"Yangfei (Felix)" writes:
> Hi,
>
> I see one gcc_assert was introduce in:
> https://gcc.gnu.org/pipermail/gcc-patches/2020-April/544271.html
> This is causing an ICE for certain cases. I have created a PR for this:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94784
> I did some check and it looks like everything works fine before the ICE.
> In the testcase we have two vectors with the same ABI identity but with
> different TYPE_MODEs.
> The proposed patch flips the assert around so that it checks that the two
> vectors have equal TYPE_VECTOR_SUBPARTS and that converting the corresponding
> element types is a useless_type_conversion_p.
>
> Bootstrap and tested on aarch64-linux-gnu. OK?
>
> Thanks,
> Felix
>
> From 518b0cf9433914f437631f16816cfb564fcce285 Mon Sep 17 00:00:00 2001
> From: Fei Yang
> Date: Mon, 27 Apr 2020 15:04:57 +0800
> Subject: [PATCH] forwprop: Fix ICE when building an identity constructor
> [PR94784]
>
> In the testcase for PR94784, we have two vectors with the same ABI identity
> but with different TYPE_MODEs. It would be better to flip the assert around
> so that it checks that the two vectors have equal TYPE_VECTOR_SUBPARTS and
> that converting the corresponding element types is a
> useless_type_conversion_p.
>
> 2020-04-27 Felix Yang
>
> gcc/
> PR tree-optimization/94784
> * tree-ssa-forwprop.c (simplify_vector_constructor): Flip the
> assert around so that it checks that the two vectors have equal
> TYPE_VECTOR_SUBPARTS and that converting the corresponding element
> types is a useless_type_conversion_p.
>
> gcc/testsuite/
> PR tree-optimization/94784
> * gcc.dg/pr94784.c: New test.
> ---
> gcc/ChangeLog | 11 +--
> gcc/testsuite/ChangeLog| 5 +
> gcc/testsuite/gcc.dg/pr94784.c | 16
> gcc/tree-ssa-forwprop.c| 14 --
> 4 files changed, 42 insertions(+), 4 deletions(-)
> create mode 100644 gcc/testsuite/gcc.dg/pr94784.c
>
> diff --git a/gcc/ChangeLog b/gcc/ChangeLog
> index 38dd5837e20..03f73c736ff 100644
> --- a/gcc/ChangeLog
> +++ b/gcc/ChangeLog
> @@ -1,3 +1,11 @@
> +2020-04-27 Felix Yang
> +
> + PR tree-optimization/94784
> + * tree-ssa-forwprop.c (simplify_vector_constructor): Flip the
> + assert around so that it checks that the two vectors have equal
> + TYPE_VECTOR_SUBPARTS and that converting the corresponding element
> + types is a useless_type_conversion_p.
> +
> 2020-04-25 David Edelsohn
>
> * config/rs6000/rs6000-logue.c (rs6000_stack_info): Don't push a
> @@ -1790,8 +1798,7 @@
>
> PR tree-optimization/94269
> * tree-ssa-math-opts.c (convert_plusminus_to_widen): Restrict
> - this
> - operation to single basic block.
> + this operation to single basic block.
Not wrong, but might as well leave the entry as-is.
>
> 2020-03-25 Jeff Law
>
> diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
> index 07fe8a68598..cfc7b13985d 100644
> --- a/gcc/testsuite/ChangeLog
> +++ b/gcc/testsuite/ChangeLog
> @@ -1,3 +1,8 @@
> +2020-04-27 Felix Yang
> +
> + PR tree-optimization/94784
> + * gcc.dg/pr94784.c: New test.
> +
> 2020-04-26 Iain Sandoe
>
> PR c++/94752
> diff --git a/gcc/testsuite/gcc.dg/pr94784.c b/gcc/testsuite/gcc.dg/pr94784.c
> new file mode 100644
> index 000..df6972f64aa
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/pr94784.c
> @@ -0,0 +1,16 @@
> +/* { dg-do compile { target aarch64*-*-* } } */
> +/* { dg-options "-O2 -ftree-slp-vectorize -march=armv8.2-a+sve
> -msve-vector-bits=256" } */
> +
> +typedef short __attribute__((vector_size (8))) v4hi;
> +
> +typedef union U4HI { v4hi v; short a[4]; } u4hi;
> +
> +short a[4];
> +
> +void pass_v4hi (v4hi v) {
> +int j;
> +u4hi u;
> +u.v = v;
> +for (j = 0; j < 4; j++)
> + a[j] = u.a[j];
> +};
> diff --git a/gcc/tree-ssa-forwprop.c b/gcc/tree-ssa-forwprop.c
> index 8ee5450b94c..7cfb18bd618 100644
> --- a/gcc/tree-ssa-forwprop.c
> +++ b/gcc/tree-ssa-forwprop.c
> @@ -2479,7 +2479,12 @@ simplify_vector_constructor (gimple_stmt_iterator *gsi)
> tree src_type = TREE_TYPE (orig[0]);
> if (!useless_type_conversion_p (type, src_type))
> {
> - gcc_assert (!targetm.compatible_vector_types_p (type, src_type));
> + gcc_assert (TYPE_VECTOR_SUBPARTS (type).is_constant ()
> + && TYPE_VECTOR_SUBPARTS (src_type).is_constant ()
> + && (TYPE_VECTOR_SUBPARTS (type).to_constant ()
> + == TYPE_VECTOR_SUBPARTS (src_type).to_constant ())
We don't need to fall back to constant comparisons here.
We should assert for known_eq between the TYPE_VECTOR_SUBPARTS instead.
Same for the other assert.
LGTM otherwise, and sorry for the breakage.
Thanks,
Richard
