[Mesa-dev] [PATCH] nir/loop_unroll: Fix out-of-bounds access handling

[Timothy Arceri]

On 13/3/19 8:30 am, Jason Ekstrand wrote:
> The previous code was completely broken when it came to constructing the
> undef values.  I'm not sure how it ever worked.  For the case of a copy
> that reads an undefined value, we can just delete the copy because the
> destination is a valid undefined value.  This saves us the effort of
> trying to construct a value for an arbitrary copy_deref intrinsic.

Yes. It turns out this was indeed not required. Thanks for the fix!

Reviewed-by: Timothy Arceri <tarceri at itsqueeze.com>

> 
> Fixes: e8a8937a04 "nir: add partial loop unrolling support"
> Cc: Timothy Arceri <tarceri at itsqueeze.com>
> ---
>   src/compiler/nir/nir_opt_loop_unroll.c | 14 ++------------
>   1 file changed, 2 insertions(+), 12 deletions(-)
> 
> diff --git a/src/compiler/nir/nir_opt_loop_unroll.c b/src/compiler/nir/nir_opt_loop_unroll.c
> index 06ec78b8068..b2696d4aadb 100644
> --- a/src/compiler/nir/nir_opt_loop_unroll.c
> +++ b/src/compiler/nir/nir_opt_loop_unroll.c
> @@ -670,11 +670,9 @@ remove_out_of_bounds_induction_use(nir_shader *shader, nir_loop *loop,
>               if (is_access_out_of_bounds(term, nir_src_as_deref(intrin->src[0]),
>                                           trip_count)) {
>                  if (intrin->intrinsic == nir_intrinsic_load_deref) {
> -                  assert(intrin->src[0].is_ssa);
> -                  nir_ssa_def *a_ssa = intrin->src[0].ssa;
>                     nir_ssa_def *undef =
> -                     nir_ssa_undef(&b, intrin->num_components,
> -                                   a_ssa->bit_size);
> +                     nir_ssa_undef(&b, intrin->dest.ssa.num_components,
> +                                   intrin->dest.ssa.bit_size);
>                     nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
>                                              nir_src_for_ssa(undef));
>                  } else {
> @@ -686,14 +684,6 @@ remove_out_of_bounds_induction_use(nir_shader *shader, nir_loop *loop,
>               if (intrin->intrinsic == nir_intrinsic_copy_deref &&
>                   is_access_out_of_bounds(term, nir_src_as_deref(intrin->src[1]),
>                                           trip_count)) {
> -               assert(intrin->src[1].is_ssa);
> -               nir_ssa_def *a_ssa = intrin->src[1].ssa;
> -               nir_ssa_def *undef =
> -                  nir_ssa_undef(&b, intrin->num_components, a_ssa->bit_size);
> -
> -               /* Replace the copy with a store of the undefined value */
> -               b.cursor = nir_before_instr(instr);
> -               nir_store_deref(&b, nir_src_as_deref(intrin->src[0]), undef, ~0);
>                  nir_instr_remove(instr);
>               }
>            }
>