diff options
| author | Krzysztof Pyrkosz <[email protected]> | 2025-09-05 19:17:48 +0200 |
|---|---|---|
| committer | jianhuaw <[email protected]> | 2025-09-09 16:37:28 +0000 |
| commit | de25cb4603f938aafc1a182b76a6216de170e7ac (patch) | |
| tree | a5fbfb000b2491a7238fec7dc32678db64573194 | |
| parent | 5f39965dd6e6769e339dd684db3be5a667f4bc69 (diff) | |
avcodec/aarch64/vvc: Optimize vvc_apply_bdof_block_8x
Before and after:
A53:
apply_bdof_8_8x16_neon: 3320.5 ( 4.02x)
apply_bdof_10_8x16_neon: 3317.8 ( 3.90x)
apply_bdof_12_8x16_neon: 3303.6 ( 3.91x)
apply_bdof_8_8x16_neon: 3168.1 ( 4.23x)
apply_bdof_10_8x16_neon: 3127.8 ( 4.13x)
apply_bdof_12_8x16_neon: 3119.3 ( 4.18x)
A72:
apply_bdof_8_8x16_neon: 1827.4 ( 5.02x)
apply_bdof_10_8x16_neon: 1838.5 ( 4.89x)
apply_bdof_12_8x16_neon: 1841.1 ( 4.83x)
apply_bdof_8_8x16_neon: 1691.6 ( 5.46x)
apply_bdof_10_8x16_neon: 1695.9 ( 5.23x)
apply_bdof_12_8x16_neon: 1695.4 ( 5.29x)
A78
apply_bdof_8_8x16_neon: 648.9 ( 7.43x)
apply_bdof_10_8x16_neon: 646.1 ( 7.04x)
apply_bdof_12_8x16_neon: 643.8 ( 7.04x)
apply_bdof_8_8x16_neon: 603.2 ( 7.97x)
apply_bdof_10_8x16_neon: 604.1 ( 7.52x)
apply_bdof_12_8x16_neon: 604.5 ( 7.52x)
| -rw-r--r-- | libavcodec/aarch64/vvc/inter.S | 42 |
1 files changed, 14 insertions, 28 deletions
diff --git a/libavcodec/aarch64/vvc/inter.S b/libavcodec/aarch64/vvc/inter.S index d59d278275..01d2ff155c 100644 --- a/libavcodec/aarch64/vvc/inter.S +++ b/libavcodec/aarch64/vvc/inter.S @@ -915,58 +915,44 @@ endfunc vy .req x7 ldr w8, [sp] - movi v7.4s, #(1 << (14 - \bit_depth)) mov x12, #(BDOF_BLOCK_SIZE * 2) mov x14, #(VVC_MAX_PB_SIZE * 2) .if \bit_depth >= 10 // clip pixel mov w15, #((1 << \bit_depth) - 1) - movi v18.8h, #0 dup v19.8h, w15 .endif 0: - ld1r {v0.8h}, [vx], #2 - ld1r {v1.8h}, [vy], #2 - ld1r {v2.8h}, [vx] - ld1r {v3.8h}, [vy] + ldr s0, [vx], #(2 * BDOF_MIN_BLOCK_SIZE) + ldr s1, [vy], #(2 * BDOF_MIN_BLOCK_SIZE) mov w13, #(BDOF_MIN_BLOCK_SIZE) - ins v0.d[1], v2.d[1] - ins v1.d[1], v3.d[1] 1: - ld1 {v2.8h}, [gh], x12 - ld1 {v4.8h}, [gv], x12 - smull v3.4s, v0.4h, v2.4h - smull2 v16.4s, v0.8h, v2.8h - smlal v3.4s, v1.4h, v4.4h - smlal2 v16.4s, v1.8h, v4.8h - ld1 {v5.8h}, [src0], x14 ld1 {v6.8h}, [src1], x14 - saddl v2.4s, v5.4h, v6.4h - add v2.4s, v2.4s, v7.4s - add v2.4s, v2.4s, v3.4s - saddl2 v4.4s, v5.8h, v6.8h - add v4.4s, v4.4s, v7.4s - add v4.4s, v4.4s, v16.4s - sqshrn v5.4h, v2.4s, #(15 - \bit_depth) - sqshrn2 v5.8h, v4.4s, #(15 - \bit_depth) + saddl v17.4s, v5.4h, v6.4h + ld1 {v4.8h}, [gv], x12 + saddl2 v16.4s, v5.8h, v6.8h + ld1 {v2.8h}, [gh], x12 + smlal v17.4s, v4.4h, v1.h[0] + smlal2 v16.4s, v4.8h, v1.h[1] + smlal v17.4s, v2.4h, v0.h[0] + smlal2 v16.4s, v2.8h, v0.h[1] + + sqrshrun v5.4h, v17.4s, #(15 - \bit_depth) + sqrshrun2 v5.8h, v16.4s, #(15 - \bit_depth) subs w13, w13, #1 .if \bit_depth == 8 sqxtun v5.8b, v5.8h - str d5, [dst] - add dst, dst, dst_stride + st1 {v5.8b}, [dst], dst_stride .else smin v5.8h, v5.8h, v19.8h - smax v5.8h, v5.8h, v18.8h st1 {v5.8h}, [dst], dst_stride .endif b.ne 1b subs w8, w8, #(BDOF_MIN_BLOCK_SIZE) - add vx, vx, #(2 * BDOF_MIN_BLOCK_SIZE - 2) - add vy, vy, #(2 * BDOF_MIN_BLOCK_SIZE - 2) b.ne 0b ret |