On 20/01/2020 14:53, Christophe Lyon wrote:
> On Thu, 9 Jan 2020 at 16:53, Matthew Malcomson
> <matthew.malcom...@arm.com> wrote:
>>
>> + (match_test "aarch64_sve_ld1ro_operand_p (op, DImode)")))
>> +
>
>
>> (define_predicate "aarch64_sve_ldff1_operand"
>> (and (match_code "mem")
>
>> (match_test "aarch64_sve_ldff1_operand_p (op)")))
>
>
> Hi,
>
>> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/acle/asm/ld1ro_f16.c
>> b/gcc/testsuite/gcc.target/aarch64/sve/acle/asm/ld1ro_f16.c
>> new file mode 100644
>> index
>> 0000000000000000000000000000000000000000..7badc75a43ab2009e9406afc04c980fc01834716
>> --- /dev/null
>> +++ b/gcc/testsuite/gcc.target/aarch64/sve/acle/asm/ld1ro_f16.c
>> @@ -0,0 +1,119 @@
>> +/* { dg-final { check-function-bodies "**" "" "-DCHECK_ASM" } } */
>> +/* { dg-additional-options "-march=armv8.6-a+sve+f64mm" } */
>
> What is the binutils version requirement for this?
> Some validations using binutils-2.33.1 exhibit failures like:
> /xgcc
> -B/aci-gcc-fsf/builds/gcc-fsf-gccsrc/obj-aarch64-none-linux-gnu/gcc3/gcc/
> -fno-diagnostics-show-caret -fno-diagnostics-show-line-numbers
> -fdiagnostics-color=never -fdiagnostics-urls=never -std=c90 -O0 -g
> -DTEST_FULL -march=armv8.2-a+sve -fno-ipa-icf
> -march=armv8.6-a+sve+f64mm -c -o ld1ro_s16.o
> /gcc/testsuite/gcc.target/aarch64/sve/acle/asm/ld1ro_s16.c
> Assembler messages:
> Error: unknown architecture `armv8.6-a+sve+f64mm'
>
> Error: unrecognized option -march=armv8.6-a+sve+f64mm
> compiler exited with status 1
> FAIL: gcc.target/aarch64/sve/acle/asm/ld1ro_s16.c -std=c90 -O0 -g
> -DTEST_FULL 1 blank line(s) in output
> FAIL: gcc.target/aarch64/sve/acle/asm/ld1ro_s16.c -std=c90 -O0 -g
> -DTEST_FULL (test for excess errors)
> Excess errors:
> Assembler messages:
> Error: unknown architecture `armv8.6-a+sve+f64mm'
> Error: unrecognized option -march=armv8.6-a+sve+f64mm
>
>
> while other configurations using 2.32 binutils seem to pass this test:
> /xgcc
> -B/home/tcwg-buildslave/workspace/tcwg-buildfarm__0/_build/builds/aarch64-unknown-linux-gnu/aarch64-unknown-linux-gnu/gcc.git~master_rev_3684bbb022cd75da55e1457673f269980aa12cdf-stage2/gcc/
> /home/tcwg-buildslave/workspace/tcwg-buildfarm__0/snapshots/gcc.git~master_rev_3684bbb022cd75da55e1457673f269980aa12cdf/gcc/testsuite/gcc.target/aarch64/sve/acle/asm/ld1ro_f16.c
> -fno-diagnostics-show-caret -fno-diagnostics-show-line-numbers
> -fdiagnostics-color=never -fdiagnostics-urls=never -std=c90 -O0 -g
> -DTEST_FULL -march=armv8.2-a+sve -fno-ipa-icf
> -march=armv8.6-a+sve+f64mm -S -o ld1ro_f16.s
> PASS: gcc.target/aarch64/sve/acle/asm/ld1ro_f16.c -std=c90 -O0 -g
> -DTEST_FULL (test for excess errors)
>
> Ha... took me a while to realize that in the latter case we stop after
> generating the .s file and do not call the assembler...
>
>
> So... do we want/need additional consistency checks between gcc and
> gas versions?
>
Ah! Yes, that should certainly be done.
I'll start that now.
Thanks for pointing it out -- it seems I really did not take enough care
with this patch...
MM
>
> Thanks,
>
> Christophe
>
>
>> +
