Re: [boost] [iterators] Proof-of-concept for a sentinel iteratoradapter

Subject: Re: [boost] [iterators] Proof-of-concept for a sentinel iteratoradapter
From: Kevin Sopp (baraclese_at_[hidden])
Date: 2009-05-13 13:05:03

On Wed, May 13, 2009 at 6:50 PM, Steven Watanabe <watanabesj_at_[hidden]> wrote:
> I'm not absolutely sure about this, but from the assembler it
> looks like the call to strlen is being pulled out of the loop.

You're right, I added __attribute__((noinline)) to the test_strlen and
test_sentinel functions and I also added -DNDEBUG this time around now
I get the following times:

enter text length: 8
strlen: 0.05 2468358720
sentinel: 0.04 2468358720

enter text length: 512
strlen: 2.46 3054888576
sentinel: 2.17 3054888576

enter text length: 8096
strlen: 38.58 2775800256
sentinel: 34.09 2775800256

which shows that the sentinel version is indeed faster. It's
comforting to know that programmer intuition works after all!