testsuite: Add vect_perm3_int guards
authorRichard Sandiford <richard.sandiford@arm.com>
Wed, 18 Nov 2020 18:40:32 +0000 (18:40 +0000)
committerRichard Sandiford <richard.sandiford@arm.com>
Wed, 18 Nov 2020 18:40:32 +0000 (18:40 +0000)
commit6a9cd27ac6721ff378210c2851fb716c73f2a750
treece83d2824e6a0036cd3ee1ad6316eeb72076378d
parent6d0ffd9a3eaeac5bc71fe00049c9faf37a72b0fa
testsuite: Add vect_perm3_int guards

SLP vectorisation of gcc.dg/vect/fast-math-vect-call-1.c involves
a group of 3 floats, which requires the same permutation as
vect_perm3_int.

The load/store_lanes XFAILs in gcc.dg/vect/slp-perm-6.c implicitly
assumed vect_perm3_int, which is true for Advanced SIMD but not for
VLA SVE.  Whether it's true for fixed-length SVE depends on the
vector length.

The xfail selector applies on top of the target selector, so it's
not necessary to make the xfail selector a strict subset of the
target selector.

gcc/testsuite/
* gcc.dg/vect/fast-math-vect-call-1.c: Only expect SLP to be used
on vect_perm3_int targets.
* gcc.dg/vect/slp-perm-6.c: Likewise.  Only XFAIL the LOAD/STORE_LANES
tests on vect_perm3_int targets.
gcc/testsuite/gcc.dg/vect/fast-math-vect-call-1.c
gcc/testsuite/gcc.dg/vect/slp-perm-6.c