llvm dev - Jul 2010 - [LLVMdev] JIT memory manager: NeedsExactSize() ?

The JIT has a codepath that supports JIT memory managers that return true for
NeedsExactSize().  This is particularly bad for X86, because the "exact
size" is not exact, and looks completely wrong in a lot of cases.  Is
anyone using this?  Can I rip it out?  This stuff should go away with the
mc'ization of the JIT in the long term.

-Chris

Oh god, rip it out.  IIRC it's just a complete parallel
reimplementation of bunch of instruction encoding stuff just to
compute the code size.

Reid

On Wed, Jul 21, 2010 at 3:52 PM, Chris Lattner <clattner at apple.com>
wrote:
> The JIT has a codepath that supports JIT memory managers that return true
for NeedsExactSize().  This is particularly bad for X86, because the "exact
size" is not exact, and looks completely wrong in a lot of cases.  Is
anyone using this?  Can I rip it out?  This stuff should go away with the
mc'ization of the JIT in the long term.
>
> -Chris
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>

On Jul 21, 2010, at 5:01 PM, Reid Kleckner wrote:
> Oh god, rip it out.  IIRC it's just a complete parallel
> reimplementation of bunch of instruction encoding stuff just to
> compute the code size.
Yeah, I know.  It's the 3rd and most incorrect copy of the instruction
emitter.  I really really want to rip it out, but we do have a unit test for the
'needsexactsize()' stuff, so I figure someone might actually care.  Does
vmkit use this?

-Chris

> 
> Reid
> 
> On Wed, Jul 21, 2010 at 3:52 PM, Chris Lattner <clattner at
apple.com> wrote:
>> The JIT has a codepath that supports JIT memory managers that return
true for NeedsExactSize().  This is particularly bad for X86, because the
"exact size" is not exact, and looks completely wrong in a lot of
cases.  Is anyone using this?  Can I rip it out?  This stuff should go away with
the mc'ization of the JIT in the long term.
>> 
>> -Chris
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>>