llvm dev - May 2018 - [clang] Running a single testcase

On Mon, May 7, 2018 at 5:19 PM, Brian Cain <brian.cain at gmail.com>
wrote:
>
>
> On Mon, May 7, 2018 at 10:13 AM, Sedat Dilek <sedat.dilek at
gmail.com> wrote:
>>
>> ...
>
>
>>
>> sdi at iniza:~/src/llvm-toolchain/build$ ./bin/llvm-lit -v
>> ./tools/clang/test/Sema/asm.c
>> llvm-lit:
>> /home/sdi/src/llvm-toolchain/llvm/utils/lit/lit/llvm/config.py:334:
>> note: using clang: /home/sdi/src/llvm-toolchain/build/bin/clang
>> -- Testing: 1 tests, 1 threads --
>> PASS: Clang :: Sema/asm.c (1 of 1)
>> Testing Time: 0.10s
>>   Expected Passes    : 1
>>
>> This now looks good to me.
>>
>> How can I force to use a different clang than
>> "/home/sdi/src/llvm-toolchain/build/bin/clang"?
>> I have no symlink to /usr/bin/clang.
>>
>
>  Can you give a little more context?  What is your ultimate goal for this
> activity?  Why would you want to run the test in this fashion and what
> meaning would you attribute to the test passing or failing?
>
When compiling a Linux-kernel I hit a problem in the
KVM/paravirtualization area and "clobbering RCX" on Debian/testing
AMD64.
The question came up, is this is a compiler bug or not.
Details see [1].

Thus, I was looking for a reproducible testcase.
I have seen that clang ships some tests.
So, I was curious to run a single one.
Not sure if clang/test/Sema/asm.c is suitable here to see "clobbering
RCX" on Debian/x86-64.

On my host I have installed a clang-7 package from <apt.llvm.org>.
I wanted to test against that version.

Does this help you?

- Sedat -

[1] https://lkml.org/lkml/2018/4/24/1227

On Mon, May 7, 2018 at 10:46 AM, Sedat Dilek <sedat.dilek at gmail.com>
wrote:
> On Mon, May 7, 2018 at 5:19 PM, Brian Cain <brian.cain at gmail.com>
wrote:
> >
> >
> ...


> >  Can you give a little more context?  What is your ultimate goal for
this
> > activity?  Why would you want to run the test in this fashion and what
> > meaning would you attribute to the test passing or failing?
> >
>
> When compiling a Linux-kernel I hit a problem in the
> KVM/paravirtualization area and "clobbering RCX" on
Debian/testing
> AMD64.
> The question came up, is this is a compiler bug or not.
> Details see [1].
>
> Thus, I was looking for a reproducible testcase.
> I have seen that clang ships some tests.
> So, I was curious to run a single one.
> Not sure if clang/test/Sema/asm.c is suitable here to see "clobbering
> RCX" on Debian/x86-64.
>
>
Yes, I agree that it might not be the right test.

> On my host I have installed a clang-7 package from <apt.llvm.org>.
> I wanted to test against that version.
>
> Does this help you?
>
> - Sedat -
>
> [1] https://lkml.org/lkml/2018/4/24/1227
>
I see, yes, this is clearer now.

IIUC, this test is testing clang's behavior regarding inline assembly and
probably won't verify things like which registers are saved and restored.

I think you might get resolution on this issue faster if you reduce
native_save_fl()
and its caller to a simple example that shows RCX not being saved.  Bonus
points if you can identify any conditions that change this behavior -- or
versions of clang that behave differently.  Then share your test and ask on
llvm-dev and/or cfe-dev whether or not this behavior is expected.

-- 
-Brian
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20180507/6b74dbd7/attachment.html>

On Mon, May 7, 2018 at 5:56 PM, Brian Cain <brian.cain at gmail.com>
wrote:
>
> On Mon, May 7, 2018 at 10:46 AM, Sedat Dilek <sedat.dilek at
gmail.com> wrote:
>>
>> On Mon, May 7, 2018 at 5:19 PM, Brian Cain <brian.cain at
gmail.com> wrote:
>> >
>> >
>> ...
>
>
>
>>
>> >  Can you give a little more context?  What is your ultimate goal
for
>> > this
>> > activity?  Why would you want to run the test in this fashion and
what
>> > meaning would you attribute to the test passing or failing?
>> >
>>
>> When compiling a Linux-kernel I hit a problem in the
>> KVM/paravirtualization area and "clobbering RCX" on
Debian/testing
>> AMD64.
>> The question came up, is this is a compiler bug or not.
>> Details see [1].
>>
>> Thus, I was looking for a reproducible testcase.
>> I have seen that clang ships some tests.
>> So, I was curious to run a single one.
>> Not sure if clang/test/Sema/asm.c is suitable here to see
"clobbering
>> RCX" on Debian/x86-64.
>>
>
> Yes, I agree that it might not be the right test.
>
>>
>> On my host I have installed a clang-7 package from
<apt.llvm.org>.
>> I wanted to test against that version.
>>
>> Does this help you?
>>
>> - Sedat -
>>
>> [1] https://lkml.org/lkml/2018/4/24/1227
>
>
> I see, yes, this is clearer now.
>
> IIUC, this test is testing clang's behavior regarding inline assembly
and
> probably won't verify things like which registers are saved and
restored.
>
> I think you might get resolution on this issue faster if you reduce
> native_save_fl() and its caller to a simple example that shows RCX not
being
> saved.  Bonus points if you can identify any conditions that change this
> behavior -- or versions of clang that behave differently.  Then share your
> test and ask on llvm-dev and/or cfe-dev whether or not this behavior is
> expected.
>
Matthias answered in [1].

[1] https://lkml.org/lkml/2018/5/7/534