[Python-Dev] Opcode cache in ceval loop

[Yury Selivanov]

On 2016-02-01 3:21 PM, Brett Cannon wrote:
>
> On Mon, 1 Feb 2016 at 12:16 Yury Selivanov <yselivanov.ml at gmail.com 
> <mailto:yselivanov.ml at gmail.com>> wrote:
>
>     Brett,
>
>     On 2016-02-01 3:08 PM, Brett Cannon wrote:
>     >
>     >
>     > On Mon, 1 Feb 2016 at 11:51 Yury Selivanov
>     <yselivanov.ml at gmail.com <mailto:yselivanov.ml at gmail.com>
>     > <mailto:yselivanov.ml at gmail.com
>     <mailto:yselivanov.ml at gmail.com>>> wrote:
>     >
>     >     Hi Brett,
>     >
>     [..]
>     >
>     >
>     >     The first two fields are used to make sure that we have
>     objects of the
>     >     same type.  If it changes, we deoptimize the opcode
>     immediately.  Then
>     >     we try the offset.  If it's successful - we have a cache
>     hit.  If not,
>     >     that's fine, we'll try another few times before deoptimizing the
>     >     opcode.
>     >
>     >
>     > So this is a third "next step" that has its own issue?
>
>     It's all in issue http://bugs.python.org/issue26219 right now.
>
>     My current plan is to implement LOAD_METHOD/CALL_METHOD (just opcodes,
>     no cache) in 26110.
>
>     Then implement caching for LOAD_METHOD, LOAD_GLOBAL, and LOAD_ATTR in
>     26219.  I'm flexible to break down 26219 in three separate issues if
>     that helps the review process (but that would take more of my time):
>
>     - implement support for opcode caching (general infrastructure) +
>     LOAD_GLOBAL optimization
>     - LOAD_METHOD optimization
>     - LOAD_ATTR optimization
>
>
> I personally don't care how you break it down, just trying to keep all 
> the moving pieces in my head. :)
>
> Anyway, it sounds like PEP 509 is blocking part of it, but the 
> LOAD_METHOD stuff can go in as-is. So are you truly blocked only on 
> getting the latest version of that patch up to 
> http://bugs.python.org/issue26110 and getting a code review?
Yep.  The initial implementation of LOAD_METHOD doesn't need PEP 509 / 
opcode caching.  I'll have to focus on something else this week, but 
early next week I can upload a new patch for 26110.

When we have 26110 committed and PEP 509 approved and committed, I can 
update the opcode cache patch (issue 26219) and we can start reviewing it.

Yury