llvm dev - Jun 2011 - [LLVMdev] ARM thumb-2 instruction used for non-thumb2 CPUs

On Jun 22, 2011, at 2:08 PM, Renato Golin wrote:
> On 22 June 2011 19:18, Anton Korobeynikov <anton at
korobeynikov.info> wrote:
>> Unfortunately, you have to live with this. Until recently binutils
>> were quite buggy wrt thumb2 code, so, most probably you will need new
>> binutils in any case.
> 
> Hi Anton,
> 
> It's not so simple. GPL3 can be quite a nuisance to change and even to
> use. This is why LLVM is becoming to popular for ARM... ;)
> 
> Jim, how much work would you reckon it'll take to have a working
assembler?
> 
You mean to patch the old binutils to support this old instruction? Very easy,
I'd expect (well, as easy as anything ever is in binutils).

To get -integrated-as working so that we don't need to go through the
assembler at all should be mostly a matter of bug fixing, modulo inline assembly
support. For non-trivial inline assembly, and to get a system assembler
replacement based on the MC assembler, it'll be a bigger task. The ARM asm
parser is currently in need of some tender loving care. That's coming soon,
but it's a non-trivial task.

For -integrated-as, I would suggest giving it a try. I'm not completely sure
how complete the support is for ELF, but I think it's in somewhat OK shape,
at least. Worth trying. That would neatly avoid the problem if it works. If it
doesn't, we should fix the bugs so it does. ;)

-Jim

On Jun 23, 2011, at 12:01 AM, Jim Grosbach wrote:
> 
> On Jun 22, 2011, at 2:08 PM, Renato Golin wrote:
> 
>> On 22 June 2011 19:18, Anton Korobeynikov <anton at
korobeynikov.info> wrote:
>>> Unfortunately, you have to live with this. Until recently binutils
>>> were quite buggy wrt thumb2 code, so, most probably you will need
new
>>> binutils in any case.
>> 
>> Hi Anton,
>> 
>> It's not so simple. GPL3 can be quite a nuisance to change and even
to
>> use. This is why LLVM is becoming to popular for ARM... ;)
>> 
>> Jim, how much work would you reckon it'll take to have a working
assembler?
>> 
> 
> You mean to patch the old binutils to support this old instruction? Very
easy, I'd expect (well, as easy as anything ever is in binutils).
> 
> To get -integrated-as working so that we don't need to go through the
assembler at all should be mostly a matter of bug fixing, modulo inline assembly
support. For non-trivial inline assembly, and to get a system assembler
replacement based on the MC assembler, it'll be a bigger task. The ARM asm
parser is currently in need of some tender loving care. That's coming soon,
but it's a non-trivial task.
> 
> For -integrated-as, I would suggest giving it a try. I'm not completely
sure how complete the support is for ELF, but I think it's in somewhat OK
shape, at least. Worth trying. That would neatly avoid the problem if it works.
If it doesn't, we should fix the bugs so it does. ;)
I tried with FreeBSD kernel. Outcome is 2 different crashes. Output is here:

http://web.me.com/dmarion/FreeBSD/integrated-as_crash.txt

I can provide more details if needed.

Damjan

On Jun 22, 2011, at 4:02 PM, Damjan Marion wrote:
>> 
>> To get -integrated-as working so that we don't need to go through
the assembler at all should be mostly a matter of bug fixing, modulo inline
assembly support. For non-trivial inline assembly, and to get a system assembler
replacement based on the MC assembler, it'll be a bigger task. The ARM asm
parser is currently in need of some tender loving care. That's coming soon,
but it's a non-trivial task.
>> 
>> For -integrated-as, I would suggest giving it a try. I'm not
completely sure how complete the support is for ELF, but I think it's in
somewhat OK shape, at least. Worth trying. That would neatly avoid the problem
if it works. If it doesn't, we should fix the bugs so it does. ;)
> 
> I tried with FreeBSD kernel. Outcome is 2 different crashes. Output is
here:
> 
> http://web.me.com/dmarion/FreeBSD/integrated-as_crash.txt
> 
> I can provide more details if needed.

Thanks, Damjan. That's good information.

Crash #1 is an assembly source file. It's running into something it
doesn't understand on the line indicated. The assertion failure is a
secondary problem, and definitely shouldn't happen. As mentioned, though,
the asm parser isn't ready for prime time yet, so I suspect this may be just
the tip of the iceberg as far as the problems that (and similar) files would
encounter.

Crash #2 is a C file, but is also asserting in the assembly parser, which
indicates that there's inline assembly in the file somewhere that the parser
doesn't know how to handle yet.

Some Makefile magic should be able to take care of issues of the sort
encountered in #1 (don't pass -integrated-as when compiling .S files).

Inline assembly is a bit trickier. Temporarily, if those files compile/assemble
OK with the system assembler (i.e., don't have the rrx instruction mnemonic
in them), we could work around by not using -integrated-as for them. That's
pretty fugly, though, and would just be a quick workaround to allow you to keep
making progress while the real issues get fixed.

How much inline asm is there in that file that's causing issues? if it's
not too much, it might not be too bad to get at least that much of the asm
parsing working. That would be general goodness and would help move you forward.

-Jim

Hi Renato,

We are *very* interested in getting ARM MC in shape. It's the core for an
ARM assembler, disassembler, and MC JIT. To me, there are three main tasks:

1. Get all of the tests (which are not using inline asm) passing with
-integrated-as.
2. Finish up ARM asm parser so integrated-as can support inline asm and to use
it to build a standalone assembler.
3. Rewrite the ARM disassembler. We don't think the current implementation
is good enough for production usage.

For Darwin, we believe #1 is very close. We'll soon find some time start
large scale testing. For ELF, it's less clear. You guys should definitely
try it out and report bugs.

We also plan on tackling #2 in the not too distant future. It's a big chunk
of work so we'd appreciate any assistance on this. #3 is less urgent, but
we'd like to gut the existing one and replace it with something that's
more auto-generated than what's currently there.

Evan

On Jun 22, 2011, at 3:01 PM, Jim Grosbach wrote:
> 
> On Jun 22, 2011, at 2:08 PM, Renato Golin wrote:
> 
>> On 22 June 2011 19:18, Anton Korobeynikov <anton at
korobeynikov.info> wrote:
>>> Unfortunately, you have to live with this. Until recently binutils
>>> were quite buggy wrt thumb2 code, so, most probably you will need
new
>>> binutils in any case.
>> 
>> Hi Anton,
>> 
>> It's not so simple. GPL3 can be quite a nuisance to change and even
to
>> use. This is why LLVM is becoming to popular for ARM... ;)
>> 
>> Jim, how much work would you reckon it'll take to have a working
assembler?
>> 
> 
> You mean to patch the old binutils to support this old instruction? Very
easy, I'd expect (well, as easy as anything ever is in binutils).
> 
> To get -integrated-as working so that we don't need to go through the
assembler at all should be mostly a matter of bug fixing, modulo inline assembly
support. For non-trivial inline assembly, and to get a system assembler
replacement based on the MC assembler, it'll be a bigger task. The ARM asm
parser is currently in need of some tender loving care. That's coming soon,
but it's a non-trivial task.
> 
> For -integrated-as, I would suggest giving it a try. I'm not completely
sure how complete the support is for ELF, but I think it's in somewhat OK
shape, at least. Worth trying. That would neatly avoid the problem if it works.
If it doesn't, we should fix the bugs so it does. ;)
> 
> -Jim
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev

On 23 June 2011 00:24, Evan Cheng <evan.cheng at apple.com>
wrote:
> 1. Get all of the tests (which are not using inline asm) passing with
-integrated-as.
> 2. Finish up ARM asm parser so integrated-as can support inline asm and to
use it to build a standalone assembler.
> 3. Rewrite the ARM disassembler. We don't think the current
implementation is good enough for production usage.
Hi Evan,

That's a good roadmap. We can definitely start testing with
integrated-as as of now. Last time I tried, all of them passed with MC
ELF, but we found some bugs on other tests when using our front-end,
which could be many things (I couldn't test with Clang). I have to get
back to it soon.

We're setting up a nightly test for LLVM, it's already running on
Linux (and I believe Windows), but we need OSX and native ARM-Linux
too. We should multiply that by three with integrated-as and MC ELF,
to get a good coverage. We'll report any bug along the way, hopefully
in the form of a patch+test. ;)

The asm parser/disassembler could take a while, though, but that's
feasible in the long run, and I agree it's the right path.

cheers,
--renato