// General setup for verbs with IRS that do not go through jtirs[12]
// A verb u["n] using this function checks to see whether it has multiple cells; if so,
// it calls here, giving a callback; we split the arguents into cells and call the callback,
// which is often the same original function that called here.
A jtrank2ex(J jt,A a,A w,A fs,I lr,I rr,AF f2){PROLOG(0042);A y,y0,ya,yw,z;B ab,b,wb;
C*u,*uu,*v,*vv;I acn,acr,af,ak,ar,*as,at,k,mn,n=1,p,q,*s,wcn,wcr,wf,wk,wr,*ws,wt,yn,yr,*ys,yt;
RZ(a&&w);
at=AT(a); wt=AT(w);
if(at&SPARSE||wt&SPARSE)R sprank2(a,w,fs,lr,rr,f2);
// ?r=rank, ?s->shape, ?cr=effective rank, ?f=#frame, ?b=relative flag, for each argument
ar=AR(a); as=AS(a); acr=efr(ar,lr); af=ar-acr; ab=ARELATIVE(a);
wr=AR(w); ws=AS(w); wcr=efr(wr,rr); wf=wr-wcr; wb=ARELATIVE(w);
if(!af&&!wf)R CALL2(f2,a,w,fs); // if there's only one cell, run on it, that's the result
// multiple cells. Loop through them.
// ?cn=number of atoms in a cell, ?k=#bytes in a cell, uv point to one cell before aw data
// Allocate y? to hold one cell of ?, with uu,vv pointing to the data of y?
RE(acn=prod(acr,as+af)); ak=acn*bp(at); u=CAV(a)-ak; NEWYA;
RE(wcn=prod(wcr,ws+wf)); wk=wcn*bp(wt); v=CAV(w)-wk; NEWYW;
// b means 'w frame is larger'; p=#larger frame; q=#shorter frame; s->larger frame
// mn=#cells in larger frame (& therefore #cells in result); n=# times to repeat each cell
// from shorter-frame argument
b=af<=wf; p=b?wf:af; q=b?af:wf; s=b?ws:as; RE(mn=prod(p,s)); RE(n=prod(p-q,s+q));
ASSERT(!ICMP(as,ws,q),EVLENGTH); // error if frames are not same as prefix
// Initialize y? to hold data for the first cell; but if ? is empty, set y? to a cell of fills
if(AN(a))MOVEYA else RZ(ya=reshape(vec(INT,acr,as+af),filler(a)));
if(AN(w))MOVEYW else RZ(yw=reshape(vec(INT,wcr,ws+wf),filler(w)));
#define VALENCE 2
#define TEMPLATE 0
#include "cr_t.h"
}
static A jtcants(J jt,A a,A w,A z){A a1,q,y;B*b,*c;I*u,wr,zr;P*wp,*zp;
RZ(a&&w&&z);
RZ(a=grade1(a));
wr=AR(w); wp=PAV(w); a1=SPA(wp,a);
zr=AR(z); zp=PAV(z);
ASSERT(wr==zr,EVNONCE);
RZ(b=bfi(wr,a1,1));
GA(q,B01,wr,1,0); c=BAV(q); u=AV(a); DO(wr, c[i]=b[u[i]];);
A jtfxeachv(J jt,I r,A w){A*wv,x,z,*zv;I n,wd;
RZ(w);
n=AN(w); wv=AAV(w); wd=(I)w*ARELATIVE(w);
ASSERT(r>=AR(w),EVRANK);
ASSERT(n,EVLENGTH);
ASSERT(BOX&AT(w),EVDOMAIN);
GA(z,BOX,n,AR(w),AS(w)); zv=AAV(z);
DO(n, RZ(zv[i]=x=fx(WVR(i))); ASSERT(VERB&AT(x),EVDOMAIN););
static A jtipprep(J jt,A a,A w,I zt,I*pm,I*pn,I*pp){A z=mark;I*as,ar,ar1,m,mn,n,p,*ws,wr,wr1;
ar=AR(a); as=AS(a); ar1=ar?ar-1:0; RE(*pm=m=prod(ar1, as));
wr=AR(w); ws=AS(w); wr1=wr?wr-1:0; RE(*pn=n=prod(wr1,1+ws)); RE(mn=mult(m,n));
*pp=p=ar?*(as+ar1):wr?*ws:1;
ASSERT(!(ar&&wr)||p==*ws,EVLENGTH);
GA(z,zt,mn,ar1+wr1,0);
ICPY(AS(z), as,ar1);
ICPY(AS(z)+ar1,1+ws,wr1);
R z;
} /* argument validation & result for an inner product */
// Analysis for inner product
// a,w are arguments
// zt is type of result
// *pm is # 1-cells of a
// *pn is # atoms in an item of w
// *pp is number of inner-product muladds
// (in each, an atom of a multiplies an item of w)
static A jtipprep(J jt,A a,A w,I zt,I*pm,I*pn,I*pp){A z=mark;I*as,ar,ar1,m,mn,n,p,*ws,wr,wr1;
ar=AR(a); as=AS(a); ar1=ar?ar-1:0; RE(*pm=m=prod(ar1,as)); // m=# 1-cells of a. It could overflow, if there are no atoms
wr=AR(w); ws=AS(w); wr1=wr?wr-1:0; RE(*pn=n=prod(wr1,1+ws)); RE(mn=mult(m,n)); // n=#atoms in 1-cell of w; mn = #atoms in result
*pp=p=ar?*(as+ar1):wr?*ws:1; // if a is an array, the length of a 1-cell; otherwise, the number of items of w
ASSERT(!(ar&&wr)||p==*ws,EVLENGTH);
GA(z,zt,mn,ar1+wr1,0); // allocate result area
ICPY(AS(z), as,ar1); // Set shape: 1-frame of a followed by shape of item of w
ICPY(AS(z)+ar1,1+ws,wr1);
R z;
} /* argument validation & result for an inner product */
static DF2(rank2) {
DECLF;
A h=sv->h;
I ar,l,r,*v=AV(h),wr;
RZ(a&&w);
ar=AR(a);
l=efr(ar,v[1]);
wr=AR(w);
r=efr(wr,v[2]);
R l<ar||r<wr?rank2ex(a,w,fs,l,r,f2):CALL2(f2,a,w,fs);
}
// IRS setup for dyads x op y
// a is x, w is y
// fs is the f field of the verb (the verb to be applied repeatedly) - or 0 if none
// l, r are nominal ranks of fs
// f2 is a setup verb (jtover, jtreshape, etc)
A jtirs2(J jt,A a,A w,A fs,I l,I r,AF f2){A z;I af,ar,*old=jt->rank,rv[2],wf,wr;
// push the jt->rank (pointer to ranks) stack. push/pop may not match, no problem
RZ(a&&w);
ar=AR(a); rv[0]=l=efr(ar,l); af=ar-l; // get rank, effective rank, length of frame...
wr=AR(w); rv[1]=r=efr(wr,r); wf=wr-r; // ...for both args
if(!(af||wf))R CALL2(f2,a,w,fs); // if no frame, call setup verb and return result
ASSERT(!ICMP(AS(a),AS(w),MIN(af,wf)),EVLENGTH); // verify agreement
/* if(af&&wf&&af!=wf)R rank2ex(a,w,fs,l,r,f2); */
jt->rank=rv; z=CALL2(f2,a,w,fs); jt->rank=old; // save ranks, call setup verb, pop rank stack
// Not all setup verbs (*f2)() use the fs argument.
R z;
}
AF jtatcompf(J jt,A a,A w,A self){AF f;I ar,at,m,wr,wt;
RZ(a&&w);
at=AT(a); ar=AR(a);
wt=AT(w); wr=AR(w);
m=VAV(self)->flag%256;
if(1<ar||1<wr){if(32<=m&&m<=37||40<=m&&m<=45||48<=m&&m<=53)R(AF)jtfslashatg; RZ(7==m%8);}
ASSERT(AN(a)==AN(w)||!ar||!wr||5<m%8,EVLENGTH);
f=atcompX[m];
if(!f){
if(at&B01+INT+FL&&wt&B01+INT+FL)f=atcompxy[9*m+3*(at&B01?0:at&INT?1:2)+(wt&B01?0:wt&INT?1:2)];
else if(at&LIT&&wt&LIT) f=atcompC[m];
else if(at&SBT&&wt&SBT) f=atcompSB[m];
}
R f;
} /* function table look-up for comp i. 1: and i.&1@:comp etc. */
static F2(jtfitct){D d;V*sv;
RZ(a&&w);
ASSERT(!AR(w),EVRANK);
sv=VAV(a);
RZ(w=cvt(FL,w)); d=*DAV(w); ASSERT(0<=d&&d<5.82076609134675e-11,EVDOMAIN);
R CDERIV(CFIT,jtfitct1,jtfitct2,sv->mr,sv->lr,sv->rr);
}
static DF1(case1){A u;V*sv;
PREF1(case1);
sv=VAV(self);
RZ(u=from(df1(w,sv->g),sv->h));
ASSERT(!AR(u),EVRANK);
R df1(w,*AV(u));
}
static DF2(con2){A h,*hv,*x,z;V*sv;
PREF2(con2);
sv=VAV(self); h=sv->h; hv=AAV(h);
GA(z,BOX,AN(h),AR(h),AS(h)); x=AAV(z);
DO(AN(h), RZ(*x++=(VAV(*hv)->f2)(a,w,*hv)); ++hv);
R ope(z);
}
A jtrank1ex(J jt,A w,A fs,I mr,AF f1) {
PROLOG;
A y,y0,yw,z;
B wb;
C*v,*vv;
I k,mn,n=1,p,*s,wcn,wcr,wf,wk,wr,*ws,wt,yn,yr,*ys,yt;
RZ(w);
wt=AT(w);
if(wt&SPARSE)R sprank1(w,fs,mr,f1);
wr=AR(w);
ws=AS(w);
wcr=efr(wr,mr);
wf=wr-wcr;
wb=ARELATIVE(w);
if(!wf)R CALL1(f1,w,fs);
RE(wcn=prod(wcr,wf+ws));
wk=wcn*bp(wt);
v=CAV(w)-wk;
NEWYW;
p=wf;
s=ws;
RE(mn=prod(wf,ws));
if(AN(w))MOVEYW else RZ(yw=reshape(vec(INT,wcr,ws+wf),filler(w)));
#define VALENCE 1
#define TEMPLATE 0
#include "cr_t.h"
}
static A jttayamp(J jt,A w,B nf,A x,A h){A y;B ng=!nf;I j,n;V*v=VAV(h);
ASSERT(AR(x)<=(nf?v->lr:v->rr),EVRANK);
switch(v->id){
case CPLUS: R tpoly(over(x,one));
case CMINUS: R tpoly(nf?over(x,num[-1]):over(negate(x),one));
case CSTAR: R tpoly(over(zero,x));
case CDIV: ASSERT(ng,EVDOMAIN); R tpoly(over(zero,recip(x)));
case CJDOT: R tpoly(nf?over(x,a0j1):over(jdot1(x),one));
case CPOLY: ASSERT(nf,EVDOMAIN); R tpoly(BOX&AT(x)?poly1(x):x);
case CHGEOM: ASSERT(nf,EVDOMAIN); RE(j=i0(x)); ASSERT(0<=j,EVDOMAIN);
y=IX(j);
R tpoly(divide(hgcoeff(y,h),fact(y)));
case CBANG: ASSERT(nf,EVDOMAIN); RE(j=i0(x)); ASSERT(0<=j,EVDOMAIN);
R tpoly(divide(poly1(box(iota(x))),fact(x)));
case CEXP: if(nf)R eva(x,"(^.x)&^ % !");
RE(n=i0(x));
R 0<=n?tpoly(over(reshape(x,zero),one)):atop(ds(CDIV),amp(h,sc(-n)));
case CFIT: ASSERT(nf&&CPOLY==ID(v->f),EVDOMAIN);
y=over(x,IX(IC(x)));
R tpoly(mdiv(df2(x,y,h),atab(CEXP,y,IX(IC(x)))));
case CCIRCLE:
switch(i0(x)){
case 1: R eval("{&0 1 0 _1@(4&|) % !");
case -3: R eval("{&0 1 0 _1@(4&|) % ]");
case 2: R eval("{&1 0 _1 0@(4&|) % !");
case 5: R eval("2&| % !");
case -7: R eval("2&| % ]");
case 6: R eval("2&|@>: % !");
case -1: R eval("(2&| % ]) * ([: */ (1&+ % 2&+)@(i.@<.&.-:))\"0");
case -5: R eval("({&0 1 0 _1@(4&|) % ]) * ([: */ (1&+ % 2&+)@(i.@<.&.-:))\"0");
}}
ASSERT(0,EVDOMAIN);
}
void get_light(t_tmp *tmp, t_obj obc, t_cod cor, t_sph *obj_a)
{
t_sph *tm;
t_sph *obj_hit;
double old_lx[3];
double old_n[3];
double r;
double d;
tm = obj_a;
obj_hit = obc.nt;
init_light(&cor, old_lx, old_n);
while (obj_a)
{
if (obj_a->bri > 0)
{
set_r_d(&r, &d, &obc, obj_a);
set_ray(&cor, obj_a, &obc);
rand_vec(&(cor.l_x), (double)AR(atan(r / d)));
calc_pt(tmp, &obc, &cor, tm);
obc.bounce -= 80;
do_lum(&obc, &cor, obj_hit, (2 * M_PI *
(1 - (d / (sqrt(pow(d, 2) + pow(r, 2)))))));
}
obj_a = obj_a->nt;
}
de_init_light(&cor, old_lx, old_n);
}
static DF2(case2){A u;V*sv;
PREF2(case2);
sv=VAV(self);
RZ(u=from(df2(a,w,sv->g),sv->h));
ASSERT(!AR(u),EVRANK);
R df2(a,w,*AV(u));
}
A ga(I t, I r, I n, I *s) { I k=WP(t,r,n); A z=a_malloc(k);
AT(z)=t; AC(z)=1; AR(z)=r; AN(z)=n;
if (r==1) { *AS(z)=n; }
else if (r&&s) { ICPY(AS(z),s,r); }
gcpush(z);
R z;
}
static F2(jtpdtspmv){A ax,b,g,x,wx,y,yi,yj,z;B*bv;I m,n,s[2],*u,*v,*yv;P*ap,*wp,*zp;
RZ(a&&w);
ap=PAV(a); y=SPA(ap,i); yv=AV(y); s[0]=n=*AS(y); s[1]=1;
GATV(yj,INT,n,2,s);
if(DENSE&AT(w)){
GATV(yi,INT,n,2,s); u=AV(yi); AR(yj)=1; v=AV(yj);
DO(n, *u++=*yv++; *v++=*yv++;);
static A jtmerge1(J jt,A w,A ind){A z;B*b;C*wc,*zc;D*wd,*zd;I c,it,j,k,m,r,*s,t,*u,*wi,*zi;
RZ(w&&ind);
r=MAX(0,AR(w)-1); s=1+AS(w); t=AT(w); k=bp(t); m=IC(w); c=aii(w);
ASSERT(!(t&SPARSE),EVNONCE);
ASSERT(r==AR(ind),EVRANK);
ASSERT(!ICMP(s,AS(ind),r),EVLENGTH);
GA(z,t,c,r,s);
if(!(AT(ind)&B01+INT))RZ(ind=cvt(INT,ind));
it=AT(ind); u=AV(ind); b=(B*)u;
ASSERT(!c||1<m||!(it&B01),EVINDEX);
ASSERT(!c||1!=m||!memchr(b,C1,c),EVINDEX);
zi=AV(z); zc=(C*)zi; zd=(D*)zc;
wi=AV(w); wc=(C*)wi; wd=(D*)wc;
switch(MCASE(it,k)){
case MCASE(B01,sizeof(C)): DO(c, *zc++=wc[*b++?i+c:i];); break;
case MCASE(B01,sizeof(I)): DO(c, *zi++=wi[*b++?i+c:i];); break;
A gcinit(VO) { I k=WP(BOX,1,NOBJS); A memory;
nmem=mtop=bytes=totbytes=0;
memory=a_malloc(k);
AT(memory)=BOX; AR(memory)=1;
AN(memory)=*AS(memory)=NOBJS;
objs=AAV(memory);
R memory;
}
static DF1(jtgsuffix){A h,*hv,z,*zv;I m,n,r;
RZ(w);
if(jt->rank&&jt->rank[1]<AR(w)){r=jt->rank[1]; jt->rank=0; R rank1ex(w,self,jt->rank[1],jtgsuffix);}
jt->rank=0;
n=IC(w);
h=VAV(self)->h; hv=AAV(h); m=AN(h);
GATV(z,BOX,n,1,0); zv=AAV(z);
DO(n, RZ(zv[i]=df1(drop(sc(i),w),hv[i%m])););
float *mv_mult(float mat[SIZE][SIZE], float vec[SIZE]) {
static float ret[SIZE];
int i, j;
v4sf m, v, r, r0={0.,0.,0.,0.};
for (i = 0; i < SIZE; i++) {
r = r0;
for (j = 0; j < SIZE; j += 4) {
r += *(v4sf*)&mat[i][j] * *(v4sf*)&vec[j];
/*m = *(v4sf*)&mat[i][j];
v = *(v4sf*)&vec[j];
v = m * v;
r = r + v;*/
}
ret[i] = AR(r)[0] + AR(r)[1] + AR(r)[2] + AR(r)[3];
}
return ret;
}
int main(){
v4sf a={1,2,3,4}, b={5,6,7,8}, c;
float mat[SIZE][SIZE], vec[SIZE], *res;
int i,j;
for(i=0;i<SIZE;i++) for(j=0;j<SIZE;j++) mat[i][j]=0.+i+j;
for(i=0;i<SIZE;i++) vec[i]=1.;
c = a * b;
printf("%f, %f, %f, %f\n", AR(c)[0], AR(c)[1], AR(c)[2], AR(c)[3] );
printf("size = %d\n",sizeof(v4sf));
res = mv_mult( mat, vec);
for(i=0;i<SIZE;i++) printf("%d %f \n", i, res[i]);
return 0;
}
A jtirs1(J jt,A w,A fs,I m,AF f1){A z;I*old=jt->rank,rv[2],wr;
RZ(w);
wr=AR(w); rv[1]=m=efr(wr,m);
if(m>=wr)R CALL1(f1,w,fs);
rv[0]=0;
jt->rank=rv; z=CALL1(f1,w,fs); jt->rank=old;
R z;
}
vector compgene(cmatrix& A,cmatrix& B)
{
int i;
int N=A.Rows;
int IFAIL=0;
int MATV=0;
int* ITER=new int[N];
vector ALFR(N);
vector ALFI(N);
vector BETA(N);
double EPS1=0.;
matrix AR(N,N);
matrix AI(N,N);
matrix BR(N,N);
matrix BI(N,N);
for (i=0;i<N;i++) for (int j=0;j<N;j++) {
AR(i,j)=real(A(i,j));
AI(i,j)=imag(A(i,j));
BR(i,j)=real(B(i,j));
BI(i,j)=imag(B(i,j));
}
matrix VR(N,N);
matrix VI(N,N);
/* f02gjf_(&N,AR.TheMatrix,&N,AI.TheMatrix,&N,BR.TheMatrix,&N,
BI.TheMatrix,&N,&EPS1,ALFR.TheVector,ALFI.TheVector,
BETA.TheVector,&MATV,VR.TheMatrix,&N,
VI.TheMatrix,&N,ITER,&IFAIL); */
cerr<<"compgene to be implemented"<<endl;
exit(1);
if (IFAIL != 0) cerr <<"error in compgene "<<endl;
if (!ALFR.TheVector) exit(1);
for (i=0;i<N;i++) ALFR(i)=ALFR(i)/BETA(i);
// sort !
char ORDER='A';
int M1=1;
int M2=N;
/* m01caf_(ALFR.TheVector,&M1,&M2,&ORDER,&IFAIL);*/
cerr<<"sort to be implemented"<<endl;
exit(1);
if (IFAIL != 0) cerr <<"error in m01caf_ "<<endl;
return ALFR;
}
inline void
SyrkLN
( T alpha, const DistMatrix<T>& A, T beta, DistMatrix<T>& C,
bool conjugate=false )
{
#ifndef RELEASE
PushCallStack("internal::SyrkLN");
if( A.Grid() != C.Grid() )
throw std::logic_error
("A and C must be distributed over the same grid");
if( A.Height() != C.Height() || A.Height() != C.Width() )
{
std::ostringstream msg;
msg << "Nonconformal SyrkLN:\n"
<< " A ~ " << A.Height() << " x " << A.Width() << "\n"
<< " C ~ " << C.Height() << " x " << C.Width() << "\n";
throw std::logic_error( msg.str().c_str() );
}
#endif
const Grid& g = A.Grid();
// Matrix views
DistMatrix<T> AL(g), AR(g),
A0(g), A1(g), A2(g);
// Temporary distributions
DistMatrix<T,MC, STAR> A1_MC_STAR(g);
DistMatrix<T,VR, STAR> A1_VR_STAR(g);
DistMatrix<T,STAR,MR > A1Trans_STAR_MR(g);
A1_MC_STAR.AlignWith( C );
A1_VR_STAR.AlignWith( C );
A1Trans_STAR_MR.AlignWith( C );
// Start the algorithm
ScaleTrapezoid( beta, LEFT, LOWER, 0, C );
LockedPartitionRight( A, AL, AR, 0 );
while( AR.Width() > 0 )
{
LockedRepartitionRight
( AL, /**/ AR,
A0, /**/ A1, A2 );
//--------------------------------------------------------------------//
A1_VR_STAR = A1_MC_STAR = A1;
A1Trans_STAR_MR.TransposeFrom( A1_VR_STAR, conjugate );
LocalTrrk( LOWER, alpha, A1_MC_STAR, A1Trans_STAR_MR, T(1), C );
//--------------------------------------------------------------------//
SlideLockedPartitionRight
( AL, /**/ AR,
A0, A1, /**/ A2 );
}
#ifndef RELEASE
PopCallStack();
#endif
}
// This is the derived verb for f/. y
static DF1(jtoblique){A x,y;I m,n,r,*u,*v;
RZ(w);
r=AR(w); // r = rank of w
// create y= ,/ w - the _2-cells of w arranged in a list
// we just create a header for y, pointing to the data from w
RZ(y=gah(MAX(r-1,1),w));
u=AS(w); v=AS(y); // u,v->shape of y
if(1>=r){*v=m=AN(w); n=1;}else{m=*u++; n=*u++; *v++=m*n; ICPY(v,u,r-2);} // set shape of y as _2-cells of w
// Create x=+"0 1&i./ 2 {. $y
RZ(x=irs2(IX(m),IX(n),0L,0L,1L,jtplus)); AR(x)=1; *AS(x)=AN(x);
// perform x f/. y, which does the requested operation
RZ(x=df2(x,y,sldot(VAV(self)->f)));
// Final tweak: the result should have (0 >. <: +/ 2 {. $y) cells. It will, as long as
// m and n are both non0: when one is 0, result has 0 cells (but that cell is the correct result
// of execution on a fill-cell). Correct the length of the 0 case, when the result length should be nonzero
// if((m==0 || n==0) && (m+n>0)){R reitem(sc(m+n-1),x);} This change withdrawn pending further deliberation
R x;
}
static DF1(rank1) {
DECLF;
A h=sv->h;
I m,*v=AV(h),wr;
RZ(w);
wr=AR(w);
m=efr(wr,v[0]);
R m<wr?rank1ex(w,fs,m,f1):CALL1(f1,w,fs);
}
A jtirs2(J jt,A a,A w,A fs,I l,I r,AF f2) {
A z;
I af,ar,*old=jt->rank,rv[2],wf,wr;
RZ(a&&w);
ar=AR(a);
rv[0]=l=efr(ar,l);
af=ar-l;
wr=AR(w);
rv[1]=r=efr(wr,r);
wf=wr-r;
if(!(af||wf))R CALL2(f2,a,w,fs);
ASSERT(!ICMP(AS(a),AS(w),MIN(af,wf)),EVLENGTH);
/* if(af&&wf&&af!=wf)R rank2ex(a,w,fs,l,r,f2); */
jt->rank=rv;
z=CALL2(f2,a,w,fs);
jt->rank=old;
R z;
}
inline void
LocalTrrkKernel
( UpperOrLower uplo,
Orientation orientationOfA,
Orientation orientationOfB,
T alpha, const DistMatrix<T,STAR,MC >& A,
const DistMatrix<T,MR, STAR>& B,
T beta, DistMatrix<T,MC, MR >& C )
{
#ifndef RELEASE
PushCallStack("LocalTrrkKernel");
CheckInput( orientationOfA, orientationOfB, A, B, C );
#endif
const Grid& g = C.Grid();
DistMatrix<T,STAR,MC> AL(g), AR(g);
DistMatrix<T,MR,STAR> BT(g),
BB(g);
DistMatrix<T,MC,MR> CTL(g), CTR(g),
CBL(g), CBR(g);
DistMatrix<T,MC,MR> DTL(g), DBR(g);
const int half = C.Height()/2;
ScaleTrapezoid( beta, LEFT, uplo, 0, C );
LockedPartitionRight( A, AL, AR, half );
LockedPartitionDown
( B, BT,
BB, half );
PartitionDownDiagonal
( C, CTL, CTR,
CBL, CBR, half );
DTL.AlignWith( CTL );
DBR.AlignWith( CBR );
DTL.ResizeTo( CTL.Height(), CTL.Width() );
DBR.ResizeTo( CBR.Height(), CBR.Width() );
//------------------------------------------------------------------------//
if( uplo == LOWER )
internal::LocalGemm
( orientationOfA, orientationOfB, alpha, AR, BT, T(1), CBL );
else
internal::LocalGemm
( orientationOfA, orientationOfB, alpha, AL, BB, T(1), CTR );
internal::LocalGemm
( orientationOfA, orientationOfB, alpha, AL, BT, T(0), DTL );
AxpyTriangle( uplo, T(1), DTL, CTL );
internal::LocalGemm
( orientationOfA, orientationOfB, alpha, AR, BB, T(0), DBR );
AxpyTriangle( uplo, T(1), DBR, CBR );
//------------------------------------------------------------------------//
#ifndef RELEASE
PopCallStack();
#endif
}
void readFastX(std::string const filename)
{
reader_type AR(filename);
typename reader_type::pattern_type pattern;
while ( AR.getNextPatternUnlocked(pattern) )
{
std::cerr << pattern.pattern << std::endl;
}
}
