PR #23524 opened by Ramiro Polla (ramiro) URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/23524 Patch URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/23524.patch
`libswscale/aarch64/ops_entries.c` has diverged a bit since !22678 has stalled. This PR fixes the file, and I'll create another PR to make it run with fate. >From dea830187b189e9c9fe3c9034964c00bcde6b82d Mon Sep 17 00:00:00 2001 From: Ramiro Polla <[email protected]> Date: Tue, 16 Jun 2026 13:49:03 +0200 Subject: [PATCH 1/2] swscale/aarch64/ops: convert single-component packed read/write to planar b488ee55 relaxed the check and stopped taking the number of components into account. This led to the generation of duplicate read/write functions for single component packed/planar. ops_entries.c wasn't regenerated to take this into account; instead some conversions were no longer being supported for the neon backend. --- libswscale/aarch64/ops_impl_conv.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/libswscale/aarch64/ops_impl_conv.c b/libswscale/aarch64/ops_impl_conv.c index 98cb89edbc..479afbb3ab 100644 --- a/libswscale/aarch64/ops_impl_conv.c +++ b/libswscale/aarch64/ops_impl_conv.c @@ -88,9 +88,9 @@ static int convert_to_aarch64_impl(SwsContext *ctx, const SwsOpList *ops, int n, out->op = AARCH64_SWS_OP_READ_NIBBLE; else if (op->rw.frac == 3) out->op = AARCH64_SWS_OP_READ_BIT; - else if (op->rw.mode == SWS_RW_PACKED) + else if (op->rw.mode == SWS_RW_PACKED && op->rw.elems > 1) out->op = AARCH64_SWS_OP_READ_PACKED; - else if (op->rw.mode == SWS_RW_PLANAR) + else if (op->rw.mode == SWS_RW_PACKED || op->rw.mode == SWS_RW_PLANAR) out->op = AARCH64_SWS_OP_READ_PLANAR; else return AVERROR(ENOTSUP); @@ -106,9 +106,9 @@ static int convert_to_aarch64_impl(SwsContext *ctx, const SwsOpList *ops, int n, out->op = AARCH64_SWS_OP_WRITE_NIBBLE; else if (op->rw.frac == 3) out->op = AARCH64_SWS_OP_WRITE_BIT; - else if (op->rw.mode == SWS_RW_PACKED) + else if (op->rw.mode == SWS_RW_PACKED && op->rw.elems > 1) out->op = AARCH64_SWS_OP_WRITE_PACKED; - else if (op->rw.mode == SWS_RW_PLANAR) + else if (op->rw.mode == SWS_RW_PACKED || op->rw.mode == SWS_RW_PLANAR) out->op = AARCH64_SWS_OP_WRITE_PLANAR; else return AVERROR(ENOTSUP); -- 2.52.0 >From 7c50b687cbfb3368bc32ef1a0863ca2065152ce8 Mon Sep 17 00:00:00 2001 From: Ramiro Polla <[email protected]> Date: Tue, 16 Jun 2026 14:24:50 +0200 Subject: [PATCH 2/2] swscale/aarch64/ops: skip filtering ops lists when generating ops_entries.c sws_ops_aarch64 would still register operations that came before the filtering ops (which are unsupported). --- libswscale/aarch64/ops_entries.c | 7 ------- libswscale/tests/sws_ops_aarch64.c | 16 ++++++++++++++++ 2 files changed, 16 insertions(+), 7 deletions(-) diff --git a/libswscale/aarch64/ops_entries.c b/libswscale/aarch64/ops_entries.c index 04a665a9f1..4b8e4bbae1 100644 --- a/libswscale/aarch64/ops_entries.c +++ b/libswscale/aarch64/ops_entries.c @@ -108,7 +108,6 @@ { .op = AARCH64_SWS_OP_SWIZZLE, .swizzle = 0x0213, .block_size = 16, .type = AARCH64_PIXEL_U8, .mask = 0x1001 }, { .op = AARCH64_SWS_OP_SWIZZLE, .swizzle = 0x0213, .block_size = 32, .type = AARCH64_PIXEL_U8, .mask = 0x1001 }, { .op = AARCH64_SWS_OP_SWIZZLE, .swizzle = 0x0231, .block_size = 8, .type = AARCH64_PIXEL_U8, .mask = 0x1011 }, -{ .op = AARCH64_SWS_OP_SWIZZLE, .swizzle = 0x0312, .block_size = 8, .type = AARCH64_PIXEL_U8, .mask = 0x1101 }, { .op = AARCH64_SWS_OP_SWIZZLE, .swizzle = 0x0312, .block_size = 16, .type = AARCH64_PIXEL_U8, .mask = 0x1101 }, { .op = AARCH64_SWS_OP_SWIZZLE, .swizzle = 0x0312, .block_size = 32, .type = AARCH64_PIXEL_U8, .mask = 0x1101 }, { .op = AARCH64_SWS_OP_SWIZZLE, .swizzle = 0x0321, .block_size = 8, .type = AARCH64_PIXEL_U8, .mask = 0x1111 }, @@ -247,7 +246,6 @@ { .op = AARCH64_SWS_OP_CLEAR, .block_size = 16, .type = AARCH64_PIXEL_U16, .mask = 0x0001 }, { .op = AARCH64_SWS_OP_CLEAR, .block_size = 16, .type = AARCH64_PIXEL_U16, .mask = 0x0010 }, { .op = AARCH64_SWS_OP_CLEAR, .block_size = 16, .type = AARCH64_PIXEL_U16, .mask = 0x1000 }, -{ .op = AARCH64_SWS_OP_CONVERT, .to_type = AARCH64_PIXEL_U8, .block_size = 8, .type = AARCH64_PIXEL_U16, .mask = 0x0111 }, { .op = AARCH64_SWS_OP_CONVERT, .to_type = AARCH64_PIXEL_U8, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x0001 }, { .op = AARCH64_SWS_OP_CONVERT, .to_type = AARCH64_PIXEL_U8, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x0011 }, { .op = AARCH64_SWS_OP_CONVERT, .to_type = AARCH64_PIXEL_U8, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x0111 }, @@ -256,8 +254,6 @@ { .op = AARCH64_SWS_OP_CONVERT, .to_type = AARCH64_PIXEL_U8, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x1111 }, { .op = AARCH64_SWS_OP_CONVERT, .to_type = AARCH64_PIXEL_U8, .block_size = 16, .type = AARCH64_PIXEL_U16, .mask = 0x0111 }, { .op = AARCH64_SWS_OP_CONVERT, .to_type = AARCH64_PIXEL_U16, .block_size = 8, .type = AARCH64_PIXEL_U8, .mask = 0x0111 }, -{ .op = AARCH64_SWS_OP_CONVERT, .to_type = AARCH64_PIXEL_U16, .block_size = 8, .type = AARCH64_PIXEL_U32, .mask = 0x0010 }, -{ .op = AARCH64_SWS_OP_CONVERT, .to_type = AARCH64_PIXEL_U16, .block_size = 8, .type = AARCH64_PIXEL_U32, .mask = 0x0100 }, { .op = AARCH64_SWS_OP_CONVERT, .to_type = AARCH64_PIXEL_U16, .block_size = 8, .type = AARCH64_PIXEL_U32, .mask = 0x0111 }, { .op = AARCH64_SWS_OP_CONVERT, .to_type = AARCH64_PIXEL_U16, .block_size = 8, .type = AARCH64_PIXEL_U32, .mask = 0x1110 }, { .op = AARCH64_SWS_OP_CONVERT, .to_type = AARCH64_PIXEL_U16, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x0001 }, @@ -312,7 +308,6 @@ { .op = AARCH64_SWS_OP_MAX, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x0011 }, { .op = AARCH64_SWS_OP_MAX, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x0111 }, { .op = AARCH64_SWS_OP_MAX, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x1001 }, -{ .op = AARCH64_SWS_OP_MAX, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x1110 }, { .op = AARCH64_SWS_OP_MAX, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x1111 }, { .op = AARCH64_SWS_OP_SCALE, .block_size = 8, .type = AARCH64_PIXEL_U32, .mask = 0x0001 }, { .op = AARCH64_SWS_OP_SCALE, .block_size = 8, .type = AARCH64_PIXEL_U32, .mask = 0x0111 }, @@ -372,11 +367,9 @@ { .op = AARCH64_SWS_OP_DITHER, .dither.y_offset = 0x3ff0, .dither.size_log2 = 4, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x1001 }, { .op = AARCH64_SWS_OP_DITHER, .dither.y_offset = 0x5023, .dither.size_log2 = 4, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x1111 }, { .op = AARCH64_SWS_OP_DITHER, .dither.y_offset = 0x5032, .dither.size_log2 = 4, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x1111 }, -{ .op = AARCH64_SWS_OP_DITHER, .dither.y_offset = 0x5203, .dither.size_log2 = 4, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x1111 }, { .op = AARCH64_SWS_OP_DITHER, .dither.y_offset = 0x5230, .dither.size_log2 = 4, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x1111 }, { .op = AARCH64_SWS_OP_DITHER, .dither.y_offset = 0x5ff0, .dither.size_log2 = 4, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x1001 }, { .op = AARCH64_SWS_OP_DITHER, .dither.y_offset = 0x5fff, .dither.size_log2 = 4, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x1000 }, -{ .op = AARCH64_SWS_OP_DITHER, .dither.y_offset = 0xf000, .dither.size_log2 = 4, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x0111 }, { .op = AARCH64_SWS_OP_DITHER, .dither.y_offset = 0xf023, .dither.size_log2 = 4, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x0111 }, { .op = AARCH64_SWS_OP_DITHER, .dither.y_offset = 0xf032, .dither.size_log2 = 4, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x0111 }, { .op = AARCH64_SWS_OP_DITHER, .dither.y_offset = 0xf203, .dither.size_log2 = 4, .block_size = 8, .type = AARCH64_PIXEL_F32, .mask = 0x0111 }, diff --git a/libswscale/tests/sws_ops_aarch64.c b/libswscale/tests/sws_ops_aarch64.c index 4fa10c7bb0..3319af0536 100644 --- a/libswscale/tests/sws_ops_aarch64.c +++ b/libswscale/tests/sws_ops_aarch64.c @@ -77,6 +77,22 @@ static int register_op(SwsContext *ctx, void *opaque, SwsOpList *ops) struct AVTreeNode **root = (struct AVTreeNode **) opaque; int ret; + /* Skip ops lists which include filtering, since this is still not + * supported. */ + for (int i = 0; i < ops->num_ops; i++) { + const SwsOp *op = &ops->ops[i]; + switch (op->op) { + case SWS_OP_READ: + case SWS_OP_WRITE: + if (op->rw.filter.op) + return 0; + break; + case SWS_OP_FILTER_H: + case SWS_OP_FILTER_V: + return 0; + } + } + /* Make on-stack copy of `ops` to iterate over */ SwsOpList rest = *ops; /* Use at most two full vregs during the widest precision section */ -- 2.52.0 _______________________________________________ ffmpeg-devel mailing list -- [email protected] To unsubscribe send an email to [email protected]
