This is an automated email from the git hooks/post-receive script.
Git pushed a commit to branch master
in repository ffmpeg.
The following commit(s) were added to refs/heads/master by this push:
new 30595cbc5d swscale/aarch64/yuv2rgb_neon: aggregate 16bpp predicates
30595cbc5d is described below
commit 30595cbc5db6b8443625747a9ce914aacd65492b
Author: DROOdotFOO <[email protected]>
AuthorDate: Sat May 23 17:53:20 2026 +0200
Commit: DROOdotFOO <[email protected]>
CommitDate: Tue May 26 19:26:28 2026 +0200
swscale/aarch64/yuv2rgb_neon: aggregate 16bpp predicates
The six .ifc cascades that gate 16bpp behavior in yuv2rgb_neon.S
(linesize padding in three load_args macros, d8/d9 save/restore,
main-loop pack dispatch) all branch on the same four output formats.
Aggregate the predicate into four GAS .set symbols emitted once per
declare_func via a new set_rgb16_predicates macro:
rgb16 - 1 for *565le and *555le outputs; 0 otherwise
r_first - 1 for rgb*le (R high); 0 for bgr*le (B high)
gshift - 2 for 565, 3 for 555 (passed as pack_rgb16's g_shr)
hshift - 11 for 565, 10 for 555 (passed as pack_rgb16's high_shl)
Call sites become a flat ".if rgb16" gate (five places) plus a 2-way
".if r_first" inside ".if rgb16" for the pack dispatch (one place).
.if/.endif count drops from 46 to 33; -88/+49 lines net.
Pure source-level refactor: the full object disassembly is byte-for-byte
identical to the pre-refactor build (MD5 2a6ac497cabc81849e0c80ec0fde0550
on Apple M1, clang). checkasm --test=sw_yuv2rgb 110/110, full checkasm
7657/7657.
Signed-off-by: DROOdotFOO <[email protected]>
---
libswscale/aarch64/yuv2rgb_neon.S | 137 ++++++++++++++------------------------
1 file changed, 49 insertions(+), 88 deletions(-)
diff --git a/libswscale/aarch64/yuv2rgb_neon.S
b/libswscale/aarch64/yuv2rgb_neon.S
index cf4b08351a..484d630998 100644
--- a/libswscale/aarch64/yuv2rgb_neon.S
+++ b/libswscale/aarch64/yuv2rgb_neon.S
@@ -63,22 +63,10 @@
add w17, w0, w0, lsl #1
sub w3, w3, w17 // w3
= linesize - width * 3 (padding)
.else
- .ifc \ofmt,rgb565le
+ .if rgb16
sub w3, w3, w0, lsl #1 // w3
= linesize - width * 2 (padding)
.else
- .ifc \ofmt,bgr565le
- sub w3, w3, w0, lsl #1 // w3
= linesize - width * 2 (padding)
- .else
- .ifc \ofmt,rgb555le
- sub w3, w3, w0, lsl #1 // w3
= linesize - width * 2 (padding)
- .else
- .ifc \ofmt,bgr555le
- sub w3, w3, w0, lsl #1 // w3
= linesize - width * 2 (padding)
- .else
sub w3, w3, w0, lsl #2 // w3
= linesize - width * 4 (padding)
- .endif
- .endif
- .endif
.endif
.endif
.endif
@@ -112,22 +100,10 @@
add w17, w0, w0, lsl #1
sub w3, w3, w17 // w3
= linesize - width * 3 (padding)
.else
- .ifc \ofmt,rgb565le
+ .if rgb16
sub w3, w3, w0, lsl #1 // w3
= linesize - width * 2 (padding)
.else
- .ifc \ofmt,bgr565le
- sub w3, w3, w0, lsl #1 // w3
= linesize - width * 2 (padding)
- .else
- .ifc \ofmt,rgb555le
- sub w3, w3, w0, lsl #1 // w3
= linesize - width * 2 (padding)
- .else
- .ifc \ofmt,bgr555le
- sub w3, w3, w0, lsl #1 // w3
= linesize - width * 2 (padding)
- .else
sub w3, w3, w0, lsl #2 // w3
= linesize - width * 4 (padding)
- .endif
- .endif
- .endif
.endif
.endif
.endif
@@ -171,22 +147,10 @@
add w17, w0, w0, lsl #1
sub w3, w3, w17 // w3
= linesize - width * 3 (padding)
.else
- .ifc \ofmt,rgb565le
+ .if rgb16
sub w3, w3, w0, lsl #1 // w3
= linesize - width * 2 (padding)
.else
- .ifc \ofmt,bgr565le
- sub w3, w3, w0, lsl #1 // w3
= linesize - width * 2 (padding)
- .else
- .ifc \ofmt,rgb555le
- sub w3, w3, w0, lsl #1 // w3
= linesize - width * 2 (padding)
- .else
- .ifc \ofmt,bgr555le
- sub w3, w3, w0, lsl #1 // w3
= linesize - width * 2 (padding)
- .else
sub w3, w3, w0, lsl #2 // w3
= linesize - width * 4 (padding)
- .endif
- .endif
- .endif
.endif
.endif
.endif
@@ -278,36 +242,50 @@
mov \a2, v29.8b //
real alpha (next 8 pixels)
.endm
-// The 16bpp output paths use v8/v9 to assemble packed pixels before the
-// final st1. v8/v9 are AAPCS callee-saved (low 64 bits must be preserved),
-// so each function spills d8/d9 to the stack on entry and reloads on exit.
-// Other output formats don't touch v8-v15, so the save/restore is gated.
-.macro save_d8_d9_if_16bpp ofmt
+// Map ofmt to .set predicates: rgb16=1 for the four 16bpp LE ofmts
+// (r_first=1 for rgb*, 0 for bgr*; gshift/hshift = 2/11 for 565,
+// 3/10 for 555), letting sibling macros branch on .if rgb16 instead of
+// repeating a four-way .ifc cascade.
+.macro set_rgb16_predicates ofmt
+ .set rgb16, 0
+ .set r_first, 0
+ .set gshift, 0
+ .set hshift, 0
.ifc \ofmt,rgb565le
- stp d8, d9, [sp, #-0x10]!
+ .set rgb16, 1
+ .set r_first, 1
+ .set gshift, 2
+ .set hshift, 11
.endif
.ifc \ofmt,bgr565le
- stp d8, d9, [sp, #-0x10]!
+ .set rgb16, 1
+ .set gshift, 2
+ .set hshift, 11
.endif
.ifc \ofmt,rgb555le
- stp d8, d9, [sp, #-0x10]!
+ .set rgb16, 1
+ .set r_first, 1
+ .set gshift, 3
+ .set hshift, 10
.endif
.ifc \ofmt,bgr555le
- stp d8, d9, [sp, #-0x10]!
+ .set rgb16, 1
+ .set gshift, 3
+ .set hshift, 10
.endif
.endm
-.macro restore_d8_d9_if_16bpp ofmt
-.ifc \ofmt,rgb565le
- ldp d8, d9, [sp], #0x10
-.endif
-.ifc \ofmt,bgr565le
- ldp d8, d9, [sp], #0x10
-.endif
-.ifc \ofmt,rgb555le
- ldp d8, d9, [sp], #0x10
+// 16bpp packing uses v8/v9 as the accumulator. AAPCS-64 requires d8/d9
+// callee-saved (low 64 bits of v8/v9); other ofmts don't touch v8-v15,
+// so the spill is gated on rgb16.
+.macro save_d8_d9_if_16bpp
+.if rgb16
+ stp d8, d9, [sp, #-0x10]!
.endif
-.ifc \ofmt,bgr555le
+.endm
+
+.macro restore_d8_d9_if_16bpp
+.if rgb16
ldp d8, d9, [sp], #0x10
.endif
.endm
@@ -333,8 +311,9 @@
.macro declare_func ifmt ofmt
function ff_\ifmt\()_to_\ofmt\()_neon, export=1
+ set_rgb16_predicates \ofmt
load_args_\ifmt \ofmt
- save_d8_d9_if_16bpp \ofmt
+ save_d8_d9_if_16bpp
movi v31.8h, #4, lsl #8 // 128
* (1<<3) (loop-invariant)
movi v30.8b, #255 //
alpha = 255 (loop-invariant)
@@ -415,39 +394,21 @@ function ff_\ifmt\()_to_\ofmt\()_neon, export=1
st1 { v6.8b, v7.8b }, [x10], #16
st1 { v18.8b, v19.8b }, [x15], #16
.else
- .ifc \ofmt,rgb565le
+ .if rgb16
compute_rgb v4.8b,v5.8b,v6.8b, v16.8b,v17.8b,v18.8b
- // RGB565 LE: (R[7:3] << 11) | (G[7:2] << 5) | B[7:3]
- pack_rgb16 v8, v6, v5, v4, 2, 11
- pack_rgb16 v9, v18, v17, v16, 2, 11
- st1 { v8.8h, v9.8h}, [x2], #32
- .else
- .ifc \ofmt,bgr565le
- compute_rgb v4.8b,v5.8b,v6.8b, v16.8b,v17.8b,v18.8b
- // BGR565 LE: (B[7:3] << 11) | (G[7:2] << 5) | R[7:3]
- pack_rgb16 v8, v4, v5, v6, 2, 11
- pack_rgb16 v9, v16, v17, v18, 2, 11
- st1 { v8.8h, v9.8h}, [x2], #32
+ .if r_first
+ // rgb*le: (R << hshift) | (G << 5) | B
+ pack_rgb16 v8, v6, v5, v4, gshift, hshift
+ pack_rgb16 v9, v18, v17, v16, gshift, hshift
.else
- .ifc \ofmt,rgb555le
- compute_rgb v4.8b,v5.8b,v6.8b, v16.8b,v17.8b,v18.8b
- // RGB555 LE: (R[7:3] << 10) | (G[7:3] << 5) | B[7:3]
- pack_rgb16 v8, v6, v5, v4, 3, 10
- pack_rgb16 v9, v18, v17, v16, 3, 10
- st1 { v8.8h, v9.8h}, [x2], #32
- .else
- .ifc \ofmt,bgr555le
- compute_rgb v4.8b,v5.8b,v6.8b, v16.8b,v17.8b,v18.8b
- // BGR555 LE: (B[7:3] << 10) | (G[7:3] << 5) | R[7:3]
- pack_rgb16 v8, v4, v5, v6, 3, 10
- pack_rgb16 v9, v16, v17, v18, 3, 10
+ // bgr*le: (B << hshift) | (G << 5) | R
+ pack_rgb16 v8, v4, v5, v6, gshift, hshift
+ pack_rgb16 v9, v16, v17, v18, gshift, hshift
+ .endif
st1 { v8.8h, v9.8h}, [x2], #32
- .else
+ .else
st4 { v4.8b, v5.8b, v6.8b, v7.8b}, [x2], #32
st4 {v16.8b,v17.8b,v18.8b,v19.8b}, [x2], #32
- .endif
- .endif
- .endif
.endif
.endif
.endif
@@ -464,7 +425,7 @@ function ff_\ifmt\()_to_\ofmt\()_neon, export=1
subs w1, w1, #1 //
height -= 1
b.gt 1b
mov w0, w9
- restore_d8_d9_if_16bpp \ofmt
+ restore_d8_d9_if_16bpp
ret
endfunc
.endm
_______________________________________________
ffmpeg-cvslog mailing list -- [email protected]
To unsubscribe send an email to [email protected]
