PR #20442 opened by welder URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20442 Patch URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20442.patch
A low hanging fruit Before and after on A78: dmvr_hv_8_12x20_neon: 205.3 ( 5.21x) dmvr_hv_8_20x12_neon: 171.8 ( 3.15x) dmvr_hv_8_20x20_neon: 282.7 ( 3.11x) dmvr_hv_8_12x20_neon: 172.7 ( 5.58x) dmvr_hv_8_20x12_neon: 133.3 ( 3.36x) dmvr_hv_8_20x20_neon: 214.6 ( 3.40x) >From 55e2f5d2661e23e5adab5351effec892294fd708 Mon Sep 17 00:00:00 2001 From: Krzysztof Pyrkosz <[email protected]> Date: Thu, 4 Sep 2025 22:56:43 +0200 Subject: [PATCH] Replace uxtl with umull in dmvr_hv_8 Before and after on A78: dmvr_hv_8_12x20_neon: 205.3 ( 5.21x) dmvr_hv_8_20x12_neon: 171.8 ( 3.15x) dmvr_hv_8_20x20_neon: 282.7 ( 3.11x) dmvr_hv_8_12x20_neon: 172.7 ( 5.58x) dmvr_hv_8_20x12_neon: 133.3 ( 3.36x) dmvr_hv_8_20x20_neon: 214.6 ( 3.40x) --- libavcodec/aarch64/vvc/inter.S | 29 +++++++++-------------------- 1 file changed, 9 insertions(+), 20 deletions(-) diff --git a/libavcodec/aarch64/vvc/inter.S b/libavcodec/aarch64/vvc/inter.S index 50fc073dc9..a6648b64fc 100644 --- a/libavcodec/aarch64/vvc/inter.S +++ b/libavcodec/aarch64/vvc/inter.S @@ -393,13 +393,10 @@ function ff_vvc_dmvr_hv_8_neon, export=1 movrel x9, X(ff_vvc_inter_luma_dmvr_filters) add x12, x9, mx, lsl #1 - ldrb w10, [x12] - ldrb w11, [x12, #1] mov tmp0, sp add tmp1, tmp0, #(VVC_MAX_PB_SIZE * 2) // We know the value are positive - dup v0.8h, w10 // filter_x[0] - dup v1.8h, w11 // filter_x[1] + ld2r {v0.16b, v1.16b}, [x12] add x12, x9, my, lsl #1 ldrb w10, [x12] @@ -424,14 +421,10 @@ function ff_vvc_dmvr_hv_8_neon, export=1 // width > 16 ldur q5, [src, #1] ldr q4, [src], #16 - uxtl v7.8h, v5.8b - uxtl2 v17.8h, v5.16b - uxtl v6.8h, v4.8b - uxtl2 v16.8h, v4.16b - mul v6.8h, v6.8h, v0.8h - mul v16.8h, v16.8h, v0.8h - mla v6.8h, v7.8h, v1.8h - mla v16.8h, v17.8h, v1.8h + umull v6.8h, v4.8b, v0.8b + umull2 v16.8h, v4.16b, v0.16b + umlal v6.8h, v5.8b, v1.8b + umlal2 v16.8h, v5.16b, v1.16b urshr v6.8h, v6.8h, #(8 - 6) urshr v7.8h, v16.8h, #(8 - 6) stp q6, q7, [x13], #32 @@ -451,10 +444,8 @@ function ff_vvc_dmvr_hv_8_neon, export=1 // width > 8 ldur d5, [src, #1] ldr d4, [src], #8 - uxtl v7.8h, v5.8b - uxtl v6.8h, v4.8b - mul v6.8h, v6.8h, v0.8h - mla v6.8h, v7.8h, v1.8h + umull v6.8h, v4.8b, v0.8b + umlal v6.8h, v5.8b, v1.8b urshr v6.8h, v6.8h, #(8 - 6) str q6, [x13], #16 @@ -468,10 +459,8 @@ function ff_vvc_dmvr_hv_8_neon, export=1 3: ldur s5, [src, #1] ldr s4, [src], #4 - uxtl v7.8h, v5.8b - uxtl v6.8h, v4.8b - mul v6.4h, v6.4h, v0.4h - mla v6.4h, v7.4h, v1.4h + umull v6.8h, v4.8b, v0.8b + umlal v6.8h, v5.8b, v1.8b urshr v6.4h, v6.4h, #(8 - 6) str d6, [x13], #8 -- 2.49.1 _______________________________________________ ffmpeg-devel mailing list -- [email protected] To unsubscribe send an email to [email protected]
