #16
+        vst1.32         {d22[0]}, [r8], r12
+        vsri.32         d6,  d4,  #16
+        vst1.32         {d22[1]}, [r8], r12
+        vsri.32         d7,  d5,  #16
+        vst1.32         {d23[0]}, [r8], r12
+        vst1.32         {d23[1]}, [r8], r12
+        beq             6f
+        vld1.32         {q8},     [r4,:128]!
+        vcvt.s32.f32    q8,  q8,  #31
+        vst1.32         {d2[0]},  [r8], r12
+        vst1.32         {d2[1]},  [r8], r12
+        vld1.32         {q9},     [r5,:128]!
+        vcvt.s32.f32    q9,  q9,  #31
+        vst1.32         {d3[0]},  [r8], r12
+        vst1.32         {d3[1]},  [r8], r12
+        vld1.32         {q10},    [r4,:128]!
+        vcvt.s32.f32    q10, q10, #31
+        vst1.32         {d6[0]},  [r8], r12
+        vst1.32         {d6[1]},  [r8], r12
+        vld1.32         {q11},    [r5,:128]!
+        vcvt.s32.f32    q11, q11, #31
+        vst1.32         {d7[0]},  [r8], r12
+        vst1.32         {d7[1]},  [r8], r12
+        bgt             6b
+6:      vst1.32         {d2[0]},  [r8], r12
+        vst1.32         {d2[1]},  [r8], r12
+        vst1.32         {d3[0]},  [r8], r12
+        vst1.32         {d3[1]},  [r8], r12
+        vst1.32         {d6[0]},  [r8], r12
+        vst1.32         {d6[1]},  [r8], r12
+        vst1.32         {d7[0]},  [r8], r12
+        vst1.32         {d7[1]},  [r8], r12
+        b               8f
+7:      vsri.32         d18, d16, #16
+        vsri.32         d19, d17, #16
+        vst1.32         {d18[0]}, [r8], r12
+        vsri.32         d22, d20, #16
+        vst1.32         {d18[1]}, [r8], r12
+        vsri.32         d23, d21, #16
+        vst1.32         {d19[0]}, [r8], r12
+        vst1.32         {d19[1]}, [r8], r12
+        vst1.32         {d22[0]}, [r8], r12
+        vst1.32         {d22[1]}, [r8], r12
+        vst1.32         {d23[0]}, [r8], r12
+        vst1.32         {d23[1]}, [r8], r12
+8:      subs            r3,  r3,  #2
+        add             r0,  r0,  #4
+        it              eq
+        popeq           {r4-r8, pc}
+
+        @ 1 channel
+4:      ldr             r4,  [r1]
+        tst             r2,  #8
+        mov             lr,  r2
+        mov             r5,  r0
+        vld1.32         {q0},     [r4,:128]!
+        vcvt.s32.f32    q0,  q0,  #31
+        vld1.32         {q1},     [r4,:128]!
+        vcvt.s32.f32    q1,  q1,  #31
+        bne             8f
+6:      subs            lr,  lr,  #16
+        vld1.32         {q2},     [r4,:128]!
+        vcvt.s32.f32    q2,  q2,  #31
+        vld1.32         {q3},     [r4,:128]!
+        vcvt.s32.f32    q3,  q3,  #31
+        vst1.16         {d0[1]},  [r5,:16], r12
+        vst1.16         {d0[3]},  [r5,:16], r12
+        vst1.16         {d1[1]},  [r5,:16], r12
+        vst1.16         {d1[3]},  [r5,:16], r12
+        vst1.16         {d2[1]},  [r5,:16], r12
+        vst1.16         {d2[3]},  [r5,:16], r12
+        vst1.16         {d3[1]},  [r5,:16], r12
+        vst1.16         {d3[3]},  [r5,:16], r12
+        beq             7f
+        vld1.32         {q0},     [r4,:128]!
+        vcvt.s32.f32    q0,  q0,  #31
+        vld1.32         {q1},     [r4,:128]!
+        vcvt.s32.f32    q1,  q1,  #31
+7:      vst1.16         {d4[1]},  [r5,:16], r12
+        vst1.16         {d4[3]},  [r5,:16], r12
+        vst1.16         {d5[1]},  [r5,:16], r12
+        vst1.16         {d5[3]},  [r5,:16], r12
+        vst1.16         {d6[1]},  [r5,:16], r12
+        vst1.16         {d6[3]},  [r5,:16], r12
+        vst1.16         {d7[1]},  [r5,:16], r12
+        vst1.16         {d7[3]},  [r5,:16], r12
+        bgt             6b
+        pop             {r4-r8, pc}
+8:      subs            lr,  lr,  #8
+        vst1.16         {d0[1]},  [r5,:16], r12
+        vst1.16         {d0[3]},  [r5,:16], r12
+        vst1.16         {d1[1]},  [r5,:16], r12
+        vst1.16         {d1[3]},  [r5,:16], r12
+        vst1.16         {d2[1]},  [r5,:16], r12
+        vst1.16         {d2[3]},  [r5,:16], r12
+        vst1.16         {d3[1]},  [r5,:16], r12
+        vst1.16         {d3[3]},  [r5,:16], r12
+        it              eq
+        popeq           {r4-r8, pc}
+        vld1.32         {q0},     [r4,:128]!
+        vcvt.s32.f32    q0,  q0,  #31
+        vld1.32         {q1},     [r4,:128]!
+        vcvt.s32.f32    q1,  q1,  #31
+        b               6b
+endfunc