[x265] [PATCH] Generating weighted full-pels in compressCTURows
Steve Borho
steve at borho.org
Mon Sep 30 22:34:25 CEST 2013
On Mon, Sep 30, 2013 at 6:01 AM, <shazeb at multicorewareinc.com> wrote:
> # HG changeset patch
> # User Shazeb Nawaz Khan <shazeb at multicorewareinc.com>
> # Date 1380285873 -19800
> # Fri Sep 27 18:14:33 2013 +0530
> # Node ID cc94f1a5c6ab1975502f6a0edfc4aa4233733597
> # Parent 55edc34e253c14d3eccb83a7d1db43774349ff9a
> Generating weighted full-pels in compressCTURows
>
> We are re-enabling weightP after introduction of frame parallelism. Not
> yet using the weighted pixels.
>
> diff -r 55edc34e253c -r cc94f1a5c6ab source/Lib/TLibCommon/TComPicYuv.cpp
> --- a/source/Lib/TLibCommon/TComPicYuv.cpp Sat Sep 28 22:54:44 2013
> -0500
> +++ b/source/Lib/TLibCommon/TComPicYuv.cpp Fri Sep 27 18:14:33 2013
> +0530
> @@ -250,7 +250,8 @@
> else if (mref->isWeighted == false)
> return mref;
> }
> - mref = new MotionReference(this, w);
> + mref = new MotionReference();
> + mref->init(this, w);
> mref->m_next = m_refList;
> m_refList = mref;
> return mref;
> diff -r 55edc34e253c -r cc94f1a5c6ab
> source/Lib/TLibCommon/TComWeightPrediction.cpp
> --- a/source/Lib/TLibCommon/TComWeightPrediction.cpp Sat Sep 28
> 22:54:44 2013 -0500
> +++ b/source/Lib/TLibCommon/TComWeightPrediction.cpp Fri Sep 27
> 18:14:33 2013 +0530
> @@ -431,7 +431,7 @@
> srcStride = srcYuv0->m_width;
> dstStride = outDstYuv->getStride();
>
> - primitives.weightpUni(srcY0, dstY, srcStride, dstStride, width,
> height, w0, round, shift, offset);
> + primitives.weightpUniShort(srcY0, dstY, srcStride, dstStride,
> width, height, w0, round, shift, offset, 0);
> }
>
> // Chroma U : --------------------------------------------
> @@ -447,7 +447,7 @@
> width >>= 1;
> height >>= 1;
>
> - primitives.weightpUni(srcU0, dstU, srcStride, dstStride, width,
> height, w0, round, shift, offset);
> + primitives.weightpUniShort(srcU0, dstU, srcStride, dstStride, width,
> height, w0, round, shift, offset, 0);
>
> // Chroma V : --------------------------------------------
> w0 = wp0[2].w;
> @@ -455,7 +455,7 @@
> shift = wp0[2].shift + shiftNum;
> round = shift ? (1 << (shift - 1)) : 0;
>
> - primitives.weightpUni(srcV0, dstV, srcStride, dstStride, width,
> height, w0, round, shift, offset);
> + primitives.weightpUniShort(srcV0, dstV, srcStride, dstStride, width,
> height, w0, round, shift, offset, 0);
> }
>
> //=======================================================
> diff -r 55edc34e253c -r cc94f1a5c6ab source/common/pixel.cpp
> --- a/source/common/pixel.cpp Sat Sep 28 22:54:44 2013 -0500
> +++ b/source/common/pixel.cpp Fri Sep 27 18:14:33 2013 +0530
> @@ -514,17 +514,19 @@
> }
> }
>
> -void weightUnidir(short *src, pixel *dst, intptr_t srcStride, intptr_t
> dstStride, int width, int height, int w0, int round, int shift, int offset)
> +template <typename T>
> +void weightUnidir(void *srcAbstract, pixel *dst, intptr_t srcStride,
> intptr_t dstStride, int width, int height, int w0, int round, int shift,
> int offset, int startHeight)
> {
> + T *src = static_cast<T *>(srcAbstract);
> int x, y;
> - for (y = height - 1; y >= 0; y--)
> + for (y = height - 1; y >= startHeight; y--)
> {
> for (x = width - 1; x >= 0; )
> {
> // note: luma min width is 4
> - dst[x] = (pixel) Clip3(0, ((1 << X265_DEPTH) - 1), ((w0 *
> (src[x] + IF_INTERNAL_OFFS) + round) >> shift) + offset);
> + dst[x] = (pixel) Clip3(0, ((1 << X265_DEPTH) - 1), ((w0 *
> (uint16_t) (src[x] + IF_INTERNAL_OFFS) + round) >> shift) + offset);
> x--;
> - dst[x] = (pixel) Clip3(0, ((1 << X265_DEPTH) - 1), ((w0 *
> (src[x] + IF_INTERNAL_OFFS) + round) >> shift) + offset);
> + dst[x] = (pixel) Clip3(0, ((1 << X265_DEPTH) - 1), ((w0 *
> (uint16_t) (src[x] + IF_INTERNAL_OFFS) + round) >> shift) + offset);
> x--;
> }
>
> @@ -842,7 +844,8 @@
> p.transpose[3] = transpose<32>;
> p.transpose[4] = transpose<64>;
>
> - p.weightpUni = weightUnidir;
> + p.weightpUniPixel = weightUnidir<pixel>;
> + p.weightpUniShort = weightUnidir<short>;
>
> p.pixelsub_sp = pixelsub_sp_c;
> p.pixeladd_pp = pixeladd_pp_c;
> diff -r 55edc34e253c -r cc94f1a5c6ab source/common/primitives.h
> --- a/source/common/primitives.h Sat Sep 28 22:54:44 2013 -0500
> +++ b/source/common/primitives.h Fri Sep 27 18:14:33 2013 +0530
> @@ -228,8 +228,7 @@
> typedef void (*filterRowV_N_t)(short *midA, intptr_t midStride, pixel
> *dstA, pixel *dstB, pixel *dstC, intptr_t dstStride, int width, int height,
> int marginX, int marginY, int row, int isLastRow);
> typedef void (*extendCURowBorder_t)(pixel* txt, intptr_t stride, int
> width, int height, int marginX);
>
> -
> -typedef void (*weightpUni_t)(short *src, pixel *dst, intptr_t srcStride,
> intptr_t dstStride, int width, int height, int w0, int round, int shift,
> int offset);
> +typedef void (*weightpUni_t)(void *src, pixel *dst, intptr_t srcStride,
> intptr_t dstStride, int width, int height, int w0, int round, int shift,
> int offset, int startHeight);
> typedef void (*scale_t)(pixel *dst, pixel *src, intptr_t stride);
> typedef void (*downscale_t)(pixel *src0, pixel *dstf, pixel *dsth, pixel
> *dstv, pixel *dstc,
> intptr_t src_stride, intptr_t dst_stride, int
> width, int height);
> @@ -286,7 +285,8 @@
> calcrecon_t calcrecon[NUM_SQUARE_BLOCKS];
> transpose_t transpose[NUM_SQUARE_BLOCKS];
>
> - weightpUni_t weightpUni;
> + weightpUni_t weightpUniPixel;
> + weightpUni_t weightpUniShort;
> pixelsub_sp_t pixelsub_sp;
> pixeladd_ss_t pixeladd_ss;
> pixeladd_pp_t pixeladd_pp;
> diff -r 55edc34e253c -r cc94f1a5c6ab source/common/reference.cpp
> --- a/source/common/reference.cpp Sat Sep 28 22:54:44 2013 -0500
> +++ b/source/common/reference.cpp Fri Sep 27 18:14:33 2013 +0530
> @@ -24,6 +24,7 @@
>
> #include "TLibCommon/TypeDef.h"
> #include "TLibCommon/TComPicYuv.h"
> +#include "TLibCommon/TComPic.h"
> #include "TLibCommon/TComSlice.h"
> #include "primitives.h"
> #include "reference.h"
> @@ -51,12 +52,17 @@
> return false;
> }
>
> -MotionReference::MotionReference(TComPicYuv* pic, wpScalingParam *w)
> +MotionReference::MotionReference()
> +{}
> +
> +int MotionReference::init(TComPicYuv* pic, wpScalingParam *w)
> {
> m_reconPic = pic;
> lumaStride = pic->getStride();
> m_startPad = pic->m_lumaMarginY * lumaStride + pic->m_lumaMarginX;
> m_next = NULL;
> + isWeighted = false;
> + m_numWeightedRows = 0;
>
> if (w)
> {
> @@ -66,17 +72,66 @@
> size_t padheight = height + pic->m_lumaMarginY * 2;
>
> setWeight(*w);
> - fpelPlane = (pixel*)X265_MALLOC(pixel, padwidth * padheight) +
> m_startPad;
> + fpelPlane = (pixel*)X265_MALLOC(pixel, padwidth * padheight);
> + if (fpelPlane) fpelPlane += m_startPad;
> + else return -1;
>
In this particular situation, the caller could decide to disable WeightP
for this frame if the malloc failed, but that would be adding a lot of
complication for what is likely a fatal condition. Once one malloc fails
there is a high probability the rest will fail so most likely the encode
must be stopped.
> }
> else
> {
> /* directly reference the pre-extended integer pel plane */
> fpelPlane = pic->m_picBufY + m_startPad;
> }
> + return 0;
> }
>
> MotionReference::~MotionReference()
> {
> if (isWeighted)
> - X265_FREE(fpelPlane);
> + X265_FREE(fpelPlane - m_startPad);
>
this is only safe if isWeighted is false if fpelPlane allocation failed
> }
> +
> +void MotionReference::applyWeight(TComPic* ref, int rows, int numRows)
> +{
> + TComPicYuv* pic = ref->getPicYuvRec();
> + int marginX = pic->m_lumaMarginX;
> + int marginY = pic->m_lumaMarginY;
> +
> + int width = pic->getWidth();
> + int height = ((rows - m_numWeightedRows) * g_maxCUHeight);
> + if (rows == numRows - 1) height = ((pic->getHeight() % g_maxCUHeight)
> ? (pic->getHeight() % g_maxCUHeight) : g_maxCUHeight);
> +
> + size_t dstStride = lumaStride;
> +
> + // Computing weighted CU rows
> + int shiftNum = IF_INTERNAL_PREC - X265_DEPTH;
> + shift = shift + shiftNum;
> + round = shift ? (1 << (shift - 1)) : 0;
> +
> + primitives.weightpUniPixel((pixel*) pic->getLumaAddr(), fpelPlane,
> lumaStride, dstStride, width, height, weight, round, shift, offset,
> m_numWeightedRows * (int)g_maxCUHeight);
> +
> + // Extending Left & Right
> + primitives.extendRowBorder(fpelPlane, dstStride, width, height,
> marginX);
> +
> + // Extending Above
> + if (m_numWeightedRows == 0)
> + {
> + pixel *pixY = fpelPlane - marginX;
> +
> + for (int y = 0; y < marginY; y++)
> + {
> + memcpy(pixY - (y + 1) * dstStride, pixY, dstStride *
> sizeof(pixel));
> + }
> + }
> +
> + // Extending Bottom
> + if (rows == (numRows - 1))
> + {
> + pixel *pixY = fpelPlane - marginX + (pic->getHeight() - 1) *
> dstStride;
> +
> + for (int y = 0; y < marginY; y++)
> + {
> + memcpy(pixY + (y + 1) * dstStride, pixY, dstStride *
> sizeof(pixel));
> + }
> + }
> + m_numWeightedRows = rows;
> +}
> diff -r 55edc34e253c -r cc94f1a5c6ab source/common/reference.h
> --- a/source/common/reference.h Sat Sep 28 22:54:44 2013 -0500
> +++ b/source/common/reference.h Fri Sep 27 18:14:33 2013 +0530
> @@ -30,6 +30,7 @@
> // private x265 namespace
>
> class TComPicYuv;
> +class TComPic;
> struct WpScalingParam;
> typedef WpScalingParam wpScalingParam;
>
> @@ -56,12 +57,17 @@
> {
> public:
>
> - MotionReference(TComPicYuv*, wpScalingParam* w = NULL);
> + MotionReference();
>
> ~MotionReference();
>
> + void applyWeight(TComPic* src, int rows, int numRows);
> + void extendRow(TComPic* pic, int row);
> + int init(TComPicYuv*, wpScalingParam* w = NULL);
> +
> MotionReference *m_next;
> TComPicYuv *m_reconPic;
> + int m_numWeightedRows;
>
> protected:
>
> diff -r 55edc34e253c -r cc94f1a5c6ab source/encoder/frameencoder.cpp
> --- a/source/encoder/frameencoder.cpp Sat Sep 28 22:54:44 2013 -0500
> +++ b/source/encoder/frameencoder.cpp Fri Sep 27 18:14:33 2013 +0530
> @@ -914,6 +914,10 @@
> {
> refpic->m_reconRowWait.wait();
> }
> + if(slice->getPPS()->getUseWP())
> + {
> + slice->m_mref[list][ref]->applyWeight(refpic, row
> + refLagRows, m_numRows);
> + }
> }
> }
>
> diff -r 55edc34e253c -r cc94f1a5c6ab source/x265opts.h
> --- a/source/x265opts.h Sat Sep 28 22:54:44 2013 -0500
> +++ b/source/x265opts.h Fri Sep 27 18:14:33 2013 +0530
> @@ -69,8 +69,8 @@
> OPT("ref", param->maxNumReferences, required_argument,
> 0, "max number of L0 references to be allowed (1 .. 16)")
> // Disabled because weighted uni-prediction was busted by not using
> // pre-generated planes in motion compensation
> -//OPT("no-weightp", param->bEnableWeightedPred,
> no_argument, 0, "Disable weighted prediction in P slices")
> -//OPT("weightp", param->bEnableWeightedPred,
> no_argument, 'w', "Enable weighted prediction in P slices")
> +OPT("no-weightp", param->bEnableWeightedPred, no_argument,
> 0, "Disable weighted prediction in P slices")
> +OPT("weightp", param->bEnableWeightedPred, no_argument,
> 'w', "Enable weighted prediction in P slices")
> // Disabled because weighted bi prediction is busted
> //OPT("no-weightbp", param->bEnableWeightedBiPred,
> no_argument, 0, "Disable weighted (bidirectional) prediction in B slices")
> //OPT("weightbp", param->bEnableWeightedBiPred,
> no_argument, 0, "Enable weighted (bidirectional) prediction in B slices")
> _______________________________________________
> x265-devel mailing list
> x265-devel at videolan.org
> https://mailman.videolan.org/listinfo/x265-devel
>
--
Steve Borho
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20130930/3ef5f4b9/attachment-0001.html>
More information about the x265-devel
mailing list