Just for record (and future hackers trying to re-think this) the trouble
maker is AFAIK:
int pdl_whichdatatype_double (NV nv) in Basic/Core/pdlcore.c
--
kmx
On 19.11.2015 1:08, dhunt wrote:
Hi Chris: Understand that for my part, I've already worked around this
(mis)feature. We should not r
Hi Chris: Understand that for my part, I've already worked around this
(mis)feature. We should not rush to make a fix that
would have unintended consequences.
Thanks for all your work on PDL!
--Doug
On 11/18/15 4:53 PM, Chris Marshall wrote:
The problem is how to avoid data type size
by hav
The problem is how to avoid data type size
by having a byte image be converted to a
double image by adding the value 2 to it.
Making an image frame 8x bigger would
be surprising and annoying in that case.
As for how perl does it, I think that perl
uses IVs to long (except when it is a
long long)
I like the idea of autopromoting all IVs to long and all NVs to double. After
all, that’s how Perl stores and treats them.
I don’t like the idea of setting this parameter anywhere other than at compile
time. Action at a distance is madness at a distance…
> On Nov 18, 2015, at 3:47 PM, Chris
Hi Craig: Thanks for exploring this further. I'm glad it's a cool bug
and not a boring bug or a nerdy bug!
--Doug
On 11/18/15 2:58 PM, Craig DeForest wrote:
Yes, I believe this is a Bug in the conversion code logic. Currently
we use the smallest possible type that can contain a value. Valu
I agree that the this is a surprising result---even
if correct by our current implementation. Maybe
the default for NV perl scalars should be type
PDL_Double. Maybe a perl scalar promotion
setting that could set a default value, double
or other specific type, smallest size, or ...
--Chris
On We
Yes, I believe this is a Bug in the conversion code logic. Currently we use
the smallest possible type that can contain a value. Values that fit in a
float get promoted to float and those that require a double get promoted to
double. Here’s a four-atan trial, showing the autopromotion using a
I think we may need something like
an option to enable/disable forced
conversion of perl scalars to double
piddles.
--Chris
On Wed, Nov 18, 2015 at 3:03 PM, Chris Marshall
wrote:
> Hi Doug-
>
> I think the problem is only when
> a perl scalar is the input to atan().
> The top two cases are the
Hi Doug-
I think the problem is only when
a perl scalar is the input to atan().
The top two cases are the expected
result. Maybe the logic in the type
conversion is off. You don't say
what version of PDL but I see the
same output from PDL-2.014_02.
--Chris
On Wed, Nov 18, 2015 at 1:29 PM, Dou
Hi all: I just noticed a thing which caused me some trouble.
The PDL 'atan' function defaults to 'float' precision instead of the
more natural 'double':
use PDL;
my $a1 = atan(double(2379.0)/double(1024.0));
my $a2 = atan(float(2379.0)/float(1024.0));
my $a3 = atan(2379.0/1024.0);
print "$a1,
10 matches
Mail list logo
