On 2026-04-02 14:51, Christophe Lyon wrote:
Hi,
On 4/2/26 13:33, Torbjörn SVENSSON wrote:
I don't know if there is any better option, but -mgeneral-regs-only makes the
test pass for Cortex-M55/M85 in my tests.
As the problem is not with the core, but rather with MVE, I decided to hook
this on `arm_mve`.
Ok for trunk and releases/gcc-15?
Looking at history and context, I came up with another way of fixing this:
https://gcc.gnu.org/pipermail/gcc-patches/2026-April/712121.html
Thoughts?
Either way is fine for me, but I think your solution has the advantage of
treating MVE and NEON similar.
Kind regards,
Torbjörn
Thanks,
Christophe
--
Test is supposed to verify that gcc is not vectorizing without target
support. By disabling FP registers, vectoriziaition is avoided.
gcc/testsuite/ChangeLog:
PR target/124704
* gcc.dg/tree-ssa/gen-vect-11b.c: Add -mgeneral-regs-only for
arm_mve targets.
Signed-off-by: Torbjörn SVENSSON <[email protected]>
Co-authored-by: Christophe Lyon <[email protected]>
---
gcc/testsuite/gcc.dg/tree-ssa/gen-vect-11b.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/gen-vect-11b.c
b/gcc/testsuite/gcc.dg/tree-ssa/gen-vect-11b.c
index d0028bd70db..43a58888048 100644
--- a/gcc/testsuite/gcc.dg/tree-ssa/gen-vect-11b.c
+++ b/gcc/testsuite/gcc.dg/tree-ssa/gen-vect-11b.c
@@ -1,5 +1,6 @@
/* { dg-do run { target vect_cmdline_needed } } */
/* { dg-options "-O2 -ftree-vectorize -fdump-tree-vect-details" } */
+/* { dg-options "-O2 -ftree-vectorize -fdump-tree-vect-details
-mgeneral-regs-only" { target { arm_mve } } } */
/* { dg-options "-O2 -ftree-vectorize -fdump-tree-vect-details -mno-vx" {
target { s390*-*-* } } } */
/* { dg-options "-O2 -ftree-vectorize -fdump-tree-vect-details -mno-sse" {
target { i?86-*-* x86_64-*-* } } } */
