> Actually I'm testing a patch to simplify the function down to just > return gimple_convert (gsi, true, GSI_SAME_STMT, loc, type, val); > which should perform the optimization by means of match.pd simplifications.
I see, thanks Richard. Do you prefer just go gimple_convert way, or commit this one first to fix ICE? Both are Ok to me. Pan -----Original Message----- From: Richard Biener <[email protected]> Sent: Friday, October 10, 2025 4:04 PM To: Li, Pan2 <[email protected]> Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Chen, Ken <[email protected]>; Liu, Hongtao <[email protected]>; [email protected]; [email protected] Subject: Re: [PATCH v1] Widen-Mul: Fix ICE when cast pointer/reference of build_and_insert_cast On Fri, Oct 10, 2025 at 9:57 AM Richard Biener <[email protected]> wrote: > > On Thu, Oct 9, 2025 at 9:37 AM <[email protected]> wrote: > > > > From: Pan Li <[email protected]> > > > > When the val of build_and_insert_cast comes from a cast already, we > > can refine it to the val casted from to avoid duplicated cast on > > condition. However, it is valid only for int types, other types > > like pointer/reference is invalid as it at least requires the > > type precision is totoally the same. > > > > The below test suites are passed for this patch: > > 1. The x86 bootstrap tests. > > 2. The x86 fully regression tests. > > 3. The pr122111.c test of arm. > > OK. Actually I'm testing a patch to simplify the function down to just return gimple_convert (gsi, true, GSI_SAME_STMT, loc, type, val); which should perform the optimization by means of match.pd simplifications. Richard. > > PR middel-end/122111 > > > > gcc/ChangeLog: > > > > * tree-ssa-math-opts.cc (build_and_insert_cast): Add the > > INTEGRAL_TYPE_P check for the cast val. > > > > gcc/testsuite/ChangeLog: > > > > * gcc.target/arm/pr122111.c: New test. > > > > Signed-off-by: Pan Li <[email protected]> > > --- > > gcc/testsuite/gcc.target/arm/pr122111.c | 14 ++++++++++++++ > > gcc/tree-ssa-math-opts.cc | 3 ++- > > 2 files changed, 16 insertions(+), 1 deletion(-) > > create mode 100644 gcc/testsuite/gcc.target/arm/pr122111.c > > > > diff --git a/gcc/testsuite/gcc.target/arm/pr122111.c > > b/gcc/testsuite/gcc.target/arm/pr122111.c > > new file mode 100644 > > index 00000000000..82d15e9a43b > > --- /dev/null > > +++ b/gcc/testsuite/gcc.target/arm/pr122111.c > > @@ -0,0 +1,14 @@ > > +/* Test that we do not have ice when compile */ > > +/* { dg-do compile } */ > > +/* { dg-options "-O3" } */ > > + > > +typedef long long wide; > > + > > +void g(void); > > +wide f(int *a, unsigned t) > > +{ > > + wide i = t; > > + i *= 4; > > + unsigned long ai = (__SIZE_TYPE__)a; > > + return i + ai; > > +} > > diff --git a/gcc/tree-ssa-math-opts.cc b/gcc/tree-ssa-math-opts.cc > > index 0db39f330ea..0bc4ea32f65 100644 > > --- a/gcc/tree-ssa-math-opts.cc > > +++ b/gcc/tree-ssa-math-opts.cc > > @@ -1639,7 +1639,8 @@ build_and_insert_cast (gimple_stmt_iterator *gsi, > > location_t loc, > > gimple *def = SSA_NAME_DEF_STMT (val); > > > > if (is_gimple_assign (def) > > - && CONVERT_EXPR_CODE_P (gimple_assign_rhs_code (def))) > > + && CONVERT_EXPR_CODE_P (gimple_assign_rhs_code (def)) > > + && INTEGRAL_TYPE_P (TREE_TYPE (gimple_assign_rhs1 (def)))) > > { > > tree cast_rhs = gimple_assign_rhs1 (def); > > tree cast_rhs_type = TREE_TYPE (cast_rhs); > > -- > > 2.43.0 > >
