PR #22381 opened by george.zaguri URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/22381 Patch URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/22381.patch
RPi4: put_chroma_h_10_2x2_c: 63.4 ( 1.00x) put_chroma_h_10_4x4_c: 151.4 ( 1.00x) put_chroma_h_10_8x8_c: 555.1 ( 1.00x) put_chroma_h_10_8x8_neon: 113.9 ( 4.88x) put_chroma_h_10_16x16_c: 1068.5 ( 1.00x) put_chroma_h_10_16x16_neon: 439.4 ( 2.43x) put_chroma_h_10_32x32_c: 3432.6 ( 1.00x) put_chroma_h_10_32x32_neon: 1878.3 ( 1.83x) put_chroma_h_10_64x64_c: 12872.2 ( 1.00x) put_chroma_h_10_64x64_neon: 7868.2 ( 1.64x) put_chroma_h_10_128x128_c: 45612.2 ( 1.00x) put_chroma_h_10_128x128_neon: 28742.1 ( 1.59x) put_chroma_h_12_2x2_c: 63.7 ( 1.00x) put_chroma_h_12_4x4_c: 151.5 ( 1.00x) put_chroma_h_12_8x8_c: 555.2 ( 1.00x) put_chroma_h_12_8x8_neon: 114.2 ( 4.86x) put_chroma_h_12_16x16_c: 1068.1 ( 1.00x) put_chroma_h_12_16x16_neon: 438.8 ( 2.43x) put_chroma_h_12_32x32_c: 3419.7 ( 1.00x) put_chroma_h_12_32x32_neon: 1878.7 ( 1.82x) put_chroma_h_12_64x64_c: 12862.2 ( 1.00x) put_chroma_h_12_64x64_neon: 7868.2 ( 1.63x) put_chroma_h_12_128x128_c: 45613.5 ( 1.00x) put_chroma_h_12_128x128_neon: 28743.3 ( 1.59x) Apple M4: put_chroma_h_10_2x2_c: 2.5 ( 1.00x) put_chroma_h_10_4x4_c: 6.5 ( 1.00x) put_chroma_h_10_8x8_c: 17.8 ( 1.00x) put_chroma_h_10_8x8_neon: 6.8 ( 2.60x) put_chroma_h_10_16x16_c: 53.3 ( 1.00x) put_chroma_h_10_16x16_neon: 30.4 ( 1.75x) put_chroma_h_10_32x32_c: 181.8 ( 1.00x) put_chroma_h_10_32x32_neon: 116.2 ( 1.56x) put_chroma_h_10_64x64_c: 684.2 ( 1.00x) put_chroma_h_10_64x64_neon: 470.3 ( 1.45x) put_chroma_h_10_128x128_c: 2567.6 ( 1.00x) put_chroma_h_10_128x128_neon: 1879.3 ( 1.37x) put_chroma_h_12_2x2_c: 1.9 ( 1.00x) put_chroma_h_12_4x4_c: 7.0 ( 1.00x) put_chroma_h_12_8x8_c: 16.8 ( 1.00x) put_chroma_h_12_8x8_neon: 7.9 ( 2.12x) put_chroma_h_12_16x16_c: 55.0 ( 1.00x) put_chroma_h_12_16x16_neon: 29.0 ( 1.90x) put_chroma_h_12_32x32_c: 182.5 ( 1.00x) put_chroma_h_12_32x32_neon: 116.9 ( 1.56x) put_chroma_h_12_64x64_c: 666.8 ( 1.00x) put_chroma_h_12_64x64_neon: 474.5 ( 1.41x) put_chroma_h_12_128x128_c: 2588.1 ( 1.00x) put_chroma_h_12_128x128_neon: 1912.2 ( 1.35x) >From b4082cfd4149ed6c8bf41cb28c6aef44447f2599 Mon Sep 17 00:00:00 2001 From: Georgii Zagoruiko <[email protected]> Date: Wed, 4 Mar 2026 18:40:21 +0000 Subject: [PATCH] aarch64/vvc: Optimisations of put_luma_h() functions for 10/12-bit RPi4: put_chroma_h_10_2x2_c: 63.4 ( 1.00x) put_chroma_h_10_4x4_c: 151.4 ( 1.00x) put_chroma_h_10_8x8_c: 555.1 ( 1.00x) put_chroma_h_10_8x8_neon: 113.9 ( 4.88x) put_chroma_h_10_16x16_c: 1068.5 ( 1.00x) put_chroma_h_10_16x16_neon: 439.4 ( 2.43x) put_chroma_h_10_32x32_c: 3432.6 ( 1.00x) put_chroma_h_10_32x32_neon: 1878.3 ( 1.83x) put_chroma_h_10_64x64_c: 12872.2 ( 1.00x) put_chroma_h_10_64x64_neon: 7868.2 ( 1.64x) put_chroma_h_10_128x128_c: 45612.2 ( 1.00x) put_chroma_h_10_128x128_neon: 28742.1 ( 1.59x) put_chroma_h_12_2x2_c: 63.7 ( 1.00x) put_chroma_h_12_4x4_c: 151.5 ( 1.00x) put_chroma_h_12_8x8_c: 555.2 ( 1.00x) put_chroma_h_12_8x8_neon: 114.2 ( 4.86x) put_chroma_h_12_16x16_c: 1068.1 ( 1.00x) put_chroma_h_12_16x16_neon: 438.8 ( 2.43x) put_chroma_h_12_32x32_c: 3419.7 ( 1.00x) put_chroma_h_12_32x32_neon: 1878.7 ( 1.82x) put_chroma_h_12_64x64_c: 12862.2 ( 1.00x) put_chroma_h_12_64x64_neon: 7868.2 ( 1.63x) put_chroma_h_12_128x128_c: 45613.5 ( 1.00x) put_chroma_h_12_128x128_neon: 28743.3 ( 1.59x) Apple M4: put_chroma_h_10_2x2_c: 2.5 ( 1.00x) put_chroma_h_10_4x4_c: 6.5 ( 1.00x) put_chroma_h_10_8x8_c: 17.8 ( 1.00x) put_chroma_h_10_8x8_neon: 6.8 ( 2.60x) put_chroma_h_10_16x16_c: 53.3 ( 1.00x) put_chroma_h_10_16x16_neon: 30.4 ( 1.75x) put_chroma_h_10_32x32_c: 181.8 ( 1.00x) put_chroma_h_10_32x32_neon: 116.2 ( 1.56x) put_chroma_h_10_64x64_c: 684.2 ( 1.00x) put_chroma_h_10_64x64_neon: 470.3 ( 1.45x) put_chroma_h_10_128x128_c: 2567.6 ( 1.00x) put_chroma_h_10_128x128_neon: 1879.3 ( 1.37x) put_chroma_h_12_2x2_c: 1.9 ( 1.00x) put_chroma_h_12_4x4_c: 7.0 ( 1.00x) put_chroma_h_12_8x8_c: 16.8 ( 1.00x) put_chroma_h_12_8x8_neon: 7.9 ( 2.12x) put_chroma_h_12_16x16_c: 55.0 ( 1.00x) put_chroma_h_12_16x16_neon: 29.0 ( 1.90x) put_chroma_h_12_32x32_c: 182.5 ( 1.00x) put_chroma_h_12_32x32_neon: 116.9 ( 1.56x) put_chroma_h_12_64x64_c: 666.8 ( 1.00x) put_chroma_h_12_64x64_neon: 474.5 ( 1.41x) put_chroma_h_12_128x128_c: 2588.1 ( 1.00x) put_chroma_h_12_128x128_neon: 1912.2 ( 1.35x) --- libavcodec/aarch64/vvc/dsp_init.c | 27 ++++++ libavcodec/aarch64/vvc/inter.S | 131 ++++++++++++++++++++++++++++++ 2 files changed, 158 insertions(+) diff --git a/libavcodec/aarch64/vvc/dsp_init.c b/libavcodec/aarch64/vvc/dsp_init.c index bc2677945e..d09dd36a43 100644 --- a/libavcodec/aarch64/vvc/dsp_init.c +++ b/libavcodec/aarch64/vvc/dsp_init.c @@ -43,6 +43,19 @@ void ff_vvc_put_luma_h16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdif void ff_vvc_put_luma_h_x16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width); +void ff_vvc_put_chroma_h8_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, + const int height, const int8_t *hf, const int8_t *vf, const int width); +void ff_vvc_put_chroma_h16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, + const int height, const int8_t *hf, const int8_t *vf, const int width); +void ff_vvc_put_chroma_h_x16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, + const int height, const int8_t *hf, const int8_t *vf, const int width); +void ff_vvc_put_chroma_h8_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, + const int height, const int8_t *hf, const int8_t *vf, const int width); +void ff_vvc_put_chroma_h16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, + const int height, const int8_t *hf, const int8_t *vf, const int width); +void ff_vvc_put_chroma_h_x16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, + const int height, const int8_t *hf, const int8_t *vf, const int width); + void ff_vvc_put_luma_v4_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width); void ff_vvc_put_luma_v8_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, @@ -274,6 +287,13 @@ void ff_vvc_dsp_init_aarch64(VVCDSPContext *const c, const int bd) c->inter.dmvr[0][1] = ff_vvc_dmvr_h_10_neon; c->inter.dmvr[1][1] = ff_vvc_dmvr_hv_10_neon; c->inter.apply_bdof = ff_vvc_apply_bdof_10_neon; + + c->inter.put[1][2][0][1] = ff_vvc_put_chroma_h8_10_neon; + c->inter.put[1][3][0][1] = ff_vvc_put_chroma_h16_10_neon; + c->inter.put[1][4][0][1] = + c->inter.put[1][5][0][1] = + c->inter.put[1][6][0][1] = ff_vvc_put_chroma_h_x16_10_neon; + c->inter.put[0][2][0][1] = ff_vvc_put_luma_h8_10_neon; c->inter.put[0][3][0][1] = ff_vvc_put_luma_h16_10_neon; c->inter.put[0][4][0][1] = @@ -297,6 +317,13 @@ void ff_vvc_dsp_init_aarch64(VVCDSPContext *const c, const int bd) c->inter.dmvr[0][1] = ff_vvc_dmvr_h_12_neon; c->inter.dmvr[1][1] = ff_vvc_dmvr_hv_12_neon; c->inter.apply_bdof = ff_vvc_apply_bdof_12_neon; + + c->inter.put[1][2][0][1] = ff_vvc_put_chroma_h8_12_neon; + c->inter.put[1][3][0][1] = ff_vvc_put_chroma_h16_12_neon; + c->inter.put[1][4][0][1] = + c->inter.put[1][5][0][1] = + c->inter.put[1][6][0][1] = ff_vvc_put_chroma_h_x16_12_neon; + c->inter.put[0][2][0][1] = ff_vvc_put_luma_h8_12_neon; c->inter.put[0][3][0][1] = ff_vvc_put_luma_h16_12_neon; c->inter.put[0][4][0][1] = diff --git a/libavcodec/aarch64/vvc/inter.S b/libavcodec/aarch64/vvc/inter.S index 887e456a66..752405e79b 100644 --- a/libavcodec/aarch64/vvc/inter.S +++ b/libavcodec/aarch64/vvc/inter.S @@ -1833,6 +1833,137 @@ function ff_vvc_put_luma_h_x16_12_neon, export=1 put_luma_h_x16_xx_neon 4 endfunc +.macro put_chroma_h_x8_horizontal_filter shift + // 4 bytes from hf loaded to v0.4h + // 24 bytes from _src loaded to v20.8h & v21.4h where v21.4h is loaded for shift to v1.8h,v2.8h,v3.8h + // v24.4h & v25.4h are output vectors to store + ext v1.16b, v20.16b, v21.16b, #2 + ext v2.16b, v20.16b, v21.16b, #4 + ext v3.16b, v20.16b, v21.16b, #6 + smull v24.4s, v20.4h, v0.h[0] + smull2 v25.4s, v20.8h, v0.h[0] + smlal v24.4s, v1.4h, v0.h[1] + smlal2 v25.4s, v1.8h, v0.h[1] + smlal v24.4s, v2.4h, v0.h[2] + smlal2 v25.4s, v2.8h, v0.h[2] + smlal v24.4s, v3.4h, v0.h[3] + smlal2 v25.4s, v3.8h, v0.h[3] + sqshrn v24.4h, v24.4s, #(\shift) + sqshrn v25.4h, v25.4s, #(\shift) +.endm + +.macro put_chroma_h8_xx_neon shift + // dst .req x0 + // _src .req x1 + // _src_stride .req x2 + // height .req x3 + // hf .req x4 + // vf .req x5 + // width .req x6 + mov x9, #(VVC_MAX_PB_SIZE * 2) + ldr s0, [x4] + sub x1, x1, #2 + sub x2, x2, #16 + sxtl v0.8h, v0.8b +1: + ld1 {v20.8h}, [x1], #16 + ld1 {v21.4h}, [x1], x2 + put_chroma_h_x8_horizontal_filter \shift + subs w3, w3, #1 + st1 {v24.4h, v25.4h}, [x0], x9 + b.gt 1b + ret +.endm + +.macro put_chroma_h16_xx_neon shift + // dst .req x0 + // _src .req x1 + // _src_stride .req x2 + // height .req x3 + // hf .req x4 + // vf .req x5 + // width .req x6 + mov x9, #(VVC_MAX_PB_SIZE * 2) + ldr s0, [x4] + sub x9, x9, #16 + sub x1, x1, #2 + sub x2, x2, #32 + sxtl v0.8h, v0.8b +1: + ld1 {v20.8h, v21.8h}, [x1], #32 + ld1 {v22.4h}, [x1], x2 + put_chroma_h_x8_horizontal_filter \shift + mov v20.16b, v21.16b + mov v21.16b, v22.16b + st1 {v24.4h, v25.4h}, [x0], #16 + put_chroma_h_x8_horizontal_filter \shift + subs w3, w3, #1 + st1 {v24.4h, v25.4h}, [x0], x9 + b.gt 1b + ret +.endm + +.macro put_chroma_h_x16_xx_neon shift + // dst .req x0 + // _src .req x1 + // _src_stride .req x2 + // height .req x3 + // hf .req x4 + // vf .req x5 + // width .req x6 + mov x9, #(VVC_MAX_PB_SIZE * 2) + ldr s0, [x4] + sub x9, x9, w6, uxtw #1 + sub x2, x2, w6, uxtw #1 + sxtl v0.8h, v0.8b + sub x1, x1, #2 + sub x2, x2, #16 +1: + ld1 {v20.8h}, [x1], #16 + mov w8, w6 +2: + ld1 {v21.8h, v22.8h}, [x1], #32 + put_chroma_h_x8_horizontal_filter \shift + mov v20.16b, v21.16b + mov v21.16b, v22.16b + st1 {v24.4h, v25.4h}, [x0], #16 + put_chroma_h_x8_horizontal_filter \shift + mov v20.16b, v21.16b + subs w8, w8, #16 + st1 {v24.4h, v25.4h}, [x0], #16 + b.gt 2b + subs w3, w3, #1 + add x0, x0, x9 + add x1, x1, x2 + b.gt 1b + ret +.endm + + +function ff_vvc_put_chroma_h8_10_neon, export=1 + put_chroma_h8_xx_neon 2 +endfunc + +function ff_vvc_put_chroma_h8_12_neon, export=1 + put_chroma_h8_xx_neon 4 +endfunc + +function ff_vvc_put_chroma_h16_10_neon, export=1 + put_chroma_h16_xx_neon 2 +endfunc + +function ff_vvc_put_chroma_h16_12_neon, export=1 + put_chroma_h16_xx_neon 4 +endfunc + +function ff_vvc_put_chroma_h_x16_10_neon, export=1 + put_chroma_h_x16_xx_neon 2 +endfunc + +function ff_vvc_put_chroma_h_x16_12_neon, export=1 + put_chroma_h_x16_xx_neon 4 +endfunc + .macro put_luma_v4_xx_neon shift mov x9, #(VVC_MAX_PB_SIZE * 2) sub x1, x1, x2, lsl #1 -- 2.52.0 _______________________________________________ ffmpeg-devel mailing list -- [email protected] To unsubscribe send an email to [email protected]
