On Thu, Nov 13, 2025 at 9:57 AM Uros Bizjak <[email protected]> wrote:
>
> On Thu, Nov 13, 2025 at 7:22 AM liuhongt <[email protected]> wrote:
> >
> > For instruction sequence like
> > kmovb %k0, %edx
> > kmovb %k1, %ecx
> > orb %cl, %dl
> > je .L5
> >
> > if only CCZ is cared, it can be optimized to
> >
> > kortestb %k1, %k0
> > je .L5
> >
> > Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32}.
> > Ready push to trunk.
> >
> > gcc/ChangeLog:
> >
> > * config/i386/i386.md (*ior<mode>_ccz_1): New define_insn.
> >
> > gcc/testsuite/ChangeLog:
> >
> > * gcc.target/i386/kortest_ccz-1.c: New test.
> > ---
> > gcc/config/i386/i386.md | 16 ++++++++++++++++
> > gcc/testsuite/gcc.target/i386/kortest_ccz-1.c | 13 +++++++++++++
> > 2 files changed, 29 insertions(+)
> > create mode 100644 gcc/testsuite/gcc.target/i386/kortest_ccz-1.c
> >
> > diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
> > index 3ea2439526b..ae26ef02c22 100644
> > --- a/gcc/config/i386/i386.md
> > +++ b/gcc/config/i386/i386.md
> > @@ -14203,6 +14203,22 @@ (define_insn "*<code>si_2_zext_imm"
> > (set_attr "isa" "*,apx_ndd")
> > (set_attr "mode" "SI")])
> >
> > +;; It must be put before *<code><mode>_3, the blow one.
> > +(define_insn "*ior<mode>_ccz_1"
> > + [(set (reg:CCZ FLAGS_REG)
> > + (compare:CCZ
> > + (ior:SWI1248_AVX512BWDQ_64
> > + (match_operand:SWI1248_AVX512BWDQ_64 1 "nonimmediate_operand"
> > "%0,?k")
> > + (match_operand:SWI1248_AVX512BWDQ_64 2 "<general_operand>" "<g>,
> > k"))
> > + (const_int 0)))
> > + (clobber (match_scratch:SWI1248_AVX512BWDQ_64 0 "=<r>, k"))]
>
> You can use 'X' constraint instead of 'k' here, so "<g>, X". The
> capital 'X' will signal the register allocator that no register is
> actually needed for output. Please see "extendsidi2_1" or
> "extendditi2" in i386.md for some examples.
This is even mentioned in the gcc internals documentation, 16.9.1
Simple Constraints:
‘X’
Any operand whatsoever is allowed, even if it does not satisfy
‘general_operand’. This is normally used in the constraint of a
‘match_scratch’ when certain alternatives will not actually require
a scratch register.
Uros.
