[x265] [PATCH] aq: bug fix.extend border of TComPic::m_origPicYuv to a multiple of 16

[Aarthi Priya Thirumalai]

On Nov 7, 2013 11:24 PM, "Steve Borho" <steve at borho.org> wrote:
>
>
>
>
> On Thu, Nov 7, 2013 at 5:17 AM, Aarthi Thirumalai <
aarthi at multicorewareinc.com> wrote:
>>
>> # HG changeset patch
>> # User Aarthi Thirumalai
>> # Date 1383823017 -19800
>> #      Thu Nov 07 16:46:57 2013 +0530
>> # Node ID 0823d243b4e2f9eee6323a9750eeb811bccbad77
>> # Parent  85002898f5b4308547af6ce464bbdff5f360fa13
>>  aq: bug fix.extend border of TComPic::m_origPicYuv to a multiple of 16
>>
>> diff -r 85002898f5b4 -r 0823d243b4e2 source/Lib/TLibCommon/TComPicYuv.cpp
>> --- a/source/Lib/TLibCommon/TComPicYuv.cpp      Thu Nov 07 14:31:05 2013
+0530
>> +++ b/source/Lib/TLibCommon/TComPicYuv.cpp      Thu Nov 07 16:46:57 2013
+0530
>> @@ -345,6 +345,10 @@
>>          /* width and height - without padsize */
>>          int width = m_picWidth - padx;
>>          int height = m_picHeight - pady;
>> +        uint8_t rem = m_picWidth % 16;
>> +        padx += rem ? 16 - rem : 0;
>> +        rem = m_picHeight % 16;
>> +        pady += rem ? 16 - rem : 0;
>
>
> if the row width was 17, padx would already be 3 in order to pad out to
the nearest multiple of 4.  You would be doing:  padx += 15, which would
result in a pad of 18, which is more than what you bargained for.  I
believe the += should just be =.  Also, Gopu just added a new else {}
clause this is missed here, so I've hoisted this operation higher in the
file.
>
>>when row width is 17, padx is already 3 and m_picWidth would be  20 at
this stage. So padx += 12 not 15..
>>
>>          // Manually copy pixels to up-size them
>>          for (int r = 0; r < height; r++)
>> @@ -475,10 +479,12 @@
>>          uint8_t *v = (uint8_t*)pic.planes[2];
>>
>>          /* width and height - without padsize */
>> -        int width = (m_picWidth * (pic.bitDepth > 8 ? 2 : 1)) - padx;
>> -        int height = m_picHeight - pady;
>> -
>> -        // copy pixels by row into encoder's buffer
>> +         int width = (m_picWidth * (pic.bitDepth > 8 ? 2 : 1)) - padx;
>> +         int height = m_picHeight - pady;
>> +         uint8_t rem = m_picWidth % 16;
>> +         padx += rem ? 16 - rem : 0;
>> +         rem = m_picHeight % 16;
>> +         pady += rem ? 16 - rem : 0;
>
>
> still white-space alignment problems here, but this code has been removed
>
>>
>>          for (int r = 0; r < height; r++)
>>          {
>>              memcpy(Y, y, width);
>> diff -r 85002898f5b4 -r 0823d243b4e2 source/common/lowres.cpp
>> --- a/source/common/lowres.cpp  Thu Nov 07 14:31:05 2013 +0530
>> +++ b/source/common/lowres.cpp  Thu Nov 07 16:46:57 2013 +0530
>> @@ -150,31 +150,6 @@
>>          intraMbs[i] = 0;
>>      }
>>
>> -    int y, extWidth = (orig->getWidth() + X265_LOWRES_CU_SIZE - 1);
>> -    int srcStride = orig->getStride();
>> -    int srcHeight = orig->getHeight();
>> -    int srcWidth  = orig->getWidth();
>> -    Pel *src;
>> -    src = orig->getLumaAddr();
>> -
>> -    /* extending right margin */
>> -    if (2 * width > srcWidth)
>> -    {
>> -        for (y = 0; y < srcHeight; y++)
>> -        {
>> -            ::memset(src + srcWidth, src[srcWidth - 1], sizeof(Pel) *
(X265_LOWRES_CU_SIZE - 1));
>> -            src += srcStride;
>> -        }
>> -    }
>> -
>> -    /* extending bottom margin */
>> -    src = orig->getLumaAddr() + (srcHeight - 1) * srcStride;
>> -
>> -    for (y = 1; y <= 2 * lines - srcHeight; y++)
>> -    {
>> -        ::memcpy(src + y * srcStride, src, sizeof(Pel) * (extWidth));
>> -    }
>> -
>>      /* downscale and generate 4 HPEL planes for lookahead */
>>      primitives.frame_init_lowres_core(orig->getLumaAddr(),
>>                                        lowresPlane[0], lowresPlane[1],
lowresPlane[2], lowresPlane[3],
>> diff -r 85002898f5b4 -r 0823d243b4e2 source/encoder/frameencoder.cpp
>> --- a/source/encoder/frameencoder.cpp   Thu Nov 07 14:31:05 2013 +0530
>> +++ b/source/encoder/frameencoder.cpp   Thu Nov 07 16:46:57 2013 +0530
>> @@ -1149,9 +1149,8 @@
>>      {
>>          /* Derive qpOffet for each CU by averaging offsets for all
16x16 blocks in the cu. */
>>          double qp_offset = 0;
>> -        int blockSize = g_maxCUWidth >> 2;
>> -        int maxBlockCols = (pic->getPicYuvOrg()->getWidth() +
(blockSize - 1)) / blockSize;
>> -        int maxBlockRows = (pic->getPicYuvOrg()->getHeight() +
(blockSize - 1)) / blockSize;
>> +        int maxBlockCols = (pic->getPicYuvOrg()->getWidth() + (16 - 1))
/ 16;
>> +        int maxBlockRows = (pic->getPicYuvOrg()->getHeight() + (16 -
1)) / 16;
>
>
> much better
>
>>
>>          int block_y = (cuAddr / pic->getPicSym()->getFrameWidthInCU())
* 4;
>>          int block_x = (cuAddr * 4) - block_y *
pic->getPicSym()->getFrameWidthInCU();
>>          int cnt = 0;
>> _______________________________________________
>> x265-devel mailing list
>> x265-devel at videolan.org
>> https://mailman.videolan.org/listinfo/x265-devel
>
>
>
>
> --
> Steve Borho
>
> _______________________________________________
> x265-devel mailing list
> x265-devel at videolan.org
> https://mailman.videolan.org/listinfo/x265-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20131108/32b7b57c/attachment-0001.html>