about the fast path and slow path

=?GB2312?B?UqimbWkgRm9yYXg=?= forax at univ-mlv.fr
Fri Jul 20 03:48:24 PDT 2012

On 07/20/2012 05:12 AM, David Holmes wrote:
> On 20/07/2012 12:39 PM, ½¯³å wrote:
>> What is the fast path and slow path?
> Generally:
> "fast path" - a code path optimized for speed because it is the common case
> "slow path" - what we have to do when the "fast path" is not applicable
> For example:
> allocation:
>  - fast path: if enough space bump a pointer
>  - slow path: allocate a new thread-local allocation buffer then bump
> pointer
> monitor acquisition:
>  - fast path: object is unlocked and unbiased so simply CAS in current
> thread ID into object header
>  - slow path: object is unlocked but biased, so revoke bias and use
> BasicLock to lock
> classloading:
>  - fast path: class already loaded, just check in system dictionary and
> return
>  - slow path: class is not loaded, so load it
> And there are many other examples. And it isn't necessarily just a
> choice of two things: the slow path typically has many branches and
> different lengths depending on the exact circumstances.

Also, the JIT itself tend to consider the user code not called in the
interpreter as a slow path too.
segregating slow path and fast path is a generic optimization technique.

> David


More information about the hotspot-runtime-dev mailing list