529849 - Assertion failed: _allocator.free & rmask(r) (../../../js/src/nanojit/Assembler.cpp:433)

Status: RESOLVED FIXED

(Core Graveyard :: Nanojit, defect, P2)

Description

[Ginn Chen]

Run 'make check' in obj/js/src of debug version.

Comment 1

[Leon Sha]

Attachment: patch

This bug happened because Assembler::argRegs[] and Assembler::savedRegs[] have same registers. Assembler::savedRegs[] is not needed for sparc because the registers are already saved automatically by "save" instruction. But since NumSavedRegs can not be zero, I just add a saved reg L1 in this array.

Comment 2

[Andreas Gal :gal]

Comment on attachment 413979 [details] [diff] [review]
patch

Leon, nanojit lives its own tree now. See https://developer.mozilla.org/en/NanojitMerge

Make sure to follow the instructions there when committing.

Comment 3

[Nicholas Nethercote [inactive]]

(In reply to comment #1)
> 
> This bug happened because Assembler::argRegs[] and Assembler::savedRegs[] have
> same registers. Assembler::savedRegs[] is not needed for sparc because the
> registers are already saved automatically by "save" instruction. But since
> NumSavedRegs can not be zero, I just add a saved reg L1 in this array.

Why can't NumSavedRegs be zero?  A comment explaining the dummy L1 value would be nice.

Comment 4

[Leon Sha]

(In reply to comment #3)
> Why can't NumSavedRegs be zero?  A comment explaining the dummy L1 value would
> be nice.
NumSavedRegs is used as the length of savedRegs in LIR.h and Assembler.h. NumSavedRegs to be zero will cause a zero length array. I'll add such a comment when I commit the patch.

Comment 5

[Nicholas Nethercote [inactive]]

What's wrong with a zero-length array?

Comment 6

[Ginn Chen]

Sun Studio C++ doesn't like zero-length array unless you add -features=zla in CXXFLAGS.

I think it might be cleaner, if we add -featurs=zla globally, and use 0 for NumSavedRegs.

BTW: For the patch, we don't need the line
  uint32_t kind = ins->paramKind();

Comment 7

[Ginn Chen]

Attachment: patch zero length array version

Comment 8

[Nicholas Nethercote [inactive]]

Comment on attachment 414224 [details] [diff] [review]
patch zero length array version

I know I originally asked what was wrong with zero-length arrays... but apparently they're not legal according to the standard, and MSVC has trouble with them.  (The fact that the Sun compiler requires a special flag to accept them also doesn't fill me with confidence.)

So I'm rejecting this patch, sorry.  Let's just stick to the first patch + an explanatory comment.

Comment 9

[Leon Sha]

http://hg.mozilla.org/projects/nanojit-central/rev/23ed78f42df2
http://hg.mozilla.org/tracemonkey/rev/6c119f0a91f9
http://hg.mozilla.org/tamarin-redux/rev/0b1f6d368985