ARM: NEON optimised float_to_int16

Originally committed as revision 16352 to svn://svn.ffmpeg.org/ffmpeg/trunk

2 files changed, 345 insertions, 0 deletions

diff --git a/libavcodec/arm/dsputil_neon.c b/libavcodec/arm/dsputil_neon.c
index fd8c73e72a..e50e16007d 100644
--- a/libavcodec/arm/dsputil_neon.c
+++ b/libavcodec/arm/dsputil_neon.c
@@ -104,6 +104,9 @@ void ff_h264_idct_add8_neon(uint8_t **dest, const int *block_offset,
                             DCTELEM *block, int stride,
                             const uint8_t nnzc[6*8]);
 
+void ff_float_to_int16_neon(int16_t *, const float *, long);
+void ff_float_to_int16_interleave_neon(int16_t *, const float **, long, int);
+
 void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx)
 {
     c->put_pixels_tab[0][0] = ff_put_pixels16_neon;
@@ -178,4 +181,9 @@ void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx)
     c->h264_idct_add16      = ff_h264_idct_add16_neon;
     c->h264_idct_add16intra = ff_h264_idct_add16intra_neon;
     c->h264_idct_add8       = ff_h264_idct_add8_neon;
+
+    if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
+        c->float_to_int16 = ff_float_to_int16_neon;
+        c->float_to_int16_interleave = ff_float_to_int16_interleave_neon;
+    }
 }
diff --git a/libavcodec/arm/dsputil_neon_s.S b/libavcodec/arm/dsputil_neon_s.S
index 1241cdc9ef..4f867149b9 100644
--- a/libavcodec/arm/dsputil_neon_s.S
+++ b/libavcodec/arm/dsputil_neon_s.S
@@ -272,3 +272,340 @@ function ff_put_h264_qpel8_mc00_neon, export=1
         pixfunc2 put_ pixels8_x2,   _no_rnd, vhadd.u8
         pixfunc2 put_ pixels8_y2,   _no_rnd, vhadd.u8
         pixfunc2 put_ pixels8_xy2,  _no_rnd, vshrn.u16, 1
+
+function ff_float_to_int16_neon, export=1
+        subs            r2,  r2,  #8
+        vld1.64         {d0-d1},  [r1,:128]!
+        vcvt.s32.f32    q8,  q0,  #16
+        vld1.64         {d2-d3},  [r1,:128]!
+        vcvt.s32.f32    q9,  q1,  #16
+        beq             3f
+        bics            ip,  r2,  #15
+        beq             2f
+1:      subs            ip,  ip,  #16
+        vshrn.s32       d4,  q8,  #16
+        vld1.64         {d0-d1},  [r1,:128]!
+        vcvt.s32.f32    q0,  q0,  #16
+        vshrn.s32       d5,  q9,  #16
+        vld1.64         {d2-d3},  [r1,:128]!
+        vcvt.s32.f32    q1,  q1,  #16
+        vshrn.s32       d6,  q0,  #16
+        vst1.64         {d4-d5},  [r0,:128]!
+        vshrn.s32       d7,  q1,  #16
+        vld1.64         {d16-d17},[r1,:128]!
+        vcvt.s32.f32    q8,  q8,  #16
+        vld1.64         {d18-d19},[r1,:128]!
+        vcvt.s32.f32    q9,  q9,  #16
+        vst1.64         {d6-d7},  [r0,:128]!
+        bne             1b
+        ands            r2,  r2,  #15
+        beq             3f
+2:      vld1.64         {d0-d1},  [r1,:128]!
+        vshrn.s32       d4,  q8,  #16
+        vcvt.s32.f32    q0,  q0,  #16
+        vld1.64         {d2-d3},  [r1,:128]!
+        vshrn.s32       d5,  q9,  #16
+        vcvt.s32.f32    q1,  q1,  #16
+        vshrn.s32       d6,  q0,  #16
+        vst1.64         {d4-d5},  [r0,:128]!
+        vshrn.s32       d7,  q1,  #16
+        vst1.64         {d6-d7},  [r0,:128]!
+        bx              lr
+3:      vshrn.s32       d4,  q8,  #16
+        vshrn.s32       d5,  q9,  #16
+        vst1.64         {d4-d5},  [r0,:128]!
+        bx              lr
+        .endfunc
+
+function ff_float_to_int16_interleave_neon, export=1
+        cmp             r3, #2
+        ldrlt           r1, [r1]
+        blt             ff_float_to_int16_neon
+        bne             4f
+
+        ldr             r3, [r1]
+        ldr             r1, [r1, #4]
+
+        subs            r2,  r2,  #8
+        vld1.64         {d0-d1},  [r3,:128]!
+        vcvt.s32.f32    q8,  q0,  #16
+        vld1.64         {d2-d3},  [r3,:128]!
+        vcvt.s32.f32    q9,  q1,  #16
+        vld1.64         {d20-d21},[r1,:128]!
+        vcvt.s32.f32    q10, q10, #16
+        vld1.64         {d22-d23},[r1,:128]!
+        vcvt.s32.f32    q11, q11, #16
+        beq             3f
+        bics            ip,  r2,  #15
+        beq             2f
+1:      subs            ip,  ip,  #16
+        vld1.64         {d0-d1},  [r3,:128]!
+        vcvt.s32.f32    q0,  q0,  #16
+        vsri.32         q10, q8,  #16
+        vld1.64         {d2-d3},  [r3,:128]!
+        vcvt.s32.f32    q1,  q1,  #16
+        vld1.64         {d24-d25},[r1,:128]!
+        vcvt.s32.f32    q12, q12, #16
+        vld1.64         {d26-d27},[r1,:128]!
+        vsri.32         q11, q9,  #16
+        vst1.64         {d20-d21},[r0,:128]!
+        vcvt.s32.f32    q13, q13, #16
+        vst1.64         {d22-d23},[r0,:128]!
+        vsri.32         q12, q0,  #16
+        vld1.64         {d16-d17},[r3,:128]!
+        vsri.32         q13, q1,  #16
+        vst1.64         {d24-d25},[r0,:128]!
+        vcvt.s32.f32    q8,  q8,  #16
+        vld1.64         {d18-d19},[r3,:128]!
+        vcvt.s32.f32    q9,  q9,  #16
+        vld1.64         {d20-d21},[r1,:128]!
+        vcvt.s32.f32    q10, q10, #16
+        vld1.64         {d22-d23},[r1,:128]!
+        vcvt.s32.f32    q11, q11, #16
+        vst1.64         {d26-d27},[r0,:128]!
+        bne             1b
+        ands            r2,  r2,  #15
+        beq             3f
+2:      vsri.32         q10, q8,  #16
+        vld1.64         {d0-d1},  [r3,:128]!
+        vcvt.s32.f32    q0,  q0,  #16
+        vld1.64         {d2-d3},  [r3,:128]!
+        vcvt.s32.f32    q1,  q1,  #16
+        vld1.64         {d24-d25},[r1,:128]!
+        vcvt.s32.f32    q12, q12, #16
+        vsri.32         q11, q9,  #16
+        vld1.64         {d26-d27},[r1,:128]!
+        vcvt.s32.f32    q13, q13, #16
+        vst1.64         {d20-d21},[r0,:128]!
+        vsri.32         q12, q0,  #16
+        vst1.64         {d22-d23},[r0,:128]!
+        vsri.32         q13, q1,  #16
+        vst1.64         {d24-d27},[r0,:128]!
+        bx              lr
+3:      vsri.32         q10, q8,  #16
+        vsri.32         q11, q9,  #16
+        vst1.64         {d20-d23},[r0,:128]!
+        bx              lr
+
+4:      push            {r4-r8,lr}
+        cmp             r3,  #4
+        lsl             ip,  r3,  #1
+        blt             4f
+
+        @ 4 channels
+5:      ldmia           r1!, {r4-r7}
+        mov             lr,  r2
+        mov             r8,  r0
+        vld1.64         {d16-d17},[r4,:128]!
+        vcvt.s32.f32    q8,  q8,  #16
+        vld1.64         {d18-d19},[r5,:128]!
+        vcvt.s32.f32    q9,  q9,  #16
+        vld1.64         {d20-d21},[r6,:128]!
+        vcvt.s32.f32    q10, q10, #16
+        vld1.64         {d22-d23},[r7,:128]!
+        vcvt.s32.f32    q11, q11, #16
+6:      subs            lr,  lr,  #8
+        vld1.64         {d0-d1},  [r4,:128]!
+        vcvt.s32.f32    q0,  q0,  #16
+        vsri.32         q9,  q8,  #16
+        vld1.64         {d2-d3},  [r5,:128]!
+        vcvt.s32.f32    q1,  q1,  #16
+        vsri.32         q11, q10, #16
+        vld1.64         {d4-d5},  [r6,:128]!
+        vcvt.s32.f32    q2,  q2,  #16
+        vzip.32         d18, d22
+        vld1.64         {d6-d7},  [r7,:128]!
+        vcvt.s32.f32    q3,  q3,  #16
+        vzip.32         d19, d23
+        vst1.64         {d18},    [r8], ip
+        vsri.32         q1,  q0,  #16
+        vst1.64         {d22},    [r8], ip
+        vsri.32         q3,  q2,  #16
+        vst1.64         {d19},    [r8], ip
+        vzip.32         d2,  d6
+        vst1.64         {d23},    [r8], ip
+        vzip.32         d3,  d7
+        beq             7f
+        vld1.64         {d16-d17},[r4,:128]!
+        vcvt.s32.f32    q8,  q8,  #16
+        vst1.64         {d2},     [r8], ip
+        vld1.64         {d18-d19},[r5,:128]!
+        vcvt.s32.f32    q9,  q9,  #16
+        vst1.64         {d6},     [r8], ip
+        vld1.64         {d20-d21},[r6,:128]!
+        vcvt.s32.f32    q10, q10, #16
+        vst1.64         {d3},     [r8], ip
+        vld1.64         {d22-d23},[r7,:128]!
+        vcvt.s32.f32    q11, q11, #16
+        vst1.64         {d7},     [r8], ip
+        b               6b
+7:      vst1.64         {d2},     [r8], ip
+        vst1.64         {d6},     [r8], ip
+        vst1.64         {d3},     [r8], ip
+        vst1.64         {d7},     [r8], ip
+        subs            r3,  r3,  #4
+        popeq           {r4-r8,pc}
+        cmp             r3,  #4
+        add             r0,  r0,  #8
+        bge             5b
+
+        @ 2 channels
+4:      cmp             r3,  #2
+        blt             4f
+        ldmia           r1!, {r4-r5}
+        mov             lr,  r2
+        mov             r8,  r0
+        tst             lr,  #8
+        vld1.64         {d16-d17},[r4,:128]!
+        vcvt.s32.f32    q8,  q8,  #16
+        vld1.64         {d18-d19},[r5,:128]!
+        vcvt.s32.f32    q9,  q9,  #16
+        vld1.64         {d20-d21},[r4,:128]!
+        vcvt.s32.f32    q10, q10, #16
+        vld1.64         {d22-d23},[r5,:128]!
+        vcvt.s32.f32    q11, q11, #16
+        beq             6f
+        subs            lr,  lr,  #8
+        beq             7f
+        vsri.32         d18, d16, #16
+        vsri.32         d19, d17, #16
+        vld1.64         {d16-d17},[r4,:128]!
+        vcvt.s32.f32    q8,  q8,  #16
+        vst1.32         {d18[0]}, [r8], ip
+        vsri.32         d22, d20, #16
+        vst1.32         {d18[1]}, [r8], ip
+        vsri.32         d23, d21, #16
+        vst1.32         {d19[0]}, [r8], ip
+        vst1.32         {d19[1]}, [r8], ip
+        vld1.64         {d18-d19},[r5,:128]!
+        vcvt.s32.f32    q9,  q9,  #16
+        vst1.32         {d22[0]}, [r8], ip
+        vst1.32         {d22[1]}, [r8], ip
+        vld1.64         {d20-d21},[r4,:128]!
+        vcvt.s32.f32    q10, q10, #16
+        vst1.32         {d23[0]}, [r8], ip
+        vst1.32         {d23[1]}, [r8], ip
+        vld1.64         {d22-d23},[r5,:128]!
+        vcvt.s32.f32    q11, q11, #16
+6:      subs            lr,  lr,  #16
+        vld1.64         {d0-d1},  [r4,:128]!
+        vcvt.s32.f32    q0,  q0,  #16
+        vsri.32         d18, d16, #16
+        vld1.64         {d2-d3},  [r5,:128]!
+        vcvt.s32.f32    q1,  q1,  #16
+        vsri.32         d19, d17, #16
+        vld1.64         {d4-d5},  [r4,:128]!
+        vcvt.s32.f32    q2,  q2,  #16
+        vld1.64         {d6-d7},  [r5,:128]!
+        vcvt.s32.f32    q3,  q3,  #16
+        vst1.32         {d18[0]}, [r8], ip
+        vsri.32         d22, d20, #16
+        vst1.32         {d18[1]}, [r8], ip
+        vsri.32         d23, d21, #16
+        vst1.32         {d19[0]}, [r8], ip
+        vsri.32         d2,  d0,  #16
+        vst1.32         {d19[1]}, [r8], ip
+        vsri.32         d3,  d1,  #16
+        vst1.32         {d22[0]}, [r8], ip
+        vsri.32         d6,  d4,  #16
+        vst1.32         {d22[1]}, [r8], ip
+        vsri.32         d7,  d5,  #16
+        vst1.32         {d23[0]}, [r8], ip
+        vst1.32         {d23[1]}, [r8], ip
+        beq             6f
+        vld1.64         {d16-d17},[r4,:128]!
+        vcvt.s32.f32    q8,  q8,  #16
+        vst1.32         {d2[0]},  [r8], ip
+        vst1.32         {d2[1]},  [r8], ip
+        vld1.64         {d18-d19},[r5,:128]!
+        vcvt.s32.f32    q9,  q9,  #16
+        vst1.32         {d3[0]},  [r8], ip
+        vst1.32         {d3[1]},  [r8], ip
+        vld1.64         {d20-d21},[r4,:128]!
+        vcvt.s32.f32    q10, q10, #16
+        vst1.32         {d6[0]},  [r8], ip
+        vst1.32         {d6[1]},  [r8], ip
+        vld1.64         {d22-d23},[r5,:128]!
+        vcvt.s32.f32    q11, q11, #16
+        vst1.32         {d7[0]},  [r8], ip
+        vst1.32         {d7[1]},  [r8], ip
+        bgt             6b
+6:      vst1.32         {d2[0]},  [r8], ip
+        vst1.32         {d2[1]},  [r8], ip
+        vst1.32         {d3[0]},  [r8], ip
+        vst1.32         {d3[1]},  [r8], ip
+        vst1.32         {d6[0]},  [r8], ip
+        vst1.32         {d6[1]},  [r8], ip
+        vst1.32         {d7[0]},  [r8], ip
+        vst1.32         {d7[1]},  [r8], ip
+        b               8f
+7:      vsri.32         d18, d16, #16
+        vsri.32         d19, d17, #16
+        vst1.32         {d18[0]}, [r8], ip
+        vsri.32         d22, d20, #16
+        vst1.32         {d18[1]}, [r8], ip
+        vsri.32         d23, d21, #16
+        vst1.32         {d19[0]}, [r8], ip
+        vst1.32         {d19[1]}, [r8], ip
+        vst1.32         {d22[0]}, [r8], ip
+        vst1.32         {d22[1]}, [r8], ip
+        vst1.32         {d23[0]}, [r8], ip
+        vst1.32         {d23[1]}, [r8], ip
+8:      subs            r3,  r3,  #2
+        add             r0,  r0,  #4
+        popeq           {r4-r8,pc}
+
+        @ 1 channel
+4:      ldr             r4,  [r1],#4
+        tst             r2,  #8
+        mov             lr,  r2
+        mov             r5,  r0
+        vld1.64         {d0-d1},  [r4,:128]!
+        vcvt.s32.f32    q0,  q0,  #16
+        vld1.64         {d2-d3},  [r4,:128]!
+        vcvt.s32.f32    q1,  q1,  #16
+        bne             8f
+6:      subs            lr,  lr,  #16
+        vld1.64         {d4-d5},  [r4,:128]!
+        vcvt.s32.f32    q2,  q2,  #16
+        vld1.64         {d6-d7},  [r4,:128]!
+        vcvt.s32.f32    q3,  q3,  #16
+        vst1.16         {d0[1]},  [r5,:16], ip
+        vst1.16         {d0[3]},  [r5,:16], ip
+        vst1.16         {d1[1]},  [r5,:16], ip
+        vst1.16         {d1[3]},  [r5,:16], ip
+        vst1.16         {d2[1]},  [r5,:16], ip
+        vst1.16         {d2[3]},  [r5,:16], ip
+        vst1.16         {d3[1]},  [r5,:16], ip
+        vst1.16         {d3[3]},  [r5,:16], ip
+        beq             7f
+        vld1.64         {d0-d1},  [r4,:128]!
+        vcvt.s32.f32    q0,  q0,  #16
+        vld1.64         {d2-d3},  [r4,:128]!
+        vcvt.s32.f32    q1,  q1,  #16
+7:      vst1.16         {d4[1]},  [r5,:16], ip
+        vst1.16         {d4[3]},  [r5,:16], ip
+        vst1.16         {d5[1]},  [r5,:16], ip
+        vst1.16         {d5[3]},  [r5,:16], ip
+        vst1.16         {d6[1]},  [r5,:16], ip
+        vst1.16         {d6[3]},  [r5,:16], ip
+        vst1.16         {d7[1]},  [r5,:16], ip
+        vst1.16         {d7[3]},  [r5,:16], ip
+        bgt             6b
+        pop             {r4-r8,pc}
+8:      subs            lr,  lr,  #8
+        vst1.16         {d0[1]},  [r5,:16], ip
+        vst1.16         {d0[3]},  [r5,:16], ip
+        vst1.16         {d1[1]},  [r5,:16], ip
+        vst1.16         {d1[3]},  [r5,:16], ip
+        vst1.16         {d2[1]},  [r5,:16], ip
+        vst1.16         {d2[3]},  [r5,:16], ip
+        vst1.16         {d3[1]},  [r5,:16], ip
+        vst1.16         {d3[3]},  [r5,:16], ip
+        popeq           {r4-r8,pc}
+        vld1.64         {d0-d1},  [r4,:128]!
+        vcvt.s32.f32    q0,  q0,  #16
+        vld1.64         {d2-d3},  [r4,:128]!
+        vcvt.s32.f32    q1,  q1,  #16
+        b               6b
+        .endfunc