PR #23527 opened by Ramiro Polla (ramiro)
URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/23527
Patch URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/23527.patch
Unused components weren't being taken into consideration.
Note this commit includes a hack to prevent checkasm from aborting,
since checkasm does {read, unpack, write} which doesn't properly mark
the unpacked components as unused.
>From 48c9b38087b91800d3f6aa94419209cb0400a750 Mon Sep 17 00:00:00 2001
From: Ramiro Polla <[email protected]>
Date: Fri, 12 Jun 2026 02:41:09 +0200
Subject: [PATCH] swscale/uops: generate more specific masks for unpack when
translating
Unused components weren't being taken into consideration.
Note this commit includes a hack to prevent checkasm from aborting,
since checkasm does {read, unpack, write} which doesn't properly mark
the unpacked components as unused.
---
libswscale/uops.c | 14 +++++++++++++-
libswscale/uops_macros.h | 8 ++++++++
tests/ref/fate/sws-ops-list | 2 +-
3 files changed, 22 insertions(+), 2 deletions(-)
diff --git a/libswscale/uops.c b/libswscale/uops.c
index 1bd3e2f763..a9dd0d41cf 100644
--- a/libswscale/uops.c
+++ b/libswscale/uops.c
@@ -799,7 +799,8 @@ static int translate_op(SwsContext *ctx, SwsUOpList *uops,
SwsUOpFlags flags,
uop.mask = 0;
for (int i = 0; i < 4 && op->pack.pattern[i]; i++) {
uop.par.pack.pattern[i] = op->pack.pattern[i];
- uop.mask |= SWS_COMP(i);
+ if (op->op == SWS_OP_PACK || SWS_OP_NEEDED(op, i))
+ uop.mask |= SWS_COMP(i);
}
break;
case SWS_OP_LSHIFT:
@@ -1021,6 +1022,17 @@ int ff_sws_uops_macros_gen(char **out_str)
}
}
+ /* HACK: add some patterns for checkasm */
+ const SwsUOp extra_uops[] = {
+ { .type = SWS_PIXEL_U32, .uop = SWS_UOP_UNPACK, .mask = 0xf,
.par.pack.pattern = {2, 10, 10, 10} },
+ { .type = SWS_PIXEL_U32, .uop = SWS_UOP_UNPACK, .mask = 0xf,
.par.pack.pattern = {10, 10, 10, 2} },
+ };
+ for (int i = 0; i < FF_ARRAY_ELEMS(extra_uops); i++) {
+ ret = register_uop(&root, &extra_uops[i]);
+ if (ret < 0)
+ goto fail;
+ }
+
#define BPRINT_STR(str) av_bprint_append_data(bp, str, strlen(str))
BPRINT_STR(
"/**\n"
diff --git a/libswscale/uops_macros.h b/libswscale/uops_macros.h
index 3c4d6b6a3e..62743a8359 100644
--- a/libswscale/uops_macros.h
+++ b/libswscale/uops_macros.h
@@ -934,9 +934,17 @@
#define SWS_FOR_U32_MAX(MACRO, ...)
#define SWS_FOR_STRUCT_U32_MAX(MACRO, ...)
#define SWS_FOR_U32_UNPACK(MACRO, ...) \
+ MACRO(__VA_ARGS__, u32_unpack_y_aaa2 ,
SWS_PIXEL_U32, SWS_UOP_UNPACK , 0x2, 10, 10, 10, 2) \
+ MACRO(__VA_ARGS__, u32_unpack_z_2aaa ,
SWS_PIXEL_U32, SWS_UOP_UNPACK , 0x4, 2, 10, 10, 10) \
+ MACRO(__VA_ARGS__, u32_unpack_xyz_aaa2 ,
SWS_PIXEL_U32, SWS_UOP_UNPACK , 0x7, 10, 10, 10, 2) \
+ MACRO(__VA_ARGS__, u32_unpack_yzw_2aaa ,
SWS_PIXEL_U32, SWS_UOP_UNPACK , 0xe, 2, 10, 10, 10) \
MACRO(__VA_ARGS__, u32_unpack_xyzw_2aaa ,
SWS_PIXEL_U32, SWS_UOP_UNPACK , 0xf, 2, 10, 10, 10) \
MACRO(__VA_ARGS__, u32_unpack_xyzw_aaa2 ,
SWS_PIXEL_U32, SWS_UOP_UNPACK , 0xf, 10, 10, 10, 2)
#define SWS_FOR_STRUCT_U32_UNPACK(MACRO, ...) \
+ MACRO(__VA_ARGS__, u32_unpack_y_aaa2 , .type =
SWS_PIXEL_U32, .uop = SWS_UOP_UNPACK , .mask = 0x2, .par.pack.pattern
= {10, 10, 10, 2}) \
+ MACRO(__VA_ARGS__, u32_unpack_z_2aaa , .type =
SWS_PIXEL_U32, .uop = SWS_UOP_UNPACK , .mask = 0x4, .par.pack.pattern
= {2, 10, 10, 10}) \
+ MACRO(__VA_ARGS__, u32_unpack_xyz_aaa2 , .type =
SWS_PIXEL_U32, .uop = SWS_UOP_UNPACK , .mask = 0x7, .par.pack.pattern
= {10, 10, 10, 2}) \
+ MACRO(__VA_ARGS__, u32_unpack_yzw_2aaa , .type =
SWS_PIXEL_U32, .uop = SWS_UOP_UNPACK , .mask = 0xe, .par.pack.pattern
= {2, 10, 10, 10}) \
MACRO(__VA_ARGS__, u32_unpack_xyzw_2aaa , .type =
SWS_PIXEL_U32, .uop = SWS_UOP_UNPACK , .mask = 0xf, .par.pack.pattern
= {2, 10, 10, 10}) \
MACRO(__VA_ARGS__, u32_unpack_xyzw_aaa2 , .type =
SWS_PIXEL_U32, .uop = SWS_UOP_UNPACK , .mask = 0xf, .par.pack.pattern
= {10, 10, 10, 2})
#define SWS_FOR_U32_PACK(MACRO, ...) \
diff --git a/tests/ref/fate/sws-ops-list b/tests/ref/fate/sws-ops-list
index 6b4003121a..f3311b643f 100644
--- a/tests/ref/fate/sws-ops-list
+++ b/tests/ref/fate/sws-ops-list
@@ -1 +1 @@
-bbe27c8c324f08d933f6397f5fb96650
+2f427f7cbfaef38013a3fbefcb5e1e57
--
2.52.0
_______________________________________________
ffmpeg-devel mailing list -- [email protected]
To unsubscribe send an email to [email protected]
