[vlc-commits] [Git][videolan/vlc][master] 2 commits: aarch64/sve: honor byte count in deinterlace merge

[Steve Lhomme (@robUx4)]

Steve Lhomme pushed to branch master at VideoLAN / VLC


Commits:
ccf38fe1 by markoub at 2026-06-22T12:31:03+02:00
aarch64/sve: honor byte count in deinterlace merge

The merge callback receives a byte length. merge16_arm_sve indexes 16-bit samples, so convert x3 from bytes to samples before the whilelt loop.

This matches Merge16BitGeneric, which iterates i_bytes / 2 words.

- - - - -
85b9e847 by markoub at 2026-06-22T12:31:03+02:00
aarch64: use rounded average for deinterlace merge

Match the generic deinterlace merge semantics in the AArch64 AdvSIMD backend.

AdvSIMD can use rounded halving add directly.

- - - - -


2 changed files:

- modules/isa/aarch64/simd/merge.S
- modules/isa/aarch64/sve/merge.S


Changes:

=====================================
modules/isa/aarch64/simd/merge.S
=====================================
@@ -43,12 +43,12 @@ function merge8_arm64
 	ld1		{v0.16b,v1.16b}, [SRC1], x10
 	ld1		{v4.16b,v5.16b}, [SRC2], x10
 	ld1		{v2.16b,v3.16b}, [x11], x10
-	uhadd		v0.16b, v0.16b, v4.16b
+	urhadd		v0.16b, v0.16b, v4.16b
 	ld1		{v6.16b,v7.16b}, [x12], x10
 	subs		x5,  x5,  #64
-	uhadd		v1.16b, v1.16b, v5.16b
-	uhadd		v2.16b, v2.16b, v6.16b
-	uhadd		v3.16b, v3.16b, v7.16b
+	urhadd		v1.16b, v1.16b, v5.16b
+	urhadd		v2.16b, v2.16b, v6.16b
+	urhadd		v3.16b, v3.16b, v7.16b
 	st1		{v0.16b,v1.16b}, [DEST], #32
 	st1		{v2.16b,v3.16b}, [DEST], #32
 	b.gt		1b
@@ -56,14 +56,14 @@ function merge8_arm64
 	tbz		SIZE, #5,  3f
 	ld1		{v0.16b,v1.16b}, [SRC1], #32
 	ld1		{v4.16b,v5.16b}, [SRC2], #32
-	uhadd		v0.16b, v0.16b, v4.16b
-	uhadd		v1.16b, v1.16b, v5.16b
+	urhadd		v0.16b, v0.16b, v4.16b
+	urhadd		v1.16b, v1.16b, v5.16b
 	st1		{v0.16b,v1.16b}, [DEST], #32
 3:
 	tbz		SIZE, #4,  4f
 	ld1		{v0.16b},  [SRC1]
 	ld1		{v4.16b},  [SRC2]
-	uhadd		v0.16b, v0.16b, v4.16b
+	urhadd		v0.16b, v0.16b, v4.16b
 	st1		{v0.16b},  [DEST]
 4:
 	ret
@@ -77,11 +77,11 @@ function merge16_arm64
 	ld1		{v0.8h,v1.8h}, [SRC1], #32
 	ld1		{v4.8h,v5.8h}, [SRC2], #32
 	ld1		{v2.8h,v3.8h}, [SRC1], #32
-	uhadd		v0.8h,  v0.8h,  v4.8h
+	urhadd		v0.8h,  v0.8h,  v4.8h
 	ld1		{v6.8h,v7.8h}, [SRC2], #32
-	uhadd		v1.8h,  v1.8h,  v5.8h
-	uhadd		v2.8h,  v2.8h,  v6.8h
-	uhadd		v3.8h,  v3.8h,  v7.8h
+	urhadd		v1.8h,  v1.8h,  v5.8h
+	urhadd		v2.8h,  v2.8h,  v6.8h
+	urhadd		v3.8h,  v3.8h,  v7.8h
 	st1		{v0.8h,v1.8h}, [DEST], #32
 	st1		{v2.8h,v3.8h}, [DEST], #32
 	subs		x5,  x5,  #64
@@ -90,14 +90,14 @@ function merge16_arm64
 	tbz		SIZE, #5, 3f
 	ld1		{v0.8h,v1.8h}, [SRC1], #32
 	ld1		{v4.8h,v5.8h}, [SRC2], #32
-	uhadd		v0.8h,  v0.8h,  v4.8h
-	uhadd		v1.8h,  v1.8h,  v5.8h
+	urhadd		v0.8h,  v0.8h,  v4.8h
+	urhadd		v1.8h,  v1.8h,  v5.8h
 	st1		{v0.8h,v1.8h}, [DEST], #32
 3:
 	tbz	        SIZE, #4,  4f
 	ld1		{v0.8h},  [SRC1]
 	ld1		{v4.8h},  [SRC2]
-	uhadd		v0.8h,  v0.8h,  v4.8h
+	urhadd		v0.8h,  v0.8h,  v4.8h
 	st1		{v0.8h},  [DEST]
 4:
 	ret


=====================================
modules/isa/aarch64/sve/merge.S
=====================================
@@ -43,6 +43,7 @@ function merge8_arm_sve
 
 function merge16_arm_sve
 	bti	c
+	lsr	x3, x3, #1
 	mov	x4, #0
 	b	2f
 1:	ld1h	{z0.s}, p0/z, [x1, x4, lsl #1]



View it on GitLab: https://code.videolan.org/videolan/vlc/-/compare/85d4a8074be8c0f26c43c55b977ef0c84db62ad6...85b9e8476423327d33f5aaee973874358957c34d

-- 
View it on GitLab: https://code.videolan.org/videolan/vlc/-/compare/85d4a8074be8c0f26c43c55b977ef0c84db62ad6...85b9e8476423327d33f5aaee973874358957c34d
You're receiving this email because of your account on code.videolan.org. Manage all notifications: https://code.videolan.org/-/profile/notifications | Help: https://code.videolan.org/help