Yury Khrustalev <[email protected]> writes:
> From: Szabolcs Nagy <[email protected]>
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/aarch64/gcspopm-1.c: New test.
> * gcc.target/aarch64/gcspr-1.c: New test.
> * gcc.target/aarch64/gcsss-1.c: New test.
> ---
> gcc/testsuite/gcc.target/aarch64/gcspopm-1.c | 69 ++++++++++++++++++++
> gcc/testsuite/gcc.target/aarch64/gcspr-1.c | 31 +++++++++
> gcc/testsuite/gcc.target/aarch64/gcsss-1.c | 49 ++++++++++++++
> 3 files changed, 149 insertions(+)
> create mode 100644 gcc/testsuite/gcc.target/aarch64/gcspopm-1.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/gcspr-1.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/gcsss-1.c
>
> diff --git a/gcc/testsuite/gcc.target/aarch64/gcspopm-1.c
> b/gcc/testsuite/gcc.target/aarch64/gcspopm-1.c
> new file mode 100644
> index 00000000000..6e6add39cf7
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/gcspopm-1.c
> @@ -0,0 +1,69 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mbranch-protection=none" } */
> +/* { dg-final { check-function-bodies "**" "" "" } } */
> +
> +/*
> +**foo1:
> +** sysl xzr, #3, c7, c7, #1 // gcspopm
> +** ret
> +*/
> +void
> +foo1 (void)
> +{
> + __builtin_aarch64_gcspopm ();
> +}
> +
> +/*
> +**foo2:
> +** mov x0, 0
> +** sysl x0, #3, c7, c7, #1 // gcspopm
> +** ret
> +*/
> +unsigned long long
> +foo2 (void)
> +{
> + return __builtin_aarch64_gcspopm ();
> +}
> +
> +/*
> +**foo3:
> +** mov x16, 1
> +** (
> +** mov x0, 0
> +** hint 40 // chkfeat x16
> +** |
> +** hint 40 // chkfeat x16
> +** mov x0, 0
> +** )
The mov could also happen first, before the mov x16, 1. It would
probably be easier to use...
> +** cbz x16, .*
> +** ret
> +** mov x0, 0
> +** sysl x0, #3, c7, c7, #1 // gcspopm
> +** ret
> +*/
> +unsigned long long
> +foo3 (void)
> +{
> + if (__builtin_aarch64_chkfeat (1) == 0)
> + return __builtin_aarch64_gcspopm ();
> + return 0;
> +}
unsigned long long
foo3 (unsigned long long x)
{
if (__builtin_aarch64_chkfeat (1) == 0)
return __builtin_aarch64_gcspopm ();
return x;
}
so that x0 is returned unmodified if the chkfeat returns nonzero.
FWIW, if we do remove the embedded moves from the .md define_insns,
we should also be able to get rid of the redundant zeroing of x0
on the gcspopm path.
> +
> +/*
> +**foo4:
> +** sysl xzr, #3, c7, c7, #1 // gcspopm
> +** mov x0, 0
> +** sysl x0, #3, c7, c7, #1 // gcspopm
> +** sysl xzr, #3, c7, c7, #1 // gcspopm
> +** ret
> +*/
> +unsigned long long
> +foo4 (void)
> +{
> + unsigned long long a = __builtin_aarch64_gcspopm ();
> + unsigned long long b = __builtin_aarch64_gcspopm ();
> + unsigned long long c = __builtin_aarch64_gcspopm ();
> + (void) a;
> + (void) c;
> + return b;
> +}
Nice test :)
> diff --git a/gcc/testsuite/gcc.target/aarch64/gcspr-1.c
> b/gcc/testsuite/gcc.target/aarch64/gcspr-1.c
> new file mode 100644
> index 00000000000..0e651979551
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/gcspr-1.c
> @@ -0,0 +1,31 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mbranch-protection=none" } */
> +/* { dg-final { check-function-bodies "**" "" "" } } */
> +
> +/*
> +**foo1:
> +** mrs x0, s3_3_c2_c5_1 // gcspr_el0
> +** ret
> +*/
> +void *
> +foo1 (void)
> +{
> + return __builtin_aarch64_gcspr ();
> +}
> +
> +/*
> +**foo2:
> +** mrs x[0-9]*, s3_3_c2_c5_1 // gcspr_el0
> +** sysl xzr, #3, c7, c7, #1 // gcspopm
> +** mrs x[0-9]*, s3_3_c2_c5_1 // gcspr_el0
> +** sub x0, x[0-9]*, x[0-9]*
> +** ret
> +*/
> +long
> +foo2 (void)
> +{
> + const char *p = __builtin_aarch64_gcspr ();
> + __builtin_aarch64_gcspopm ();
> + const char *q = __builtin_aarch64_gcspr ();
> + return p - q;
> +}
> diff --git a/gcc/testsuite/gcc.target/aarch64/gcsss-1.c
> b/gcc/testsuite/gcc.target/aarch64/gcsss-1.c
> new file mode 100644
> index 00000000000..025c7fee647
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/gcsss-1.c
> @@ -0,0 +1,49 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mbranch-protection=none" } */
> +/* { dg-final { check-function-bodies "**" "" "" } } */
> +
> +/*
> +**foo1:
> +** sys #3, c7, c7, #2, x0 // gcsss1
> +** mov x[0-9]*, 0
> +** sysl x[0-9]*, #3, c7, c7, #3 // gcsss2
Might as well make this:
** mov (x[0-9]+), 0
** sysl \1, #3, c7, c7, #3 // gcsss2
> +** ret
> +*/
> +void
> +foo1 (void *p)
> +{
> + __builtin_aarch64_gcsss (p);
> +}
> +
> +/*
> +**foo2:
> +** sys #3, c7, c7, #2, x0 // gcsss1
> +** mov x0, 0
> +** sysl x0, #3, c7, c7, #3 // gcsss2
> +** ret
> +*/
> +void *
> +foo2 (void *p)
> +{
> + return __builtin_aarch64_gcsss (p);
> +}
> +
> +/*
> +**foo3:
> +** mov x16, 1
> +** hint 40 // chkfeat x16
> +** cbnz x16, .*
> +** sys #3, c7, c7, #2, x0 // gcsss1
> +** mov x0, 0
> +** sysl x0, #3, c7, c7, #3 // gcsss2
> +** ret
> +** mov x0, 0
> +** ret
> +*/
> +void *
> +foo3 (void *p)
> +{
> + if (__builtin_aarch64_chkfeat (1) == 0)
> + return __builtin_aarch64_gcsss (p);
> + return 0;
> +}
Similarly to the above, it would probably be good to return p instead
of 0 here, to avoid awkwardness with the positioning of the return 0 path.
Thanks,
Richard
