[x264-devel] [PATCH 17/24] arm: Implement x264_deblock_h_chroma_mbaff_neon

[Martin Storsjö]

checkasm timing        Cortex-A7      A8     A9
deblock_chroma_420_mbaff_c    1944    1706   1526
deblock_chroma_420_mbaff_neon 1210    873    865
---
 common/arm/deblock-a.S |   57 ++++++++++++++++++++++++++++++++++++++++++++++++
 common/deblock.c       |    4 ++--
 2 files changed, 59 insertions(+), 2 deletions(-)

diff --git a/common/arm/deblock-a.S b/common/arm/deblock-a.S
index 26e95ed..d5210e0 100644
--- a/common/arm/deblock-a.S
+++ b/common/arm/deblock-a.S
@@ -309,6 +309,63 @@ function x264_deblock_h_chroma_422_neon
     pop             {pc}
 endfunc
 
+.macro h264_loop_filter_chroma8
+    vdup.8          d22, r2         @ alpha
+    vmovl.u8        q12, d24
+    vabd.u8         d26, d16, d0    @ abs(p0 - q0)
+    vabd.u8         d28, d18, d16   @ abs(p1 - p0)
+    vsubl.u8        q2,  d0,  d16
+    vsli.16         d24, d24, #8
+    vshl.i16        q2,  q2,  #2
+    vabd.u8         d30, d2,  d0    @ abs(q1 - q0)
+    vaddw.u8        q2,  q2,  d18
+    vclt.u8         d26, d26, d22   @ < alpha
+    vsubw.u8        q2,  q2,  d2
+    vdup.8          d22, r3         @ beta
+    vclt.s8         d20, d24, #0
+    vrshrn.i16      d4,  q2,  #3
+    vclt.u8         d28, d28, d22   @ < beta
+    vbic            d26, d26, d20
+    vclt.u8         d30, d30, d22   @ < beta
+    vand            d26, d26, d28
+    vneg.s8         d20, d24
+    vand            d26, d26, d30
+    vmin.s8         d4,  d4,  d24
+    vmovl.u8        q14, d16
+    vand            d4,  d4,  d26
+    vmax.s8         d4,  d4,  d20
+    vmovl.u8        q11, d0
+    vaddw.s8        q14, q14, d4
+    vsubw.s8        q11, q11, d4
+    vqmovun.s16     d16, q14
+    vqmovun.s16     d0,  q11
+.endm
+
+function x264_deblock_h_chroma_mbaff_neon
+    h264_loop_filter_start
+
+    sub             r0,  r0,  #4
+    vld1.8          {d18}, [r0], r1
+    vld1.8          {d16}, [r0], r1
+    vld1.8          {d0},  [r0], r1
+    vld1.8          {d2},  [r0], r1
+
+    TRANSPOSE4x4_16 d18, d16, d0, d2
+
+    h264_loop_filter_chroma8
+
+    vtrn.16         d16, d0
+
+    sub             r0,  r0,  r1, lsl #2
+    add             r0,  r0,  #2
+    vst1.32         {d16[0]}, [r0], r1
+    vst1.32         {d0[0]},  [r0], r1
+    vst1.32         {d16[1]}, [r0], r1
+    vst1.32         {d0[1]},  [r0]
+
+    bx              lr
+endfunc
+
 function x264_deblock_strength_neon
     ldr             ip,  [sp]
     vmov.i8         q8,  #0
diff --git a/common/deblock.c b/common/deblock.c
index 83bda62..1d398ad 100644
--- a/common/deblock.c
+++ b/common/deblock.c
@@ -740,8 +740,8 @@ void x264_deblock_strength_neon( uint8_t nnz[X264_SCAN8_SIZE], int8_t ref[2][X26
                                  int16_t mv[2][X264_SCAN8_LUMA_SIZE][2], uint8_t bs[2][8][4],
                                  int mvy_limit, int bframe );
 void x264_deblock_h_chroma_422_neon( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 );
-#if ARCH_AARCH64
 void x264_deblock_h_chroma_mbaff_neon( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 );
+#if ARCH_AARCH64
 void x264_deblock_h_chroma_intra_mbaff_neon( uint8_t *pix, intptr_t stride, int alpha, int beta );
 void x264_deblock_h_chroma_intra_neon( uint8_t *pix, intptr_t stride, int alpha, int beta );
 void x264_deblock_h_chroma_422_intra_neon( uint8_t *pix, intptr_t stride, int alpha, int beta );
@@ -874,8 +874,8 @@ void x264_deblock_init( int cpu, x264_deblock_function_t *pf, int b_mbaff )
         pf->deblock_chroma[1] = x264_deblock_v_chroma_neon;
         pf->deblock_h_chroma_420 = x264_deblock_h_chroma_neon;
         pf->deblock_h_chroma_422 = x264_deblock_h_chroma_422_neon;
-#if ARCH_AARCH64
         pf->deblock_chroma_420_mbaff = x264_deblock_h_chroma_mbaff_neon;
+#if ARCH_AARCH64
         pf->deblock_chroma_420_intra_mbaff = x264_deblock_h_chroma_intra_mbaff_neon;
         pf->deblock_h_chroma_420_intra = x264_deblock_h_chroma_intra_neon;
         pf->deblock_h_chroma_422_intra = x264_deblock_h_chroma_422_intra_neon;
-- 
1.7.10.4