On Tue, Jun 3, 2025 at 10:43 PM H.J. Lu <hjl.to...@gmail.com> wrote:
>
> On Tue, Jun 3, 2025 at 10:51 AM Andrew Pinski <quic_apin...@quicinc.com>
> wrote:
> >
> > When we have a smallish CSWTCH, it could be placed in the rodata.cst16
> > section so it can be merged with other constants across TUs.
> >
> > The fix is simple; just mark the decl as mergable (DECL_MERGEABLE).
> > DECL_MERGEABLE was added with r14-1500-g4d935f52b0d5c0 specifically
> > to improve these kind of decls.
> >
> > PR tree-optimization/120451
> >
> > gcc/ChangeLog:
> >
> > * tree-switch-conversion.cc (switch_conversion::build_one_array):
> > Mark
> > the newly created decl as mergable.
> >
> > gcc/testsuite/ChangeLog:
> >
> > * gcc.dg/tree-ssa/cswtch-6.c: New test.
> >
> > Signed-off-by: Andrew Pinski <quic_apin...@quicinc.com>
> > ---
> > gcc/testsuite/gcc.dg/tree-ssa/cswtch-6.c | 43 ++++++++++++++++++++++++
> > gcc/tree-switch-conversion.cc | 3 ++
> > 2 files changed, 46 insertions(+)
> > create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/cswtch-6.c
> >
> > diff --git a/gcc/testsuite/gcc.dg/tree-ssa/cswtch-6.c
> > b/gcc/testsuite/gcc.dg/tree-ssa/cswtch-6.c
> > new file mode 100644
> > index 00000000000..d765a03aa19
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.dg/tree-ssa/cswtch-6.c
> > @@ -0,0 +1,43 @@
> > +/* PR tree-optimization/120451 */
> > +/* { dg-do compile { target elf } } */
> > +/* { dg-options "-O2" } */
> > +
> > +void foo (int, int);
> > +
> > +__attribute__((noinline, noclone)) void
> > +f1 (int v, int w)
> > +{
> > + int i, j;
> > + if (w)
> > + {
> > + i = 129;
> > + j = i - 1;
> > + goto lab;
> > + }
> > + switch (v)
> > + {
> > + case 170:
> > + j = 7;
> > + i = 27;
> > + break;
> > + case 171:
> > + i = 8;
> > + j = 122;
> > + break;
> > + case 172:
> > + i = 21;
> > + j = -19;
> > + break;
> > + case 173:
> > + i = 18;
> > + j = 17;
> > + break;
> > + default:
> > + __builtin_abort ();
> > + }
> > +
> > + lab:
> > + foo (i, j);
> > +}
> > +
> > +/* { dg-final { scan-assembler ".rodata.cst16" } } */
> > diff --git a/gcc/tree-switch-conversion.cc b/gcc/tree-switch-conversion.cc
> > index bd4de966892..d0882879e61 100644
> > --- a/gcc/tree-switch-conversion.cc
> > +++ b/gcc/tree-switch-conversion.cc
> > @@ -1030,6 +1030,9 @@ switch_conversion::build_one_array (int num, tree
> > arr_index_type,
> > TREE_CONSTANT (decl) = 1;
> > TREE_READONLY (decl) = 1;
> > DECL_IGNORED_P (decl) = 1;
> > + /* The decl is mergable since we don't take the address ever and
> > + just reading from it. */
> > + DECL_MERGEABLE (decl) = 1;
> > if (offloading_function_p (cfun->decl))
> > DECL_ATTRIBUTES (decl)
> > = tree_cons (get_identifier ("omp declare target"), NULL_TREE,
> > --
> > 2.43.0
> >
>
> On x86-64/Linux, I got
>
> spawn -ignore SIGHUP
> /export/build/gnu/tools-build/gcc-gitlab-test/build-x86_64-linux/gcc/xgcc
> -B/export/build/gnu/tools-build/gcc-gitlab-test/build-x86_64-linux/gcc/
> /export/gnu/import/git/gitlab/x86-gcc-test/gcc/testsuite/gcc.dg/tree-ssa/cswtch-6.c
> -m32 -fdiagnostics-plain-output -O2 -ffat-lto-objects -fno-ident -S -o
> cswtch-6.s
> PASS: gcc.dg/tree-ssa/cswtch-6.c (test for excess errors)
> FAIL: gcc.dg/tree-ssa/cswtch-6.c scan-assembler .rodata.cst16
Yes I know it fails there, I already filed
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120523 for it and I am
trying to figure out how the best way of improving the situtation and
not have it fail on anything besides x86_64(!ia32).
It is just a minor optimization you can ignore the testcase failure
for now. I will have a fix by the end of July (I am on
vacation/traveling the second half of June).
Thanks,
Andrew
>
> --
> H.J.
