[Numpy-discussion] python numpy code many times slower than c++
Wed Jan 21 15:34:35 CST 2009
Neal Becker wrote:
> Ravi wrote:
>> On Wednesday 21 January 2009 13:55:49 Neal Becker wrote:
>>> I'm only interested in simple strided 1-d vectors. In that case, I
>>> think your code already works. If you have c++ code using the iterator
>>> interface, the iterators dereference will use (*array )[index]. This
>>> will use operator, which will call PyArray_GETPTR. So I think this
>>> will obey strides.
>> You are right. I had forgotten that I had simple strided vectors working.
>>> Unfortunately, it will also be slow. I suggest something like the
>>> enclosed. I have done some simple tests, and it seems to work.
>> I wonder why PyArray_GETPTR1 is slow. Is it because of the implied
>> integer multiplication? Unfortunately, your approach means that iterators
>> can become invalid if the underlying array is resized to a larger size.
>> Hmmh, perhaps we could make this configurable at compile-time ...
Iterators almost always become invalid under those sorts of changes, so I
don't think that's a surprise.
GETPTR1 has to do:
There's several memory references there, and I don't think the compiler can
assume that this value doesn't change from one access to another - so it
can't be cached.
That said, I have tried a few benchmarks. Surprisingly, I'm not seeing any
difference in a few quick tests.
I do have one cosmetic patch for you. This will shutup gcc giving the
longest warning message ever about an unused variable:
--- numpy.new.orig/numpyregister.hpp 2009-01-21 15:59:00.000000000 -0500
+++ numpy.new/numpyregister.hpp 2009-01-21 14:11:00.000000000 -0500
@@ -257,7 +257,8 @@
storage_t *the_storage = reinterpret_cast<storage_t*>( data );
void *memory_chunk = the_storage->storage.bytes;
array_storage_t dd( obj );
- array_t *v = new ( memory_chunk ) array_t( dd.size(), dd );
+ //array_t *v = new ( memory_chunk ) array_t( dd.size(), dd );
+ new ( memory_chunk ) array_t( dd.size(), dd );
data->convertible = memory_chunk;
More information about the Numpy-discussion