The Perl and Raku Conference 2025: Greenville, South Carolina - June 27-29 Learn more

use strict;
use Math::MPFR qw(:mpfr);
use Config;
my $t = 21;
print "1..$t\n";
my $why;
my $proceed = Math::MPFR::_MPFR_WANT_FLOAT128();
unless($proceed) {
####################################################
if($Config{nvtype} eq '__float128') {
Rmpfr_set_default_prec(114);
my $ok = 1;
my $it;
for $it(1 .. 10000) {
my $nv = rand(1024) / (1 + rand(1024));
#$ $larg_1 and $larg_2 will be complementary Rounding modes.
my $larg_1 = int(rand(5));
my $larg_2 = $larg_1 ? 5 - $larg_1 : $larg_1;
my $f128_1 = $nv;
my $fr_1 = Math::MPFR->new();
Rmpfr_set_NV($fr_1, $f128_1, $larg_1);
my $f128_2 = Rmpfr_get_NV($fr_1, $larg_2);
unless($f128_1 == $f128_2) {
$ok = 0;
warn "$it: $f128_1 != $f128_2\n $larg_1 : $larg_2\n\n";
}
}
if($ok) {print "ok 1\n"}
else {print "not ok 1\n"}
$ok = 1;
Rmpfr_set_default_prec(115);
for $it(1 .. 10000) {
my $nv = rand(1024) / (1 + rand(1024));
my $f128_1 = $nv;
my $fr_1 = Math::MPFR->new();
Rmpfr_set_NV($fr_1, $f128_1, 0);
my $f128_2 = Rmpfr_get_NV($fr_1, 0);
unless($f128_1 == $f128_2) {
$ok = 0;
warn "$it: $f128_1 != $f128_2\n";
}
}
if($ok) {print "ok 2\n"}
else {print "not ok 2\n"}
my $nan = Rmpfr_get_NV(Math::MPFR->new(), MPFR_RNDN);
my $pinf = 999 ** (999 ** 999);
my $ninf = $pinf * -1.0;
my $zero = 0.0;
my $nzero = -1.0 / $pinf;
my $fr = Math::MPFR->new();
Rmpfr_set_NV($fr, $nan, MPFR_RNDN);
my $rop = Rmpfr_get_NV($fr, MPFR_RNDN);
if($rop != $rop) {print "ok 3\n"}
else {
warn "\$rop: $rop\n";
print "not ok 3\n";
}
Rmpfr_set_NV($fr, $pinf, MPFR_RNDN);
$rop = Rmpfr_get_NV($fr, MPFR_RNDN);
if($rop != 0.0 && $rop > 0.0 && $rop / $rop != 1.0) {print "ok 4\n"}
else {
warn "\$rop: $rop\n";
print "not ok 4\n";
}
Rmpfr_set_NV($fr, $ninf, MPFR_RNDN);
$rop = Rmpfr_get_NV($fr, MPFR_RNDN);
if($rop != 0.0 && $rop < 0.0 && $rop / $rop != 1.0) {print "ok 5\n"}
else {
warn "\$rop: $rop\n";
print "not ok 5\n";
}
Rmpfr_set_NV($fr, $zero, MPFR_RNDN);
$rop = Rmpfr_get_NV($fr, MPFR_RNDN);
if($rop == 0 && substr("$rop", 0, 1) ne '-') {print "ok 6\n"}
else {
warn "\$rop: $rop\n";
print "not ok 6\n";
}
Rmpfr_set_NV($fr, $nzero, MPFR_RNDN);
$rop = Rmpfr_get_NV($fr, MPFR_RNDN);
if($rop == 0 && substr(sprintf("%e", $rop), 0, 1) eq '-') {print "ok 7\n"}
else {
warn "\nExpected -0\nGot: ", sprintf("%e", $rop), "\n";
if($rop == 0) {
warn "Problem with signed zero - not registering a failure for this\n";
print "ok 7\n";
}
else {print "not ok 7\n"}
}
my $bigpos = Math::MPFR->new('1.4e4932');
my $bigneg = $bigpos * -1;
$rop = Rmpfr_get_NV($bigpos, MPFR_RNDN);
if($rop != 0.0 && $rop > 0.0 && $rop / $rop != 1.0) {print "ok 8\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 8\n";
}
$rop = Rmpfr_get_NV($bigneg, MPFR_RNDN);
if($rop != 0.0 && $rop < 0.0 && $rop / $rop != 1.0) {print "ok 9\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 9\n";
}
$rop = Rmpfr_get_NV(Math::MPFR->new('1.18973149535723176508575932662800702e4932'), MPFR_RNDZ);
if($rop == '1.18973149535723176508575932662800702e4932') {print "ok 10\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 10\n";
}
if(-$rop == '-1.18973149535723176508575932662800702e4932') {print "ok 11\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 11\n";
}
print "ok 12\n";
my $littlepos = Math::MPFR->new('7e-4967');
my $littleneg = $littlepos * -1;
$rop = Rmpfr_get_NV($littlepos, MPFR_RNDZ);
if($rop == 0.0) {print "ok 13\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 13\n";
}
$rop = Rmpfr_get_NV($littleneg, MPFR_RNDZ);
if($rop == 0) {print "ok 14\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 14\n";
}
# Any +ve non-zero value (no matter how small) will be rounded to a non-zero value under RNDA.
# Values anywhere between zero and the minimum subnormal value (as is the case here) will be
# rounded to that minimum subnormal value (6.47517511943802511092443895822764655e-4966)
$rop = Rmpfr_get_NV($littlepos, MPFR_RNDA);
if($rop == 6.475175119438025110924438958227646552e-4966) {print "ok 15\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 15\n";
}
# Any -ve non-zero value (no matter how close to zero) will be rounded to a non-zero value under RNDA.
# Values anywhere between zero and the negated minimum subnormal value (as is the case here) will be
# be rounded to that negated minimum subnormal value (-6.47517511943802511092443895822764655e-4966)
$rop = Rmpfr_get_NV($littleneg, MPFR_RNDA);
if($rop == -6.475175119438025110924438958227646552e-4966) {print "ok 16\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 16\n";
}
print "ok 17\n";
my $fr_F128 = Rmpfr_init2(115);
my $f128_1 = 1e-298;
Rmpfr_set_NV($fr_F128, $f128_1, MPFR_RNDN);
my $f128_2 = Rmpfr_get_NV($fr_F128, MPFR_RNDN);
if($f128_1 == $f128_2) {print "ok 18\n"}
else {
warn "\n $f128_1: $f128_1\n \$f128_2: $f128_2\n";
print "not ok 18\n";
}
print "ok 19\n";
$f128_1 = 1e-360;
Rmpfr_set_NV($fr_F128, $f128_1, MPFR_RNDN);
$f128_2 = Rmpfr_get_NV($fr_F128, MPFR_RNDN);
if($f128_1 == $f128_2) {print "ok 20\n"}
else {
warn "\n $f128_1: $f128_1\n \$f128_2: $f128_2\n";
print "not ok 20\n";
}
print "ok 21\n";
}
####################################################
else {
$why = "__float128 tests not applicable to this build of perl\n";
warn "\n Skipping all tests: $why";
print "ok $_\n" for 1..$t;
exit 0;
}
}
else {
eval {require Math::Float128; Math::Float128->import (qw(:all));};
if($@) {
$why = "Couldn't load Math::Float128\n";
warn "\n Skipping all tests: $why: $@\n";
print "ok $_\n" for 1..$t;
exit 0;
}
if(196866 >= MPFR_VERSION) {
$why = "No float128 support with this version of the mpfr library\n";
warn "\n Skipping all tests: $why";
print "ok $_\n" for 1..$t;
exit 0;
}
if($proceed) {
Rmpfr_set_default_prec(114);
my $ok = 1;
my $it;
for $it(1 .. 10000) {
my $nv = rand(1024) / (1 + rand(1024));
#$ $larg_1 and $larg_2 will be complementary Rounding modes.
my $larg_1 = int(rand(5));
my $larg_2 = $larg_1 ? 5 - $larg_1 : $larg_1;
my $f128_1 = NVtoF128($nv);
my $fr_1 = Math::MPFR->new();
Rmpfr_set_FLOAT128($fr_1, $f128_1, $larg_1);
my $f128_2 = NVtoF128(0);
Rmpfr_get_FLOAT128($f128_2, $fr_1, $larg_2);
unless($f128_1 == $f128_2) {
$ok = 0;
warn "$it: $f128_1 != $f128_2\n $larg_1 : $larg_2\n\n";
}
}
if($ok) {print "ok 1\n"}
else {print "not ok 1\n"}
$ok = 1;
Rmpfr_set_default_prec(115);
for $it(1 .. 10000) {
my $nv = rand(1024) / (1 + rand(1024));
my $f128_1 = NVtoF128($nv);
my $fr_1 = Math::MPFR->new();
Rmpfr_set_FLOAT128($fr_1, $f128_1, 0);
my $f128_2 = NVtoF128(0);
Rmpfr_get_FLOAT128($f128_2, $fr_1, 0);
unless($f128_1 == $f128_2) {
$ok = 0;
warn "$it: $f128_1 != $f128_2\n";
}
}
if($ok) {print "ok 2\n"}
else {print "not ok 2\n"}
my $nanF128 = NaNF128();
my $pinfF128 = InfF128(1);
my $ninfF128 = InfF128(-1);
my $zeroF128 = ZeroF128(1);
my $nzeroF128 = ZeroF128(-1);
my $rop = Math::Float128->new();
my $fr = Math::MPFR->new();
Rmpfr_set_FLOAT128($fr, $nanF128, MPFR_RNDN);
Rmpfr_get_FLOAT128($rop, $fr, MPFR_RNDN);
if(is_NaNF128($rop)) {print "ok 3\n"}
else {
warn "\$rop: $rop\n";
print "not ok 3\n";
}
Rmpfr_set_FLOAT128($fr, $pinfF128, MPFR_RNDN);
Rmpfr_get_FLOAT128($rop, $fr, MPFR_RNDN);
if(is_InfF128($rop) > 0) {print "ok 4\n"}
else {
warn "\$rop: $rop\n";
print "not ok 4\n";
}
Rmpfr_set_FLOAT128($fr, $ninfF128, MPFR_RNDN);
Rmpfr_get_FLOAT128($rop, $fr, MPFR_RNDN);
if(is_InfF128($rop) < 0) {print "ok 5\n"}
else {
warn "\$rop: $rop\n";
print "not ok 5\n";
}
Rmpfr_set_FLOAT128($fr, $zeroF128, MPFR_RNDN);
Rmpfr_get_FLOAT128($rop, $fr, MPFR_RNDN);
if(is_ZeroF128($rop) > 0) {print "ok 6\n"}
else {
warn "\$rop: $rop\n";
print "not ok 6\n";
}
Rmpfr_set_FLOAT128($fr, $nzeroF128, MPFR_RNDN);
Rmpfr_get_FLOAT128($rop, $fr, MPFR_RNDN);
if(is_ZeroF128($rop) < 0) {print "ok 7\n"}
else {
warn "\$rop: $rop\n";
print "not ok 7\n";
}
my $bigpos = Math::MPFR->new('1.4e4932');
my $bigneg = $bigpos * -1;
Rmpfr_get_FLOAT128($rop, $bigpos, MPFR_RNDN);
if(is_InfF128($rop) > 0) {print "ok 8\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 8\n";
}
Rmpfr_get_FLOAT128($rop, $bigneg, MPFR_RNDN);
if(is_InfF128($rop) < 0) {print "ok 9\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 9\n";
}
Rmpfr_get_FLOAT128($rop, Math::MPFR->new('1.18973149535723176508575932662800702e4932'), MPFR_RNDZ);
if($rop == Math::Float128->new('1.18973149535723176508575932662800702e4932')) {print "ok 10\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 10\n";
}
if(-$rop == Math::Float128->new('-1.18973149535723176508575932662800702e4932')) {print "ok 11\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 11\n";
}
if($rop == Math::Float128::FLT128_MAX()) {print "ok 12\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 12\n";
}
my $littlepos = Math::MPFR->new('7e-4967');
my $littleneg = $littlepos * -1;
Rmpfr_get_FLOAT128($rop, $littlepos, MPFR_RNDZ);
if(is_ZeroF128($rop) > 0) {print "ok 13\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 13\n";
}
Rmpfr_get_FLOAT128($rop, $littleneg, MPFR_RNDZ);
if(is_ZeroF128($rop) < 0) {print "ok 14\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 14\n";
}
# Any +ve non-zero value (no matter how small) will be rounded to a non-zero value under RNDA.
# Values anywhere between zero and the minimum subnormal value (as is the case here) will be
# rounded to that minimum subnormal value (6.47517511943802511092443895822764655e-4966)
Rmpfr_get_FLOAT128($rop, $littlepos, MPFR_RNDA);
if(!is_ZeroF128($rop)) {print "ok 15\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 15\n";
}
# Any -ve non-zero value (no matter how close to zero) will be rounded to a non-zero value under RNDA.
# Values anywhere between zero and the negated minimum subnormal value (as is the case here) will be
# be rounded to that negated minimum subnormal value (-6.47517511943802511092443895822764655e-4966)
Rmpfr_get_FLOAT128($rop, $littleneg, MPFR_RNDA);
if(!is_ZeroF128($rop)) {print "ok 16\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 16\n";
}
Rmpfr_get_FLOAT128($rop, Math::MPFR->new('6.475175119438025110924438958227646552e-4966'), MPFR_RNDN);
if($rop == Math::Float128::FLT128_DENORM_MIN()) {print "ok 17\n"}
else {
warn "\n\$rop: $rop\n";
print "not ok 17\n";
}
my $fr_F128 = Rmpfr_init2(115);
my $f128_1 = STRtoF128('1e-298');
my $f128_2 = Math::Float128->new();
Rmpfr_set_FLOAT128($fr_F128, $f128_1, MPFR_RNDN);
Rmpfr_get_FLOAT128($f128_2, $fr_F128, MPFR_RNDN);
if($f128_1 == $f128_2) {print "ok 18\n"}
else {
warn "\n $f128_1: $f128_1\n \$f128_2: $f128_2\n";
print "not ok 18\n";
}
$f128_1 = NVtoF128(1e-298);
Rmpfr_set_FLOAT128($fr_F128, $f128_1, MPFR_RNDN);
Rmpfr_get_FLOAT128($f128_2, $fr_F128, MPFR_RNDN);
if($f128_1 == $f128_2) {print "ok 19\n"}
else {
warn "\n $f128_1: $f128_1\n \$f128_2: $f128_2\n";
print "not ok 19\n";
}
$f128_1 = STRtoF128('1e-360');
Rmpfr_set_FLOAT128($fr_F128, $f128_1, MPFR_RNDN);
Rmpfr_get_FLOAT128($f128_2, $fr_F128, MPFR_RNDN);
if($f128_1 == $f128_2) {print "ok 20\n"}
else {
warn "\n $f128_1: $f128_1\n \$f128_2: $f128_2\n";
print "not ok 20\n";
}
$f128_1 = NVtoF128(1e-360);
Rmpfr_set_FLOAT128($fr_F128, $f128_1, MPFR_RNDN);
Rmpfr_get_FLOAT128($f128_2, $fr_F128, MPFR_RNDN);
if($f128_1 == $f128_2) {print "ok 21\n"}
else {
warn "\n $f128_1: $f128_1\n \$f128_2: $f128_2\n";
print "not ok 21\n";
}
}
else {
warn "Skipping all tests - Math::MPFR not built for Float128 support";
print "ok $_\n" for 1..$t;
}
}