RE: [PATCH][vect]: Fix null dereference in boolean reductions [PR122475]
Tamar Christina <[email protected]>
| Newsgroups | gmane.comp.gcc.patches |
|---|---|
| Message-ID | <VI0PR08MB10392250A366132BE72D6D148FFDB2@VI0PR08MB10392.eurprd08.prod.outlook.com> |
> -----Original Message----- > From: Andrea Pinski <[email protected]> > Sent: 13 August 2026 21:27 > To: Tamar Christina <[email protected]> > Cc: [email protected]; nd <[email protected]>; [email protected]; > [email protected] > Subject: Re: [PATCH][vect]: Fix null dereference in boolean reductions > [PR122475] > > 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..ed229c5475cc96fc0ef > 3beea71b101db89e08916 > > --- /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. Hmm it's not meant to be one (hence also no tree-vect.h import), but the vect.exp testsuite misdetected it as one because of the "main" function. I'll fix tomorrow. Thanks, Tamar > > 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..6492c44c065159d06d > 7e69499dfc9834721d62ef > > --- /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..83c5c206dd2597c377 > 6317005a3dd38ee5991f16 > > --- /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..576a69c82d25aa74ce9cc8 > dbb89d0b340ed9ba39 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)) > > { > > > > > > --