Re: [FFmpeg-devel] [PATCH 3/6] avcodec/pngdec: Optimize has_trns code

[Michael Niedermayer]

On Sun, Aug 18, 2019 at 08:49:06AM +0200, Reimar Döffinger wrote:
> On 18.08.2019, at 01:28, Michael Niedermayer  wrote:
> 
> > 30M cycles -> 5M cycles
> 
> I see nothing wrong with it, but:
> You could save reviewers a lot of time if you gave them a hint of what the 
> change
> contains instead of them having to reverse-engineer.
> In this case for example something like
> "add inner loop specialisations for 2 bpp and 4 bpp"

added


> 
> > 
> > @@ -1363,19 +1364,38 @@ exit_loop:
> > unsigned x, y;
> > 
> > av_assert0(s->bit_depth > 1);
> > -
> 
> Cosmetic?

removed and reposted 

thx

[...]
-- 
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

When you are offended at any man's fault, turn to yourself and study your
own failings. Then you will forget your anger. -- Epictetus


signature.asc
Description: PGP signature
___
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH 3/6] avcodec/pngdec: Optimize has_trns code

[Reimar Döffinger]

On 18.08.2019, at 01:28, Michael Niedermayer  wrote:

> 30M cycles -> 5M cycles

I see nothing wrong with it, but:
You could save reviewers a lot of time if you gave them a hint of what the 
change
contains instead of them having to reverse-engineer.
In this case for example something like
"add inner loop specialisations for 2 bpp and 4 bpp"

> 
> @@ -1363,19 +1364,38 @@ exit_loop:
> unsigned x, y;
> 
> av_assert0(s->bit_depth > 1);
> -

Cosmetic?
___
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".

[FFmpeg-devel] [PATCH 3/6] avcodec/pngdec: Optimize has_trns code

[Michael Niedermayer]

30M cycles -> 5M cycles

Testcase: fate-rgbapng-4816
Testcase: 
16097/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_APNG_fuzzer-5664690889293824

Found-by: continuous fuzzing process 
https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
---
 libavcodec/pngdec.c | 38 +-
 1 file changed, 29 insertions(+), 9 deletions(-)

diff --git a/libavcodec/pngdec.c b/libavcodec/pngdec.c
index cad5796545..4ca4f7bdc1 100644
--- a/libavcodec/pngdec.c
+++ b/libavcodec/pngdec.c
@@ -24,6 +24,7 @@
 #include "libavutil/avassert.h"
 #include "libavutil/bprint.h"
 #include "libavutil/imgutils.h"
+#include "libavutil/intreadwrite.h"
 #include "libavutil/stereo3d.h"
 #include "libavutil/mastering_display_metadata.h"
 
@@ -1363,19 +1364,38 @@ exit_loop:
 unsigned x, y;
 
 av_assert0(s->bit_depth > 1);
-
 for (y = 0; y < s->height; ++y) {
 uint8_t *row = >image_buf[s->image_linesize * y];
 
-/* since we're updating in-place, we have to go from right to left 
*/
-for (x = s->width; x > 0; --x) {
-uint8_t *pixel = [s->bpp * (x - 1)];
-memmove(pixel, [raw_bpp * (x - 1)], raw_bpp);
+if (s->bpp == 2 && byte_depth == 1) {
+uint8_t *pixel = [2 * s->width - 1];
+uint8_t *rowp  = [1 * s->width - 1];
+int tcolor = s->transparent_color_be[0];
+for (x = s->width; x > 0; --x) {
+*pixel-- = *rowp == tcolor ? 0 : 0xff;
+*pixel-- = *rowp--;
+}
+} else if (s->bpp == 4 && byte_depth == 1) {
+uint8_t *pixel = [4 * s->width - 1];
+uint8_t *rowp  = [3 * s->width - 1];
+int tcolor = AV_RL24(s->transparent_color_be);
+for (x = s->width; x > 0; --x) {
+*pixel-- = AV_RL24(rowp-2) == tcolor ? 0 : 0xff;
+*pixel-- = *rowp--;
+*pixel-- = *rowp--;
+*pixel-- = *rowp--;
+}
+} else {
+/* since we're updating in-place, we have to go from right to 
left */
+for (x = s->width; x > 0; --x) {
+uint8_t *pixel = [s->bpp * (x - 1)];
+memmove(pixel, [raw_bpp * (x - 1)], raw_bpp);
 
-if (!memcmp(pixel, s->transparent_color_be, raw_bpp)) {
-memset([raw_bpp], 0, byte_depth);
-} else {
-memset([raw_bpp], 0xff, byte_depth);
+if (!memcmp(pixel, s->transparent_color_be, raw_bpp)) {
+memset([raw_bpp], 0, byte_depth);
+} else {
+memset([raw_bpp], 0xff, byte_depth);
+}
 }
 }
 }
-- 
2.22.1

___
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".