[x265-commits] [x265] fix SATD32x32 16bits cumulate sum overflow (x86 version o...
Min Chen
chenm003 at 163.com
Mon Jan 27 19:45:39 CET 2014
details: http://hg.videolan.org/x265/rev/83767892376d
branches: stable
changeset: 5910:83767892376d
user: Min Chen <chenm003 at 163.com>
date: Fri Jan 24 18:10:13 2014 +0800
description:
fix SATD32x32 16bits cumulate sum overflow (x86 version only)
Subject: [x265] wavefront: add missing initializer
details: http://hg.videolan.org/x265/rev/4fcfa56420fb
branches: stable
changeset: 5911:4fcfa56420fb
user: Deepthi Nandakumar <deepthi at multicorewareinc.com>
date: Mon Jan 27 23:21:06 2014 +0530
description:
wavefront: add missing initializer
Subject: [x265] WeightPrediction: check difPoc <= bframes+1 to allow weight analysis for valid references
details: http://hg.videolan.org/x265/rev/a02a028afea9
branches: stable
changeset: 5912:a02a028afea9
user: Kavitha Sampath <kavitha at multicorewareinc.com>
date: Fri Jan 24 10:45:41 2014 +0530
description:
WeightPrediction: check difPoc <= bframes+1 to allow weight analysis for valid references
Subject: [x265] common: do not allow encodes other than 4:2:0 to start
details: http://hg.videolan.org/x265/rev/10fc60881bbf
branches: stable
changeset: 5913:10fc60881bbf
user: Steve Borho <steve at borho.org>
date: Mon Jan 27 11:54:44 2014 -0600
description:
common: do not allow encodes other than 4:2:0 to start
Subject: [x265] me: add one more pixel lag for DIA search's relaxed search bounds
details: http://hg.videolan.org/x265/rev/dc4e57833aae
branches: stable
changeset: 5914:dc4e57833aae
user: Steve Borho <steve at borho.org>
date: Mon Jan 27 12:30:05 2014 -0600
description:
me: add one more pixel lag for DIA search's relaxed search bounds
diffstat:
source/common/common.cpp | 6 ++-
source/common/wavefront.h | 2 +-
source/common/x86/pixel-a.asm | 48 ++++++++++++++++-----------------
source/encoder/frameencoder.cpp | 1 +
source/encoder/weightPrediction.cpp | 52 ++++++++++++++++++------------------
source/encoder/weightPrediction.h | 4 ++-
6 files changed, 58 insertions(+), 55 deletions(-)
diffs (239 lines):
diff -r b173809575c6 -r dc4e57833aae source/common/common.cpp
--- a/source/common/common.cpp Mon Jan 27 00:04:22 2014 -0600
+++ b/source/common/common.cpp Mon Jan 27 12:30:05 2014 -0600
@@ -174,7 +174,7 @@ void x265_param_default(x265_param *para
/* Inter Coding tools */
param->searchMethod = X265_HEX_SEARCH;
param->subpelRefine = 2;
- param->searchRange = 58;
+ param->searchRange = 57;
param->maxNumMergeCand = 2;
param->bEnableWeightedPred = 1;
param->bEnableWeightedBiPred = 0;
@@ -277,7 +277,7 @@ int x265_param_default_preset(x265_param
{
param->lookaheadDepth = 10;
param->maxCUSize = 32;
- param->searchRange = 26;
+ param->searchRange = 25;
param->bFrameAdaptive = 0;
param->subpelRefine = 0;
param->maxNumMergeCand = 2;
@@ -438,6 +438,8 @@ int x265_check_params(x265_param *param)
CHECK(param->inputBitDepth > x265_max_bit_depth,
"inputBitDepth must be <= x265_max_bit_depth");
+ CHECK(param->internalCsp != X265_CSP_I420,
+ "Only 4:2:0 color space is supported at this time");
CHECK(param->rc.qp < -6 * (param->inputBitDepth - 8) || param->rc.qp > 51,
"QP exceeds supported range (-QpBDOffsety to 51)");
diff -r b173809575c6 -r dc4e57833aae source/common/wavefront.h
--- a/source/common/wavefront.h Mon Jan 27 00:04:22 2014 -0600
+++ b/source/common/wavefront.h Mon Jan 27 12:30:05 2014 -0600
@@ -49,7 +49,7 @@ private:
public:
- WaveFront(ThreadPool *pool) : JobProvider(pool), m_queuedBitmap(0) {}
+ WaveFront(ThreadPool *pool) : JobProvider(pool), m_queuedBitmap(0), m_enableBitmap(0) {}
virtual ~WaveFront();
diff -r b173809575c6 -r dc4e57833aae source/common/x86/pixel-a.asm
--- a/source/common/x86/pixel-a.asm Mon Jan 27 00:04:22 2014 -0600
+++ b/source/common/x86/pixel-a.asm Mon Jan 27 12:30:05 2014 -0600
@@ -1361,43 +1361,41 @@ cglobal pixel_satd_32x32, 4,8,8 ;if W
SATD_END_SSE2 m6, m7
%else
cglobal pixel_satd_32x32, 4,7,8,0-gprsize ;if !WIN64
-
SATD_START_SSE2 m6, m7
mov r6, r0
mov [rsp], r2
- call pixel_satd_8x8_internal
- call pixel_satd_8x8_internal
- call pixel_satd_8x8_internal
- call pixel_satd_8x8_internal
-%if HIGH_BIT_DEPTH
- pxor m7, m7
-%endif
- SATD_ACCUM m6, m0, m7
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
lea r0, [r6 + 8*SIZEOF_PIXEL]
mov r2, [rsp]
add r2, 8*SIZEOF_PIXEL
- call pixel_satd_8x8_internal
- call pixel_satd_8x8_internal
- call pixel_satd_8x8_internal
- call pixel_satd_8x8_internal
- SATD_ACCUM m6, m0, m7
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
lea r0, [r6 + 16*SIZEOF_PIXEL]
mov r2, [rsp]
add r2, 16*SIZEOF_PIXEL
- call pixel_satd_8x8_internal
- call pixel_satd_8x8_internal
- call pixel_satd_8x8_internal
- call pixel_satd_8x8_internal
- SATD_ACCUM m6, m0, m7
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
lea r0, [r6 + 24*SIZEOF_PIXEL]
mov r2, [rsp]
add r2, 24*SIZEOF_PIXEL
- call pixel_satd_8x8_internal
- call pixel_satd_8x8_internal
- call pixel_satd_8x8_internal
- call pixel_satd_8x8_internal
- SATD_END_SSE2 m6, m7
-
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ pxor m7, m7
+ movhlps m7, m6
+ paddd m6, m7
+ pshufd m7, m6, 1
+ paddd m6, m7
+ movd eax, m6
+ RET
%endif
%if WIN64
diff -r b173809575c6 -r dc4e57833aae source/encoder/frameencoder.cpp
--- a/source/encoder/frameencoder.cpp Mon Jan 27 00:04:22 2014 -0600
+++ b/source/encoder/frameencoder.cpp Mon Jan 27 12:30:05 2014 -0600
@@ -945,6 +945,7 @@ void FrameEncoder::compressCTURows()
}
int range = m_cfg->param.searchRange + /* fpel search */
+ 1 + /* diamond search range check lag */
2 + /* subpel refine */
NTAPS_LUMA / 2; /* subpel filter half-length */
uint32_t refLagRows = 1 + ((range + g_maxCUHeight - 1) / g_maxCUHeight);
diff -r b173809575c6 -r dc4e57833aae source/encoder/weightPrediction.cpp
--- a/source/encoder/weightPrediction.cpp Mon Jan 27 00:04:22 2014 -0600
+++ b/source/encoder/weightPrediction.cpp Mon Jan 27 12:30:05 2014 -0600
@@ -195,9 +195,11 @@ bool WeightPrediction::checkDenom(int de
ref = &m_slice->getRefPic(list, refIdxTemp)->m_lowres;
refPoc = m_slice->getRefPic(list, refIdxTemp)->getPOC();
difPoc = abs(curPoc - refPoc);
- m_mvs = fenc->lowresMvs[list][difPoc - 1];
- if (m_mvs)
+ if (difPoc > m_bframes + 1)
+ continue;
+ else
{
+ m_mvs = fenc->lowresMvs[list][difPoc - 1];
if (m_mvs[0].x == 0x7FFF)
continue;
else
@@ -245,44 +247,42 @@ bool WeightPrediction::checkDenom(int de
switch (yuv)
{
case 0:
-
+ {
m_mcbuf = ref->fpelPlane;
m_inbuf = fenc->lowresPlane[0];
- if (m_mvs)
+ pixel *tempm_buf;
+ pixel m_buf8[8 * 8];
+ int pixoff = 0, cu = 0;
+ intptr_t strd;
+ for (int y = 0; y < m_frmHeight; y += 8, pixoff = y * m_refStride)
{
- pixel *tempm_buf;
- pixel m_buf8[8 * 8];
- int pixoff = 0, cu = 0;
- intptr_t strd;
- for (int y = 0; y < m_frmHeight; y += 8, pixoff = y * m_refStride)
+ for (int x = 0; x < m_frmWidth; x += 8, pixoff += 8, cu++)
{
- for (int x = 0; x < m_frmWidth; x += 8, pixoff += 8, cu++)
+ if (m_mvCost[cu] > fenc->intraCost[cu])
{
- if (fenc->lowresMvCosts[0][difPoc - 1][cu] > fenc->intraCost[cu])
- {
- strd = m_refStride;
- tempm_buf = m_inbuf + pixoff;
- }
- else
- {
- strd = 8;
- tempm_buf = ref->lowresMC(pixoff, m_mvs[cu], m_buf8, strd);
- ic++;
- }
- primitives.blockcpy_pp(8, 8, m_buf + (y * m_refStride) + x, m_refStride, tempm_buf, strd);
+ strd = m_refStride;
+ tempm_buf = m_inbuf + pixoff;
}
+ else
+ {
+ strd = 8;
+ tempm_buf = ref->lowresMC(pixoff, m_mvs[cu], m_buf8, strd);
+ ic++;
+ }
+ primitives.blockcpy_pp(8, 8, m_buf + (y * m_refStride) + x, m_refStride, tempm_buf, strd);
}
+ }
- m_mcbuf = m_buf;
- }
+ m_mcbuf = m_buf;
break;
+ }
case 1:
m_mcbuf = m_slice->getRefPic(list, refIdxTemp)->getPicYuvOrg()->getCbAddr();
m_inbuf = m_slice->getPic()->getPicYuvOrg()->getCbAddr();
m_blockSize = 8;
- if (m_mvs) mcChroma();
+ mcChroma();
break;
case 2:
@@ -290,7 +290,7 @@ bool WeightPrediction::checkDenom(int de
m_mcbuf = m_slice->getRefPic(list, refIdxTemp)->getPicYuvOrg()->getCrAddr();
m_inbuf = m_slice->getPic()->getPicYuvOrg()->getCrAddr();
m_blockSize = 8;
- if (m_mvs) mcChroma();
+ mcChroma();
break;
}
diff -r b173809575c6 -r dc4e57833aae source/encoder/weightPrediction.h
--- a/source/encoder/weightPrediction.h Mon Jan 27 00:04:22 2014 -0600
+++ b/source/encoder/weightPrediction.h Mon Jan 27 12:30:05 2014 -0600
@@ -42,10 +42,11 @@ private:
pixel *m_mcbuf, *m_inbuf, *m_buf;
int32_t *m_intraCost;
MV *m_mvs;
+ int m_bframes;
public:
- WeightPrediction(TComSlice *slice)
+ WeightPrediction(TComSlice *slice, x265_param param)
{
this->m_slice = slice;
m_csp = m_slice->getPic()->getPicYuvOrg()->m_picCsp;
@@ -56,6 +57,7 @@ public:
m_dstStride = m_frmWidth;
m_refStride = m_slice->getPic()->m_lowres.lumaStride;
m_intraCost = m_slice->getPic()->m_lowres.intraCost;
+ m_bframes = param.bframes;
m_mcbuf = NULL;
m_inbuf = NULL;
More information about the x265-commits
mailing list