Re: [Qemu-devel] [PULL 0/7] check-softfloat, fp-bench and clang compile fixes

[Alex Bennée]

Peter Maydell <address@hidden> writes:

> On Thu, 17 Jan 2019 at 18:30, Emilio G. Cota <address@hidden> wrote:
>>
>> On Thu, Jan 17, 2019 at 17:37:54 +0000, Peter Maydell wrote:
>> > On Thu, 17 Jan 2019 at 13:27, Alex Bennée <address@hidden> wrote:
>> > >
>> > > The following changes since commit 
>> > > 4b9f0b0f7c84eea2dfb0d5be3e0254bc91319dbc:
>> > >
>> > >   Merge remote-tracking branch 
>> > > 'remotes/stefanha/tags/block-pull-request' into staging (2019-01-15 
>> > > 17:24:00 +0000)
>> > >
>> > > are available in the Git repository at:
>> > >
>> > >   https://github.com/stsquad/qemu.git tags/pull-fpu-next-170119-1
>> (snip)
>> >
>> > FreeBSD, OSX, x86-64 Linux clang builds:
>> (snip)
>> > PPC64, AArch64:
>> (snip)
>> > NetBSD:
>> (snip)
>>
>> I have added a few commits to fix these -- diff below.
>> Note that one fix requires a small change in the testfloat
>> submodule, which I'm pulling here.
>>
>> Alex, can you cherry-pick the commits in this branch?
>>
>>   https://github.com/cota/qemu/tree/bennee-pull
>>
>> They should go in before adding fp-test to the automated build,
>> of course. I left them at the top to make it easier for
>> you to cherry-pick.
>>
>> Also, since this will require a respin, you might want to fix the
>> "Makfile" typo in patch 7's title.
>>
>> > S390X host:
>> > Looks like a failure running the tests, but no diagnostics about
>> > what exactly went wrong or clear "test failed" indicator:
>> >
>> > cd /home/linux1/qemu/build/all/tests/fp && ./fp-test -s -l 1
>> > i32_to_f16 i64_to_f16 i32_to_f32 i64_t
>> > o_f32 i32_to_f64 i64_to_f64 i32_to_f128 i64_to_f128 >
>> > int-to-float.out 2>  int-to-float.err
>> > /home/linux1/qemu/tests/Makefile.include:913: recipe for target
>> > 'check-softfloat-conv' failed
>> > make: *** [check-softfloat-conv] Error 1
>>
>> What are the contents of "int-to-float.err"?
>
<snip>
>>> Testing i32_to_f128
> 372 tests total.
> 21 tests performed; 20 errors found.

It's probably not a regression but a failure in the old f128 code. When
I was enabling the tests I basically turned them all on and then
disabled what failed (all of which was exercising non-refactored code).
But of course this was all on x86_64 (although I also ran them on a
aarch64 host). It looks like more errors come out on a s390x host for
some reason. You can run manually for a more complete test with more
coverage:

  ./fp-test -l 2 -r all i32_to_f128

>
>
>> diff --git a/tests/fp/Makefile b/tests/fp/Makefile
>> index 5019dcdca0..5a35e7c210 100644
>> --- a/tests/fp/Makefile
>> +++ b/tests/fp/Makefile
>> @@ -65,8 +65,7 @@ QEMU_CFLAGS += $(TF_OPTS)
>>  TF_CFLAGS :=
>>  TF_CFLAGS += -Wno-strict-prototypes
>>  TF_CFLAGS += -Wno-unknown-pragmas
>> -TF_CFLAGS += -Wno-discarded-qualifiers
>> -TF_CFLAGS += -Wno-maybe-uninitialized
>> +TF_CFLAGS += -Wno-uninitialized
>>  TF_CFLAGS += -Wno-missing-prototypes
>>  TF_CFLAGS += -Wno-return-type
>>  TF_CFLAGS += -Wno-unused-function
>
> configure has logic to check whether it can use particular
> warning enable/disable flags. Newer gcc (and I hope clang
> but forget) will happily silently allow -Wno-random-new-thing
> even if they don't support -Wrandom-new-thing) but I'm not
> sure our minimum compiler version is yet new enough to
> be able to rely on that (indeed the warning messages suggest
> it is not).
>
>> diff --git a/tests/fp/berkeley-testfloat-3 b/tests/fp/berkeley-testfloat-3
>> index ca9fa2ba05..5a59dcec19 160000
>> --- a/tests/fp/berkeley-testfloat-3
>> +++ b/tests/fp/berkeley-testfloat-3
>> @@ -1 +1 @@
>> -Subproject commit ca9fa2ba05625ba929958f163b01747e07dd39cc
>> +Subproject commit 5a59dcec19327396a011a17fd924aed4fec416b3
>> diff --git a/tests/fp/fp-test.c b/tests/fp/fp-test.c
>> index fca576309c..2a35ef601d 100644
>> --- a/tests/fp/fp-test.c
>> +++ b/tests/fp/fp-test.c
>> @@ -789,7 +789,7 @@ static int set_init_flags(const char *flags)
>>      return 0;
>>  }
>>
>> -static uint8_t slow_clear_flags(void)
>> +static uint_fast8_t slow_clear_flags(void)
>>  {
>>      uint8_t prev = slowfloat_exceptionFlags;
>>
>> @@ -797,7 +797,7 @@ static uint8_t slow_clear_flags(void)
>>      return prev;
>>  }
>>
>> -static uint8_t qemu_clear_flags(void)
>> +static uint_fast8_t qemu_clear_flags(void)
>>  {
>>      uint8_t prev = qemu_flags_to_sf(qsf.float_exception_flags);
>
> Why are we using uint_fast8_t here anyway? We switched
> softfloat to using plain old uint8_t everywhere a while
> back.
>
> thanks
> -- PMM


--
Alex Bennée