[x265] [PATCH] max num bmv parameterized
chen
chenm003 at 163.com
Fri Aug 5 17:17:41 CEST 2016
At 2016-08-05 19:32:53,nvijay.anand at trispacetech.com wrote:
># HG changeset patch
># User N Vijay Anand <nvijay.anand at trispacetech.com>
># Date 1470396737 -19800
># Fri Aug 05 17:02:17 2016 +0530
># Node ID f46e843a27cbaa4a1c79f1a43d41a04d63f601c4
># Parent 72f16a34946f5f03da875c67a75124dfec1b4ecb
>max num bmv parameterized
>
>diff -r 72f16a34946f -r f46e843a27cb source/encoder/motion.cpp
>--- a/source/encoder/motion.cpp Sat Jul 30 09:48:59 2016 +0530
>+++ b/source/encoder/motion.cpp Fri Aug 05 17:02:17 2016 +0530
>@@ -100,17 +100,17 @@
>
> }
>
>-#define MAX_NUM_BESTVECTORS (4)
>+#define MAX_NUM_BESTVECTORS (16)
in you new patch, you never use this constant anymore. I guess Pradeep want to keep this because pass maxNumBmv throughput function need extra instructions to set up parameters registers.
>
>-inline void PushToBMVStack(MV *bStack, MV & bv, int *bCostStack, int bcost)
>+inline void PushToBMVStack(MV *bStack, MV & bv, int *bCostStack, int bcost, int maxNumBmv)
> {
>- for (int i=0; i<MAX_NUM_BESTVECTORS; i++)
>+ for (int i=0; i<maxNumBmv; i++)
> {
> if((bCostStack[i] == bcost) && (bv == bStack[i]))
> break;
>- if((bCostStack[i] > bcost) && (bv != bStack[i]))
>+ if((bCostStack[i] >= bcost) && (bv != bStack[i]))
> {
>- for (int j=MAX_NUM_BESTVECTORS-1; j>i; j--)
>+ for (int j=maxNumBmv-1; j>i; j--)
> {
> bStack[j] = bStack[j-1];
> bCostStack[j] = bCostStack[j-1];
>@@ -247,7 +247,7 @@
> bmv = tmv; \
> bPointNr = point; \
> bDistance = dist; \
>- PushToBMVStack(bmvStack, tmv, bmvCostStack, cost); \
>+ PushToBMVStack(bmvStack, tmv, bmvCostStack, cost, maxNumBmv); \
> } \
> } while (0)
>
>@@ -257,7 +257,7 @@
> int cost = sad(fenc, FENC_STRIDE, fref + (mx) + (my) * stride, stride); \
> cost += mvcost(MV(mx, my) << 2); \
> COPY2_IF_LT(bcost, cost, bmv, MV(mx, my)); \
>- PushToBMVStack(bmvStack, MV(mx,my), bmvCostStack, cost); \
>+ PushToBMVStack(bmvStack, MV(mx,my), bmvCostStack, cost, maxNumBmv); \
> } while (0)
>
> #define COST_MV_X3_DIR(m0x, m0y, m1x, m1y, m2x, m2y, costs) \
>@@ -271,9 +271,9 @@
> (costs)[0] += mvcost((bmv + MV(m0x, m0y)) << 2); \
> (costs)[1] += mvcost((bmv + MV(m1x, m1y)) << 2); \
> (costs)[2] += mvcost((bmv + MV(m2x, m2y)) << 2); \
>- PushToBMVStack(bmvStack, bmv+MV(m0x,m0y), bmvCostStack, (costs)[0]); \
>- PushToBMVStack(bmvStack, bmv+MV(m1x,m1y), bmvCostStack, (costs)[1]); \
>- PushToBMVStack(bmvStack, bmv+MV(m2x,m2y), bmvCostStack, (costs)[2]); \
>+ PushToBMVStack(bmvStack, bmv+MV(m0x,m0y), bmvCostStack, (costs)[0], maxNumBmv); \
>+ PushToBMVStack(bmvStack, bmv+MV(m1x,m1y), bmvCostStack, (costs)[1], maxNumBmv); \
>+ PushToBMVStack(bmvStack, bmv+MV(m2x,m2y), bmvCostStack, (costs)[2], maxNumBmv); \
> }
>
> #define COST_MV_PT_DIST_X4(m0x, m0y, p0, d0, m1x, m1y, p1, d1, m2x, m2y, p2, d2, m3x, m3y, p3, d3) \
>@@ -289,13 +289,13 @@
> (costs)[2] += mvcost(MV(m2x, m2y) << 2); \
> (costs)[3] += mvcost(MV(m3x, m3y) << 2); \
> COPY4_IF_LT(bcost, costs[0], bmv, MV(m0x, m0y), bPointNr, p0, bDistance, d0); \
>- PushToBMVStack(bmvStack, MV(m0x,m0y), bmvCostStack, (costs)[0]); \
>+ PushToBMVStack(bmvStack, MV(m0x,m0y), bmvCostStack, (costs)[0], maxNumBmv); \
> COPY4_IF_LT(bcost, costs[1], bmv, MV(m1x, m1y), bPointNr, p1, bDistance, d1); \
>- PushToBMVStack(bmvStack, MV(m1x,m1y), bmvCostStack, (costs)[1]); \
>+ PushToBMVStack(bmvStack, MV(m1x,m1y), bmvCostStack, (costs)[1], maxNumBmv); \
> COPY4_IF_LT(bcost, costs[2], bmv, MV(m2x, m2y), bPointNr, p2, bDistance, d2); \
>- PushToBMVStack(bmvStack, MV(m2x,m2y), bmvCostStack, (costs)[2]); \
>+ PushToBMVStack(bmvStack, MV(m2x,m2y), bmvCostStack, (costs)[2], maxNumBmv); \
> COPY4_IF_LT(bcost, costs[3], bmv, MV(m3x, m3y), bPointNr, p3, bDistance, d3); \
>- PushToBMVStack(bmvStack, MV(m3x,m3y), bmvCostStack, (costs)[3]); \
>+ PushToBMVStack(bmvStack, MV(m3x,m3y), bmvCostStack, (costs)[3], maxNumBmv); \
> }
>
> #define COST_MV_X4(m0x, m0y, m1x, m1y, m2x, m2y, m3x, m3y) \
>@@ -312,13 +312,13 @@
> costs[2] += mvcost((omv + MV(m2x, m2y)) << 2); \
> costs[3] += mvcost((omv + MV(m3x, m3y)) << 2); \
> COPY2_IF_LT(bcost, costs[0], bmv, omv + MV(m0x, m0y)); \
>- PushToBMVStack(bmvStack, omv+MV(m0x,m0y), bmvCostStack, (costs)[0]); \
>+ PushToBMVStack(bmvStack, omv+MV(m0x,m0y), bmvCostStack, costs[0], maxNumBmv); \
> COPY2_IF_LT(bcost, costs[1], bmv, omv + MV(m1x, m1y)); \
>- PushToBMVStack(bmvStack, omv+MV(m1x,m1y), bmvCostStack, (costs)[1]); \
>+ PushToBMVStack(bmvStack, omv+MV(m1x,m1y), bmvCostStack, costs[1], maxNumBmv); \
> COPY2_IF_LT(bcost, costs[2], bmv, omv + MV(m2x, m2y)); \
>- PushToBMVStack(bmvStack, omv+MV(m2x,m2y), bmvCostStack, (costs)[2]); \
>+ PushToBMVStack(bmvStack, omv+MV(m2x,m2y), bmvCostStack, costs[2], maxNumBmv); \
> COPY2_IF_LT(bcost, costs[3], bmv, omv + MV(m3x, m3y)); \
>- PushToBMVStack(bmvStack, omv+MV(m3x,m3y), bmvCostStack, (costs)[3]); \
>+ PushToBMVStack(bmvStack, omv+MV(m3x,m3y), bmvCostStack, costs[3], maxNumBmv); \
> }
>
> #define COST_MV_X4_DIR(m0x, m0y, m1x, m1y, m2x, m2y, m3x, m3y, costs) \
>@@ -334,10 +334,10 @@
> (costs)[1] += mvcost((bmv + MV(m1x, m1y)) << 2); \
> (costs)[2] += mvcost((bmv + MV(m2x, m2y)) << 2); \
> (costs)[3] += mvcost((bmv + MV(m3x, m3y)) << 2); \
>- PushToBMVStack(bmvStack, bmv+MV(m0x,m0y), bmvCostStack, (costs)[0]); \
>- PushToBMVStack(bmvStack, bmv+MV(m1x,m1y), bmvCostStack, (costs)[1]); \
>- PushToBMVStack(bmvStack, bmv+MV(m2x,m2y), bmvCostStack, (costs)[2]); \
>- PushToBMVStack(bmvStack, bmv+MV(m3x,m3y), bmvCostStack, (costs)[3]); \
>+ PushToBMVStack(bmvStack, bmv+MV(m0x,m0y), bmvCostStack, (costs)[0], maxNumBmv); \
>+ PushToBMVStack(bmvStack, bmv+MV(m1x,m1y), bmvCostStack, (costs)[1], maxNumBmv); \
>+ PushToBMVStack(bmvStack, bmv+MV(m2x,m2y), bmvCostStack, (costs)[2], maxNumBmv); \
>+ PushToBMVStack(bmvStack, bmv+MV(m3x,m3y), bmvCostStack, (costs)[3], maxNumBmv); \
> }
>
> #define DIA1_ITER(mx, my) \
>@@ -379,6 +379,7 @@
> int & bcost,
> MV *bmvStack,
> int *bmvCostStack,
>+ int maxNumBmv,
> int & bPointNr,
> int & bDistance,
> int earlyExitIters,
>@@ -657,9 +658,19 @@
>
> /* re-measure full pel rounded MVP with SAD as search start point */
> MV bmv = pmv.roundToFPel();
>- MV bmvStack[4] = {bmv, bmv, bmv, bmv};
>- int bmvCostStack[4] = {bprecost, 0x7fffffff, 0x7fffffff, 0x7fffffff};
>- int bcost = bprecost;
>+ MV bmvStack[MAX_NUM_BESTVECTORS];
>+ int bmvCostStack[MAX_NUM_BESTVECTORS];
>+ int bcost = bprecost;
>+ const int maxNumBmv = 1 << searchMethod;
>+
>+ bmvStack[0] = bmv;
>+ bmvCostStack[0] = bprecost;
>+ for (int i=1 ; i < maxNumBmv; i++)
>+ {
>+ bmvStack[i] = bmv;
>+ bmvCostStack[i] = 0x7fffffff;
>+ }
>+
> if (pmv.isSubpel())
> {
> bcost = sad(fenc, FENC_STRIDE, fref + bmv.x + bmv.y * stride, stride) + mvcost(bmv << 2);
>@@ -934,7 +945,7 @@
> do \
> { \
> COPY2_IF_LT(bcost, costs[k], dir, x * 16 + (y & 15)); \
>- PushToBMVStack(bmvStack, bmv+MV(x*i,y*i), bmvCostStack, costs[k]); \
>+ PushToBMVStack(bmvStack, omv+MV(x*i,y*i), bmvCostStack, costs[k], maxNumBmv); \
> } while (0)
>
> SADS(0, +0, -4, +0, +4, -2, -3, +2, -3);
>@@ -996,7 +1007,7 @@
> int bDistance = 0;
>
> const int EarlyExitIters = 3;
>- StarPatternSearch(ref, mvmin, mvmax, bmv, bcost, bmvStack, bmvCostStack, bPointNr, bDistance, EarlyExitIters, merange);
>+ StarPatternSearch(ref, mvmin, mvmax, bmv, bcost, bmvStack, bmvCostStack, maxNumBmv, bPointNr, bDistance, EarlyExitIters, merange);
>
> if (bDistance == 1)
> {
>@@ -1048,19 +1059,19 @@
> stride, costs);
> costs[0] += mvcost(tmv << 2);
> COPY2_IF_LT(bcost, costs[0], bmv, tmv);
>- PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[0]);
>+ PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[0], maxNumBmv);
> tmv.x += RasterDistance;
> costs[1] += mvcost(tmv << 2);
> COPY2_IF_LT(bcost, costs[1], bmv, tmv);
>- PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[1]);
>+ PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[1], maxNumBmv);
> tmv.x += RasterDistance;
> costs[2] += mvcost(tmv << 2);
> COPY2_IF_LT(bcost, costs[2], bmv, tmv);
>- PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[2]);
>+ PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[2], maxNumBmv);
> tmv.x += RasterDistance;
> costs[3] += mvcost(tmv << 3);
> COPY2_IF_LT(bcost, costs[3], bmv, tmv);
>- PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[3]);
>+ PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[3], maxNumBmv);
> }
> else
> COST_MV(tmv.x, tmv.y);
>@@ -1074,7 +1085,7 @@
> bDistance = 0;
> bPointNr = 0;
> const int MaxIters = 32;
>- StarPatternSearch(ref, mvmin, mvmax, bmv, bcost, bmvStack, bmvCostStack, bPointNr, bDistance, MaxIters, merange);
>+ StarPatternSearch(ref, mvmin, mvmax, bmv, bcost, bmvStack, bmvCostStack, maxNumBmv, bPointNr, bDistance, MaxIters, merange);
>
> if (bDistance == 1)
> {
>@@ -1124,19 +1135,19 @@
> stride, costs);
> costs[0] += mvcost(tmv << 2);
> COPY2_IF_LT(bcost, costs[0], bmv, tmv);
>- PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[0]);
>+ PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[0], maxNumBmv);
> tmv.x++;
> costs[1] += mvcost(tmv << 2);
> COPY2_IF_LT(bcost, costs[1], bmv, tmv);
>- PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[1]);
>+ PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[1], maxNumBmv);
> tmv.x++;
> costs[2] += mvcost(tmv << 2);
> COPY2_IF_LT(bcost, costs[2], bmv, tmv);
>- PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[2]);
>+ PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[2], maxNumBmv);
> tmv.x++;
> costs[3] += mvcost(tmv << 2);
> COPY2_IF_LT(bcost, costs[3], bmv, tmv);
>- PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[3]);
>+ PushToBMVStack(bmvStack, tmv, bmvCostStack, (costs)[3], maxNumBmv);
> }
> else
> COST_MV(tmv.x, tmv.y);
>@@ -1155,12 +1166,12 @@
> {
> bmv = bestpre;
> bcost = bprecost;
>- PushToBMVStack(bmvStack, bmv, bmvCostStack, bcost);
>+ PushToBMVStack(bmvStack, bmv, bmvCostStack, bcost, maxNumBmv);
> }
> else
> {
> bmv = bmv.toQPel(); // promote search bmv to qpel
>- for (int i=0; i<4; i++)
>+ for (int i=0; i<maxNumBmv; i++)
> {
> bmvStack[i] = bmvStack[i].toQPel();
> }
>@@ -1177,7 +1188,7 @@
> else if (ref->isLowres)
> {
> int bdir = 0;
>- for (int nBmv=0; nBmv<MAX_NUM_BESTVECTORS; nBmv++)
>+ for (int nBmv=0; nBmv<maxNumBmv; nBmv++)
> {
> bdir = 0;
> bmv = bmvStack[nBmv];
>@@ -1208,7 +1219,7 @@
>
> bmv = bmvStack[0];
> bcost = bmvCostStack[0];
>- for (int i=1; i<MAX_NUM_BESTVECTORS; i++)
>+ for (int i=1; i<maxNumBmv; i++)
> {
> if (bmvCostStack[i]<bcost)
> {
>@@ -1221,9 +1232,10 @@
> {
> pixelcmp_t hpelcomp;
>
>- for (int nBmv=0; nBmv<MAX_NUM_BESTVECTORS; nBmv++)
>+ for (int nBmv=0; nBmv<maxNumBmv; nBmv++)
> {
> bmv = bmvStack[nBmv];
>+ bcost = bmvCostStack[nBmv];
>
> if (wl.hpel_satd)
> {
>@@ -1275,7 +1287,7 @@
>
> bmv = bmvStack[0];
> bcost = bmvCostStack[0];
>- for (int i=1; i<MAX_NUM_BESTVECTORS; i++)
>+ for (int i=1; i<maxNumBmv; i++)
> {
> if (bmvCostStack[i]<bcost)
> {
>@@ -1287,6 +1299,7 @@
>
> x265_emms();
> outQMv = bmv;
>+
> return bcost;
> }
>
>diff -r 72f16a34946f -r f46e843a27cb source/encoder/motion.h
>--- a/source/encoder/motion.h Sat Jul 30 09:48:59 2016 +0530
>+++ b/source/encoder/motion.h Fri Aug 05 17:02:17 2016 +0530
>@@ -103,6 +103,7 @@
> int & bcost,
> MV *bmvStack,
> int *bCostStack,
>+ int maxNumBmv,
> int & bPointNr,
> int & bDistance,
> int earlyExitIters,
>_______________________________________________
>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/20160805/dc331651/attachment-0001.html>
More information about the x265-devel
mailing list