On 08/01/2025 13:37, Christophe Lyon wrote:
> On Wed, 8 Jan 2025 at 14:17, Richard Earnshaw (lists)
> <richard.earns...@arm.com> wrote:
>>
>> On 19/12/2024 12:17, Christophe Lyon wrote:
>>> Without this patch, testcases using arm_v8_3a_fp16_complex_neon fail
>>> to compile on arm-linux-gnueabihf with
>>> fatal error: gnu/stubs-soft.h: No such file or directory
>>> because they are actually compiled with
>>> -mfloat-abi=softfp -mfpu=auto -mcpu=unset -march=armv8.3-a+fp16
>>>
>>> Fix this by including stdint.h in the sample code for the effective-target.
>>>
>>> This makes these tests PASS instead of being UNRESOLVED:
>>> fast-math-bb-slp-complex-add-half-float.c
>>> fast-math-bb-slp-complex-mla-half-float.c
>>> fast-math-bb-slp-complex-mls-half-float.c
>>> fast-math-bb-slp-complex-mul-half-float.c
>>> fast-math-complex-add-half-float.c
>>> fast-math-complex-mla-half-float.c
>>> fast-math-complex-mls-half-float.c
>>> fast-math-complex-mul-half-float.c
>>>
>>> except for two new
>>> FAIL: gcc.dg/vect/complex/fast-math-complex-mls-half-float.c scan-tree-dump
>>> vect "Found COMPLEX_ADD_ROT270"
>>> (and the same with -flto -ffat-lto-objects)
>>>
>>> gcc/testsuite/ChangeLog:
>>>
>>> * lib/target-supports.exp
>>> (check_effective_target_arm_v8_3a_fp16_complex_neon_ok_nocache):
>>> Include stdint.h.
>>> ---
>>> gcc/testsuite/lib/target-supports.exp | 1 +
>>> 1 file changed, 1 insertion(+)
>>>
>>> diff --git a/gcc/testsuite/lib/target-supports.exp
>>> b/gcc/testsuite/lib/target-supports.exp
>>> index a16e9534ccd..9f4e2700dd2 100644
>>> --- a/gcc/testsuite/lib/target-supports.exp
>>> +++ b/gcc/testsuite/lib/target-supports.exp
>>> @@ -13298,6 +13298,7 @@ proc
>>> check_effective_target_arm_v8_3a_fp16_complex_neon_ok_nocache { } {
>>> #if !defined (__ARM_FEATURE_COMPLEX)
>>> #error "__ARM_FEATURE_COMPLEX not defined"
>>> #endif
>>> + #include <stdint.h>
>>> } "$flags -mcpu=unset -march=armv8.3-a+fp16"] } {
>>> set et_arm_v8_3a_fp16_complex_neon_flags \
>>> "$flags -mcpu=unset -march=armv8.3-a+fp16"
>>
>> I'm somewhat surprised that a test in gcc.dg/vect/complex is using an
>> unguarded arm_... set of options. Tamar, I think this test is one of yours:
>> how is this supposed to work on non-arm platforms?
>>
>
> IIUC add_options_for_arm_v8_3a_fp16_complex_neon calls
> check_effective_target_arm_v8_3a_fp16_complex_neon_ok which returns 0
> on non-arm / non-aaarch64 platforms, thus keeps "$flags" unchanged.
>
> I was a bit surprised too :-)
>
> Many (not most) of the check_effective_target_arm* do NOT have the:
> if { ![istarget arm*-*-*] && ![istarget aarch64*-*-*] } {
> return 0;
> }
>
Euk! This is all quite messy. One of the things that is making testsuite
maintenance quite tricky is that we have many similarly named tests with subtly
different semantics. I think if we're to sort this out and regain some sanity
we're going to have to come up with some convention that makes things like this
more explicit in the naming.
R.
>
> Thanks,
>
> Christophe
>
>> R.
