PR #20933 opened by nots1dd
URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20933
Patch URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20933.patch
>From what I understand, we only need to be moving N bytes as window_size
- s->hop_size after inverse fft to the buffer for overlap. Currently the
compiler throws a warning regarding possible overflow and this patch
fixs it.
I ran FATE and tests are passing
>From 93b8c7d3acdb09c8c9514c513bfb734cac8cbe48 Mon Sep 17 00:00:00 2001
From: nots1dd
Date: Sun, 16 Nov 2025 16:12:24 +0530
Subject: [PATCH] avfilter/af_afftfilt: fix the overlap buffer size
>From what I understand, we only need to be moving N bytes as window_size
- s->hop_size after inverse fft to the buffer for overlap. Currently the
compiler throws a warning regarding possible overflow and this patch
fixs it.
---
libavfilter/af_afftfilt.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/libavfilter/af_afftfilt.c b/libavfilter/af_afftfilt.c
index 08cdcae2f7..4fc28c6143 100644
--- a/libavfilter/af_afftfilt.c
+++ b/libavfilter/af_afftfilt.c
@@ -250,6 +250,7 @@ static int filter_channel(AVFilterContext *ctx, void *arg,
int jobnr, int nb_job
{
AFFTFiltContext *s = ctx->priv;
const int window_size = s->window_size;
+const int hop_size = s->hop_size;
const float *window_lut = s->window_func_lut;
const float f = sqrtf(1.f - s->overlap);
const int channels = s->channels;
@@ -294,7 +295,7 @@ static int filter_channel(AVFilterContext *ctx, void *arg,
int jobnr, int nb_job
s->itx_fn(s->ifft[ch], fft_out, fft_temp, sizeof(*fft_temp));
-memmove(buf, buf + s->hop_size, window_size * sizeof(float));
+memmove(buf, buf + hop_size, (window_size - hop_size) * sizeof(float));
for (int i = 0; i < window_size; i++)
buf[i] += fft_out[i].re * window_lut[i] * f;
}
--
2.49.1
___
ffmpeg-devel mailing list -- [email protected]
To unsubscribe send an email to [email protected]