diff mbox

[FFmpeg-devel] avcodec/dpx: Support for RGBA 12-bit packed decoding

Message ID e8afa2d1-1d07-4553-ad4c-a0c77d2dbb7a@mediaarea.net
State Accepted
Commit 5205b3289e46b680c314e44a9243929b6d00146a
Headers show

Commit Message

Jerome Martinez June 1, 2018, 9:12 a.m. UTC 
On 08/02/2018 11:28, Jerome Martinez wrote:
> Currently RGB and RGBA 12-bit are supported by DPX decoder only if 
> component values are padded (packing "Filled to 32-bit words, method A").
> This patch adds decoding of RGB and RGBA 12-bit with no padding 
> (packing "Packed into 32-bit words").
>
> As I have no file with line boundaries not aligned on 32-bit, I can 
> not have good tests about the stride computing (so code about non 
> aligned boundaries is theory) so I preferred to limit risks by 
> decoding only if line boundaries are aligned on 32-bit words:
> - 8 pixels for RGB (8 pixels x 3 components x 12 bits = 288 bits = 9 x 
> 32-bit words)
> - 2 pixels for RGBA (2 pixels x 4 components x 12 bits = 3 x 32-bit 
> words)
>
> I think Little Endian parsing is fine thanks to the generic code about 
> Big vs Little endian but I have no Little Endian test file so I also 
> limited the decoding to Big Endian files.
>
> Would be happy to check with cases I was not able to check if someone 
> provides files. 


Since the email, RGB 12-bit packed decoding was pushed.
Attached is the patch for RGBA 12-bit packed decoding, rebased.

Test file:
https://trac.ffmpeg.org/attachment/ticket/5639/12bit_a.dpx
From b22c71ed0d81bce8a0a0116052970f8a9bcc9008 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?J=C3=A9r=C3=B4me=20Martinez?= <jerome@mediaarea.net>
Date: Fri, 1 Jun 2018 10:09:01 +0200
Subject: [PATCH] avcodec/dpx: Support for RGBA 12-bit packed decoding

Limited to widths multiple of 2 due to lack of test files for such corner case

This partially fixes ticket #5639
---
 libavcodec/dpx.c | 3 +++
 1 file changed, 3 insertions(+)

Comments

Paul B Mahol June 1, 2018, 9:53 a.m. UTC | #1 
On 6/1/18, Jerome Martinez <jerome@mediaarea.net> wrote:
> On 08/02/2018 11:28, Jerome Martinez wrote:
>> Currently RGB and RGBA 12-bit are supported by DPX decoder only if
>> component values are padded (packing "Filled to 32-bit words, method A").
>> This patch adds decoding of RGB and RGBA 12-bit with no padding
>> (packing "Packed into 32-bit words").
>>
>> As I have no file with line boundaries not aligned on 32-bit, I can
>> not have good tests about the stride computing (so code about non
>> aligned boundaries is theory) so I preferred to limit risks by
>> decoding only if line boundaries are aligned on 32-bit words:
>> - 8 pixels for RGB (8 pixels x 3 components x 12 bits = 288 bits = 9 x
>> 32-bit words)
>> - 2 pixels for RGBA (2 pixels x 4 components x 12 bits = 3 x 32-bit
>> words)
>>
>> I think Little Endian parsing is fine thanks to the generic code about
>> Big vs Little endian but I have no Little Endian test file so I also
>> limited the decoding to Big Endian files.
>>
>> Would be happy to check with cases I was not able to check if someone
>> provides files.
>
>
> Since the email, RGB 12-bit packed decoding was pushed.
> Attached is the patch for RGBA 12-bit packed decoding, rebased.
>
> Test file:
> https://trac.ffmpeg.org/attachment/ticket/5639/12bit_a.dpx
>

applied.
diff mbox

Patch

diff --git a/libavcodec/dpx.c b/libavcodec/dpx.c
index 026fb10e90..fb388b6e52 100644
--- a/libavcodec/dpx.c
+++ b/libavcodec/dpx.c
@@ -237,6 +237,9 @@  static int decode_frame(AVCodecContext *avctx,
             if (descriptor == 50 && endian && (avctx->width%8) == 0) { // Little endian and widths not a multiple of 8 need tests
                 tested = 1;
             }
+            if (descriptor == 51 && endian && (avctx->width%2) == 0) { // Little endian and widths not a multiple of 2 need tests
+                tested = 1;
+            }
             if (!tested) {
                 av_log(avctx, AV_LOG_ERROR, "Packing to 16bit required\n");
                 return -1;