[x264-devel] [PATCH] x264_denoise_dct_neon

[George Stephanos]

---
 common/arm/quant-a.S |   28 ++++++++++++++++++++++++++++
 common/arm/quant.h   |    2 ++
 common/quant.c       |    1 +
 3 files changed, 31 insertions(+), 0 deletions(-)

diff --git a/common/arm/quant-a.S b/common/arm/quant-a.S
index 3d435cf..2b23604 100644
--- a/common/arm/quant-a.S
+++ b/common/arm/quant-a.S
@@ -680,3 +680,31 @@ function x264_decimate_score64_neon
     bx          lr
 .endfunc
 
+function x264_denoise_dct_neon
+1:
+    vld1.64     {d0-d3}, [r0,:128]
+    vshr.s16    q8, q0, #15
+    vshr.s16    q9, q1, #15
+    vabs.s16    q0, q0
+    vabs.s16    q1, q1
+    vld1.64     {d4-d7}, [r1,:128]!
+    vld1.64     {d20-d23}, [r1,:128]
+    vaddw.s16   q2, d0
+    vaddw.s16   q3, d1
+    sub         r1, #32
+    vaddw.s16   q10, d2
+    vaddw.s16   q11, d3
+    vst1.64     {d4-d7}, [r1,:128]!
+    vld1.64     {d4-d7}, [r2,:128]!
+    vst1.64     {d20-d23}, [r1,:128]!
+    vqsub.u16   q0, q2
+    vqsub.u16   q1, q3
+    veor        q0, q8
+    veor        q1, q9
+    vsub.s16    q0, q8
+    vsub.s16    q1, q9
+    subs        r3, #16
+    vst1.64     {d0-d3}, [r0,:128]!
+    bgt         1b
+    bx          lr
+.endfunc
diff --git a/common/arm/quant.h b/common/arm/quant.h
index ee78d1c..2079abc 100644
--- a/common/arm/quant.h
+++ b/common/arm/quant.h
@@ -51,4 +51,6 @@ int x264_decimate_score15_neon( int16_t * );
 int x264_decimate_score16_neon( int16_t * );
 int x264_decimate_score64_neon( int16_t * );
 
+void x264_denoise_dct_neon( int16_t *, uint32_t *, udctcoef *offset, int size );
+
 #endif
diff --git a/common/quant.c b/common/quant.c
index 5d85c81..a986666 100644
--- a/common/quant.c
+++ b/common/quant.c
@@ -713,6 +713,7 @@ void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf )
         pf->decimate_score15 = x264_decimate_score15_neon;
         pf->decimate_score16 = x264_decimate_score16_neon;
         pf->decimate_score64 = x264_decimate_score64_neon;
+        pf->denoise_dct = x264_denoise_dct_neon;
     }
 #endif
 #endif // HIGH_BIT_DEPTH
-- 
1.7.4.1