On 9/21/26 12:51 PM, Dominic P wrote:
Four of these tests ask for vect_int from gcc.dg/tree-ssa, and the fifth,
minmax-swar-5.c, needs the same guard. The flags those selectors are
answered under come from vect.exp, which is also what adds -msse2,
-maltivec or the NEON options and what skips a target that cannot
vectorize at all. Move all five there, keep the requirement, and drop the
ia32 option line vect.exp supplies itself.
The rules ask target_supports_op_p for the vector MIN_EXPR, so the vector
spelling folds only where the target has one: on cortex-m0 all six
functions of minmax-swar-5.c stop at VEC_COND_EXPR. The two tests that
assume a 16-byte vector gate their scans on vect128, and the widths come
from sizeof.
Tested on x86_64-pc-linux-gnu and aarch64-linux-gnu, and with an
arm-none-eabi cross on cortex-a15 NEON and cortex-m0 soft.
Assisted-by: Claude Opus 5 (Anthropic)
gcc/testsuite/ChangeLog:
* gcc.dg/tree-ssa/minmax-swar-5.c: Move to...
* gcc.dg/vect/minmax-swar-5.c: ...here. Require vect_int, take the
flags from vect.exp, size the vector with sizeof and gate the scans.
* gcc.dg/tree-ssa/minmax-swar-6.c: Move to...
* gcc.dg/vect/minmax-swar-6.c: ...here. Take the flags from vect.exp.
* gcc.dg/tree-ssa/minmax-swar-7.c: Move to...
* gcc.dg/vect/minmax-swar-7.c: ...here. Likewise.
* gcc.dg/tree-ssa/minmax-negate-3.c: Move to...
* gcc.dg/vect/minmax-negate-3.c: ...here. Likewise.
* gcc.dg/tree-ssa/vector-alignup-1.c: Move to...
* gcc.dg/vect/vector-alignup-1.c: ...here. Likewise, size the
vectors with sizeof and gate the scans.
Signed-off-by: Dominic P <[email protected]>
OK
jeff
