On Wed, May 14, 2025 at 4:06 PM Spencer Abson <spencer.ab...@arm.com> wrote:
>
> Floating-point to integer conversions can be inexact or invalid (e.g., due to
> overflow or NaN). However, since users of operation_could_trap_p infer the
> bool FP_OPERATION argument from the expression's type, FIX_TRUNC_EXPR is
> considered non-trapping here.
>
> This patch handles FIX_TRUNC_EXPR explicitly.
>
> gcc/ChangeLog:
>
> * tree-eh.cc (operation_could_trap_helper_p): Cover FIX_TRUNC_EXPR
> explicitly.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.dg/tree-ssa/ifcvt-fix-trunc-1.c: New test.
> * gcc.dg/tree-ssa/ifcvt-fix-trunc-2.c: New test.
> ---
> .../gcc.dg/tree-ssa/ifcvt-fix-trunc-1.c | 18 ++++++++++++++++++
> .../gcc.dg/tree-ssa/ifcvt-fix-trunc-2.c | 6 ++++++
> gcc/tree-eh.cc | 5 +++++
> 3 files changed, 29 insertions(+)
> create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/ifcvt-fix-trunc-1.c
> create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/ifcvt-fix-trunc-2.c
>
> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/ifcvt-fix-trunc-1.c
> b/gcc/testsuite/gcc.dg/tree-ssa/ifcvt-fix-trunc-1.c
> new file mode 100644
> index 00000000000..c65441ad679
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/tree-ssa/ifcvt-fix-trunc-1.c
> @@ -0,0 +1,18 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -ftree-vectorize -fdump-tree-ifcvt-stats" } */
> +
> +void test (int *dst, float *arr, int *pred, int n)
> +{
> + for (int i = 0; i < n; i++)
> + {
> + int pred_i = pred[i];
> + float arr_i = arr[i];
> +
> + dst[i] = pred_i ? (int)arr_i : 5;
> + }
> +}
> +
> +/* We expect this to fail if_convertible_loop_p so long as we have no
> + conditional IFN for FIX_TRUNC_EXPR. */
> +
> +/* { dg-final { scan-tree-dump-times "Applying if-conversion" 0 "ifcvt" } }
> */
> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/ifcvt-fix-trunc-2.c
> b/gcc/testsuite/gcc.dg/tree-ssa/ifcvt-fix-trunc-2.c
> new file mode 100644
> index 00000000000..628b754e94d
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/tree-ssa/ifcvt-fix-trunc-2.c
> @@ -0,0 +1,6 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -ftree-vectorize -fno-trapping-math
> -fdump-tree-ifcvt-stats" } */
> +
> +#include "ifcvt-fix-trunc-1.c"
> +
> +/* { dg-final { scan-tree-dump-times "Applying if-conversion" 1 "ifcvt" } }
> */
> diff --git a/gcc/tree-eh.cc b/gcc/tree-eh.cc
> index a4d59954c05..154e8fafb12 100644
> --- a/gcc/tree-eh.cc
> +++ b/gcc/tree-eh.cc
> @@ -2538,6 +2538,11 @@ operation_could_trap_helper_p (enum tree_code op,
> /* Constructing an object cannot trap. */
> return false;
>
> + case FIX_TRUNC_EXPR:
> + /* Don't rely on FP_OPERATION to cover FP->INT conversions,
> + since the complete expression does not have a FP type. */
I'll note that we are careful for comparisons in callers. But I guess handling
FIX_TRUNC_EXPR explicitly as always FP operation is sensible. But can
you also add VEC_UNPACK_FIX_TRUNC_HI_EXPR,
VEC_UNPACK_FIX_TRUNC_LO_EXPR and VEC_PACK_FIX_TRUNC_EXPR
to this block for completeness?
OK with that change.
Richard.
> + return flag_trapping_math;
> +
> case COND_EXPR:
> case VEC_COND_EXPR:
> /* Whether *COND_EXPR can trap depends on whether the
> --
> 2.34.1
>
