On Wed, Aug 12, 2026 at 11:01 AM Daniel Barboza
<[email protected]> wrote:
>
> We're not trying to fold builtin_strnlen calls into constants like we do
> with builtin_strlen. This results in missed optimizations where we
> have a strnlen that uses a PHI as input:
>
> const char a[4] = "123";
>
> int g (int i)
> {
> return __builtin_strnlen (i ? a : "", 4);
> }
>
> The strnlen call could be folded into a single PHI <0, 3> but instead
> we're calling strnlen with the PHI result:
>
> # iftmp.1_2 = PHI <&aD.4472(3), ""(2)>
> # VUSE <.MEM_4(D)>
> # RANGE [irange] long unsigned int [0, 4] MASK 0x7 VALUE 0x0
> # USE = nonlocal escaped const-pool { D.4472 } (nonlocal)
> _1 = strnlenD.1862 (iftmp.1_2, 4);
> ---
> gcc/fold-const-call.cc | 11 ++++++++++
> gcc/testsuite/gcc.dg/tree-ssa/pr86937.c | 15 +++++++++++++
> gcc/testsuite/gcc.dg/warn-strnlen-no-nul.c | 25 +++++++++++-----------
> 3 files changed, 39 insertions(+), 12 deletions(-)
> create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr86937.c
>
> diff --git a/gcc/fold-const-call.cc b/gcc/fold-const-call.cc
> index b3a9d7715d6..c483b80e704 100644
> --- a/gcc/fold-const-call.cc
> +++ b/gcc/fold-const-call.cc
> @@ -1911,6 +1911,17 @@ fold_const_call (combined_fn fn, tree type, tree arg0,
> tree arg1)
> }
> return NULL_TREE;
>
> + case CFN_BUILT_IN_STRNLEN:
> + if ((p0 = c_getstr (arg0)))
> + {
> + unsigned HOST_WIDE_INT s1 = 0;
> + if (!size_t_cst_p (arg1, &s1))
> + return NULL_TREE;
> +
> + return build_int_cst (type, strnlen (p0, s1));
> + }
> + return NULL_TREE;
> +
> case CFN_FOLD_LEFT_PLUS:
> return fold_const_fold_left (type, arg0, arg1, PLUS_EXPR);
>
> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr86937.c
> b/gcc/testsuite/gcc.dg/tree-ssa/pr86937.c
> new file mode 100644
> index 00000000000..fcd1c9d0eb7
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr86937.c
> @@ -0,0 +1,15 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -fdump-tree-pre" } */
> +const char a[4] = "123";
> +
> +int f (int i)
> +{
> + return __builtin_strlen (i ? a : "");
> +}
> +
> +int g (int i)
> +{
> + return __builtin_strnlen (i ? a : "", 4);
> +}
> +
> +/* { dg-final { scan-tree-dump-times "strnlen" 0 "pre" } } */
Please add a space after strnlen just in case it matches something
else. Basically there should be no calls left to strnlen. The dumps
are always `functionname (...)`.
> diff --git a/gcc/testsuite/gcc.dg/warn-strnlen-no-nul.c
> b/gcc/testsuite/gcc.dg/warn-strnlen-no-nul.c
> index 70f6a432b97..fde2fa0e189 100644
> --- a/gcc/testsuite/gcc.dg/warn-strnlen-no-nul.c
> +++ b/gcc/testsuite/gcc.dg/warn-strnlen-no-nul.c
> @@ -147,14 +147,14 @@ T (v0 ? b[3] : "", bsz);
> the strnlen calls are safe because the reads are bounded by
> the length of the constant arguments. Most of the calls are
> not diagnosed anymore as a result of the fix for PR 103215. */
> -T (v0 ? "" : b[0], bsz + 1); /* { dg-warning "bound 6 exceeds source
> size 5" } */
> +T (v0 ? "" : b[0], bsz + 1); /* { dg-warning "bound 6 exceeds source
> size 5" "pr86937" { xfail *-*-* } } */
So if I read the comment, I think the dg-warning can just be removed.
> T (v0 ? "" : b[1], bsz + 1);
> T (v0 ? "" : b[2], bsz + 1);
> -T (v0 ? "" : b[3], bsz + 1);
> -T (v0 ? b[0] : "", bsz + 1); /* { dg-warning "bound 6 exceeds source
> size 5" } */
Remove this warning rather than changing xfailing.
> +T (v0 ? "" : b[3], bsz + 1); /* { dg-warning "unterminated" } */
This new warning is correct as b[3] is "54321" in a array size of 5.
> +T (v0 ? b[0] : "", bsz + 1); /* { dg-warning "bound 6 exceeds source
> size 5" "pr86937" { xfail *-*-* } } */
> T (v0 ? b[1] : "", bsz + 1);
> T (v0 ? b[2] : "", bsz + 1);
> -T (v0 ? b[3] : "", bsz + 1);
> +T (v0 ? b[3] : "", bsz + 1); /* { dg-warning "unterminated" } */
Likewise.
>
> T (v0 ? "" : b[i0], bsz);
> T (v0 ? "" : b[i1], bsz);
> @@ -168,11 +168,11 @@ T (v0 ? b[i3] : "", bsz);
> T (v0 ? "" : b[i0], bsz + 1);
> T (v0 ? "" : b[i1], bsz + 1);
> T (v0 ? "" : b[i2], bsz + 1);
> -T (v0 ? "" : b[i3], bsz + 1);
> +T (v0 ? "" : b[i3], bsz + 1); /* { dg-warning "unterminated" } */
Likewise.
> T (v0 ? b[i0] : "", bsz + 1);
> T (v0 ? b[i1] : "", bsz + 1);
> T (v0 ? b[i2] : "", bsz + 1);
> -T (v0 ? b[i3] : "", bsz + 1);
> +T (v0 ? b[i3] : "", bsz + 1); /* { dg-warning "unterminated" } */
Likewise.
>
> T (v0 ? "1234" : b[3], bsz);
> T (v0 ? "1234" : b[i3], bsz);
> @@ -184,17 +184,18 @@ T (v0 ? b[0] : b[2], bsz);
> T (v0 ? b[2] : b[3], bsz);
> T (v0 ? b[3] : b[2], bsz);
>
> -T (v0 ? "1234" : b[3], bsz + 1);
> -T (v0 ? "1234" : b[i3], bsz + 1);
> -T (v0 ? b[3] : "1234", bsz + 1);
> -T (v0 ? b[i3] : "1234", bsz + 1);
> +/* New warnings being thrown after PR86937. */
> +T (v0 ? "1234" : b[3], bsz + 1); /* { dg-warning "unterminated" } */
> +T (v0 ? "1234" : b[i3], bsz + 1); /* { dg-warning "unterminated" } */
> +T (v0 ? b[3] : "1234", bsz + 1); /* { dg-warning "unterminated" } */
> +T (v0 ? b[i3] : "1234", bsz + 1); /* { dg-warning "unterminated" } */
These are all valid warnings.
>
> /* That the following are not diagnosed is a bug/limitation resulting from
> the fix for PR 103215. */
> T (v0 ? a : b[3], bsz + 1); /* { dg-warning "bound 6 exceeds source
> size 5" "pr103215" { xfail *-*-* } } */
> T (v0 ? b[0] : b[2], bsz + 1); /* { dg-warning "bound 6 exceeds source
> size 5" "pr103215" { xfail *-*-* } } */
> -T (v0 ? b[2] : b[3], bsz + 1); /* { dg-warning "bound 6 exceeds source
> size 5" "pr103215" { xfail *-*-* } } */
> -T (v0 ? b[3] : b[2], bsz + 1); /* { dg-warning "bound 6 exceeds source
> size 5" "pr103215" { xfail *-*-* } } */
> +T (v0 ? b[2] : b[3], bsz + 1); /* { dg-warning "unterminated" } */
> +T (v0 ? b[3] : b[2], bsz + 1); /* { dg-warning "unterminated" } */
Yes the same new valid warning b[3] is unterminated.
So ok with the removal of the newly xfailed dg-warning since those
warnings won't show up ever since the values are all calculated at
compile time.
Thanks,
Andrea
>
> struct A { char a[5], b[5]; };
>
> --
> 2.43.0
>
