On 14/03/2025 09:59, Richard Sandiford wrote:
"Andre Vieira (lists)" <andre.simoesdiasvie...@arm.com> writes:
diff --git a/gcc/testsuite/gcc.target/aarch64/no-sve-with-sme-3.c
b/gcc/testsuite/gcc.target/aarch64/no-sve-with-sme-3.c
new file mode 100644
index
0000000000000000000000000000000000000000..7a873fa2396a4129225bae8ac8797c86fac17261
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/no-sve-with-sme-3.c
@@ -0,0 +1,9 @@
+/* { dg-do compile } */
+/* { dg-skip-if "Do not override mcpu or march" { *-*-* } { -mcpu=* -march=* } {
"" } } */
+/* { dg-message "sorry, unimplemented: no support for 'sme' without 'sve2'" ""
{ target *-*-* } 0 } */
+
+int __attribute__ ((target( "arch=armv8.2-a+ssve-fp8fma"))) main (void)
+{
+ return 0;
+}
Is the skip necessary for these two tests? I would expect them to work
even with -march and -mcpu flags, given that the tests override the
architecture completely.
The tests look good with the skips removed.
No they don't, I just copied the 1 test that does need them, adding a
4th now to test -march=armv8-a with a pragma GCC target "+sme", that one
will need the skips.
diff --git a/gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/mlalb_lane_mf8.c
b/gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/mlalb_lane_mf8.c
index
39a272f60e0e0d6b8cd396ac1131f68352182670..f952af025efcc1865c3711cebbac4cae57c934a3
100644
--- a/gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/mlalb_lane_mf8.c
+++ b/gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/mlalb_lane_mf8.c
@@ -6,7 +6,7 @@
#pragma GCC target "+fp8fma"
#ifdef STREAMING_COMPATIBLE
-#pragma GCC target "+ssve-fp8fma"
+#pragma GCC target "+sve2+ssve-fp8fma"
Is the +sve2 necessary here? The test is in the sve2/ directory and
so I would expect sve2 to be enabled already.
Same for the other sve2/ tests.
Yeah that may have been overzealous processing of the testcases on my
part...
LGTM otherwise, with the change mentioned yesterday.
Thanks,
Richard
