On Wed, Apr 27, 2011 at 10:04 PM, Mihai Badoiu wrote:
> I'm using intel xeon harpertown (E5450) and Python 2.6.4.
> In the cython code, when I use fpclassify, in the slow case I get 3
> (FP_SUBNORMAL)
> In the pure-C code, when I use fpclassify, in the case that's supposed to be
> slow but it's no
I'm using intel xeon harpertown (E5450) and Python 2.6.4.
In the cython code, when I use fpclassify, in the slow case I get 3
(FP_SUBNORMAL)
In the pure-C code, when I use fpclassify, in the case that's supposed to be
slow but it's not, I get a 2 (FP_ZERO)
Somehow, the FPU's have different result
Alec Taylor writes:
> What's an FPU?
Wikipedia not working for you today? :-)
http://en.wikipedia.org/wiki/Floating-point_unit>
I take it as a promising sign that computer-savvy people don't need to
know the term anymore, since FPUs have long been integrated parts of the
CPU on most computers.
On Apr 27, 10:11 am, Alec Taylor wrote:
> What's an FPU?
http://lmgtfy.com/?q=fpu
--
http://mail.python.org/mailman/listinfo/python-list
On Wed, Apr 27, 2011 at 3:11 PM, Alec Taylor wrote:
> What's an FPU?
Floating Point Unit, the part of your computer's processor that
handles floating-point mathematics. Integer calculations are done in
the main CPU, but the FPU (which these days is part of the same hunk
of silicon, but used to be
What's an FPU?
On Tue, Apr 26, 2011 at 11:40 PM, Mihai Badoiu wrote:
> Hi,
> I have terrible performance for multiplication when one number gets very
> close to zero. I'm using cython by writing the following code:
> cdef int i
> cdef double x = 1.0
> for 0 <= i < 1000:
>
On Wed, Apr 27, 2011 at 4:14 AM, Dan Goodman wrote:
> Hi,
>
> On 26/04/2011 15:40, Mihai Badoiu wrote:
>> I have terrible performance for multiplication when one number gets very
>> close to zero. I'm using cython by writing the following code:
>
> This might be an issue with denormal numbers:
>
On 4/26/2011 3:27 PM, Dan Stromberg wrote:
for 0<= i< 1000:
x *= 0.8
#x += 0.01
print x
In my WinXP (Athlon), 3.2 standard install
x=1.0
print(x)
for i in range(1000):
x *= 0.8
x += 0.01
print(x)
takes about 3 1/2 secs with addition comment
On Tue, Apr 26, 2011 at 6:40 AM, Mihai Badoiu wrote:
> Hi,
> I have terrible performance for multiplication when one number gets very
> close to zero. I'm using cython by writing the following code:
> cdef int i
> cdef double x = 1.0
> for 0 <= i < 1000:
> x *= 0.8
>
Yes, running on pure python has the same issue (but overall only a factor 3
away):
i = 0
x = 1.0
while i < 1000:
x *= 0.8
#x += 0.01
i += 1
print x
On Tue, Apr 26, 2011 at 1:44 PM, Philip Semanchuk wrote:
>
> On Apr 26, 2011, at 1:34 PM, Mihai Badoiu wrote:
>
> > Already did. T
Hi,
On 26/04/2011 15:40, Mihai Badoiu wrote:
> I have terrible performance for multiplication when one number gets very
> close to zero. I'm using cython by writing the following code:
This might be an issue with denormal numbers:
http://en.wikipedia.org/wiki/Denormal_number
I don't know much
On Apr 26, 2011, at 1:34 PM, Mihai Badoiu wrote:
> Already did. They suggested the python list, because the asm generated code
> is really correct and the problem might be with the python running on top.
Does the same timing in consistency appear when you use pure Python?
bye
Philip
>
> On
Already did. They suggested the python list, because the asm generated code
is really correct and the problem might be with the python running on top.
On Tue, Apr 26, 2011 at 1:04 PM, Chris Colbert wrote:
>
>
> On Tue, Apr 26, 2011 at 8:40 AM, Mihai Badoiu wrote:
>
>> Hi,
>>
>> I have terrible
On Tue, Apr 26, 2011 at 8:40 AM, Mihai Badoiu wrote:
> Hi,
>
> I have terrible performance for multiplication when one number gets very
> close to zero. I'm using cython by writing the following code:
>
>
You should ask this question on the Cython users mailing list.
--
http://mail.python.org/m
Hi,
I have terrible performance for multiplication when one number gets very
close to zero. I'm using cython by writing the following code:
cdef int i
cdef double x = 1.0
for 0 <= i < 1000:
x *= 0.8
#x += 0.01
print x
This code runs much much slower (20+ time
15 matches
Mail list logo
