On 07/10/2026 15:10, Christophe Lyon wrote:
> The test checks that we don't accidentally vectorize when fp16 is NOT
> enabled, so the test would fail when using
> e.g. -march=armv8.1-m.main+mve.fp.
>
> Introduce arm_mve_fp effective-target, similar to arm_mve, but check
> if MVE floating-point is supported.
>
> Skip the test in that case.
>
> Tested on:
> -mthumb/-march=armv8.1-m.main+mve.fp+fp.dp/-mtune=cortex-m55/-mfloat-abi=hard/-mfpu=auto
>
> gcc/ChangeLog:
>
> * doc/sourcebuild.texi: Document new arm_mve_fp effective-target.
>
> gcc/testsuite/ChangeLog:
>
> * lib/target-supports.exp: Add arm_mve_fp effective-target.
> * gcc.target/arm/vect-half-floats.c: Add dg-skip-if for
> arm_mve_fp.
> ---
> gcc/doc/sourcebuild.texi | 5 +++++
> gcc/testsuite/gcc.target/arm/vect-half-floats.c | 1 +
> gcc/testsuite/lib/target-supports.exp | 16 ++++++++++++++++
> 3 files changed, 22 insertions(+)
>
> diff --git a/gcc/doc/sourcebuild.texi b/gcc/doc/sourcebuild.texi
> index 3aa9fcbb577..9302259960a 100644
> --- a/gcc/doc/sourcebuild.texi
> +++ b/gcc/doc/sourcebuild.texi
> @@ -2360,6 +2360,11 @@ ARM target supports the @code{-mfloat-abi=hard} option.
> @anchor{arm_mve}
> ARM target supports generating MVE instructions.
>
> +@item arm_mve_fp
> +@anchor{arm_mve_fp}
> +ARM target supports generating MVE instructions with floating-point
> +support.
> +
> @item arm_v8_1_lob_ok
> @anchor{arm_v8_1_lob_ok}
> ARM Target supports executing the Armv8.1-M Mainline Low Overhead Loop
> diff --git a/gcc/testsuite/gcc.target/arm/vect-half-floats.c
> b/gcc/testsuite/gcc.target/arm/vect-half-floats.c
> index 49ca2e9b083..dc14688e9c1 100644
> --- a/gcc/testsuite/gcc.target/arm/vect-half-floats.c
> +++ b/gcc/testsuite/gcc.target/arm/vect-half-floats.c
> @@ -1,4 +1,5 @@
> /* { dg-do compile } */
> +/* { dg-skip-if "mve.fp supports fp16 vectorization" { arm_mve_fp } } */
This doesn't feel right. We already test float16 and arm_fp16_ok below, but
shouldn't the check be a generic arm_fp16_vect that handles both MVE and neon
variants that also support fp16 vectorized operations?
R.
> /* { dg-options "-Ofast -ftree-vectorize -fdump-tree-vect-all -std=c11" } */
> /* { dg-require-effective-target float16 } */
> /* { dg-require-effective-target arm_fp16_ok } */
> diff --git a/gcc/testsuite/lib/target-supports.exp
> b/gcc/testsuite/lib/target-supports.exp
> index ae04b6a8daa..803376006ae 100644
> --- a/gcc/testsuite/lib/target-supports.exp
> +++ b/gcc/testsuite/lib/target-supports.exp
> @@ -7019,6 +7019,22 @@ proc check_effective_target_arm_mve { } {
> }]
> }
>
> +# Return 1 if this is an ARM target supporting MVE with floating-point
> +# instructions.
> +proc check_effective_target_arm_mve_fp { } {
> + if { ![istarget arm*-*-*] } {
> + return 0
> + }
> + return [check_no_compiler_messages arm_mve_fp assembly {
> + #if !defined (__ARM_FEATURE_MVE)
> + #error "__ARM_FEATURE_MVE not defined"
> + #endif
> + #if !(__ARM_FEATURE_MVE & 2)
> + #error "__ARM_FEATURE_MVE for floating point not defined"
> + #endif
> + }]
> +}
> +
> # Return 1 if the target supports ARMv8.1-M MVE with floating point
> # instructions, 0 otherwise. The test is valid for ARM.
> # Record the command line options needed.
