Avoid double memory initilization in PMT unified vectors


#1

Hi,

in PMT, there is code like:

pmt_u8vector::pmt_u8vector(size_t k, uint8_t fill)
: d_v(k)
{
for (size_t i = 0; i < k; i++)
d_v[i] = fill;
}

As std::vector initializes the memory by itself when called with an size
argument in the constructor, the memory is initialized twice.

A faster version with the same result is:

pmt_u8vector::pmt_u8vector(size_t k, uint8_t fill)
: d_v(k, fill)
{}

Same is true for the other data types.

Another question:
Why are the classes for the different types written “by hand”, and not
using
templates, eg:

template
class pmt_uniform_vector
{

}
typedef pmt_uniform_vector< uint8_t > pmt_u8vector;

Stefan


Stefan Brüns / Bergstraße 21 / 52062 Aachen
mailto:lurch at gmx.li http://www.kawo1.rwth-aachen.de/~lurchi/
phone: +49 241 53809034 mobile: +49 151 50412019


#2

On Tue, Oct 07, 2008 at 07:26:08PM +0200, Stefan Brüns wrote:

As std::vector initializes the memory by itself when called with an size
argument in the constructor, the memory is initialized twice.

A faster version with the same result is:

pmt_u8vector::pmt_u8vector(size_t k, uint8_t fill)
: d_v(k, fill)
{}

Thanks. I’ll fix.

}
typedef pmt_uniform_vector< uint8_t > pmt_u8vector;

Stefan

My goal was to keep the interface completely opaque. It’s likely that
the underlying implementation will be completely rewritten at some
point. Although we’re currently using a C++ style implementation,
it’s quite possible that we’ll move to a more lisp-style
implementation with type info stored in the low pointer bits, GC
instead of reference counting, etc.

Using templates in the .cc files to reduce replicated code would be a
good idea with the current implementation.

Eric