From 0e983a060451bc46f320274b5a0c10f492b555ac Mon Sep 17 00:00:00 2001
From: Niklas Haas <git@haasn.dev>
Date: Wed, 25 Mar 2026 14:09:30 +0100
Subject: [PATCH] swscale: align allocated frame buffers to SwsPass hints

This avoids hitting the slow memcpy fallback paths altogether, whenever
swscale.c is handling plane allocation.

Signed-off-by: Niklas Haas <git@haasn.dev>
---
 libswscale/graph.c   | 11 +++++++++++
 libswscale/graph.h   |  5 +++++
 libswscale/swscale.c | 24 +++++++++++++++++-------
 3 files changed, 33 insertions(+), 7 deletions(-)

diff --git a/libswscale/graph.c b/libswscale/graph.c
index 6836f8cf60..2020fcf8c7 100644
--- a/libswscale/graph.c
+++ b/libswscale/graph.c
@@ -38,6 +38,17 @@
 #include "graph.h"
 #include "ops.h"
 
+int ff_sws_pass_aligned_width(const SwsPass *pass, int width)
+{
+    if (!pass)
+        return width;
+
+    size_t aligned_w = width;
+    aligned_w = FFALIGN(aligned_w, pass->output->width_align);
+    aligned_w += pass->output->width_pad;
+    return aligned_w <= INT_MAX ? aligned_w : width;
+}
+
 /* Allocates one buffer per plane */
 static int frame_alloc_planes(AVFrame *dst)
 {
diff --git a/libswscale/graph.h b/libswscale/graph.h
index 886f0d2996..116f2a89bf 100644
--- a/libswscale/graph.h
+++ b/libswscale/graph.h
@@ -110,6 +110,11 @@ struct SwsPass {
     void *priv;
 };
 
+/**
+ * Align `width` to the optimal size for `pass`.
+ */
+int ff_sws_pass_aligned_width(const SwsPass *pass, int width);
+
 /**
  * Filter graph, which represents a 'baked' pixel format conversion.
  */
diff --git a/libswscale/swscale.c b/libswscale/swscale.c
index d5c91e541c..3ebbcd5d3f 100644
--- a/libswscale/swscale.c
+++ b/libswscale/swscale.c
@@ -1479,14 +1479,9 @@ int sws_frame_setup(SwsContext *ctx, const AVFrame *dst, const AVFrame *src)
         if (ret < 0)
             return ret;
 #endif
-    } else {
-        /* Software frames */
-        ret = ff_frame_pool_video_reinit(&s->frame_pool, dst->width, dst->height,
-                                         dst->format, av_cpu_max_align());
-        if (ret < 0)
-            return ret;
     }
 
+    int dst_width = dst->width;
     for (int field = 0; field < 2; field++) {
         SwsFormat src_fmt = ff_fmt_from_frame(src, field);
         SwsFormat dst_fmt = ff_fmt_from_frame(dst, field);
@@ -1512,12 +1507,20 @@ int sws_frame_setup(SwsContext *ctx, const AVFrame *dst, const AVFrame *src)
             goto fail;
         }
 
-        if (s->graph[field]->incomplete && ctx->flags & SWS_STRICT) {
+        const SwsGraph *graph = s->graph[field];
+        if (graph->incomplete && ctx->flags & SWS_STRICT) {
             err_msg = "Incomplete scaling graph";
             ret = AVERROR(EINVAL);
             goto fail;
         }
 
+        if (!graph->noop) {
+            av_assert0(graph->num_passes);
+            const SwsPass *last_pass = graph->passes[graph->num_passes - 1];
+            const int aligned_w = ff_sws_pass_aligned_width(last_pass, dst->width);
+            dst_width = FFMAX(dst_width, aligned_w);
+        }
+
         if (!src_fmt.interlaced) {
             ff_sws_graph_free(&s->graph[FIELD_BOTTOM]);
             break;
@@ -1540,6 +1543,13 @@ int sws_frame_setup(SwsContext *ctx, const AVFrame *dst, const AVFrame *src)
         return ret;
     }
 
+    if (!dst->hw_frames_ctx) {
+        ret = ff_frame_pool_video_reinit(&s->frame_pool, dst_width, dst->height,
+                                         dst->format, av_cpu_max_align());
+        if (ret < 0)
+            return ret;
+    }
+
     return 0;
 }