On Mon, 2009-01-12 at 11:59 +0200, Pekka Enberg wrote:
> The current blocker bug for a simple "hello, world" application is in
> the way Jato handles mimic stack on basic block boundaries. The problem
> is that a mimic stack can be non-empty at the end of a basic block
> because of, for example, the ternary operator.
There's commit 5cf876d9fbd7ea1d2ba6f5d679d9e24a43002a1d ("jit: mimic
stack spilling and reloading") in topic/ostack now that fixes up the new
regression test. Unfortunately I missed the fact that there's another
case we need to take care of.
In the first case, we have a basic block A with two _successors_ B and C
(we handle this case now). In the second one, we have a basic block X
with _predecessor_ Y and Z. The problem with the current patch is that
we don't use the same temporaries in Y and Z which makes the new inverse
case of the original regression test to fail.
There's also some "interesting" things going on in the generated
assembly:
TRACE: jamvm/ObjectStackTest.assertIsGreaterThanOne(I)V
Length: 14
Code: 1a 04 a4 00 07 04 a7 00 04 03 b8 00 0d b1
Disassembler Listing:
0x09339000: 57 push %edi
0x09339001: 56 push %esi
0x09339002: 53 push %ebx
0x09339003: 55 push %ebp
0x09339004: 89 e5 mov%esp,%ebp
0x09339006: 8b 45 14 mov0x14(%ebp),%eax
0x09339009: 83 f8 01 cmp$0x1,%eax
0x0933900c: 0f 8e 0c 00 00 00 jle0x0933901e
---
0x09339012: e9 0e 00 00 00 jmp0x09339025 # (1)
0x09339017: b8 01 00 00 00 mov$0x1,%eax
0x0933901c: 89 c3 mov%eax,%ebx
---
0x0933901e: b8 00 00 00 00 mov$0x0,%eax
0x09339023: 89 c1 mov%eax,%ecx # (2)
0x09339025: 53 push %ebx
0x09339026: e8 d5 bf fe ff call 0x09325000
0x0933902b: 83 c4 04 add$0x4,%esp
0x0933902e: e9 00 00 00 00 jmp0x09339033
0x09339033: 5d pop%ebp
0x09339034: 5b pop%ebx
0x09339035: 5e pop%esi
0x09339036: 5f pop%edi
0x09339037: c3 ret
So in (1), we're adding the store instructions *after* the jmp from the
basic block and in (2) we should be using "ebx" just like in the other
predecessor of the basic block.
Pekka
--
This SF.net email is sponsored by:
SourcForge Community
SourceForge wants to tell your story.
http://p.sf.net/sfu/sf-spreadtheword
___
Jatovm-devel mailing list
Jatovm-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/jatovm-devel
