[x264-devel] [PATCH 18/23] aarch64: x264_denoise_dct_neon

[Janne Grunau]

3.5 times faster.
---
 common/aarch64/quant-a.S | 27 ++++++++++++++++++++++++++-
 common/aarch64/quant.h   |  2 ++
 common/quant.c           |  1 +
 3 files changed, 29 insertions(+), 1 deletion(-)

diff --git a/common/aarch64/quant-a.S b/common/aarch64/quant-a.S
index f80946d..9dfe0b6 100644
--- a/common/aarch64/quant-a.S
+++ b/common/aarch64/quant-a.S
@@ -572,4 +572,29 @@ endfunc
 
 X264_COEFF_LEVEL_RUN 8
 X264_COEFF_LEVEL_RUN 15
-X264_COEFF_LEVEL_RUN 16
\ No newline at end of file
+X264_COEFF_LEVEL_RUN 16
+
+function x264_denoise_dct_neon, export=1
+1:  subs        w3,  w3,  #16
+    ld1         {v0.8h,v1.8h}, [x0]
+    ld1         {v4.4s,v5.4s,v6.4s,v7.4s}, [x1]
+    abs         v16.8h,  v0.8h
+    abs         v17.8h,  v1.8h
+    ld1         {v2.8h,v3.8h}, [x2], #32
+    cmlt        v18.8h,  v0.8h,   #0
+    cmlt        v19.8h,  v1.8h,   #0
+    uaddw       v4.4s,   v4.4s,   v16.4h
+    uaddw2      v5.4s,   v5.4s,   v16.8h
+    uqsub       v20.8h,  v16.8h,  v2.8h
+    uqsub       v21.8h,  v17.8h,  v3.8h
+    uaddw       v6.4s,   v6.4s,   v17.4h
+    uaddw2      v7.4s,   v7.4s,   v17.8h
+    neg         v22.8h,  v20.8h
+    neg         v23.8h,  v21.8h
+    bsl         v18.16b, v22.16b, v20.16b
+    bsl         v19.16b, v23.16b, v21.16b
+    st1         {v4.4s,v5.4s,v6.4s,v7.4s}, [x1], #64
+    st1         {v18.8h,v19.8h}, [x0], #32
+    b.gt        1b
+    ret
+endfunc
diff --git a/common/aarch64/quant.h b/common/aarch64/quant.h
index ef9e200..6858e54 100644
--- a/common/aarch64/quant.h
+++ b/common/aarch64/quant.h
@@ -52,4 +52,6 @@ int x264_coeff_level_run4_aarch64( int16_t *, x264_run_level_t * );
 int x264_coeff_level_run8_neon( int16_t *, x264_run_level_t * );
 int x264_coeff_level_run15_neon( int16_t *, x264_run_level_t * );
 int x264_coeff_level_run16_neon( int16_t *, x264_run_level_t * );
+
+void x264_denoise_dct_neon( dctcoef *, uint32_t *, udctcoef *, int );
 #endif
diff --git a/common/quant.c b/common/quant.c
index d15c6b7..4fdaf40 100644
--- a/common/quant.c
+++ b/common/quant.c
@@ -763,6 +763,7 @@ 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
-- 
2.1.3