On Mon, Nov 3, 2025 at 12:10 AM Tamar Christina <[email protected]> wrote:
>
> neutral_op can be null, so guard against that.
>
> Bootstrapped Regtested on aarch64-none-linux-gnu,
> arm-none-linux-gnueabihf, x86_64-pc-linux-gnu
> -m32, -m64 and no issues
>
> Pushed to master.
>
> Thanks,
> Tamar
>
> gcc/ChangeLog:
>
> PR tree-optimization/122475
> * tree-vect-loop.cc (vectorizable_reduction): Check for neutral_op.
>
> gcc/testsuite/ChangeLog:
>
> PR tree-optimization/122475
> * gcc.dg/vect/pr122475.c: New test.
> * gcc.target/aarch64/sve/vect-reduc-bool-19.c: New test.
> * gcc.target/aarch64/sve/vect-reduc-bool-20.c: New test.
>
> ---
> diff --git a/gcc/testsuite/gcc.dg/vect/pr122475.c
> b/gcc/testsuite/gcc.dg/vect/pr122475.c
> new file mode 100644
> index
> 0000000000000000000000000000000000000000..ed229c5475cc96fc0ef3beea71b101db89e08916
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/vect/pr122475.c
> @@ -0,0 +1,13 @@
> +/* { dg-additional-options "-march=armv8-a+sve" { target aarch64*-*-* } } */
> +/* Check that we don't ICE. */
> +int a;
> +int b;
> +int main() {
> + for (char t = 0; t < 14; t += 2)
> + for (int u = 0; u < 242; u += 4) {
> + a = a < 0 ? a : 0;
> + b = b < 0 ? b : 0;
> + }
> +}
> +
> +/* { dg-final { scan-tree-dump-times "optimized: loop vectorized" 1 "vect" {
> target aarch64*-*-* } } } */
I only noticed this failure now (sorry) But the above testcase fails
at execution if the target does not have sve support on it.
Thanks,
Andrea
> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/vect-reduc-bool-19.c
> b/gcc/testsuite/gcc.target/aarch64/sve/vect-reduc-bool-19.c
> new file mode 100644
> index
> 0000000000000000000000000000000000000000..6492c44c065159d06d7e69499dfc9834721d62ef
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/sve/vect-reduc-bool-19.c
> @@ -0,0 +1,17 @@
> +/* { dg-do compile } */
> +/* { dg-additional-options "-mautovec-preference=sve-only
> -fdump-tree-vect-details -O3 --param vect-epilogues-nomask=0" } */
> +
> +int p[128];
> +
> +bool __attribute__((noipa))
> +fand (int n, bool r1, bool r2)
> +{
> + bool r = true;
> + for (int i = 0; i < (n/2); i+=2)
> + {
> + r &= (p[i] != 0) & r1;
> + r &= (p[i+1] != 0) & r2;
> + }
> + return r;
> +}
> +/* { dg-final { scan-tree-dump-times "optimized: loop vectorized" 1 "vect" }
> } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/vect-reduc-bool-20.c
> b/gcc/testsuite/gcc.target/aarch64/sve/vect-reduc-bool-20.c
> new file mode 100644
> index
> 0000000000000000000000000000000000000000..83c5c206dd2597c3776317005a3dd38ee5991f16
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/sve/vect-reduc-bool-20.c
> @@ -0,0 +1,32 @@
> +/* { dg-do compile } */
> +/* { dg-additional-options "-mautovec-preference=sve-only
> -fdump-tree-vect-details -O3 --param vect-epilogues-nomask=0" } */
> +
> +#include <stdbool.h>
> +#include <stdint.h>
> +
> +void vec_slp_cmp (char* restrict a, char* restrict b, int n) {
> + bool x0 = b[0] != 0;
> + bool x1 = b[1] != 0;
> + bool x2 = b[2] != 0;
> + bool x3 = b[3] != 0;
> + for (int i = 0; i < n; ++i) {
> + x0 &= (a[i * 4] != 0);
> + x1 &= (a[i * 4 + 1] != 0);
> + x2 &= (a[i * 4 + 2] != 0);
> + x3 &= (a[i * 4 + 3] != 0);
> + }
> + b[0] = x0;
> + b[1] = x1;
> + b[2] = x2;
> + b[3] = x3;
> +}
> +
> +void vec_slp_cmp1 (char* restrict a, char* restrict b, int n) {
> + bool x0 = b[0] != 0;
> + for (int i = 0; i < n; ++i) {
> + x0 &= (a[i] != 0);
> + }
> + b[0] = x0;
> +}
> +
> +/* { dg-final { scan-tree-dump-times "optimized: loop vectorized" 2 "vect" {
> target aarch64*-*-* } } } */
> diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc
> index
> 50cdc2a90fa29c1e0d116c0589bc246e6d8fcc84..576a69c82d25aa74ce9cc8dbb89d0b340ed9ba39
> 100644
> --- a/gcc/tree-vect-loop.cc
> +++ b/gcc/tree-vect-loop.cc
> @@ -7578,7 +7578,9 @@ vectorizable_reduction (loop_vec_info loop_vinfo,
> if ((double_reduc || neutral_op)
> && !nunits_out.is_constant ()
> && (SLP_TREE_LANES (slp_node) != 1 && !reduc_chain)
> - && !operand_equal_p (neutral_op, vect_phi_initial_value
> (reduc_def_phi))
> + && (!neutral_op
> + || !operand_equal_p (neutral_op,
> + vect_phi_initial_value (reduc_def_phi)))
> && !direct_internal_fn_supported_p (IFN_VEC_SHL_INSERT,
> vectype_out, OPTIMIZE_FOR_SPEED))
> {
>
>
> --
