[FFmpeg-devel] [PATCH v3 6/6] lavc/aarch64: add hevc sao band 8x8 tiling

[J. Dekker]

--bench on AWS Graviton:

hevc_sao_band_8x8_8_c: 317.5
hevc_sao_band_8x8_8_neon: 97.5
hevc_sao_band_16x16_8_c: 1115.0
hevc_sao_band_16x16_8_neon: 322.7
hevc_sao_band_32x32_8_c: 4599.2
hevc_sao_band_32x32_8_neon: 1246.2
hevc_sao_band_48x48_8_c: 10021.7
hevc_sao_band_48x48_8_neon: 2740.5
hevc_sao_band_64x64_8_c: 17635.0
hevc_sao_band_64x64_8_neon: 4875.7

Signed-off-by: J. Dekker <jdek at itanimul.li>
---
 libavcodec/aarch64/hevcdsp_init_aarch64.c |  6 +++++-
 libavcodec/aarch64/hevcdsp_sao_neon.S     | 11 +++++++----
 2 files changed, 12 insertions(+), 5 deletions(-)

diff --git a/libavcodec/aarch64/hevcdsp_init_aarch64.c b/libavcodec/aarch64/hevcdsp_init_aarch64.c
index b93cec9e44..2002530266 100644
--- a/libavcodec/aarch64/hevcdsp_init_aarch64.c
+++ b/libavcodec/aarch64/hevcdsp_init_aarch64.c
@@ -77,7 +77,11 @@ av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)
         c->idct_dc[1]                  = ff_hevc_idct_8x8_dc_8_neon;
         c->idct_dc[2]                  = ff_hevc_idct_16x16_dc_8_neon;
         c->idct_dc[3]                  = ff_hevc_idct_32x32_dc_8_neon;
-        c->sao_band_filter[0]          = ff_hevc_sao_band_filter_8x8_8_neon;
+        c->sao_band_filter[0]          =
+        c->sao_band_filter[1]          =
+        c->sao_band_filter[2]          =
+        c->sao_band_filter[3]          =
+        c->sao_band_filter[4]          = ff_hevc_sao_band_filter_8x8_8_neon;
         c->sao_edge_filter[0]          = ff_hevc_sao_edge_filter_8x8_8_neon;
         c->sao_edge_filter[1]          =
         c->sao_edge_filter[2]          =
diff --git a/libavcodec/aarch64/hevcdsp_sao_neon.S b/libavcodec/aarch64/hevcdsp_sao_neon.S
index 73b0b3b056..d524323fe8 100644
--- a/libavcodec/aarch64/hevcdsp_sao_neon.S
+++ b/libavcodec/aarch64/hevcdsp_sao_neon.S
@@ -3,7 +3,7 @@
  *
  * AArch64 NEON optimised SAO functions for HEVC decoding
  *
- * Copyright (c) 2020 Josh Dekker <josh at itanimul.li>
+ * Copyright (c) 2020-2021  J. Dekker <jdek at itanimul.li>
  *
  * This file is part of FFmpeg.
  *
@@ -35,6 +35,7 @@ function ff_hevc_sao_band_filter_8x8_8_neon, export=1
         stp             xzr, xzr, [sp, #32]
         stp             xzr, xzr, [sp, #48]
         mov             w8,  #4
+        sxtw            x6,  w6
 0:      ldrsh           x9, [x4,  x8, lsl #1]      // sao_offset_val[k+1]
         subs            w8,  w8,  #1
         add             w10, w8,  w5               // k + sao_left_class
@@ -43,7 +44,9 @@ function ff_hevc_sao_band_filter_8x8_8_neon, export=1
         bne             0b
         ld1             {v16.16b-v19.16b}, [sp], #64
         movi            v20.8h,   #1
-1:      mov             w8,  w6                    // beginning of line
+        sub             x2,  x2, x6                // stride_dst - width
+        sub             x3,  x3, x6                // stride_src - width
+1:      mov             x8,  x6                    // beginning of line
 2:      // Simple layout for accessing 16bit values
         // with 8bit LUT.
         //
@@ -52,7 +55,7 @@ function ff_hevc_sao_band_filter_8x8_8_neon, export=1
         // |xDE#xAD|xCA#xFE|xBE#xEF|xFE#xED|....
         // +----------------------------------->
         //    i-0     i-1     i-2     i-3
-        ld1             {v2.8b}, [x1]              // dst[x] = av_clip_pixel(src[x] + offset_table[src[x] >> shift]);
+        ld1             {v2.8b}, [x1], #8          // dst[x] = av_clip_pixel(src[x] + offset_table[src[x] >> shift]);
         uxtl            v0.8h,  v2.8b              // load src[x]
         ushr            v2.8h,  v0.8h, #3          // >> BIT_DEPTH - 3
         shl             v1.8h,  v2.8h, #1          // low (x2, accessing short)
@@ -61,7 +64,7 @@ function ff_hevc_sao_band_filter_8x8_8_neon, export=1
         tbx             v2.16b, {v16.16b-v19.16b}, v1.16b // table
         add             v1.8h,  v0.8h, v2.8h       // src[x] + table
         sqxtun          v4.8b,  v1.8h              // clip + narrow
-        st1             {v4.8b}, [x0]              // store
+        st1             {v4.8b}, [x0], #8          // store
         subs            w8, w8,  #8                // done 8 pixels
         bne             2b
         subs            w7, w7,  #1                // finished line, prep. new
-- 
2.32.0 (Apple Git-132)