Skip to content

Rewrite ROTMG to address cases not covered by the netlib algorithm #1480

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 6 commits into from
Mar 4, 2018
Merged

Rewrite ROTMG to address cases not covered by the netlib algorithm #1480

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
81f81cb
Rewrite ROTMG based on the new implementation in GONUM
martin-frbg Mar 3, 2018
3fb3a74
Correct ROTMG utest for issue1452 and add another from gonum
martin-frbg Mar 3, 2018
471832e
Update rotmg tests in utest_main2.c
martin-frbg Mar 3, 2018
f8b52aa
Fix spelling in comment
martin-frbg Mar 4, 2018
25f5ca7
Fix spelling errors in comments
martin-frbg Mar 4, 2018
f29c74f
More spelling fixes in comments
martin-frbg Mar 4, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
126 changes: 58 additions & 68 deletions interface/rotmg.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,6 +64,13 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){

FLOAT du, dp1, dp2, dq2, dq1, dh11=ZERO, dh21=ZERO, dh12=ZERO, dh22=ZERO, dflag=-ONE, dtemp;

if (*dd2 == ZERO || dy1 == ZERO)
{
dflag = -TWO;
dparam[0] = dflag;
return;
}

if(*dd1 < ZERO)
{
dflag = -ONE;
Expand All @@ -76,6 +83,16 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){
*dd2 = ZERO;
*dx1 = ZERO;
}
else if ((*dd1 == ZERO || *dx1 == ZERO) && *dd2 > ZERO)
{
dflag = ONE;
dh12 = 1;
dh21 = -1;
*dx1 = dy1;
dtemp = *dd1;
*dd1 = *dd2;
*dd2 = dtemp;
}
else
{
dp2 = *dd2 * dy1;
Expand All @@ -90,6 +107,9 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){
dq1 = dp1 * *dx1;
if(ABS(dq1) > ABS(dq2))
{
dflag = ZERO;
dh11 = ONE;
dh22 = ONE;
dh21 = - dy1 / *dx1;
dh12 = dp2 / dp1;

Expand All @@ -100,8 +120,19 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){
*dd1 = *dd1 / du;
*dd2 = *dd2 / du;
*dx1 = *dx1 * du;
} else {
dflag = -ONE;

dh11 = ZERO;
dh12 = ZERO;
dh21 = ZERO;
dh22 = ZERO;

*dd1 = ZERO;
*dd2 = ZERO;
*dx1 = ZERO;
}

}
else
{
Expand All @@ -120,7 +151,9 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){
}
else
{
dflag = ONE;
dflag = ONE;
dh21 = -ONE;
dh12 = ONE;

dh11 = dp1 / dp2;
dh22 = *dx1 / dy1;
Expand All @@ -134,76 +167,33 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){
}


if(*dd1 != ZERO)
while ( *dd1 <= RGAMSQ && *dd1 != ZERO)
{
if( (*dd1 <= RGAMSQ) || (*dd1 >= GAMSQ) )
{
if(dflag == ZERO)
{
dh11 = ONE;
dh22 = ONE;
dflag = -ONE;
}
else
{
dh21 = -ONE;
dh12 = ONE;
dflag = -ONE;
}
if( *dd1 <= RGAMSQ )
{
while (ABS(*dd1) <= RGAMSQ) {
*dd1 = *dd1 * (GAM * GAM);
*dx1 = *dx1 / GAM;
dh11 = dh11 / GAM;
dh12 = dh12 / GAM;
}
}
else
{
while (ABS(*dd1) >= GAMSQ) {
*dd1 = *dd1 / (GAM * GAM);
*dx1 = *dx1 * GAM;
dh11 = dh11 * GAM;
dh12 = dh12 * GAM;
}
}
}
dflag = -ONE;
*dd1 = *dd1 * (GAM * GAM);
*dx1 = *dx1 / GAM;
dh11 = dh11 / GAM;
dh12 = dh12 / GAM;
}
while (ABS(*dd1) > GAMSQ) {
dflag = -ONE;
*dd1 = *dd1 / (GAM * GAM);
*dx1 = *dx1 * GAM;
dh11 = dh11 * GAM;
dh12 = dh12 * GAM;
}

if(*dd2 != ZERO)
{
if( (ABS(*dd2) <= RGAMSQ) || (ABS(*dd2) >= GAMSQ) )
{
if(dflag == ZERO)
{
dh11 = ONE;
dh22 = ONE;
dflag = -ONE;
}
else
{
dh21 = -ONE;
dh12 = ONE;
dflag = -ONE;
}
if( ABS(*dd2) <= RGAMSQ )
{
while (ABS(*dd2) <= RGAMSQ) {
*dd2 = *dd2 * (GAM * GAM);
dh21 = dh21 / GAM;
dh22 = dh22 / GAM;
}
}
else
{
while (ABS(*dd2) >= GAMSQ) {
*dd2 = *dd2 / (GAM * GAM);
dh21 = dh21 * GAM;
dh22 = dh22 * GAM;
}
}
}
while (ABS(*dd2) <= RGAMSQ && *dd2 != ZERO) {
dflag = -ONE;
*dd2 = *dd2 * (GAM * GAM);
dh21 = dh21 / GAM;
dh22 = dh22 / GAM;
}
while (ABS(*dd2) > GAMSQ) {
dflag = -ONE;
*dd2 = *dd2 / (GAM * GAM);
dh21 = dh21 * GAM;
dh22 = dh22 * GAM;
}

}
Expand Down
77 changes: 56 additions & 21 deletions utest/test_rotmg.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,7 +53,7 @@ CTEST (drotmg,rotmg)
te_param[i]=tr_param[i]=0.0;
}

//reference values as calulated by netlib blas
//reference values as calculated by netlib blas

tr_d1= 0.1732048;
tr_d2= 0.03840234;
Expand All @@ -71,13 +71,13 @@ CTEST (drotmg,rotmg)
tr_param[4]= 0.0;

BLASFUNC(drotmg)(&te_d1, &te_d2, &te_x1, &te_y1, te_param);
ASSERT_DBL_NEAR_TOL(te_d1, tr_d1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(te_d2, tr_d2, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(te_x1, tr_x1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(te_y1, tr_y1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_d1, te_d1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_d2, te_d2, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_x1, te_x1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_y1, te_y1, DOUBLE_EPS);

for(i=0; i<5; i++){
ASSERT_DBL_NEAR_TOL(te_param[i], tr_param[i], DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_param[i], te_param[i], DOUBLE_EPS);
}
}

Expand All @@ -91,7 +91,7 @@ CTEST (drotmg,rotmg_issue1452)
double tr_param[5];
int i=0;

// from issue #1452, buggy version returned 0.000244 for param[3]
// from issue #1452
te_d1 = 5.9e-8;
te_d2 = 5.960464e-8;
te_x1 = 1.0;
Expand All @@ -100,8 +100,8 @@ CTEST (drotmg,rotmg_issue1452)
for(i=0; i<5; i++){
te_param[i]=tr_param[i]=0.0;
}

//reference values as calulated by netlib blas
te_param[3]=1./4096.;
//reference values as calculated by gonum blas with rotmg rewritten to Hopkins' algorithm
tr_d1= 0.99995592822897;
tr_d2= 0.98981219860583;
tr_x1= 0.03662270484346;
Expand All @@ -110,19 +110,19 @@ CTEST (drotmg,rotmg_issue1452)
tr_param[0]= -1.0;
tr_param[1]= 0.00000161109346;
tr_param[2]= -0.00024414062500;
tr_param[3]= 1.0;
tr_param[3]= 0.00024414062500;
tr_param[4]= 0.00000162760417;

//OpenBLAS
BLASFUNC(drotmg)(&te_d1, &te_d2, &te_x1, &te_y1, te_param);

ASSERT_DBL_NEAR_TOL(te_d1, tr_d1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(te_d2, tr_d2, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(te_x1, tr_x1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(te_y1, tr_y1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_d1, te_d1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_d2, te_d2, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_x1, te_x1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_y1, te_y1, DOUBLE_EPS);

for(i=0; i<5; i++){
ASSERT_DBL_NEAR_TOL(te_param[i], tr_param[i], DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_param[i], te_param[i], DOUBLE_EPS);
}

}
Expand All @@ -145,7 +145,7 @@ CTEST(drotmg, rotmg_D1eqD2_X1eqX2)
te_param[i]=tr_param[i]=0.0;
}

//reference values as calulated by netlib blas
//reference values as calculated by netlib blas
tr_d1= 1.0;
tr_d2= 1.0;
tr_x1= 16.0;
Expand All @@ -160,12 +160,47 @@ CTEST(drotmg, rotmg_D1eqD2_X1eqX2)
//OpenBLAS
BLASFUNC(drotmg)(&te_d1, &te_d2, &te_x1, &te_y1, te_param);

ASSERT_DBL_NEAR_TOL(te_d1, tr_d1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(te_d2, tr_d2, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(te_x1, tr_x1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(te_y1, tr_y1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_d1, te_d1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_d2, te_d2, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_x1, te_x1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_y1, te_y1, DOUBLE_EPS);

for(i=0; i<5; i++){
ASSERT_DBL_NEAR_TOL(tr_param[i], te_param[i], DOUBLE_EPS);
}
}

CTEST(drotmg, drotmg_D1_big_D2_big_flag_zero)
{
double te_d1, tr_d1;
double te_d2, tr_d2;
double te_x1, tr_x1;
double te_y1, tr_y1;
double te_param[5]={1.,4096.,-4096.,1.,4096.};
double tr_param[5]={-1.,4096.,-3584.,1792.,4096.};
int i=0;
te_d1= tr_d1=1600000000.;
te_d2= tr_d2=800000000.;
te_x1= tr_x1=8.;
te_y1= tr_y1=7.;


//reference values as calculated by gonum
tr_d1= 68.96627824858757;
tr_d2= 34.483139124293785;
tr_x1= 45312.;
tr_y1= 7.0;


//OpenBLAS
BLASFUNC(drotmg)(&te_d1, &te_d2, &te_x1, &te_y1, te_param);

ASSERT_DBL_NEAR_TOL(tr_d1, te_d1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_d2, te_d2, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_x1, te_x1, DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_y1, te_y1, DOUBLE_EPS);

for(i=0; i<5; i++){
ASSERT_DBL_NEAR_TOL(te_param[i], tr_param[i], DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(tr_param[i], te_param[i], DOUBLE_EPS);
}
}
Loading