Re: [PATCH 2/2] middle-end/126788 - more vector folding

On Sat, 15 Aug 2026, H.J. Lu wrote:

> On Fri, Aug 14, 2026 at 8:13 PM Richard Biener <[email protected]> wrote:
> >
> > The last match.pd update for double-conversion folding didn't handle
> > the case where a vector conversion would go away because
> > supportable_convert_operation doesn't consider a noop conversion
> > (or a sign conversion).  The following rectifies this, allowing
> > a NOP_EXPR and VIEW_CONVERT_EXPR for same mode types as supportable
> > conversion as we can RTL expand that just fine.
> >
> > This shows that the vectorizer routed sign-conversions through
> > vectorizable_assignment but now vectorizable_conversion would
> > handle it, emitting NOP_EXPRs (which is fine) instead of
> > VIEW_CONVERT_EXPRs (which we declared canonical for vectors).
> > This would confuse some foldings, leading to testsuite FAILs,
> > so make vectorizable_conversion also prefer VIEW_CONVERT_EXPRs
> > here.
> >
> > Bootstrapped and tested on x86_64-unknown-linux-gnu.
> >
> > I'll wait on comments on 1/2 before pushing.
> >
> > Richard.
> >
> >         PR middle-end/126788
> >         * optabs-tree.cc (supportable_convert_operation): For same
> >         modes allow NOP_EXPR and VIEW_CONVERT_EXPR.
> >         * tree-ssa-forwprop.cc (simplify_vector_constructor): Avoid
> >         converting the vector to be used for blending into the
> >         result to the type of the permutation.
> >         * tree-vect-stmts.cc (vectorizable_conversion): Prefer
> >         VIEW_CONVERT_EXPR for sign changes.
> >
> >         * gcc.target/i386/pr126788.c: New testcase.
> > ---
> >  gcc/optabs-tree.cc                       |  4 ++++
> >  gcc/testsuite/gcc.target/i386/pr126788.c | 24 ++++++++++++++++++++++++
> >  gcc/tree-ssa-forwprop.cc                 |  1 +
> >  gcc/tree-vect-stmts.cc                   |  9 ++++++++-
> >  4 files changed, 37 insertions(+), 1 deletion(-)
> >  create mode 100644 gcc/testsuite/gcc.target/i386/pr126788.c
> >
> 
> It failed with this:
> 
> [hjl@gnu-zen4-1 gcc]$ cat /tmp/x.C
> extern int _M_current;
> extern int end ();
> struct color
> {
>   char r;
>   struct
>   {
>     unsigned char g;
>     unsigned char b;
>   } m_24bit;
>   color (char g, char b)
>   {
>     m_24bit.g = g;
>     m_24bit.b = b;
>   }
> };
> void
> set_style_bg_color (color)
> {
>   for (;;)
>     {
>       const unsigned char g = _M_current;
>       if (end ())
> break;
>       set_style_bg_color (color (g, _M_current));
>     }
> }
> [hjl@gnu-zen4-1 gcc]$ ./xgcc -B./ -S -O2 -march=x86-64-v4 /tmp/x.C
> /tmp/x.C: In function ‘void set_style_bg_color(color)’:
> /tmp/x.C:18:1: error: type mismatch between an SSA_NAME and its symbol
>    18 | set_style_bg_color (color)
>       | ^~~~~~~~~~~~~~~~~~
> /tmp/x.C:18:1: error: type mismatch between an SSA_NAME and its symbol
> while verifying SSA_NAME vect_g_49.10_100 in statement
> vect_g_49.10_100 = (vector(2) unsigned char) _101;
> during GIMPLE pass: slp
> /tmp/x.C:18:1: internal compiler error: verify_ssa failed

Interesting.  This is because the vectorizer seems to preserve
qualifiers when building vector types for scalar types,
building const vector(2) unsigned char, but make_ssa_name (<type>)
will strip qualifiers from <type>.

I'll robustify this for now.


> 0x4fa8116 internal_error(char const*, ...)
> /export/gnu/import/git/gitlab/x86-gcc-test/gcc/diagnostic-global-context.cc:787
> 0x313c855 verify_ssa(bool, bool)
> /export/gnu/import/git/gitlab/x86-gcc-test/gcc/tree-ssa.cc:1226
> 0x2c64811 execute_function_todo
> /export/gnu/import/git/gitlab/x86-gcc-test/gcc/passes.cc:2107
> 0x2c6345a do_per_function
> /export/gnu/import/git/gitlab/x86-gcc-test/gcc/passes.cc:1699
> 0x2c649a8 execute_todo
> /export/gnu/import/git/gitlab/x86-gcc-test/gcc/passes.cc:2152
> ./cc1plus -quiet -iprefix
> /export/build/gnu/tools-build/gcc-gitlab-test-debug/build-x86_64-linux/gcc/../lib/gcc/x86_64-pc-linux-gnu/17.0.0/
> -isystem ./include -isystem ./include-fixed -D_GNU_SOURCE /tmp/x.C
> -quiet -dumpbase x.C -dumpbase-ext .C -march=x86-64-v4
> -mtls-dialect=gnu2 -O2 -o x.s
> Please submit a full bug report, with preprocessed source (by using
> -freport-bug).
> Please include the complete backtrace with any bug report.
> See <https://gcc.gnu.org/bugs/> for instructions.
> [hjl@gnu-zen4-1 gcc]$
> 
> 

-- 
Richard Biener <[email protected]>
SUSE Software Solutions Germany GmbH,
Frankenstrasse 146, 90461 Nuernberg, Germany;
GF: Jochen Jaser, Andrew McDonald, Abhinav Puri; (HRB 36809, AG Nuernberg)