From: Zhao Zhili <zhiliz...@tencent.com>
hevc_sao_edge_8_8_c: 124.5 ( 1.00x)
hevc_sao_edge_8_8_simd128: 18.1 ( 6.89x)
hevc_sao_edge_16_8_c: 478.6 ( 1.00x)
hevc_sao_edge_16_8_simd128: 48.9 ( 9.79x)
hevc_sao_edge_32_8_c: 2021.1 ( 1.00x)
hevc_sao_edge_32_8_simd128: 187.4 (10.79x)
hevc_sao_edge_48_8_c: 4295.5 ( 1.00x)
hevc_sao_edge_48_8_simd128: 397.4 (10.81x)
hevc_sao_edge_64_8_c: 7245.5 ( 1.00x)
hevc_sao_edge_64_8_simd128: 709.5 (10.21x)
Signed-off-by: Zhao Zhili <zhiliz...@tencent.com>
---
libavcodec/wasm/hevc/dsp_init.c | 6 ++
libavcodec/wasm/hevc/sao.c | 140 ++++++++++++++++++++++++++++++++
libavcodec/wasm/hevc/sao.h | 10 +++
3 files changed, 156 insertions(+)
diff --git a/libavcodec/wasm/hevc/dsp_init.c b/libavcodec/wasm/hevc/dsp_init.c
index 76a1031ff4..8672bbc2e1 100644
--- a/libavcodec/wasm/hevc/dsp_init.c
+++ b/libavcodec/wasm/hevc/dsp_init.c
@@ -42,6 +42,12 @@ av_cold void ff_hevc_dsp_init_wasm(HEVCDSPContext *c, const
int bit_depth)
c->sao_band_filter[2] =
c->sao_band_filter[3] =
c->sao_band_filter[4] = ff_hevc_sao_band_filter_16x16_8_simd128;
+
+ c->sao_edge_filter[0] = ff_hevc_sao_edge_filter_8x8_8_simd128;
+ c->sao_edge_filter[1] =
+ c->sao_edge_filter[2] =
+ c->sao_edge_filter[3] =
+ c->sao_edge_filter[4] = ff_hevc_sao_edge_filter_16x16_8_simd128;
} else if (bit_depth == 10) {
c->idct[0] = ff_hevc_idct_4x4_10_simd128;
c->idct[1] = ff_hevc_idct_8x8_10_simd128;
diff --git a/libavcodec/wasm/hevc/sao.c b/libavcodec/wasm/hevc/sao.c
index 82134af7f3..a863b8e720 100644
--- a/libavcodec/wasm/hevc/sao.c
+++ b/libavcodec/wasm/hevc/sao.c
@@ -22,6 +22,10 @@
#include <wasm_simd128.h>
+#include "libavcodec/defs.h"
+
+#define HEVC_MAX_PB_SIZE 64
+
void ff_hevc_sao_band_filter_8x8_8_simd128(uint8_t *dst, const uint8_t *src,
ptrdiff_t stride_dst,
ptrdiff_t stride_src,
@@ -111,3 +115,139 @@ void ff_hevc_sao_band_filter_16x16_8_simd128(uint8_t
*dst, const uint8_t *src,
src += stride_src;
}
}
+
+void ff_hevc_sao_edge_filter_8x8_8_simd128(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride_dst,
+ const int16_t *sao_offset_val,
+ int eo, int width, int height)
+{
+ static const int8_t pos[4][2][2] = {
+ { { -1, 0 }, { 1, 0 } }, // horizontal
+ { { 0, -1 }, { 0, 1 } }, // vertical
+ { { -1, -1 }, { 1, 1 } }, // 45 degree
+ { { 1, -1 }, { -1, 1 } }, // 135 degree
+ };
+ int a_stride, b_stride;
+ ptrdiff_t stride_src = (2 * HEVC_MAX_PB_SIZE +
AV_INPUT_BUFFER_PADDING_SIZE);
+ const v128_t edge_idx = wasm_u8x16_make(1, 2, 0, 3,
+ 4, 0, 0, 0,
+ 0, 0, 0, 0,
+ 0, 0, 0, 0);
+ v128_t sao_offset = wasm_v128_load(sao_offset_val);
+ v128_t one = wasm_i8x16_const_splat(1);
+ v128_t two = wasm_i8x16_const_splat(2);
+
+ a_stride = pos[eo][0][0] + pos[eo][0][1] * stride_src;
+ b_stride = pos[eo][1][0] + pos[eo][1][1] * stride_src;
+ for (int y = height; y > 0; y -= 2) {
+ v128_t v0, v1, v2;
+ v128_t diff0, diff1;
+
+ v0 = wasm_v128_load64_zero(src);
+ v1 = wasm_v128_load64_zero(src + a_stride);
+ v2 = wasm_v128_load64_zero(src + b_stride);
+ src += stride_src;
+ v0 = wasm_v128_load64_lane(src, v0, 1);
+ v1 = wasm_v128_load64_lane(src + a_stride, v1, 1);
+ v2 = wasm_v128_load64_lane(src + b_stride, v2, 1);
+ src += stride_src;
+
+ diff0 = wasm_u8x16_gt(v0, v1);
+ v1 = wasm_u8x16_lt(v0, v1);
+ diff0 = wasm_i8x16_sub(v1, diff0);
+
+ diff1 = wasm_u8x16_gt(v0, v2);
+ v2 = wasm_u8x16_lt(v0, v2);
+ diff1 = wasm_i8x16_sub(v2, diff1);
+
+ v1 = wasm_i8x16_add(diff0, two);
+ v1 = wasm_i8x16_add(v1, diff1);
+
+ v2 = wasm_i8x16_swizzle(edge_idx, v1); // offset_val
+ v1 = wasm_i8x16_shl(v2, 1); // Access int16_t
+ v2 = wasm_i8x16_add(v1, one); // Access upper half of int16_t
+ diff0 = wasm_i8x16_shuffle(v1, v2, 0, 16, 1, 17, 2, 18, 3, 19, 4,
+ 20, 5, 21, 6, 22, 7, 23);
+ diff1 = wasm_i8x16_shuffle(v1, v2, 8, 24, 9, 25, 10, 26, 11, 27,
+ 12, 28, 13, 29, 14, 30, 15, 31);
+ v1 = wasm_u16x8_extend_high_u8x16(v0);
+ v0 = wasm_u16x8_extend_low_u8x16(v0);
+ diff0 = wasm_i8x16_swizzle(sao_offset, diff0);
+ diff1 = wasm_i8x16_swizzle(sao_offset, diff1);
+
+ v0 = wasm_i16x8_add_sat(v0, diff0);
+ v1 = wasm_i16x8_add_sat(v1, diff1);
+ v0 = wasm_u8x16_narrow_i16x8(v0, v1);
+
+ wasm_v128_store64_lane(dst, v0, 0);
+ dst += stride_dst;
+ wasm_v128_store64_lane(dst, v0, 1);
+ dst += stride_dst;
+ }
+}
+
+void ff_hevc_sao_edge_filter_16x16_8_simd128(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride_dst,
+ const int16_t *sao_offset_val,
+ int eo, int width, int height)
+{
+ static const int8_t pos[4][2][2] = {
+ { { -1, 0 }, { 1, 0 } }, // horizontal
+ { { 0, -1 }, { 0, 1 } }, // vertical
+ { { -1, -1 }, { 1, 1 } }, // 45 degree
+ { { 1, -1 }, { -1, 1 } }, // 135 degree
+ };
+ int a_stride, b_stride;
+ ptrdiff_t stride_src = (2 * HEVC_MAX_PB_SIZE +
AV_INPUT_BUFFER_PADDING_SIZE);
+ const v128_t edge_idx = wasm_u8x16_make(1, 2, 0, 3,
+ 4, 0, 0, 0,
+ 0, 0, 0, 0,
+ 0, 0, 0, 0);
+ v128_t sao_offset = wasm_v128_load(sao_offset_val);
+ v128_t one = wasm_i8x16_const_splat(1);
+ v128_t two = wasm_i8x16_const_splat(2);
+
+ a_stride = pos[eo][0][0] + pos[eo][0][1] * stride_src;
+ b_stride = pos[eo][1][0] + pos[eo][1][1] * stride_src;
+ for (int y = height; y > 0; y--) {
+ for (int x = 0; x < width; x += 16) {
+ v128_t v0, v1, v2;
+ v128_t diff0, diff1;
+
+ v0 = wasm_v128_load(&src[x]);
+ v1 = wasm_v128_load(&src[x + a_stride]);
+ v2 = wasm_v128_load(&src[x + b_stride]);
+
+ diff0 = wasm_u8x16_gt(v0, v1);
+ v1 = wasm_u8x16_lt(v0, v1);
+ diff0 = wasm_i8x16_sub(v1, diff0);
+
+ diff1 = wasm_u8x16_gt(v0, v2);
+ v2 = wasm_u8x16_lt(v0, v2);
+ diff1 = wasm_i8x16_sub(v2, diff1);
+
+ v1 = wasm_i8x16_add(diff0, two);
+ v1 = wasm_i8x16_add(v1, diff1);
+
+ v2 = wasm_i8x16_swizzle(edge_idx, v1); // offset_val
+ v1 = wasm_i8x16_shl(v2, 1); // Access int16_t
+ v2 = wasm_i8x16_add(v1, one); // Access upper half of
int16_t
+ diff0 = wasm_i8x16_shuffle(v1, v2, 0, 16, 1, 17, 2, 18, 3, 19, 4,
+ 20, 5, 21, 6, 22, 7, 23);
+ diff1 = wasm_i8x16_shuffle(v1, v2, 8, 24, 9, 25, 10, 26, 11, 27,
+ 12, 28, 13, 29, 14, 30, 15, 31);
+ v1 = wasm_u16x8_extend_high_u8x16(v0);
+ v0 = wasm_u16x8_extend_low_u8x16(v0);
+ diff0 = wasm_i8x16_swizzle(sao_offset, diff0);
+ diff1 = wasm_i8x16_swizzle(sao_offset, diff1);
+
+ v0 = wasm_i16x8_add_sat(v0, diff0);
+ v1 = wasm_i16x8_add_sat(v1, diff1);
+ v0 = wasm_u8x16_narrow_i16x8(v0, v1);
+ wasm_v128_store(&dst[x], v0);
+ }
+
+ src += stride_src;
+ dst += stride_dst;
+ }
+}
diff --git a/libavcodec/wasm/hevc/sao.h b/libavcodec/wasm/hevc/sao.h
index 6119ec90f1..052420d7de 100644
--- a/libavcodec/wasm/hevc/sao.h
+++ b/libavcodec/wasm/hevc/sao.h
@@ -38,4 +38,14 @@ void ff_hevc_sao_band_filter_16x16_8_simd128(uint8_t *_dst,
const uint8_t *_src,
int sao_left_class, int width,
int height);
+void ff_hevc_sao_edge_filter_8x8_8_simd128(uint8_t *_dst, const uint8_t *_src,
+ ptrdiff_t stride_dst,
+ const int16_t *sao_offset_val,
+ int eo, int width, int height);
+
+void ff_hevc_sao_edge_filter_16x16_8_simd128(uint8_t *_dst, const uint8_t
*_src,
+ ptrdiff_t stride_dst,
+ const int16_t *sao_offset_val,
+ int eo, int width, int height);
+
#endif
\ No newline at end of file
--
2.43.0
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
