void Jit::ApplyPrefixD(const u8 *vregs, VectorSize sz) {
_assert_(js.prefixDFlag & ArmJitState::PREFIX_KNOWN);
if (!js.prefixD) return;
int n = GetNumVectorElements(sz);
for (int i = 0; i < n; i++) {
if (js.VfpuWriteMask(i))
continue;
int sat = (js.prefixD >> (i * 2)) & 3;
if (sat == 1) {
// clamped = fabs(x) - fabs(x-0.5f) + 0.5f; // [ 0, 1]
fpr.MapRegV(vregs[i], MAP_DIRTY);
MOVI2F(S0, 0.5, R0);
VABS(S1, fpr.V(vregs[i])); // S1 = fabs(x)
VSUB(S2, fpr.V(vregs[i]), S0); // S2 = fabs(x-0.5f) {VABD}
VABS(S2, S2);
VSUB(fpr.V(vregs[i]), S1, S2); // v[i] = S1 - S2 + 0.5f
VADD(fpr.V(vregs[i]), fpr.V(vregs[i]), S0);
} else if (sat == 3) {
// clamped = fabs(x) - fabs(x-1.0f); // [-1, 1]
fpr.MapRegV(vregs[i], MAP_DIRTY);
MOVI2F(S0, 1.0, R0);
VABS(S1, fpr.V(vregs[i])); // S1 = fabs(x)
VSUB(S2, fpr.V(vregs[i]), S0); // S2 = fabs(x-1.0f) {VABD}
VABS(S2, S2);
VSUB(fpr.V(vregs[i]), S1, S2); // v[i] = S1 - S2
}
}
}
void JitArm::ps_abs(UGeckoInstruction inst)
{
INSTRUCTION_START
JITDISABLE(bJITPairedOff);
FALLBACK_IF(inst.Rc);
u32 b = inst.FB, d = inst.FD;
ARMReg vB0 = fpr.R0(b);
ARMReg vB1 = fpr.R1(b);
ARMReg vD0 = fpr.R0(d, false);
ARMReg vD1 = fpr.R1(d, false);
VABS(vD0, vB0);
VABS(vD1, vB1);
}
/* ///////////////////////////////////////////////////////////////////////////
// Routine: Ju
//
// Purpose: Evaluate the integrand J_k(u) of the energy functional J(u)
// at the single point x. This is your nonlinear energy
// functional for which your weak form PDE below in Fu_v() is the
// Euler condition. (There may not be such a J(u) in all cases.)
//
// /\ /\
// J(u) = \ J_0(u) dx + \ J_1(u) ds
// \/m \/dm
//
// Input: PDE = pointer to the PDE object
// key = integrand to evaluate (0=J_0, 1=J_1)
//
// Output: Value of the integrand is returned
//
// Speed: This function is called by MC once times for a single
// quadrature point during assembly, and needs to be fast.
//
// Author: Michael Holst
/////////////////////////////////////////////////////////////////////////// */
VPUBLIC double Ju(PDE *thee, int key)
{
double value = 0.0;
double mytime = PDE_getTime( thee );
int ekey = PDE_getEnergyKey( thee );
switch( ekey ) {
case 0:
/* interior form case */
if (key == 0) {
value = my_US(thee->dim, thee->vec, xq, mytime);
value = VABS(value - U[0]);
/* boundary form case */
} else if (key == 1) {
value = 0.0;
} else { VASSERT(0); }
break;
case 1:
value = 1.0;
break;
default:
value = 0.0;
break;
}
return value;
}
void JitArm::fabsx(UGeckoInstruction inst)
{
INSTRUCTION_START
JITDISABLE(FloatingPoint)
ARMReg vD = fpr.R0(inst.FD);
ARMReg vB = fpr.R0(inst.FB);
VABS(vD, vB);
if (inst.Rc) Helper_UpdateCR1(vD);
}
void Jit::ApplyPrefixST(u8 *vregs, u32 prefix, VectorSize sz) {
if (prefix == 0xE4) return;
int n = GetNumVectorElements(sz);
u8 origV[4];
static const float constantArray[8] = {0.f, 1.f, 2.f, 0.5f, 3.f, 1.f/3.f, 0.25f, 1.f/6.f};
for (int i = 0; i < n; i++)
origV[i] = vregs[i];
for (int i = 0; i < n; i++)
{
int regnum = (prefix >> (i*2)) & 3;
int abs = (prefix >> (8+i)) & 1;
int negate = (prefix >> (16+i)) & 1;
int constants = (prefix >> (12+i)) & 1;
// Unchanged, hurray.
if (!constants && regnum == i && !abs && !negate)
continue;
// This puts the value into a temp reg, so we won't write the modified value back.
vregs[i] = fpr.GetTempV();
fpr.MapRegV(vregs[i], MAP_NOINIT | MAP_DIRTY);
if (!constants) {
// Prefix may say "z, z, z, z" but if this is a pair, we force to x.
// TODO: But some ops seem to use const 0 instead?
if (regnum >= n) {
ERROR_LOG_REPORT(CPU, "Invalid VFPU swizzle: %08x / %d", prefix, sz);
regnum = 0;
}
if (abs) {
VABS(fpr.V(vregs[i]), fpr.V(origV[regnum]));
} else {
VMOV(fpr.V(vregs[i]), fpr.V(origV[regnum]));
}
} else {
// TODO: There is VMOV s, imm on ARM, that can generate some of these constants. Not 1/3 or 1/6 though.
MOVI2F(fpr.V(vregs[i]), constantArray[regnum + (abs<<2)], R0);
}
// TODO: This can be integrated into the VABS / VMOV above, and also the constants.
if (negate)
VNEG(fpr.V(vregs[i]), fpr.V(vregs[i]));
// TODO: This probably means it will swap out soon, inefficiently...
fpr.ReleaseSpillLockV(vregs[i]);
}
}
void Jit::ApplyPrefixST(u8 *vregs, u32 prefix, VectorSize sz) {
if (prefix == 0xE4) return;
int n = GetNumVectorElements(sz);
u8 origV[4];
static const float constantArray[8] = {0.f, 1.f, 2.f, 0.5f, 3.f, 1.f/3.f, 0.25f, 1.f/6.f};
for (int i = 0; i < n; i++)
origV[i] = vregs[i];
for (int i = 0; i < n; i++)
{
int regnum = (prefix >> (i*2)) & 3;
int abs = (prefix >> (8+i)) & 1;
int negate = (prefix >> (16+i)) & 1;
int constants = (prefix >> (12+i)) & 1;
// Unchanged, hurray.
if (!constants && regnum == i && !abs && !negate)
continue;
// This puts the value into a temp reg, so we won't write the modified value back.
vregs[i] = fpr.GetTempV();
if (!constants) {
fpr.MapDirtyInV(vregs[i], origV[regnum]);
fpr.SpillLockV(vregs[i]);
// Prefix may say "z, z, z, z" but if this is a pair, we force to x.
// TODO: But some ops seem to use const 0 instead?
if (regnum >= n) {
WARN_LOG(CPU, "JIT: Invalid VFPU swizzle: %08x : %d / %d at PC = %08x (%s)", prefix, regnum, n, js.compilerPC, currentMIPS->DisasmAt(js.compilerPC));
regnum = 0;
}
if (abs) {
VABS(fpr.V(vregs[i]), fpr.V(origV[regnum]));
if (negate)
VNEG(fpr.V(vregs[i]), fpr.V(vregs[i]));
} else {
if (negate)
VNEG(fpr.V(vregs[i]), fpr.V(origV[regnum]));
else
VMOV(fpr.V(vregs[i]), fpr.V(origV[regnum]));
}
} else {
fpr.MapRegV(vregs[i], MAP_DIRTY | MAP_NOINIT);
fpr.SpillLockV(vregs[i]);
MOVI2F(fpr.V(vregs[i]), constantArray[regnum + (abs<<2)], R0, negate);
}
}
}
VPUBLIC double Vxnrm1(int *nx, int *ny, int *nz,
double *x) {
double xnrm1 = 0.0; ///< Accumulates the calculated normal value
MAT3(x, *nx, *ny, *nz);
// The indices used to traverse the matrices
int i, j, k;
/// @todo parallel optimization
for(k=2; k<=*nz-1; k++)
for(j=2; j<=*ny-1; j++)
for(i=2; i<=*nx-1; i++)
xnrm1 += VABS(VAT3(x, i, j, k));
return xnrm1;
}
/*
* ***************************************************************************
* Routine: Slu_lnDet
*
* Purpose: Calculate the log of the determinant of a factored matrix.
*
* Notes: UMFPACK has a built-in routine to compute the determinant,
* which returns both the mantissa and the exponent separately.
* This avoids most overflow and underflow problems.
*
* Author: Stephen Bond
* ***************************************************************************
*/
VPUBLIC double Slu_lnDet(Slu *thee)
{
int status;
double Mx, Ex, lndet;
void *Numeric = thee->work;
VASSERT( thee != VNULL );
VASSERT( thee->statLU );
/* Determinant = Mx * 10^Ex */
status = umfpack_di_get_determinant( &Mx, &Ex, Numeric, VNULL );
/* LOG(DET) = LOG(Mantissa) + Exponent*LOG(10) */
lndet = VLOG(VABS(Mx)) + Ex*VLOG(10);
if (UMFPACK_OK == status) {
Vnm_print(0, "Slu_lnDet: ln(det(A)) = %g\n", lndet);
return lndet;
} else {
Vnm_print(0, "Slu_lnDet: Failed! Returning 1.0\n");
return 1.0;
}
}
/*
* ***************************************************************************
* Routine: Gem_formFix
*
* Purpose: Make some specified hacked fix to a given mesh.
*
* Notes: key==0 --> ?
*
* Author: Michael Holst
* ***************************************************************************
*/
VPUBLIC void Gem_formFix(Gem *thee, int key)
{
int i, j, k, l, m, nabors, btype;
double radk, radl, radm, myTol;
VV *v[4];
SS *sm, *sm0, *sm1, *sm2;
/* input check and some i/o */
btype = key;
VASSERT( (0 <= btype) && (btype <= 2) );
/* go through all simplices and zero all boundary faces */
Vnm_print(0,"Gem_makeBnd: zeroing boundary faces/vertices..");
Gem_setNumBF(thee, 0);
Gem_setNumBV(thee, 0);
for (i=0; i<Gem_numSS(thee); i++) {
sm = Gem_SS(thee,i);
if ( (i>0) && (i % VPRTKEY) == 0 ) Vnm_print(0,"[BS:%d]",i);
/* get local vertices */
for (j=0; j<Gem_dimVV(thee); j++)
v[j] = SS_vertex(sm,j);
/* reset all vertices and faces to interior type */
for (j=0; j<Gem_dimVV(thee); j++) {
/* the other three local vertex/face numbers besides "j" */
k=(j+1) % Gem_dimVV(thee);
l=(k+1) % Gem_dimVV(thee);
m=(l+1) % Gem_dimVV(thee);
SS_setFaceType(sm, j, 0);
VV_setType(v[k], 0);
VV_setType(v[l], 0);
if (Gem_dim(thee) == 3) VV_setType(v[m], 0);
}
}
Vnm_print(0,"..done.\n");
/* are we done */
/* if (btype == 0) return; */
/* okay now make a boundary */
Vnm_print(0,"Gem_makeBnd: rebuilding boundary faces/vertices..");
for (i=0; i<Gem_numSS(thee); i++) {
sm = Gem_SS(thee,i);
if ( (i>0) && (i % VPRTKEY) == 0 ) Vnm_print(0,"[BS:%d]",i);
/* get local vertices */
for (j=0; j<Gem_dimVV(thee); j++)
v[j] = SS_vertex(sm,j);
/* rebuild everything */
for (j=0; j<Gem_dimVV(thee); j++) {
/* the other three local vertex/face numbers besides "j" */
k=(j+1) % Gem_dimVV(thee);
l=(k+1) % Gem_dimVV(thee);
m=(l+1) % Gem_dimVV(thee);
/* look for a face nabor sharing face "j" (opposite vertex "j") */
nabors = 0;
for (sm0=VV_firstSS(v[k]); sm0!=VNULL;sm0=SS_link(sm0,v[k])) {
for (sm1=VV_firstSS(v[l]); sm1!=VNULL; sm1=SS_link(sm1,v[l])) {
if (Gem_dim(thee) == 2) {
if ((sm0!=sm) && (sm0==sm1)) nabors++;
} else {
for (sm2=VV_firstSS(v[m]); sm2!=VNULL;
sm2=SS_link(sm2,v[m])) {
if ((sm0!=sm) && (sm0==sm1) && (sm0==sm2)) {
nabors++;
}
}
}
}
}
/* if no one there, then face "j" is actually a boundary face */
if (nabors == 0) {
myTol = 1.0e-2;
if ( ( VABS(VV_coord(v[k],2) - 0.0) < myTol)
&& ( VABS(VV_coord(v[l],2) - 0.0) < myTol)
&& ( VABS(VV_coord(v[m],2) - 0.0) < myTol) ) {
btype = 1;
} else if ( ( VABS(VV_coord(v[k],2) - 68.03512) < myTol)
&& ( VABS(VV_coord(v[l],2) - 68.03512) < myTol)
&& ( VABS(VV_coord(v[m],2) - 68.03512) < myTol) ) {
btype = 3;
} else {
radk = VSQRT( VSQR( VV_coord(v[k],0) )
+ VSQR( VV_coord(v[k],1) ) );
radl = VSQRT( VSQR( VV_coord(v[l],0) )
+ VSQR( VV_coord(v[l],1) ) );
radm = VSQRT( VSQR( VV_coord(v[m],0) )
+ VSQR( VV_coord(v[m],1) ) );
if ( ( VABS(radk - 1.5) < myTol)
&& ( VABS(radl - 1.5) < myTol)
&& ( VABS(radm - 1.5) < myTol) ) {
btype = 2;
} else if ( ( VABS(radk - 2.0) < myTol)
&& ( VABS(radl - 2.0) < myTol)
&& ( VABS(radm - 2.0) < myTol) ){
btype = 4;
} else {
btype = 0;
}
}
SS_setFaceType(sm, j, btype);
Gem_numBFpp(thee);
if (VINTERIOR( VV_type(v[k])) ) {
VV_setType(v[k], btype);
Gem_numBVpp(thee);
}
if (VINTERIOR( VV_type(v[l])) ) {
VV_setType(v[l], btype);
Gem_numBVpp(thee);
}
if (Gem_dim(thee) == 3) {
if (VINTERIOR( VV_type(v[m])) ) {
VV_setType(v[m], btype);
Gem_numBVpp(thee);
}
}
}
}
}
Vnm_print(0,"..done.\n");
}
/*
* ***************************************************************************
* Routine: Mat_printLN
*
* Purpose: Print an LN format matrix as a DENSE matrix in MATLAB format.
*
* Author: Stephen Bond and Michael Holst
* ***************************************************************************
*/
VPUBLIC void Mat_printLN(Mat *thee)
{
int i, j;
int numR, numC;
char rn[80];
const int MaxRows = 30;
const int MaxCols = 30;
double matrix[30][30];
LinkA *mt;
LinkRC *mtX;
numR = thee->numR;
numC = thee->numC;
strncpy(rn,"Mat_printLN:",80);
/* some i/o */
Vnm_print(0, "%s printing <%s>" " [dim=(%dx%d),sym=%d,numA=%d]\n",
rn, thee->name, numR, numC, thee->sym, thee->numA);
/* size check */
if ((numR > MaxRows) || (numC > MaxCols)) {
Vnm_print(0, "%smatrix too large to view....skipping.\n", rn);
return;
}
/* make a dense matrix first */
for (i=0; i<numR; i++)
for (j=0; j<numC; j++)
matrix[i][j] = 0.0;
if (thee->state != NULL_STATE) {
switch (thee->format) {
case RLN_FORMAT:
for (i=0; i<numR; i++) {
for (mt=(LinkA*)Vset_access(thee->lnkU,i);
mt!=VNULL; mt=mt->next) {
if (mt->idx >= 0) {
j = mt->idx;
matrix[i][j] = mt->val;
}
}
}
break;
case CLN_FORMAT:
for (j=0; j<numC; j++) {
for (mt=(LinkA*)Vset_access(thee->lnkL,j);
mt!=VNULL; mt=mt->next) {
if (mt->idx >= 0) {
i = mt->idx;
matrix[i][j] = mt->val;
}
}
}
break;
case XLN_FORMAT:
for (i=0; i<numR; i++) {
if ( thee->sym == ISNOT_SYM ) {
mtX = ((LinkRC**) thee->xln)[i];
} else {
mtX = (LinkRC*) &( ((LinkRCS*) thee->xln)[i] );
matrix[i][i] = ((LinkRCS*) mtX)->val;
mtX = mtX->next;
}
for ( /* no-op */ ; mtX!=VNULL; mtX=mtX->next) {
j = mtX->idx;
if (j < numC) {
if ( thee->sym == ISNOT_SYM ) {
matrix[i][j] = ((LinkRCS*) mtX)->val;
} else if ( thee->sym == IS_SYM ) {
matrix[i][j] = ((LinkRCS*) mtX)->val;
matrix[j][i] = ((LinkRCS*) mtX)->val;
} else {
matrix[i][j] = ((LinkRCN*) mtX)->val;
matrix[j][i] = ((LinkRCN*) mtX)->valT;
}
}
}
}
break;
default:
Vnm_print(0,
"%smatrix not in correct format to print....skipping.\n", rn);
break;
}
}
/* print the matrix */
Vnm_print(0, "%s = [\n", thee->name);
for (i=0; i<numR; i++) {
for (j=0; j<numC; j++) {
if (VABS(matrix[i][j]) < 0.0001) {
Vnm_print(0, " 0.0 ");
} else {
Vnm_print(0, "%7.3f", matrix[i][j]);
}
}
Vnm_print(0, "\n");
}
Vnm_print(0, "];\n");
}
VPUBLIC void Vpower(int *nx, int *ny, int *nz,
int *iz, int *ilev,
int *ipc, double *rpc, double *ac, double *cc,
double *w1, double *w2, double *w3, double *w4,
double *eigmax, double *eigmax_model, double *tol,
int *itmax, int *iters, int *iinfo) {
int lev, level;
double denom, fac, rho, oldrho, error, relerr;
/// @todo Just use a constant definition of PI here
double pi = 4.0 * atan( 1.0 );
// Utility variables
int skipIters = 0;
double alpha;
MAT2(iz, 50, 1);
WARN_UNTESTED;
// Recover level information
level = 1;
lev = (*ilev - 1) + level;
// Seed vector: random to contain all components
Vaxrand(nx, ny, nz, w1);
Vazeros(nx, ny, nz, w2);
Vazeros(nx, ny, nz, w3);
Vazeros(nx, ny, nz, w4);
// Compute raleigh quotient with the seed vector
denom = Vxnrm2(nx, ny, nz, w1);
fac = 1.0 / denom;
Vxscal(nx, ny, nz, &fac, w1);
Vmatvec(nx, ny, nz,
RAT(ipc, VAT2(iz, 5, lev)), RAT(rpc, VAT2(iz, 6,lev)),
RAT(ac, VAT2(iz, 7, lev)), RAT(cc, VAT2(iz, 1,lev)), w1, w2);
oldrho = Vxdot(nx, ny, nz, w1, w2);
// I/O
if (oldrho == 0.0) {
if (*iinfo > 3) {
Vnm_print(2, "POWER: iter: estimate = %d %g\n", *iters, oldrho);
}
rho = oldrho;
} else {
// Main iteration
*iters = 0;
while(1) {
(*iters)++;
// Apply the matrix A
Vmatvec(nx, ny, nz,
RAT(ipc, VAT2(iz, 5, lev)), RAT(rpc, VAT2(iz, 6, lev)),
RAT(ac, VAT2(iz, 7, lev)), RAT(cc, VAT2(iz, 1, lev)), w1, w2);
Vxcopy(nx, ny, nz, w2, w1);
// Normalize the new vector
denom = Vxnrm2(nx, ny, nz, w1);
fac = 1.0 / denom;
Vxscal(nx, ny, nz, &fac, w1);
// Compute the new raleigh quotient
Vmatvec(nx, ny, nz,
RAT(ipc, VAT2(iz, 5, lev)), RAT(rpc, VAT2(iz, 6, lev)),
RAT( ac, VAT2(iz, 7, lev)), RAT( cc, VAT2(iz, 1, lev)), w1, w2);
rho = Vxdot(nx, ny, nz, w1, w2);
// Stopping test ***
// w2=A*x, w1=x, stop = 2-norm(A*x-lamda*x)
Vxcopy(nx, ny, nz, w1, w3);
Vxcopy(nx, ny, nz, w2, w4);
Vxscal(nx, ny, nz, &rho, w3);
alpha = -1.0;
Vxaxpy(nx, ny, nz, &alpha, w3, w4);
error = Vxnrm2(nx, ny, nz, w4);
relerr = VABS(rho - oldrho ) / VABS( rho );
// I/O
if (*iinfo > 3) {
Vnm_print(2, "POWER: iters =%d\n", *iters);
Vnm_print(2, " error =%g\n", error);
Vnm_print(2, " relerr =%g\n", relerr);
Vnm_print(2, " rho =%g\n", rho);
}
if( relerr < *tol || *iters == *itmax)
break;
oldrho = rho;
}
}
// Return some stuff ***
*eigmax = rho;
fac = VPOW(2.0, *ilev - 1);
*eigmax_model = fac * (6.0 - 2.0 * VCOS((*nx - 2) * pi / (*nx - 1))
- 2.0 * VCOS((*ny - 2) * pi / (*ny - 1)));
}
VPUBLIC void Vipower(int *nx,int *ny,int *nz,
double *u, int *iz,
double *w0, double *w1, double *w2, double *w3, double *w4,
double *eigmin, double *eigmin_model, double *tol,
int *itmax, int *iters,
int *nlev, int *ilev, int *nlev_real, int *mgsolv,
int *iok, int *iinfo, double *epsiln, double *errtol, double *omega,
int *nu1, int *nu2, int *mgsmoo,
int *ipc, double *rpc,
double *pc, double *ac, double *cc, double *tru) {
int level, lev;
double denom, fac, rho, oldrho;
double error, relerr, errtol_s;
int itmax_s, iters_s, ierror_s, iok_s, iinfo_s, istop_s;
int nu1_s, nu2_s, mgsmoo_s;
/// @todo Just use a constant definition of PI here
double pi = 4.0 * atan( 1.0 );
// Utility variables
double alpha;
MAT2(iz, 50, 1);
WARN_UNTESTED;
// Recover level information
level = 1;
lev = (*ilev - 1) + level;
// Seed vector: random to contain all components
Vaxrand(nx, ny, nz, w1);
Vazeros(nx, ny, nz, w2);
Vazeros(nx, ny, nz, w3);
Vazeros(nx, ny, nz, w4);
Vazeros(nx, ny, nz, RAT(w0, VAT2(iz, 1, lev)));
Vazeros(nx, ny, nz, RAT( u, VAT2(iz, 1, lev)));
// Compute raleigh quotient with the seed vector ***
denom = Vxnrm2(nx, ny, nz, w1);
fac = 1.0 / denom;
Vxscal(nx, ny, nz, &fac, w1);
Vmatvec(nx, ny, nz,
RAT(ipc, VAT2(iz, 5, lev)), RAT(rpc, VAT2(iz, 6, lev)),
RAT( ac, VAT2(iz, 7, lev)), RAT( cc, VAT2(iz, 1, lev)), w1, w2);
oldrho = Vxdot(nx, ny, nz, w1, w2);
// I/O
if (oldrho == 0.0) {
if (*iinfo > 3) {
Vnm_print(2, "Vipower: iters=%d\n", *iters);
Vnm_print(2, " estimate=%f\n", oldrho);
}
rho = oldrho;
} else {
//main iteration
*iters = 0;
while (1) {
(*iters)++;
// Apply the matrix A^{-1} (using MG solver)
itmax_s = 100;
iters_s = 0;
ierror_s = 0;
iok_s = 0;
iinfo_s = 0;
istop_s = 0;
mgsmoo_s = 1;
nu1_s = 1;
nu2_s = 1;
errtol_s = *epsiln;
Vxcopy(nx, ny, nz, w1, RAT(w0, VAT2(iz, 1,lev)));
Vmvcs(nx, ny, nz, u, iz,
w1, w2, w3, w4,
&istop_s, &itmax_s, &iters_s, &ierror_s,
nlev, ilev, nlev_real, mgsolv,
&iok_s, &iinfo_s, epsiln,
&errtol_s, omega, &nu1_s, &nu2_s, &mgsmoo_s,
ipc, rpc, pc, ac, cc, w0, tru);
Vxcopy(nx, ny, nz, RAT(u, VAT2(iz, 1, lev)), w1);
// Normalize the new vector
denom = Vxnrm2(nx, ny, nz, w1);
fac = 1.0 / denom;
Vxscal(nx, ny, nz, &fac, w1);
// Compute the new raleigh quotient
Vmatvec(nx, ny, nz,
RAT(ipc, VAT2(iz, 5, lev)), RAT(rpc, VAT2(iz, 6, lev)),
RAT(ac, VAT2(iz, 7,lev)), RAT(cc, VAT2(iz, 1, lev)), w1, w2);
rho = Vxdot(nx, ny, nz, w1, w2);
// Stopping test
// w2=A*x, w1=x, stop = 2-norm(A*x-lamda*x) ***
Vxcopy(nx, ny, nz, w1, w3);
Vxcopy(nx, ny, nz, w2, w4);
Vxscal(nx, ny, nz, &rho, w3);
alpha = -1.0;
Vxaxpy(nx, ny, nz, &alpha, w3, w4);
error = Vxnrm2(nx, ny, nz, w4);
relerr = VABS(rho - oldrho ) / VABS( rho );
// I/O
if (*iinfo > 3) {
Vnm_print(2, "POWER: iters =%d\n", *iters);
Vnm_print(2, " error =%g\n", error);
Vnm_print(2, " relerr =%g\n", relerr);
Vnm_print(2, " rho =%g\n", rho);
}
if (relerr < *tol || *iters == *itmax)
break;
oldrho = rho;
}
}
// Return some stuff
*eigmin = rho;
fac = VPOW(2.0, *ilev - 1);
*eigmin_model = fac * (6.0 - 2.0 * VCOS(pi / (*nx - 1))
- 2.0 * VCOS(pi / (*ny - 1))
- 2.0 * VCOS(pi / (*nz - 1)));
}
VEXTERNC void Vmpower(int *nx, int *ny, int *nz,
double *u, int *iz,
double *w0, double *w1, double *w2, double *w3, double *w4,
double *eigmax, double *tol,
int *itmax, int *iters, int *nlev, int *ilev, int *nlev_real,
int *mgsolv, int *iok, int *iinfo,
double *epsiln, double *errtol, double *omega,
int *nu1, int *nu2, int *mgsmoo, int *ipc, double *rpc,
double *pc, double *ac, double *cc, double *fc, double *tru) {
// Local variables
int lev, level;
double denom, fac, rho, oldrho, error;
double relerr;
int itmax_s, iters_s, ierror_s, iok_s, iinfo_s, istop_s;
double alpha;
MAT2(iz, 50, 1);
// Recover level information
level = 1;
lev = (*ilev - 1) + level;
// Seed vector: random to contain all components
Vaxrand(nx, ny, nz, w1);
Vazeros(nx, ny, nz, w2);
Vazeros(nx, ny, nz, w3);
Vazeros(nx, ny, nz, w4);
Vazeros(nx, ny, nz, RAT(u, VAT2(iz, 1, lev)));
// NOTE: we destroy "fc" on this level due to lack of vectors... ***
Vazeros(nx,ny,nz,RAT(fc, VAT2(iz, 1, lev)));
// Normalize the seed vector
denom = Vxnrm2(nx, ny, nz, w1);
fac = 1.0 / denom;
Vxscal(nx, ny, nz, &fac, w1);
// Compute raleigh quotient with the seed vector
Vxcopy(nx, ny, nz, w1, RAT(u, VAT2(iz, 1, lev)));
itmax_s = 1;
iters_s = 0;
ierror_s = 0;
iok_s = 0;
iinfo_s = 0;
istop_s = 1;
Vmvcs(nx, ny, nz,
u, iz, w0, w2, w3, w4,
&istop_s, &itmax_s, &iters_s, &ierror_s,
nlev, ilev, nlev_real, mgsolv,
&iok_s, &iinfo_s,
epsiln, errtol, omega, nu1, nu2, mgsmoo,
ipc, rpc,
pc, ac, cc, fc, tru);
oldrho = Vxdot(nx, ny, nz, w1, RAT(u, VAT2(iz, 1, lev)));
// I/O
if (oldrho == 0.0) {
if (*iinfo > 3) {
Vnm_print(2, "Vmp0ower: iter=%d, estimate=%f", *iters, oldrho);
}
rho = oldrho;
} else {
// Main iteration
*iters = 0;
while (1) {
(*iters)++;
// Apply the matrix M
Vxcopy(nx, ny, nz, w1, RAT(u, VAT2(iz, 1, lev)));
itmax_s = 1;
iters_s = 0;
ierror_s = 0;
iok_s = 0;
iinfo_s = 0;
istop_s = 1;
Vmvcs(nx, ny, nz,
u, iz, w1, w2, w3, w4,
&istop_s, &itmax_s, &iters_s, &ierror_s,
nlev, ilev, nlev_real, mgsolv,
&iok_s, &iinfo_s,
epsiln, errtol, omega, nu1, nu2, mgsmoo,
ipc, rpc,
pc, ac, cc, fc, tru);
Vxcopy(nx, ny, nz, RAT(u, VAT2(iz, 1, lev)), w1);
// Normalize the new vector
denom = Vxnrm2(nx, ny, nz, w1);
fac = 1.0 / denom;
Vxscal(nx, ny, nz, &fac, w1);
// Compute the new raleigh quotient
Vxcopy(nx, ny, nz, w1, RAT(u, VAT2(iz, 1, lev)));
itmax_s = 1;
iters_s = 0;
ierror_s = 0;
iok_s = 0;
iinfo_s = 0;
istop_s = 1;
Vmvcs(nx, ny, nz,
u, iz, w0, w2, w3, w4,
&istop_s, &itmax_s, &iters_s, &ierror_s,
nlev, ilev, nlev_real, mgsolv,
&iok_s, &iinfo_s,
epsiln, errtol, omega, nu1, nu2, mgsmoo,
ipc, rpc,
pc, ac, cc, fc, tru);
Vxcopy(nx, ny, nz, RAT(u, VAT2(iz, 1, lev)), w2);
rho = Vxdot(nx, ny, nz, w1, w2);
// Stopping test
// w2=A*x, w1=x, stop = 2-norm(A*x-lamda*x)
alpha = -1.0;
Vxcopy(nx, ny, nz, w1, w3);
Vxcopy(nx, ny, nz, w2, w4);
Vxscal(nx, ny, nz, &rho, w3);
Vxaxpy(nx, ny, nz, &alpha, w3, w4);
error = Vxnrm2(nx, ny, nz, w4);
relerr = VABS( rho - oldrho ) / VABS( rho );
// I/O
if (*iinfo > 3) {
Vnm_print(2, "Vmpower: iter=%d; error=%f; relerr=%f; estimate=%f",
*iters, error, relerr, rho);
}
if ((relerr < *tol) || (*iters == *itmax)) {
break;
}
oldrho = rho;
}
}
*eigmax = rho;
}
void
tunAutoconfigDeviceRec (LocalDevicePtr local, TunDevicePtr tun, TunDeviceInfo info)
{
if (TUN_DEVICE_TEST_VAL_REL (info, REL_X) &&
TUN_DEVICE_TEST_VAL_REL (info, REL_Y) &&
TUN_DEVICE_TEST_KEY (info, BTN_LEFT))
{
OTLOG (local, "I found Rel(X,Y), Button(Left)");
OTLOG (local, "I think it is a mouse! :)");
VREL (REL_X, 0);
VREL (REL_Y, 1);
if (TUN_DEVICE_TEST_VAL_REL (info, REL_WHEEL))
{
OTLOG (local, "I found mouse wheel - mapping to BUTTON 4 & 5");
RVALUATOR (REL_WHEEL). mouse_wheel_hack = TRUE;
}
BUTTON_TEST_AND_ASSIGN (BTN_LEFT, 1);
BUTTON_TEST_AND_ASSIGN (BTN_RIGHT, 2);
BUTTON_TEST_AND_ASSIGN (BTN_MIDDLE, 3);
tun->is_absolute = FALSE;
return;
}
if (TUN_DEVICE_TEST_VAL_ABS (info, ABS_X) &&
TUN_DEVICE_TEST_VAL_ABS (info, ABS_Y) &&
TUN_DEVICE_TEST_KEY (info, BTN_TOUCH))
{
OTLOG (local, "I found Abs (X,Y), Button(Touch)");
OTLOG (local, "I think it is a Tablet! :)");
VABS (ABS_X,0);
VABS (ABS_Y,1);
TLOG ("Reverse Y coordinate (tablets have 0,0 in left lower corner)");
AVALUATOR (ABS_Y).upsidedown = TRUE;
if (TUN_DEVICE_TEST_VAL_ABS (info, ABS_PRESSURE))
VABS (ABS_PRESSURE, 2);
if (TUN_DEVICE_TEST_VAL_ABS (info, ABS_TILT_X))
VABS (ABS_TILT_X, 3);
if (TUN_DEVICE_TEST_VAL_ABS (info, ABS_TILT_Y))
VABS (ABS_TILT_Y, 4);
BUTTON (BTN_TOUCH, 1);
BUTTON_TEST_AND_ASSIGN (BTN_STYLUS, 2);
if (TUN_DEVICE_TEST_KEY (info, BTN_TOOL_PEN))
{
OTLOG (local, "Proximity event using ToolPen button");
tun->lbut_to_xbut_tbl [BTN_TOOL_PEN - tun->first_lbutton] = TUN_BUTTON_PROXIMITY;
}
tun->is_absolute = TRUE;
return;
}
if (TUN_DEVICE_TEST_VAL_ABS (info, ABS_X) &&
TUN_DEVICE_TEST_VAL_ABS (info, ABS_Y) &&
TUN_DEVICE_TEST_VAL_ABS (info, ABS_Z) &&
TUN_DEVICE_TEST_VAL_ABS (info, ABS_RX) &&
TUN_DEVICE_TEST_VAL_ABS (info, ABS_RY) &&
TUN_DEVICE_TEST_VAL_ABS (info, ABS_RZ) &&
AVALUATOR(ABS_X).min == - AVALUATOR(ABS_X).max &&
AVALUATOR(ABS_Y).min == - AVALUATOR(ABS_Y).max &&
AVALUATOR(ABS_Z).min == - AVALUATOR(ABS_Z).max &&
AVALUATOR(ABS_RX).min == - AVALUATOR(ABS_RX).max &&
AVALUATOR(ABS_RY).min == - AVALUATOR(ABS_RY).max &&
AVALUATOR(ABS_RZ).min == - AVALUATOR(ABS_RZ).max)
{
OTLOG (local, "found Abs (X,Y,Z,RX,RY,RZ)");
OTLOG (local, "I think it is some sort of 6DO device! :)");
VABSASREL (ABS_X, 0);
VABSASREL (ABS_Y, 1);
VABSASREL (ABS_Z, 2);
VABSASREL (ABS_RX, 3);
VABSASREL (ABS_RY, 4);
VABSASREL (ABS_RZ, 5);
tun->is_absolute = FALSE;
return;
}
}
void pbdirectpolforce_(double uind[maxatm][3], double uinp[maxatm][3],
double rff[maxatm][3], double rft[maxatm][3]) {
Vpmg *pmg[NOSH_MAXCALC];
Vpmgp *pmgp[NOSH_MAXCALC];
Vpbe *pbe[NOSH_MAXCALC];
MGparm *mgparm = VNULL;
PBEparm *pbeparm = VNULL;
Vatom *atom = VNULL;
double kT, force[3], torque[3];
double sign, zkappa2, epsp, epsw;
int i,j;
for (i=0; i<NOSH_MAXCALC; i++) {
pmg[i] = VNULL;
pmgp[i] = VNULL;
pbe[i] = VNULL;
}
// Read the converged induced dipole data into APBS Vatom structures.
for (i=0; i < alist[0]->number; i++){
atom = Valist_getAtom(alist[0],i);
Vatom_setInducedDipole(atom, uind[i]);
Vatom_setNLInducedDipole(atom, uinp[i]);
for (j=0;j<3;j++){
rff[i][j] = 0.0;
rft[i][j] = 0.0;
}
}
for (i=0; i<2; i++) {
VASSERT(permU[i] != VNULL);
VASSERT(indU[i] != VNULL);
VASSERT(nlIndU[i] != VNULL);
pmg[i] = VNULL;
pmgp[i] = VNULL;
pbe[i] = VNULL;
/* Useful local variables */
mgparm = nosh->calc[i]->mgparm;
pbeparm = nosh->calc[i]->pbeparm;
/* Set up problem */
if (!initMG(i, nosh, mgparm, pbeparm, realCenter, pbe,
alist, dielXMap, dielYMap, dielZMap,
kappaMap, chargeMap, pmgp, pmg, potMap)) {
Vnm_tprint( 2, "Error setting up MG calculation!\n");
return;
}
if (i == 0) {
sign = -1.0;
} else {
sign = 1.0;
}
// Q-Phi Force & Torque
if (!pmg[i]->pmgp->nonlin &&
(pmg[i]->surfMeth == VSM_SPLINE ||
pmg[i]->surfMeth == VSM_SPLINE3 ||
pmg[i]->surfMeth == VSM_SPLINE4)) {
for (j=0; j < alist[0]->number; j++){
Vpmg_qfDirectPolForce(pmg[i], permU[i], indU[i], j, force, torque);
rff[j][0] += sign * force[0];
rff[j][1] += sign * force[1];
rff[j][2] += sign * force[2];
rft[j][0] += sign * torque[0];
rft[j][1] += sign * torque[1];
rft[j][2] += sign * torque[2];
Vpmg_qfNLDirectPolForce(pmg[i], permU[i],
nlIndU[i], j,force,torque);
rff[j][0] += sign * force[0];
rff[j][1] += sign * force[1];
rff[j][2] += sign * force[2];
rft[j][0] += sign * torque[0];
rft[j][1] += sign * torque[1];
rft[j][2] += sign * torque[2];
}
// Dieletric Boundary Force
epsp = Vpbe_getSoluteDiel(pmg[i]->pbe);
epsw = Vpbe_getSolventDiel(pmg[i]->pbe);
if (VABS(epsp-epsw) > VPMGSMALL) {
for (j=0; j < alist[0]->number; j++){
Vpmg_dbDirectPolForce(pmg[i], permU[i], indU[i], j, force);
rff[j][0] += sign * force[0];
rff[j][1] += sign * force[1];
rff[j][2] += sign * force[2];
Vpmg_dbNLDirectPolForce(pmg[i], permU[i], nlIndU[i], j, force);
rff[j][0] += sign * force[0];
rff[j][1] += sign * force[1];
rff[j][2] += sign * force[2];
}
}
// Ionic Boundary Force
zkappa2 = Vpbe_getZkappa2(pmg[i]->pbe);
if (zkappa2 > VPMGSMALL) {
for (j=0; j < alist[0]->number; j++){
Vpmg_ibDirectPolForce(pmg[i], permU[i], indU[i], j, force);
rff[j][0] += sign * force[0];
rff[j][1] += sign * force[1];
rff[j][2] += sign * force[2];
Vpmg_ibNLDirectPolForce(pmg[i], permU[i],
nlIndU[i], j, force);
rff[j][0] += sign * force[0];
rff[j][1] += sign * force[1];
rff[j][2] += sign * force[2];
}
}
}
}
// kT in kcal/mol
kT = Vunit_kb * (1e-3) * Vunit_Na * 298.15 / 4.184;
for (i=0; i<alist[0]->number; i++){
rff[i][0] *= kT;
rff[i][1] *= kT;
rff[i][2] *= kT;
rft[i][0] *= kT;
rft[i][1] *= kT;
rft[i][2] *= kT;
}
killMG(nosh, pbe, pmgp, pmg);
}
void JitArm::fctiwx(UGeckoInstruction inst)
{
INSTRUCTION_START
JITDISABLE(bJITFloatingPointOff)
u32 b = inst.FB;
u32 d = inst.FD;
ARMReg vB = fpr.R0(b);
ARMReg vD = fpr.R0(d);
ARMReg V0 = fpr.GetReg();
ARMReg V1 = fpr.GetReg();
ARMReg V2 = fpr.GetReg();
ARMReg rA = gpr.GetReg();
ARMReg fpscrReg = gpr.GetReg();
FixupBranch DoneMax, DoneMin;
LDR(fpscrReg, R9, PPCSTATE_OFF(fpscr));
MOVI2R(rA, (u32)minmaxFloat);
// Check if greater than max float
{
VLDR(V0, rA, 8); // Load Max
VCMPE(vB, V0);
VMRS(_PC); // Loads in to APSR
FixupBranch noException = B_CC(CC_LE);
VMOV(vD, V0); // Set to max
SetFPException(fpscrReg, FPSCR_VXCVI);
DoneMax = B();
SetJumpTarget(noException);
}
// Check if less than min float
{
VLDR(V0, rA, 0);
VCMPE(vB, V0);
VMRS(_PC);
FixupBranch noException = B_CC(CC_GE);
VMOV(vD, V0);
SetFPException(fpscrReg, FPSCR_VXCVI);
DoneMin = B();
SetJumpTarget(noException);
}
// Within ranges, convert to integer
// Set rounding mode first
// PPC <-> ARM rounding modes
// 0, 1, 2, 3 <-> 0, 3, 1, 2
ARMReg rB = gpr.GetReg();
VMRS(rA);
// Bits 22-23
BIC(rA, rA, Operand2(3, 5));
LDR(rB, R9, PPCSTATE_OFF(fpscr));
AND(rB, rB, 0x3); // Get the FPSCR rounding bits
CMP(rB, 1);
SetCC(CC_EQ); // zero
ORR(rA, rA, Operand2(3, 5));
SetCC(CC_NEQ);
CMP(rB, 2); // +inf
SetCC(CC_EQ);
ORR(rA, rA, Operand2(1, 5));
SetCC(CC_NEQ);
CMP(rB, 3); // -inf
SetCC(CC_EQ);
ORR(rA, rA, Operand2(2, 5));
SetCC();
VMSR(rA);
ORR(rA, rA, Operand2(3, 5));
VCVT(vD, vB, TO_INT | IS_SIGNED);
VMSR(rA);
gpr.Unlock(rB);
VCMPE(vD, vB);
VMRS(_PC);
SetCC(CC_EQ);
BIC(fpscrReg, fpscrReg, FRFIMask);
FixupBranch DoneEqual = B();
SetCC();
SetFPException(fpscrReg, FPSCR_XX);
ORR(fpscrReg, fpscrReg, FIMask);
VABS(V1, vB);
VABS(V2, vD);
VCMPE(V2, V1);
VMRS(_PC);
SetCC(CC_GT);
ORR(fpscrReg, fpscrReg, FRMask);
SetCC();
SetJumpTarget(DoneEqual);
SetJumpTarget(DoneMax);
SetJumpTarget(DoneMin);
MOVI2R(rA, (u32)&doublenum);
VLDR(V0, rA, 0);
NEONXEmitter nemit(this);
nemit.VORR(vD, vD, V0);
if (inst.Rc) Helper_UpdateCR1(fpscrReg, rA);
STR(fpscrReg, R9, PPCSTATE_OFF(fpscr));
gpr.Unlock(rA);
gpr.Unlock(fpscrReg);
fpr.Unlock(V0);
fpr.Unlock(V1);
fpr.Unlock(V2);
}
int ArmJit::Replace_fabsf() {
fpr.MapDirtyIn(0, 12);
VABS(fpr.R(0), fpr.R(12));
return 4; // Number of instructions in the MIPS function
}
/* ///////////////////////////////////////////////////////////////////////////
// Routine: Vpee_markRefine
//
// Author: Nathan Baker (and Michael Holst: the author of AM_markRefine, on
// which this is based)
/////////////////////////////////////////////////////////////////////////// */
VPUBLIC int Vpee_markRefine(Vpee *thee,
AM *am,
int level,
int akey,
int rcol,
double etol,
int bkey
) {
Aprx *aprx;
int marked = 0,
markMe,
i,
smid,
count,
currentQ;
double minError = 0.0,
maxError = 0.0,
errEst = 0.0,
mlevel,
barrier;
SS *sm;
VASSERT(thee != VNULL);
/* Get the Aprx object from AM */
aprx = am->aprx;
/* input check and some i/o */
if ( ! ((-1 <= akey) && (akey <= 4)) ) {
Vnm_print(0,"Vpee_markRefine: bad refine key; simplices marked = %d\n",
marked);
return marked;
}
/* For uniform markings, we have no effect */
if ((-1 <= akey) && (akey <= 0)) {
marked = Gem_markRefine(thee->gm, akey, rcol);
return marked;
}
/* Informative I/O */
if (akey == 2) {
Vnm_print(0,"Vpee_estRefine: using Aprx_estNonlinResid().\n");
} else if (akey == 3) {
Vnm_print(0,"Vpee_estRefine: using Aprx_estLocalProblem().\n");
} else if (akey == 4) {
Vnm_print(0,"Vpee_estRefine: using Aprx_estDualProblem().\n");
} else {
Vnm_print(0,"Vpee_estRefine: bad key given; simplices marked = %d\n",
marked);
return marked;
}
if (thee->killFlag == 0) {
Vnm_print(0, "Vpee_markRefine: No error attenuation -- simplices in all partitions will be marked.\n");
} else if (thee->killFlag == 1) {
Vnm_print(0, "Vpee_markRefine: Maximum error attenuation -- only simplices in local partition will be marked.\n");
} else if (thee->killFlag == 2) {
Vnm_print(0, "Vpee_markRefine: Spherical error attenutation -- simplices within a sphere of %4.3f times the size of the partition will be marked\n",
thee->killParam);
} else if (thee->killFlag == 2) {
Vnm_print(0, "Vpee_markRefine: Neighbor-based error attenuation -- simplices in the local and neighboring partitions will be marked [NOT IMPLEMENTED]!\n");
VASSERT(0);
} else {
Vnm_print(2,"Vpee_markRefine: bogus killFlag given; simplices marked = %d\n",
marked);
return marked;
}
/* set the barrier type */
mlevel = (etol*etol) / Gem_numSS(thee->gm);
if (bkey == 0) {
barrier = (etol*etol);
Vnm_print(0,"Vpee_estRefine: forcing [err per S] < [TOL] = %g\n",
barrier);
} else if (bkey == 1) {
barrier = mlevel;
Vnm_print(0,"Vpee_estRefine: forcing [err per S] < [(TOL^2/numS)^{1/2}] = %g\n",
VSQRT(barrier));
} else {
Vnm_print(0,"Vpee_estRefine: bad bkey given; simplices marked = %d\n",
marked);
return marked;
}
/* timer */
Vnm_tstart(30, "error estimation");
/* count = num generations to produce from marked simplices (minimally) */
count = 1; /* must be >= 1 */
/* check the refinement Q for emptyness */
currentQ = 0;
if (Gem_numSQ(thee->gm,currentQ) > 0) {
Vnm_print(0,"Vpee_markRefine: non-empty refinement Q%d....clearing..",
currentQ);
Gem_resetSQ(thee->gm,currentQ);
Vnm_print(0,"..done.\n");
}
if (Gem_numSQ(thee->gm,!currentQ) > 0) {
Vnm_print(0,"Vpee_markRefine: non-empty refinement Q%d....clearing..",
!currentQ);
Gem_resetSQ(thee->gm,!currentQ);
Vnm_print(0,"..done.\n");
}
VASSERT( Gem_numSQ(thee->gm,currentQ) == 0 );
VASSERT( Gem_numSQ(thee->gm,!currentQ) == 0 );
/* clear everyone's refinement flags */
Vnm_print(0,"Vpee_markRefine: clearing all simplex refinement flags..");
for (i=0; i<Gem_numSS(thee->gm); i++) {
if ( (i>0) && (i % VPRTKEY) == 0 ) Vnm_print(0,"[MS:%d]",i);
sm = Gem_SS(thee->gm,i);
SS_setRefineKey(sm,currentQ,0);
SS_setRefineKey(sm,!currentQ,0);
SS_setRefinementCount(sm,0);
}
Vnm_print(0,"..done.\n");
/* NON-ERROR-BASED METHODS */
/* Simplex flag clearing */
if (akey == -1) return marked;
/* Uniform & user-defined refinement*/
if ((akey == 0) || (akey == 1)) {
smid = 0;
while ( smid < Gem_numSS(thee->gm)) {
/* Get the simplex and find out if it's markable */
sm = Gem_SS(thee->gm,smid);
markMe = Vpee_ourSimp(thee, sm, rcol);
if (markMe) {
if (akey == 0) {
marked++;
Gem_appendSQ(thee->gm,currentQ, sm);
SS_setRefineKey(sm,currentQ,1);
SS_setRefinementCount(sm,count);
} else if (Vpee_userDefined(thee, sm)) {
marked++;
Gem_appendSQ(thee->gm,currentQ, sm);
SS_setRefineKey(sm,currentQ,1);
SS_setRefinementCount(sm,count);
}
}
smid++;
}
}
/* ERROR-BASED METHODS */
/* gerror = global error accumulation */
aprx->gerror = 0.;
/* traverse the simplices and process the error estimates */
Vnm_print(0,"Vpee_markRefine: estimating error..");
smid = 0;
while ( smid < Gem_numSS(thee->gm)) {
/* Get the simplex and find out if it's markable */
sm = Gem_SS(thee->gm,smid);
markMe = Vpee_ourSimp(thee, sm, rcol);
if ( (smid>0) && (smid % VPRTKEY) == 0 ) Vnm_print(0,"[MS:%d]",smid);
/* Produce an error estimate for this element if it is in the set */
if (markMe) {
if (akey == 2) {
errEst = Aprx_estNonlinResid(aprx, sm, am->u,am->ud,am->f);
} else if (akey == 3) {
errEst = Aprx_estLocalProblem(aprx, sm, am->u,am->ud,am->f);
} else if (akey == 4) {
errEst = Aprx_estDualProblem(aprx, sm, am->u,am->ud,am->f);
}
VASSERT( errEst >= 0. );
/* if error estimate above tol, mark element for refinement */
if ( errEst > barrier ) {
marked++;
Gem_appendSQ(thee->gm,currentQ, sm); /*add to refinement Q*/
SS_setRefineKey(sm,currentQ,1); /* note now on refine Q */
SS_setRefinementCount(sm,count); /* refine X many times? */
}
/* keep track of min/max errors over the mesh */
minError = VMIN2( VSQRT(VABS(errEst)), minError );
maxError = VMAX2( VSQRT(VABS(errEst)), maxError );
/* store the estimate */
Bvec_set( aprx->wev, smid, errEst );
/* accumlate into global error (errEst is SQUAREd already) */
aprx->gerror += errEst;
/* otherwise store a zero for the estimate */
} else {
Bvec_set( aprx->wev, smid, 0. );
}
smid++;
}
/* do some i/o */
Vnm_print(0,"..done. [marked=<%d/%d>]\n",marked,Gem_numSS(thee->gm));
Vnm_print(0,"Vpee_estRefine: TOL=<%g> Global_Error=<%g>\n",
etol, aprx->gerror);
Vnm_print(0,"Vpee_estRefine: (TOL^2/numS)^{1/2}=<%g> Max_Ele_Error=<%g>\n",
VSQRT(mlevel),maxError);
Vnm_tstop(30, "error estimation");
/* check for making the error tolerance */
if ((bkey == 1) && (aprx->gerror <= etol)) {
Vnm_print(0,
"Vpee_estRefine: *********************************************\n");
Vnm_print(0,
"Vpee_estRefine: Global Error criterion met; setting marked=0.\n");
Vnm_print(0,
"Vpee_estRefine: *********************************************\n");
marked = 0;
}
/* return */
return marked;
}
void Jit::Comp_VV2Op(u32 op) {
CONDITIONAL_DISABLE;
DISABLE;
if (js.HasUnknownPrefix())
DISABLE;
VectorSize sz = GetVecSize(op);
int n = GetNumVectorElements(sz);
u8 sregs[4], dregs[4];
GetVectorRegsPrefixS(sregs, sz, _VS);
GetVectorRegsPrefixD(dregs, sz, _VD);
ARMReg tempxregs[4];
for (int i = 0; i < n; ++i)
{
if (!IsOverlapSafeAllowS(dregs[i], i, n, sregs))
{
int reg = fpr.GetTempV();
fpr.MapRegV(reg, MAP_NOINIT | MAP_DIRTY);
fpr.SpillLockV(reg);
tempxregs[i] = fpr.V(reg);
}
else
{
fpr.MapRegV(dregs[i], (dregs[i] == sregs[i] ? 0 : MAP_NOINIT) | MAP_DIRTY);
fpr.SpillLockV(dregs[i]);
tempxregs[i] = fpr.V(dregs[i]);
}
}
// Warning: sregs[i] and tempxregs[i] may be the same reg.
// Helps for vmov, hurts for vrcp, etc.
for (int i = 0; i < n; ++i)
{
switch ((op >> 16) & 0x1f)
{
case 0: // d[i] = s[i]; break; //vmov
// Probably for swizzle.
VMOV(tempxregs[i], fpr.V(sregs[i]));
break;
case 1: // d[i] = fabsf(s[i]); break; //vabs
//if (!fpr.V(sregs[i]).IsSimpleReg(tempxregs[i]))
VABS(tempxregs[i], fpr.V(sregs[i]));
break;
case 2: // d[i] = -s[i]; break; //vneg
VNEG(tempxregs[i], fpr.V(sregs[i]));
break;
case 4: // if (s[i] < 0) d[i] = 0; else {if(s[i] > 1.0f) d[i] = 1.0f; else d[i] = s[i];} break; // vsat0
DISABLE;
break;
case 5: // if (s[i] < -1.0f) d[i] = -1.0f; else {if(s[i] > 1.0f) d[i] = 1.0f; else d[i] = s[i];} break; // vsat1
DISABLE;
break;
case 16: // d[i] = 1.0f / s[i]; break; //vrcp
MOVI2F(S0, 1.0f, R0);
VDIV(tempxregs[i], S0, fpr.V(sregs[i]));
break;
case 17: // d[i] = 1.0f / sqrtf(s[i]); break; //vrsq
MOVI2F(S0, 1.0f, R0);
VSQRT(S1, fpr.V(sregs[i]));
VDIV(tempxregs[i], S0, S1);
break;
case 18: // d[i] = sinf((float)M_PI_2 * s[i]); break; //vsin
DISABLE;
break;
case 19: // d[i] = cosf((float)M_PI_2 * s[i]); break; //vcos
DISABLE;
break;
case 20: // d[i] = powf(2.0f, s[i]); break; //vexp2
DISABLE;
break;
case 21: // d[i] = logf(s[i])/log(2.0f); break; //vlog2
DISABLE;
break;
case 22: // d[i] = sqrtf(s[i]); break; //vsqrt
VSQRT(tempxregs[i], fpr.V(sregs[i]));
VABS(tempxregs[i], tempxregs[i]);
break;
case 23: // d[i] = asinf(s[i] * (float)M_2_PI); break; //vasin
DISABLE;
break;
case 24: // d[i] = -1.0f / s[i]; break; // vnrcp
MOVI2F(S0, -1.0f, R0);
VDIV(tempxregs[i], S0, fpr.V(sregs[i]));
break;
case 26: // d[i] = -sinf((float)M_PI_2 * s[i]); break; // vnsin
DISABLE;
break;
case 28: // d[i] = 1.0f / expf(s[i] * (float)M_LOG2E); break; // vrexp2
DISABLE;
break;
}
}
fpr.MapRegsV(dregs, sz, MAP_NOINIT | MAP_DIRTY);
for (int i = 0; i < n; ++i)
{
VMOV(fpr.V(dregs[i]), tempxregs[i]);
}
ApplyPrefixD(dregs, sz);
fpr.ReleaseSpillLocks();
}
/*
* ***************************************************************************
* Routine: Aprx_partInert
*
* Purpose: Partition the domain using inertial bisection.
* Partition sets of points in R^d (d=2 or d=3) by viewing them
* as point masses of a rigid body, and by then employing the
* classical mechanics ideas of inertia and Euler axes.
*
* Notes: We first locate the center of mass, then change the coordinate
* system so that the center of mass is located at the origin.
* We then form the (symmetric) dxd inertia tensor, and then find
* the set of (real) eigenvalues and (orthogonal) eigenvectors.
* The eigenvectors represent the principle inertial rotation axes,
* and the eigenvalues represent the inertial strength in those
* principle directions. The smallest inerial component along an
* axis represents a direction along which the rigid body is most
* "line-like" (assuming all the points have the same mass).
*
* For our purposes, it makes sense to using the axis (eigenvector)
* corresponding to the smallest inertia (eigenvalue) as the line to
* bisect with a line (d=2) or a plane (d=3). We know the center of
* mass, and once we also have this particular eigenvector, we can
* effectively bisect the point set into the two regions separated
* by the line/plane simply by taking an inner-product of the
* eigenvector with each point (or rather the 2- or 3-vector
* representing the point). A positive inner-product represents one
* side of the cutting line/plane, and a negative inner-product
* represents the other side (a zero inner-product is right on the
* cutting line/plane, so we arbitrarily assign it to one region or
* the other).
*
* Author: Michael Holst
* ***************************************************************************
*/
VPUBLIC int Aprx_partInert(Aprx *thee, int pcolor,
int numC, double *evec, simHelper *simH)
{
int i, j, k, lambdaI;
double rad, sca, lambda, normal, caxis[3];
Mat3 I, II, V, D;
Vnm_print(0,"Aprx_partInert: WARNING: assuming single-chart manifold.\n");
Vnm_print(0,"Aprx_partInert: [pc=%d] partitioning:\n", pcolor);
/* form the inertia tensors */
Mat3_eye(I);
Mat3_init(II, 0.);
for (i=0; i<numC; i++) {
/* get vector length (squared!) */
rad = 0.;
for (j=0; j<3; j++) {
rad += ( simH[i].bc[j] * simH[i].bc[j] );
}
/* add contribution to the inertia tensor */
for (j=0; j<3; j++) {
for (k=0; k<3; k++) {
II[j][k] += ( simH[i].mass *
(I[j][k]*rad - simH[i].bc[j]*simH[i].bc[k]) );
}
}
}
/* find the d-principle axes, and isolate the single axis we need */
/* (the principle axis we want is the one with SMALLEST moment) */
sca = Mat3_nrm8(II);
Mat3_scal(II, 1./sca);
(void)Mat3_qri(V, D, II);
lambda = VLARGE;
lambdaI = -1;
for (i=0; i<3; i++) {
if ( VABS(D[i][i]) < lambda ) {
lambda = VABS(D[i][i]);
lambdaI = i;
}
}
VASSERT( lambda > 0. );
VASSERT( lambda != VLARGE );
VASSERT( lambdaI >= 0 );
for (i=0; i<3; i++) {
caxis[i] = V[i][lambdaI];
}
normal = Vec3_nrm2(caxis);
VASSERT( normal > 0. );
Vec3_scal(caxis,1./normal);
/* decompose points based on bisecting principle axis with a line or */
/* plane; we do this using an inner-product test with normal vec "caxis" */
normal = 0;
for (i=0; i<numC; i++) {
evec[i] = Vec3_dot( simH[i].bc, caxis );
normal += (evec[i]*evec[i]);
}
normal = VSQRT( normal );
/* normalize the final result */
for (i=0; i<numC; i++) {
evec[i] = evec[i] / normal;
}
return 0;
}
void JitArm::fctiwzx(UGeckoInstruction inst)
{
INSTRUCTION_START
JITDISABLE(bJITFloatingPointOff)
u32 b = inst.FB;
u32 d = inst.FD;
ARMReg vB = fpr.R0(b);
ARMReg vD = fpr.R0(d);
ARMReg V0 = fpr.GetReg();
ARMReg V1 = fpr.GetReg();
ARMReg V2 = fpr.GetReg();
ARMReg rA = gpr.GetReg();
ARMReg fpscrReg = gpr.GetReg();
FixupBranch DoneMax, DoneMin;
LDR(fpscrReg, R9, PPCSTATE_OFF(fpscr));
MOVI2R(rA, (u32)minmaxFloat);
// Check if greater than max float
{
VLDR(V0, rA, 8); // Load Max
VCMPE(vB, V0);
VMRS(_PC); // Loads in to APSR
FixupBranch noException = B_CC(CC_LE);
VMOV(vD, V0); // Set to max
SetFPException(fpscrReg, FPSCR_VXCVI);
DoneMax = B();
SetJumpTarget(noException);
}
// Check if less than min float
{
VLDR(V0, rA, 0);
VCMPE(vB, V0);
VMRS(_PC);
FixupBranch noException = B_CC(CC_GE);
VMOV(vD, V0);
SetFPException(fpscrReg, FPSCR_VXCVI);
DoneMin = B();
SetJumpTarget(noException);
}
// Within ranges, convert to integer
VCVT(vD, vB, TO_INT | IS_SIGNED | ROUND_TO_ZERO);
VCMPE(vD, vB);
VMRS(_PC);
SetCC(CC_EQ);
BIC(fpscrReg, fpscrReg, FRFIMask);
FixupBranch DoneEqual = B();
SetCC();
SetFPException(fpscrReg, FPSCR_XX);
ORR(fpscrReg, fpscrReg, FIMask);
VABS(V1, vB);
VABS(V2, vD);
VCMPE(V2, V1);
VMRS(_PC);
SetCC(CC_GT);
ORR(fpscrReg, fpscrReg, FRMask);
SetCC();
SetJumpTarget(DoneEqual);
SetJumpTarget(DoneMax);
SetJumpTarget(DoneMin);
MOVI2R(rA, (u32)&doublenum);
VLDR(V0, rA, 0);
NEONXEmitter nemit(this);
nemit.VORR(vD, vD, V0);
if (inst.Rc) Helper_UpdateCR1(fpscrReg, rA);
STR(fpscrReg, R9, PPCSTATE_OFF(fpscr));
gpr.Unlock(rA);
gpr.Unlock(fpscrReg);
fpr.Unlock(V0);
fpr.Unlock(V1);
fpr.Unlock(V2);
}
void apbsempole_(int *natom, double x[maxatm][3],
double rad[maxatm], double rpole[maxatm][13],
double *total,
double energy[maxatm], double fld[maxatm][3],
double rff[maxatm][3], double rft[maxatm][3]) {
/* Misc. pointers to APBS data structures */
Vpmg *pmg[NOSH_MAXCALC];
Vpmgp *pmgp[NOSH_MAXCALC];
Vpbe *pbe[NOSH_MAXCALC];
MGparm *mgparm = VNULL;
PBEparm *pbeparm = VNULL;
Vatom *atom = VNULL;
/* Vgrid configuration for the kappa and dielectric maps */
double nx,ny,nz,hx,hy,hzed,xmin,ymin,zmin;
double *data;
double zkappa2, epsp, epsw;
/* Loop indeces */
int i,j;
/* Observables and unit conversion */
double sign, force[3], torque[3], field[3];
double kT,electric,debye;
double charge, dipole[3], quad[9];
debye = 4.8033324;
for (i=0; i<NOSH_MAXCALC; i++) {
pmg[i] = VNULL;
pmgp[i] = VNULL;
pbe[i] = VNULL;
}
/* Kill the saved potential Vgrids */
for (i=0; i<2; i++){
if (permU[i] != VNULL) Vgrid_dtor(&permU[i]);
if (indU[i] != VNULL) Vgrid_dtor(&indU[i]);
if (nlIndU[i] != VNULL) Vgrid_dtor(&nlIndU[i]);
}
/* Kill the old atom list */
if (alist[0] != VNULL) {
Valist_dtor(&alist[0]);
}
/* Create a new atom list (mol == 1) */
if (alist[0] == VNULL) {
alist[0] = Valist_ctor();
alist[0]->atoms = Vmem_malloc(alist[0]->vmem, *natom, (sizeof(Vatom)));
alist[0]->number = *natom;
}
/* Read TINKER input data into Vatom instances. */
for (i=0; i < alist[0]->number; i++){
atom = Valist_getAtom(alist[0],i);
Vatom_setAtomID(atom, i);
Vatom_setPosition(atom, x[i]);
Vatom_setRadius(atom, rad[i]);
charge = rpole[i][0];
Vatom_setCharge(atom, charge);
dipole[0] = rpole[i][1];
dipole[1] = rpole[i][2];
dipole[2] = rpole[i][3];
Vatom_setDipole(atom, dipole);
quad[0] = rpole[i][4];
quad[1] = rpole[i][5];
quad[2] = rpole[i][6];
quad[3] = rpole[i][7];
quad[4] = rpole[i][8];
quad[5] = rpole[i][9];
quad[6] = rpole[i][10];
quad[7] = rpole[i][11];
quad[8] = rpole[i][12];
Vatom_setQuadrupole(atom, quad);
/* Useful check
printf(" %i %f (%f,%f,%f)\n",i,rad[i], x[i][0], x[i][1], x[i][2]);
printf(" %f\n %f,%f,%f\n", charge, dipole[0], dipole[1], dipole[2]);
printf(" %f\n", quad[0]);
printf(" %f %f\n", quad[3], quad[4]);
printf(" %f %f %f\n", quad[6], quad[7], quad[8]); */
energy[i] = 0.0;
for (j=0;j<3;j++){
fld[i][j] = 0.0;
rff[i][j] = 0.0;
rft[i][j] = 0.0;
}
}
nosh->nmol = 1;
Valist_getStatistics(alist[0]);
/* Only call the setupCalc routine once, so that we can
reuse this nosh object */
if (nosh->ncalc < 2) {
if (NOsh_setupElecCalc(nosh, alist) != 1) {
printf("Error setting up calculations\n");
exit(-1);
}
}
/* Solve the LPBE for the homogeneous and then solvated states */
for (i=0; i<2; i++) {
/* Useful local variables */
mgparm = nosh->calc[i]->mgparm;
pbeparm = nosh->calc[i]->pbeparm;
/* Just to be robust */
if (!MGparm_check(mgparm)){
printf("MGparm Check failed\n");
printMGPARM(mgparm, realCenter);
exit(-1);
}
if (!PBEparm_check(pbeparm)){
printf("PBEparm Check failed\n");
printPBEPARM(pbeparm);
exit(-1);
}
/* Set up the problem */
mgparm->chgs = VCM_PERMANENT;
if (!initMG(i, nosh, mgparm, pbeparm, realCenter, pbe,
alist, dielXMap, dielYMap, dielZMap,
kappaMap, chargeMap, pmgp, pmg, potMap)) {
Vnm_tprint( 2, "Error setting up MG calculation!\n");
return;
}
/* Solve the PDE */
if (solveMG(nosh, pmg[i], mgparm->type) != 1) {
Vnm_tprint(2, "Error solving PDE!\n");
return;
}
/* Set partition information for observables and I/O */
/* Note - parallel operation has NOT been tested. */
if (setPartMG(nosh, mgparm, pmg[i]) != 1) {
Vnm_tprint(2, "Error setting partition info!\n");
return;
}
nx = pmg[i]->pmgp->nx;
ny = pmg[i]->pmgp->ny;
nz = pmg[i]->pmgp->nz;
hx = pmg[i]->pmgp->hx;
hy = pmg[i]->pmgp->hy;
hzed = pmg[i]->pmgp->hzed;
xmin = pmg[i]->pmgp->xmin;
ymin = pmg[i]->pmgp->ymin;
zmin = pmg[i]->pmgp->zmin;
/* Save dielectric/kappa maps into Vgrids, then change the nosh
* data structure to think it read these maps in from a file.
* The goal is to save setup time during convergence of the
* induced dipoles. This is under consideration...
* */
/*
// X (shifted)
data = Vmem_malloc(mem, nx*ny*nz, sizeof(double));
Vpmg_fillArray(pmg[i], data, VDT_DIELX, 0.0, pbeparm->pbetype);
dielXMap[i] = Vgrid_ctor(nx,ny,nz,hx,hy,hzed,
xmin + 0.5*hx,ymin,zmin,data);
dielXMap[i]->readdata = 1;
// Y (shifted)
data = Vmem_malloc(mem, nx*ny*nz, sizeof(double));
Vpmg_fillArray(pmg[i], data, VDT_DIELY, 0.0, pbeparm->pbetype);
dielYMap[i] = Vgrid_ctor(nx,ny,nz,hx,hy,hzed,
xmin,ymin + 0.5*hy,zmin,data);
dielYMap[i]->readdata = 1;
// Z (shifted)
data = Vmem_malloc(mem, nx*ny*nz, sizeof(double));
Vpmg_fillArray(pmg[i], data, VDT_DIELZ, 0.0, pbeparm->pbetype);
dielZMap[i] = Vgrid_ctor(nx,ny,nz,hx,hy,hzed,
xmin,ymin,zmin + 0.5*hzed,data);
dielZMap[i]->readdata = 1;
// Kappa
data = Vmem_malloc(mem, nx*ny*nz, sizeof(double));
Vpmg_fillArray(pmg[i], data, VDT_KAPPA, 0.0, pbeparm->pbetype);
kappaMap[i] = Vgrid_ctor(nx,ny,nz,hx,hy,hzed,xmin,ymin,zmin,data);
kappaMap[i]->readdata = 1;
// Update the pbeparam structure, since we now have
// dielectric and kappap maps
pbeparm->useDielMap = 1;
pbeparm->dielMapID = i + 1;
pbeparm->useKappaMap = 1;
pbeparm->kappaMapID = i + 1;
*/
data = Vmem_malloc(mem, nx*ny*nz, sizeof(double));
Vpmg_fillArray(pmg[i], data, VDT_POT, 0.0, pbeparm->pbetype, pbeparm);
permU[i] = Vgrid_ctor(nx,ny,nz,hx,hy,hzed,xmin,ymin,zmin,data);
permU[i]->readdata = 1;
// set readdata flag to have the dtor to free data
if (i == 0){
sign = -1.0;
} else {
sign = 1.0;
}
/* Calculate observables */
for (j=0; j < alist[0]->number; j++){
energy[j] += sign * Vpmg_qfPermanentMultipoleEnergy(pmg[i], j);
Vpmg_fieldSpline4(pmg[i], j, field);
fld[j][0] += sign * field[0];
fld[j][1] += sign * field[1];
fld[j][2] += sign * field[2];
}
if (!pmg[i]->pmgp->nonlin &&
(pmg[i]->surfMeth == VSM_SPLINE ||
pmg[i]->surfMeth == VSM_SPLINE3 ||
pmg[i]->surfMeth == VSM_SPLINE4)) {
for (j=0; j < alist[0]->number; j++){
Vpmg_qfPermanentMultipoleForce(pmg[i], j, force, torque);
rff[j][0] += sign * force[0];
rff[j][1] += sign * force[1];
rff[j][2] += sign * force[2];
rft[j][0] += sign * torque[0];
rft[j][1] += sign * torque[1];
rft[j][2] += sign * torque[2];
}
kT = Vunit_kb * (1e-3) * Vunit_Na * 298.15 * 1.0/4.184;
epsp = Vpbe_getSoluteDiel(pmg[i]->pbe);
epsw = Vpbe_getSolventDiel(pmg[i]->pbe);
if (VABS(epsp-epsw) > VPMGSMALL) {
for (j=0; j < alist[0]->number; j++){
Vpmg_dbPermanentMultipoleForce(pmg[i], j, force);
rff[j][0] += sign * force[0];
rff[j][1] += sign * force[1];
rff[j][2] += sign * force[2];
}
}
zkappa2 = Vpbe_getZkappa2(pmg[i]->pbe);
if (zkappa2 > VPMGSMALL) {
for (j=0; j < alist[0]->number; j++) {
Vpmg_ibPermanentMultipoleForce(pmg[i], j, force);
rff[j][0] += sign * force[0];
rff[j][1] += sign * force[1];
rff[j][2] += sign * force[2];
}
}
}
}
//nosh->ndiel = 2;
//nosh->nkappa = 2;
/*
printf("Energy (multipole) %f Kcal/mol\n", *energy);
printf("Energy (volume) %f Kcal/mol\n", evol * 0.5 * kT);
*/
// Convert results into kcal/mol units
kT = Vunit_kb * (1e-3) * Vunit_Na * 298.15 * 1.0/4.184;
// Electric converts from electron**2/Angstrom to kcal/mol
electric = 332.063709;
*total = 0.0;
for (i=0; i<alist[0]->number; i++){
/* starting with the field in KT/e/Ang^2 multiply by kcal/mol/KT
the field is then divided by "electric" to convert to e/Ang^2 */
energy[i] *= 0.5 * kT;
*total += energy[i];
fld[i][0] *= kT / electric;
fld[i][1] *= kT / electric;
fld[i][2] *= kT / electric;
rff[i][0] *= kT;
rff[i][1] *= kT;
rff[i][2] *= kT;
rft[i][0] *= kT;
rft[i][1] *= kT;
rft[i][2] *= kT;
}
killMG(nosh, pbe, pmgp, pmg);
}
void Jit::Comp_FPU2op(u32 op)
{
CONDITIONAL_DISABLE;
int fs = _FS;
int fd = _FD;
// logBlocks = 1;
switch (op & 0x3f)
{
case 4: //F(fd) = sqrtf(F(fs)); break; //sqrt
fpr.MapDirtyIn(fd, fs);
VSQRT(fpr.R(fd), fpr.R(fs));
break;
case 5: //F(fd) = fabsf(F(fs)); break; //abs
fpr.MapDirtyIn(fd, fs);
VABS(fpr.R(fd), fpr.R(fs));
break;
case 6: //F(fd) = F(fs); break; //mov
fpr.MapDirtyIn(fd, fs);
VMOV(fpr.R(fd), fpr.R(fs));
break;
case 7: //F(fd) = -F(fs); break; //neg
fpr.MapDirtyIn(fd, fs);
VNEG(fpr.R(fd), fpr.R(fs));
break;
case 12: //FsI(fd) = (int)floorf(F(fs)+0.5f); break; //round.w.s
fpr.MapDirtyIn(fd, fs);
VCVT(fpr.R(fd), fpr.R(fs), TO_INT | IS_SIGNED);
break;
case 13: //FsI(fd) = Rto0(F(fs))); break; //trunc.w.s
fpr.MapDirtyIn(fd, fs);
VCVT(fpr.R(fd), fpr.R(fs), TO_INT | IS_SIGNED | ROUND_TO_ZERO);
break;
case 14: //FsI(fd) = (int)ceilf (F(fs)); break; //ceil.w.s
fpr.MapDirtyIn(fd, fs);
MOVI2F(S0, 0.5f, R0);
VADD(S0,fpr.R(fs),S0);
VCVT(fpr.R(fd), S0, TO_INT | IS_SIGNED);
break;
case 15: //FsI(fd) = (int)floorf(F(fs)); break; //floor.w.s
fpr.MapDirtyIn(fd, fs);
MOVI2F(S0, 0.5f, R0);
VSUB(S0,fpr.R(fs),S0);
VCVT(fpr.R(fd), S0, TO_INT | IS_SIGNED);
break;
case 32: //F(fd) = (float)FsI(fs); break; //cvt.s.w
fpr.MapDirtyIn(fd, fs);
VCVT(fpr.R(fd), fpr.R(fs), TO_FLOAT | IS_SIGNED);
break;
case 36: //FsI(fd) = (int) F(fs); break; //cvt.w.s
fpr.MapDirtyIn(fd, fs);
LDR(R0, CTXREG, offsetof(MIPSState, fcr31));
AND(R0, R0, Operand2(3));
// MIPS Rounding Mode:
// 0: Round nearest
// 1: Round to zero
// 2: Round up (ceil)
// 3: Round down (floor)
CMP(R0, Operand2(2));
SetCC(CC_GE); MOVI2F(S0, 0.5f, R1);
SetCC(CC_GT); VSUB(S0,fpr.R(fs),S0);
SetCC(CC_EQ); VADD(S0,fpr.R(fs),S0);
SetCC(CC_GE); VCVT(fpr.R(fd), S0, TO_INT | IS_SIGNED); /* 2,3 */
SetCC(CC_AL);
CMP(R0, Operand2(1));
SetCC(CC_EQ); VCVT(fpr.R(fd), fpr.R(fs), TO_INT | IS_SIGNED | ROUND_TO_ZERO); /* 1 */
SetCC(CC_LT); VCVT(fpr.R(fd), fpr.R(fs), TO_INT | IS_SIGNED); /* 0 */
SetCC(CC_AL);
break;
default:
DISABLE;
}
}
/*
* ***************************************************************************
* Routine: Gem_makeBndExt
*
* Purpose: Mark selected boundary faces in a special way.
*
* Author: Michael Holst
* ***************************************************************************
*/
VPUBLIC void Gem_makeBndExt(Gem *thee, int key)
{
int i, j, k, l, m, p, q, nabors, btype, done, btypeGeneric;
VV *v[4];
SS *sm, *sm0, *sm1, *sm2;
double x[4][3], xchk;
/* go through all simplices and zero all boundary faces */
Vnm_print(0,"Gem_makeBnd: zeroing boundary faces/vertices..");
Gem_setNumBF(thee, 0);
Gem_setNumBV(thee, 0);
for (i=0; i<Gem_numSS(thee); i++) {
sm = Gem_SS(thee,i);
if ( (i>0) && (i % VPRTKEY) == 0 ) Vnm_print(0,"[BS:%d]",i);
/* get local vertices */
for (j=0; j<Gem_dimVV(thee); j++)
v[j] = SS_vertex(sm,j);
/* reset all vertices and faces to interior type */
for (j=0; j<Gem_dimVV(thee); j++) {
/* the other three local vertex/face numbers besides "j" */
k=(j+1) % Gem_dimVV(thee);
l=(k+1) % Gem_dimVV(thee);
m=(l+1) % Gem_dimVV(thee);
SS_setFaceType(sm, j, 0);
VV_setType(v[k], 0);
VV_setType(v[l], 0);
if (Gem_dim(thee) == 3) VV_setType(v[m], 0);
}
}
Vnm_print(0,"..done.\n");
/* okay now make a boundary */
Vnm_print(0,"Gem_makeBnd: rebuilding boundary faces/vertices..");
for (i=0; i<Gem_numSS(thee); i++) {
sm = Gem_SS(thee,i);
if ( (i>0) && (i % VPRTKEY) == 0 ) Vnm_print(0,"[BS:%d]",i);
/* get local vertices */
for (j=0; j<Gem_dimVV(thee); j++)
v[j] = SS_vertex(sm,j);
/* rebuild everything */
for (j=0; j<Gem_dimVV(thee); j++) {
/* the other three local vertex/face numbers besides "j" */
k=(j+1) % Gem_dimVV(thee);
l=(k+1) % Gem_dimVV(thee);
m=(l+1) % Gem_dimVV(thee);
/* look for a face nabor sharing face "j" (opposite vertex "j") */
nabors = 0;
for (sm0=VV_firstSS(v[k]); sm0!=VNULL;sm0=SS_link(sm0,v[k])) {
for (sm1=VV_firstSS(v[l]); sm1!=VNULL; sm1=SS_link(sm1,v[l])) {
if (Gem_dim(thee) == 2) {
if ((sm0!=sm) && (sm0==sm1)) nabors++;
} else {
for (sm2=VV_firstSS(v[m]); sm2!=VNULL;
sm2=SS_link(sm2,v[m])) {
if ((sm0!=sm) && (sm0==sm1) && (sm0==sm2)) {
nabors++;
}
}
}
}
}
/* if no one there, then face "j" is actually a boundary face */
if (nabors == 0) {
/* grab coordinates of the vertices of this face */
for (q=0; q<Gem_dim(thee); q++) {
x[0][q] = VV_coord(v[k],q);
}
for (q=0; q<Gem_dim(thee); q++) {
x[1][q] = VV_coord(v[l],q);
}
if (Gem_dim(thee) == 3) {
for (q=0; q<Gem_dim(thee); q++) {
x[2][q] = VV_coord(v[m],q);
}
}
/* default is interior; should not occur! */
btypeGeneric = 18;
done = 0;
btype = btypeGeneric;
/* ---------- check for base marking ---------- */
xchk = 0.0;
for (p=0; p<Gem_dim(thee); p++) {
xchk += VABS( x[p][1] - (-1.0) );
}
if (xchk < VSMALL) {
done = 1;
btype = 1;
}
/* ---------- check for base marking again ---------- */
xchk = 0.0;
for (p=0; p<Gem_dim(thee); p++) {
xchk += VABS( x[p][1] - ( 0.0) );
}
if (xchk < VSMALL) {
done = 1;
btype = 18;
}
/* ---------- check for first section ---------- */
if (!done) {
done = 1;
btype = 2;
for (p=0; p<Gem_dim(thee); p++) {
if (! ( ( 1.9 <= x[p][0])
&& ( 6.1 >= x[p][0])
&& (-VSMALL <= x[p][1])
&& (-1.1 <= x[p][2])
&& ( 1.1 >= x[p][2]) )) {
done = 0;
btype = btypeGeneric;
}
}
if (done) {
xchk = 0.0;
for (p=0; p<Gem_dim(thee); p++) {
xchk += VABS( x[p][1] - 10.0 );
}
if (xchk < VSMALL) {
btype = 10;
}
}
}
/* ---------- check for second section ---------- */
if (!done) {
done = 1;
btype = 4;
for (p=0; p<Gem_dim(thee); p++) {
if (! ( ( 7.9 <= x[p][0])
&& (12.1 >= x[p][0])
&& (-VSMALL <= x[p][1])
&& (-1.1 <= x[p][2])
&& ( 1.1 >= x[p][2]) )) {
done = 0;
btype = btypeGeneric;
}
}
if (done) {
xchk = 0.0;
for (p=0; p<Gem_dim(thee); p++) {
xchk += VABS( x[p][1] - 10.0 );
}
if (xchk < VSMALL) {
btype = 12;
}
}
}
/* ---------- check for third section ---------- */
if (!done) {
done = 1;
btype = 6;
for (p=0; p<Gem_dim(thee); p++) {
if (! ( (13.9 <= x[p][0])
&& (18.1 >= x[p][0])
&& (-VSMALL <= x[p][1])
&& (-1.1 <= x[p][2])
&& ( 1.1 >= x[p][2]) )) {
done = 0;
btype = btypeGeneric;
}
}
if (done) {
xchk = 0.0;
for (p=0; p<Gem_dim(thee); p++) {
xchk += VABS( x[p][1] - 10.0 );
}
if (xchk < VSMALL) {
btype = 14;
}
}
}
/* ---------- check for fourth section ---------- */
if (!done) {
done = 1;
btype = 8;
for (p=0; p<Gem_dim(thee); p++) {
if (! ( (19.9 <= x[p][0])
&& (24.1 >= x[p][0])
&& (-VSMALL <= x[p][1])
&& (-1.1 <= x[p][2])
&& ( 1.1 >= x[p][2]) )) {
done = 0;
btype = btypeGeneric;
}
}
if (done) {
xchk = 0.0;
for (p=0; p<Gem_dim(thee); p++) {
xchk += VABS( x[p][1] - 10.0 );
}
if (xchk < VSMALL) {
btype = 16;
}
}
}
/* should have been marked with SOME boundary type */
VASSERT( 0 != btype );
/* set the facetype */
SS_setFaceType(sm, j, btype);
Gem_numBFpp(thee);
/* set the vertex types (dirichlet overrides robin) */
if (!VDIRICHLET( VV_type(v[k])) ) {
if (VINTERIOR( VV_type(v[k])) ) {
Gem_numBVpp(thee);
}
VV_setType(v[k], btype);
}
if (!VDIRICHLET( VV_type(v[l])) ) {
if (VINTERIOR( VV_type(v[l])) ) {
Gem_numBVpp(thee);
}
VV_setType(v[l], btype);
}
if (Gem_dim(thee) == 3) {
if (!VDIRICHLET( VV_type(v[m])) ) {
if (VINTERIOR( VV_type(v[m])) ) {
Gem_numBVpp(thee);
}
VV_setType(v[m], btype);
}
}
}
}
}
Vnm_print(0,"..done.\n");
}
