Message ID | tencent_E931D38C21E051D99E7927CB2E4268D9E908@qq.com |
---|---|
State | New |
Headers | show |
Series | [FFmpeg-devel] avutil/imgutils: optimize image copy efficiency | expand |
Context | Check | Description |
---|---|---|
yinshiyou/make_loongarch64 | success | Make finished |
yinshiyou/make_fate_loongarch64 | success | Make fate finished |
andriy/make_x86 | success | Make finished |
andriy/make_fate_x86 | success | Make fate finished |
On Thu, 4 May 2023, xufuji456 wrote: > It makes sense when copying 4K/8K video, if linesize > equals to aligned linesize. > > Signed-off-by: xufuji456 <839789740@qq.com> > --- > libavutil/imgutils.c | 15 +++++++++++---- > 1 file changed, 11 insertions(+), 4 deletions(-) > > diff --git a/libavutil/imgutils.c b/libavutil/imgutils.c > index 9ab5757cf6..1d432e7a57 100644 > --- a/libavutil/imgutils.c > +++ b/libavutil/imgutils.c > @@ -525,10 +525,17 @@ int av_image_copy_to_buffer(uint8_t *dst, int dst_size, > const uint8_t *src = src_data[i]; > h = (height + (1 << shift) - 1) >> shift; > > - for (j = 0; j < h; j++) { > - memcpy(dst, src, linesize[i]); > - dst += FFALIGN(linesize[i], align); > - src += src_linesize[i]; > + if (FFALIGN(linesize[i], align) == linesize[i] && src_linesize[i] == linesize[i]) { > + int size = linesize[i] * h; For cases like these, I would prefer to use "linesize[i] * (h - 1) + w" instead. For cases if copying e.g. into an offsetted position within a buffer, writing the last trailing padding would end up writing out of bounds. That said, I'm unsure about how much gain you get from this optimization here - some numbers to back it up would be useful. // Martin
Thank you for your review, Martin.
I try add some data, and submit again.
------------------ Original ------------------
From: "FFmpeg development discussions and patches" <martin@martin.st>;
Date: Thu, May 4, 2023 04:22 PM
To: "FFmpeg development discussions and patches"<ffmpeg-devel@ffmpeg.org>;
Cc: "徐福隆"<839789740@qq.com>;
Subject: Re: [FFmpeg-devel] [PATCH] avutil/imgutils: optimize image copy efficiency
On Thu, 4 May 2023, xufuji456 wrote:
> It makes sense when copying 4K/8K video, if linesize
> equals to aligned linesize.
>
> Signed-off-by: xufuji456 <839789740@qq.com>
> ---
> libavutil/imgutils.c | 15 +++++++++++----
> 1 file changed, 11 insertions(+), 4 deletions(-)
>
> diff --git a/libavutil/imgutils.c b/libavutil/imgutils.c
> index 9ab5757cf6..1d432e7a57 100644
> --- a/libavutil/imgutils.c
> +++ b/libavutil/imgutils.c
> @@ -525,10 +525,17 @@ int av_image_copy_to_buffer(uint8_t *dst, int dst_size,
> const uint8_t *src = src_data[i];
> h = (height + (1 << shift) - 1) >> shift;
>
> - for (j = 0; j < h; j++) {
> - memcpy(dst, src, linesize[i]);
> - dst += FFALIGN(linesize[i], align);
> - src += src_linesize[i];
> + if (FFALIGN(linesize[i], align) == linesize[i] && src_linesize[i] == linesize[i]) {
> + int size = linesize[i] * h;
For cases like these, I would prefer to use "linesize[i] * (h - 1) + w"
instead. For cases if copying e.g. into an offsetted position within a
buffer, writing the last trailing padding would end up writing out of
bounds.
That said, I'm unsure about how much gain you get from this optimization
here - some numbers to back it up would be useful.
// Martin
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
There is an if condition "FFALIGN(linesize[i], align) == linesize[i] && src_linesize[i] == linesize[i]".
When src's linesize equals to dst, it has no padding in this case I think. If has padding, it will execute
previous process. As a result, we could ignore the situation of padding in this case.
Thank you
//frank
------------------ Original ------------------
From: "FFmpeg development discussions and patches" <martin@martin.st>;
Date: Thu, May 4, 2023 04:22 PM
To: "FFmpeg development discussions and patches"<ffmpeg-devel@ffmpeg.org>;
Cc: "徐福隆"<839789740@qq.com>;
Subject: Re: [FFmpeg-devel] [PATCH] avutil/imgutils: optimize image copy efficiency
On Thu, 4 May 2023, xufuji456 wrote:
> It makes sense when copying 4K/8K video, if linesize
> equals to aligned linesize.
>
> Signed-off-by: xufuji456 <839789740@qq.com>
> ---
> libavutil/imgutils.c | 15 +++++++++++----
> 1 file changed, 11 insertions(+), 4 deletions(-)
>
> diff --git a/libavutil/imgutils.c b/libavutil/imgutils.c
> index 9ab5757cf6..1d432e7a57 100644
> --- a/libavutil/imgutils.c
> +++ b/libavutil/imgutils.c
> @@ -525,10 +525,17 @@ int av_image_copy_to_buffer(uint8_t *dst, int dst_size,
> const uint8_t *src = src_data[i];
> h = (height + (1 << shift) - 1) >> shift;
>
> - for (j = 0; j < h; j++) {
> - memcpy(dst, src, linesize[i]);
> - dst += FFALIGN(linesize[i], align);
> - src += src_linesize[i];
> + if (FFALIGN(linesize[i], align) == linesize[i] && src_linesize[i] == linesize[i]) {
> + int size = linesize[i] * h;
For cases like these, I would prefer to use "linesize[i] * (h - 1) + w"
instead. For cases if copying e.g. into an offsetted position within a
buffer, writing the last trailing padding would end up writing out of
bounds.
That said, I'm unsure about how much gain you get from this optimization
here - some numbers to back it up would be useful.
// Martin
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
diff --git a/libavutil/imgutils.c b/libavutil/imgutils.c index 9ab5757cf6..1d432e7a57 100644 --- a/libavutil/imgutils.c +++ b/libavutil/imgutils.c @@ -525,10 +525,17 @@ int av_image_copy_to_buffer(uint8_t *dst, int dst_size, const uint8_t *src = src_data[i]; h = (height + (1 << shift) - 1) >> shift; - for (j = 0; j < h; j++) { - memcpy(dst, src, linesize[i]); - dst += FFALIGN(linesize[i], align); - src += src_linesize[i]; + if (FFALIGN(linesize[i], align) == linesize[i] && src_linesize[i] == linesize[i]) { + int size = linesize[i] * h; + memcpy(dst, src, size); + dst += size; + src += size; + } else { + for (j = 0; j < h; j++) { + memcpy(dst, src, linesize[i]); + dst += FFALIGN(linesize[i], align); + src += src_linesize[i]; + } } }
It makes sense when copying 4K/8K video, if linesize equals to aligned linesize. Signed-off-by: xufuji456 <839789740@qq.com> --- libavutil/imgutils.c | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-)