[FFmpeg-devel] [PATCH] swscale/aarch64: Fix rgb24toyv12 only works with aligned width

[Zhao Zhili]

> On Sep 18, 2024, at 20:51, Martin Storsjö <martin at martin.st> wrote:
> 
> On Wed, 18 Sep 2024, Zhao Zhili wrote:
> 
>> From: Zhao Zhili <zhilizhao at tencent.com>
>> 
>> Since c0666d8b, rgb24toyv12 is broken for width non-aligned to 16.
>> Add a simple wrapper to handle the non-aligned part.
>> 
>> Signed-off-by: Zhao Zhili <zhilizhao at tencent.com>
>> Co-authored-by: johzzy <hellojinqiang at gmail.com>
>> ---
>> libswscale/aarch64/rgb2rgb.c | 23 ++++++++++++++++++++++-
>> tests/checkasm/sw_rgb.c      |  2 +-
>> 2 files changed, 23 insertions(+), 2 deletions(-)
>> 
>> diff --git a/libswscale/aarch64/rgb2rgb.c b/libswscale/aarch64/rgb2rgb.c
>> index d978a6f173..20a25033cb 100644
>> --- a/libswscale/aarch64/rgb2rgb.c
>> +++ b/libswscale/aarch64/rgb2rgb.c
>> @@ -27,9 +27,30 @@
>> #include "libswscale/swscale.h"
>> #include "libswscale/swscale_internal.h"
>> 
>> +// Only handle width aligned to 16
>> void ff_rgb24toyv12_neon(const uint8_t *src, uint8_t *ydst, uint8_t *udst,
>>                         uint8_t *vdst, int width, int height, int lumStride,
>>                         int chromStride, int srcStride, int32_t *rgb2yuv);
>> +
>> +static void rgb24toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst,
>> +                        uint8_t *vdst, int width, int height, int lumStride,
>> +                        int chromStride, int srcStride, int32_t *rgb2yuv)
>> +{
>> +    int width_align = width & (~15);
>> +
>> +    if (width_align > 0)
>> +        ff_rgb24toyv12_neon(src, ydst, udst, vdst, width_align, height,
>> +                lumStride, chromStride, srcStride, rgb2yuv);
>> +    if (width_align < width) {
>> +        src += width_align * 3;
>> +        ydst += width_align;
>> +        udst += width_align / 2;
>> +        vdst += width_align / 2;
>> +        ff_rgb24toyv12_c(src, ydst, udst, vdst, width - width_align, height,
>> +                lumStride, chromStride, srcStride, rgb2yuv);
>> +    }
>> +}
>> +
>> void ff_interleave_bytes_neon(const uint8_t *src1, const uint8_t *src2,
>>                              uint8_t *dest, int width, int height,
>>                              int src1Stride, int src2Stride, int dstStride);
>> @@ -42,7 +63,7 @@ av_cold void rgb2rgb_init_aarch64(void)
>>    int cpu_flags = av_get_cpu_flags();
>> 
>>    if (have_neon(cpu_flags)) {
>> -        ff_rgb24toyv12  = ff_rgb24toyv12_neon;
>> +        ff_rgb24toyv12  = rgb24toyv12;
>>        interleaveBytes = ff_interleave_bytes_neon;
>>        deinterleaveBytes = ff_deinterleave_bytes_neon;
>>    }
>> diff --git a/tests/checkasm/sw_rgb.c b/tests/checkasm/sw_rgb.c
>> index af9434073a..a57c471e3b 100644
>> --- a/tests/checkasm/sw_rgb.c
>> +++ b/tests/checkasm/sw_rgb.c
>> @@ -129,7 +129,7 @@ static int cmp_off_by_n(const uint8_t *ref, const uint8_t *test, size_t n, int a
>> 
>> static void check_rgb24toyv12(struct SwsContext *ctx)
>> {
>> -    static const int input_sizes[] = {16, 128, 512, MAX_LINE_SIZE, -MAX_LINE_SIZE};
>> +    static const int input_sizes[] = {4, 16, 128, 512, MAX_LINE_SIZE, -MAX_LINE_SIZE};
>> 
> 
> I think it would be good to test a case which isn't mod16, but bigger than 16 as well.

Patch v2 changed width 512 to 540.

https://ffmpeg.org/pipermail/ffmpeg-devel/2024-September/333636.html

I have thought about handle the non-aligned width in asm. It’s cache friendly
than current C wrapper. But current implementation is simple, and the lose of
performance may be negligible.

> 
> Other than that, I think this change looks reasonable.
> 
> // Martin
> 
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".