harmony-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Xiaoming Gu (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HARMONY-5022) [drlvm][jit][opt] incorrect codegeneration of compressed obj arithmetics
Date Fri, 06 Feb 2009 08:03:59 GMT

    [ https://issues.apache.org/jira/browse/HARMONY-5022?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12671037#action_12671037
] 

Xiaoming Gu commented on HARMONY-5022:
--------------------------------------

I read the code about heap base address in Ia32InstCodeSelector.cpp and checked the operands
created by those parts of code with a simple test case on a Windows64 machine. The code itself
and the practical results both show all related operands created in HIR2LIR pass are in 64-bit
size. The bug is gone and I'll close this JIRA.

> [drlvm][jit][opt] incorrect codegeneration of compressed obj arithmetics
> ------------------------------------------------------------------------
>
>                 Key: HARMONY-5022
>                 URL: https://issues.apache.org/jira/browse/HARMONY-5022
>             Project: Harmony
>          Issue Type: Bug
>          Components: DRLVM
>         Environment: x86_64
>            Reporter: Alexey Varlamov
>            Assignee: Mikhail Fursov
>
> Codeselector of Jitrino.OPT always (except ldind) handles heap base as 64-bit immediate
operand for SUB/ADD/CMP instructions, while CPU only supports imm32 for these instruction
group:
> working_vm\vm\jitrino\src\codegenerator\ia32\Ia32InstCodeSelector.cpp(1305):        zeroType,
op, irManager.newImmOpnd(op->getType(),(zeroType == CompareOp::Ref) || (zeroType == CompareOp::CompRef)
? (POINTER_SIZE_INT)VMInterface::getHeapBase() : 0));
> working_vm\vm\jitrino\src\codegenerator\ia32\Ia32InstCodeSelector.cpp(1324):        zeroType,
op, irManager.newImmOpnd(op->getType(),(zeroType == CompareOp::Ref) || (zeroType == CompareOp::CompRef)
? (POINTER_SIZE_INT)VMInterface::getHeapBase() : 0));
> working_vm\vm\jitrino\src\codegenerator\ia32\Ia32InstCodeSelector.cpp(1479):        return
irManager.newImmOpnd(typeManager.getNullObjectType(), (POINTER_SIZE_INT)VMInterface::getHeapBase());
> working_vm\vm\jitrino\src\codegenerator\ia32\Ia32InstCodeSelector.cpp(1939):        Opnd
* heap_base = irManager.newImmOpnd(typeManager.getIntPtrType(), (POINTER_SIZE_INT)VMInterface::getHeapBase());
> working_vm\vm\jitrino\src\codegenerator\ia32\Ia32InstCodeSelector.cpp(2205):        copyOpnd(base,
irManager.newImmOpnd(base->getType(), (POINTER_SIZE_INT)VMInterface::getHeapBase()));
> However encoder silently truncates such operands to 32 bit - that is separate issue,
I'll dig into it.
> In fact the heap base must be loaded to reg64 first.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message