[Bug target/82518] [6/7/8 regression] gfortran.fortran-torture/execute/in-pack.f90 fails on armeb
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82518 --- Comment #52 from ktkachov at gcc dot gnu.org --- Author: ktkachov Date: Tue Mar 20 17:13:16 2018 New Revision: 258687 URL: https://gcc.gnu.org/viewcvs?rev=258687&root=gcc&view=rev Log: This PR shows that we get the load/store_lanes logic wrong for arm big-endian. It is tricky to get right. Aarch64 does it by adding the appropriate lane-swapping operations during expansion. I'd like to do the same on arm eventually, but we'd need to port and validate the VTBL-generating code and add it to all the right places and I'm not comfortable enough doing it for GCC 8, but I am keen in getting the wrong-code fixed. As I say in the PR, vectorisation on armeb is already severely restricted (we disable many patterns on BYTES_BIG_ENDIAN) and the load/store_lanes patterns really were not working properly at all, so disabling them is not a radical approach. The way to do that is to return false in ARRAY_MODE_SUPPORTED_P for BYTES_BIG_ENDIAN. Bootstrapped and tested on arm-none-linux-gnueabihf. Also tested on armeb-none-eabi. PR target/82518 * config/arm/arm.c (arm_array_mode_supported_p): Return false for BYTES_BIG_ENDIAN. * lib/target-supports.exp (check_effective_target_vect_load_lanes): Disable for armeb targets. * gcc.target/arm/pr82518.c: New test. Added: trunk/gcc/testsuite/gcc.target/arm/pr82518.c Modified: trunk/gcc/config/arm/arm.c trunk/gcc/testsuite/lib/target-supports.exp
[Bug target/82518] [6/7/8 regression] gfortran.fortran-torture/execute/in-pack.f90 fails on armeb
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82518 ktkachov at gcc dot gnu.org changed: What|Removed |Added Keywords||wrong-code Status|NEW |ASSIGNED Assignee|unassigned at gcc dot gnu.org |ktkachov at gcc dot gnu.org --- Comment #51 from ktkachov at gcc dot gnu.org --- Testing a simple patch along the lines described in comment #50
[Bug target/82518] [6/7/8 regression] gfortran.fortran-torture/execute/in-pack.f90 fails on armeb
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82518
ktkachov at gcc dot gnu.org changed:
What|Removed |Added
CC||ktkachov at gcc dot gnu.org
--- Comment #50 from ktkachov at gcc dot gnu.org ---
(In reply to Wilco from comment #49)
> AArch64 does this:
>
> (define_expand "vec_store_lanesoi"
> [(set (match_operand:OI 0 "aarch64_simd_struct_operand" "=Utv")
> (unspec:OI [(match_operand:OI 1 "register_operand" "w")
> (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
>UNSPEC_ST2))]
> "TARGET_SIMD"
> {
> if (BYTES_BIG_ENDIAN)
> {
> rtx tmp = gen_reg_rtx (OImode);
> rtx mask = aarch64_reverse_mask (mode, );
> emit_insn (gen_aarch64_rev_reglistoi (tmp, operands[1], mask));
> emit_insn (gen_aarch64_simd_st2 (operands[0], tmp));
> }
> else
> emit_insn (gen_aarch64_simd_st2 (operands[0], operands[1]));
> DONE;
> })
>
> ARM seems to be missing the swap:
>
> (define_expand "vec_store_lanesoi"
> [(set (match_operand:OI 0 "neon_struct_operand")
> (unspec:OI [(match_operand:OI 1 "s_register_operand")
> (unspec:VQ2 [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
>UNSPEC_VST2))]
> "TARGET_NEON")
>
> So clearly looks like a backend issue.
Indeed, and arm is missing the equivalent logic, including the reverse_mask,
rev_reglist etc.
For GCC 8 and the branches the least invasive fix would be to return false for
BYTES_BIG_ENDIAN in arm_array_mode_supported_p. That will disable the use of
the vec_load, vec_store lanes on big-endian. vectorisation on arm NEON is
already severely restricted (look at all the patterns in neon.md gated on
!BYTES_BIG_ENDIAN) and the vec_load/store_lanes has never worked correctly on
that target as far as I can see, so switching it off properly is not a radical
change.
At some point we'll want to take a holistic approach for NEON big-endian and
fix up (and document!) the lane-ordering everywhere, but the priority at this
stage is to fix the wrong-code in a not-too-invasive way.
[Bug target/82518] [6/7/8 regression] gfortran.fortran-torture/execute/in-pack.f90 fails on armeb
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82518 Wilco changed: What|Removed |Added Priority|P1 |P2 Component|tree-optimization |target Version|8.0 |6.4.1 Summary|[8 regression] |[6/7/8 regression] |gfortran.fortran-torture/ex |gfortran.fortran-torture/ex |ecute/in-pack.f90 fails on |ecute/in-pack.f90 fails on |armeb since r252917 |armeb Known to fail||6.4.1
