[GCC][PATCH][Aarch64] Added pattern to match zero extended bfxil

[Kyrill Tkachov]

Hi all,

On 18/09/18 00:00, Ramana Radhakrishnan wrote:
> On Mon, 17 Sep 2018, 23:56 Christophe Lyon, <christophe.lyon@linaro.org>
> wrote:
>
> > On Fri, 14 Sep 2018 at 12:04, Sam Tebbs <sam.tebbs@arm.com> wrote:
> > >
> > >
> > >
> > > On 08/28/2018 11:54 PM, James Greenhalgh wrote:
> > >
> > > <snip>
> > > >
> > > > OK once the other one is approved, with the obvious rebase over the
> > renamed
> > > > function.
> > > >
> > > > James
> > >
> > > Here is the rebased patch. Still OK for me to commit to trunk now that
> > > the other patch has been committed?
> > >
> > > Sam
> > >
> > > >
> > > >> gcc/
> > > >> 2018-07-31  Sam Tebbs <sam.tebbs@arm.com>
> > > >>
> > > >>           PR target/85628
> > > >>           * config/aarch64/aarch64.md (*aarch64_bfxilsi_uxtw): Define.
> > > >>
> > > >> gcc/testsuite
> > > >> 2018-07-31  Sam Tebbs <sam.tebbs@arm.com>
> > > >>
> > > >>           PR target/85628
> > > >>           * gcc.target/aarch64/combine_bfxil.c
> > > >> (combine_zero_extended_int, foo6):
> > > >>           New functions.
> > >
> >
> > Hi Sam,
> >
> > This patch causes a regression when using -mabi=ilp32:
> > FAIL: gcc.target/aarch64/combine_bfxil.c scan-assembler-not uxtw\\t
> >
> > How much do we care about ilp32?
> >
>
>
> Lets keep the testsuite clean please.
>
>

This is due to the foo* functions in the test that take a pointer argument and store results into it.
In ILP32 the callee clears the top bits with a uxtw. This is irrelevant to the pattern this test is exercising.
I'll fix up the test.

Thanks,
Kyrill

> Ramana
>
> >
> > Christophe
> >