Re: [PATCH][ARM/AArch64] PR 68088: Fix RTL checking ICE due to subregs inside accumulator forwarding check

[Nikolai Bozhenov <n dot bozhenov at samsung dot com>]

On 10/28/2015 01:07 PM, Kyrill Tkachov wrote:
> Hi all,
>
> This RTL checking error occurs on aarch64 in 
> aarch_accumulator_forwarding when processing an msubsi insn
> with subregs:
> (insn 15 14 16 3 (set (reg/v:SI 78 [ i ])
>         (minus:SI (subreg:SI (reg/v:DI 76 [ aul ]) 0)
>             (mult:SI (subreg:SI (reg:DI 83) 0)
>                 (subreg:SI (reg:DI 75 [ _20 ]) 0)))) schedice.c:10 357 
> {*msubsi}
>
> The register_operand predicate for that pattern allows subregs (I 
> think correctly).
> The code in aarch_accumulator_forwarding doesn't take that into 
> account and ends up
> taking a REGNO of a SUBREG, causing a checking error.
>
> This patch fixes that by stripping the subregs off the accumulator rtx 
> before
> checking that the inner expression is a REG and taking its REGNO.
>
> The testcase now works fine with an aarch64-none-elf toolchain 
> configure for RTL checking.
>
> The testcase is taken verbatim from the BZ entry for PR 68088.
> Since this function is shared between arm and aarch64 I've 
> bootstrapped and tested it on both
> and I'll need ok's for both ports.
>
> Ok for trunk?
>
> Thanks,
> Kyrill
>
> 2015-10-28  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
>
>     PR target/68088
>     * config/arm/aarch-common.c (aarch_strip_subreg): New function.
>     (aarch_accumulator_forwarding): Strip subregs from accumulator rtx
>     when appropriate.
>
> 2015-10-28  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
>
>     * gcc.target/aarch64/pr68088_1.c: New test.

Hi!

I faced the same issue but I had somewhat different RTL for the consumer:

    (insn 20 15 21 2 (set (reg/i:SI 0 r0)
            (minus:SI (subreg:SI (reg:DI 117) 4)
                (mult:SI (reg:SI 123)
                    (reg:SI 114)))) gasman.c:4 48 {*mulsi3subsi})

where (reg:DI 117) is produced by umulsidi3_v6 instruction. Is it
really true that (subreg:SI (reg:DI 117) 4) may be forwarded in one
cycle in this case?

Thanks,
Nikolai