[PATCH] tree-optimization/98117 - fix range set by vectorization on niter IVs

Richard Sandiford richard.sandiford@arm.com
Mon Dec 7 10:30:12 GMT 2020


Richard Biener <rguenther@suse.de> writes:
> This avoids the degenerate case of a TYPE_MAX_VALUE latch iteration
> count value causing wrong range info for the vector IV.  There's
> still the case of VF == 1 where if we don't know whether we hit the
> above case we cannot emit a range.
>
> Bootstrap and regtest running on x86_64-unknown-linux-gnu.
>
> 2020-12-07  Richard Biener  <rguenther@suse.de>
>
> 	PR tree-optimization/98117
> 	* tree-vect-loop-manip.c (vect_gen_vector_loop_niters):
> 	Properly handle degenerate niter when setting the vector
> 	loop IV range.
>
> 	* gcc.dg/torture/pr98117.c: New testcase.
> ---
>  gcc/testsuite/gcc.dg/torture/pr98117.c | 19 +++++++++++++++++
>  gcc/tree-vect-loop-manip.c             | 28 ++++++++++++++++++++------
>  2 files changed, 41 insertions(+), 6 deletions(-)
>  create mode 100644 gcc/testsuite/gcc.dg/torture/pr98117.c
>
> diff --git a/gcc/testsuite/gcc.dg/torture/pr98117.c b/gcc/testsuite/gcc.dg/torture/pr98117.c
> new file mode 100644
> index 00000000000..f2160257263
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/torture/pr98117.c
> @@ -0,0 +1,19 @@
> +/* { dg-do run } */
> +/* { dg-additional-options "-fno-tree-scev-cprop" } */
> +
> +unsigned char c;
> +void __attribute__((noipa))
> +e()
> +{
> +  do
> +    {
> +    }
> +  while (++c);
> +}
> +int main()
> +{
> +  e();
> +  if (c != 0)
> +    __builtin_abort ();
> +  return 0;
> +}
> diff --git a/gcc/tree-vect-loop-manip.c b/gcc/tree-vect-loop-manip.c
> index 36179188f6d..2370b879b21 100644
> --- a/gcc/tree-vect-loop-manip.c
> +++ b/gcc/tree-vect-loop-manip.c
> @@ -2034,13 +2034,29 @@ vect_gen_vector_loop_niters (loop_vec_info loop_vinfo, tree niters,
>        niters_vector = force_gimple_operand (niters_vector, &stmts, true, var);
>        gsi_insert_seq_on_edge_immediate (pe, stmts);
>        /* Peeling algorithm guarantees that vector loop bound is at least ONE,
> -	 we set range information to make niters analyzer's life easier.  */
> +	 we set range information to make niters analyzer's life easier.
> +	 Note the number of latch iteration value can be TYPE_MAX_VALUE so
> +	 we have to represent the vector niter TYPE_MAX_VALUE + 1 >> log_vf.  */
>        if (stmts != NULL && log_vf)
> -	set_range_info (niters_vector, VR_RANGE,
> -			wi::to_wide (build_int_cst (type, 1)),
> -			wi::to_wide (fold_build2 (RSHIFT_EXPR, type,
> -						  TYPE_MAX_VALUE (type),
> -						  log_vf)));
> +	{
> +	  if (niters_no_overflow)
> +	    set_range_info (niters_vector, VR_RANGE,
> +			    wi::one (TYPE_PRECISION (type)),
> +			    wi::rshift (wi::max_value (TYPE_PRECISION (type),
> +						       TYPE_SIGN (type)),
> +					exact_log2 (const_vf),
> +					TYPE_SIGN (type)));
> +	  /* For VF == 1 the vector IV might also overflow so we cannot
> +	     assert a minimum value of 1.  */
> +	  else if (const_vf > 1)

This code is common to const vf and non-const vf, so const_vf isn't
guaranteed to be valid.

Thanks,
Richard

> +	    set_range_info (niters_vector, VR_RANGE,
> +			    wi::one (TYPE_PRECISION (type)),
> +			    wi::rshift (wi::max_value (TYPE_PRECISION (type),
> +						       TYPE_SIGN (type))
> +					- (const_vf - 1),
> +					exact_log2 (const_vf), TYPE_SIGN (type))
> +			    + 1);
> +	}
>      }
>    *niters_vector_ptr = niters_vector;
>    *step_vector_ptr = step_vector;


More information about the Gcc-patches mailing list