[PATCH] sra: Only verify sizes of scalar accesses (PR 93845)

[PATCH] sra: Only verify sizes of scalar accesses (PR 93845)

[Martin Jambor]

Hi,

the testcase is another example - in addition to recent PR 93516 - where
the SRA access verifier is confused by the fact that get_ref_base_extent
can return different sizes for the same type, depending whether they are
COMPONENT_REF or not.  In the previous bug I decided to keep the
verifier check for aggregate type even though it is not really important
and instead avoid easily detectable type-within-the-same-type situation.
This testcase is however a result of a fairly random looking type cast
and so cannot be handled in the same way.

Because the check is not really important for aggregates, this patch
simply disables it for non-register types.

Bootstrapped and tested on x86_64-linux.  OK for trunk?

Thanks,

Martin

2020-02-20  Martin Jambor  <mjambor@suse.cz>

	PR tree-optimization/93845
	* tree-sra.c (verify_sra_access_forest): Only test access size of
	scalar types.

	testsuite/
	* g++.dg/tree-ssa/pr93845.C: New test.
---
 gcc/ChangeLog                           |  6 +++++
 gcc/testsuite/ChangeLog                 |  5 +++++
 gcc/testsuite/g++.dg/tree-ssa/pr93845.C | 30 +++++++++++++++++++++++++
 gcc/tree-sra.c                          |  3 ++-
 4 files changed, 43 insertions(+), 1 deletion(-)
 create mode 100644 gcc/testsuite/g++.dg/tree-ssa/pr93845.C

diff --git a/gcc/testsuite/g++.dg/tree-ssa/pr93845.C b/gcc/testsuite/g++.dg/tree-ssa/pr93845.C
new file mode 100644
index 00000000000..72e473fffcd
--- /dev/null
+++ b/gcc/testsuite/g++.dg/tree-ssa/pr93845.C
@@ -0,0 +1,30 @@
+/* { dg-do compile } */
+/* { dg-options "-O1" } */
+
+struct g;
+struct h {
+  g *operator->();
+};
+class i {
+  void *a;
+  int b;
+
+public:
+  template <typename f> f j() { return *static_cast<f *>(this); }
+};
+struct k : i {};
+struct l : k {};
+struct m {
+  i n();
+  i o(l, l, int);
+};
+struct g {
+  void m_fn4(k);
+};
+h a;
+i b;
+i m::n() {
+  l c, d, e = o(d, c, 0).j<l>();
+  a->m_fn4(e);
+  return b;
+}
diff --git a/gcc/tree-sra.c b/gcc/tree-sra.c
index 0cfac0a8192..1439f11f15a 100644
--- a/gcc/tree-sra.c
+++ b/gcc/tree-sra.c
@@ -2339,7 +2339,8 @@ verify_sra_access_forest (struct access *root)
       gcc_assert (offset == access->offset);
       gcc_assert (access->grp_unscalarizable_region
 		  || size == max_size);
-      gcc_assert (max_size == access->size);
+      gcc_assert (!is_gimple_reg_type (access->type)
+		  || max_size == access->size);
       gcc_assert (reverse == access->reverse);
 
       if (access->first_child)
-- 
2.25.0

Re: [PATCH] sra: Only verify sizes of scalar accesses (PR 93845)

[Richard Biener]

[-- Attachment #1: Type: text/plain, Size: 2870 bytes --]

On Fri, 21 Feb 2020, Martin Jambor wrote:

> Hi,
> 
> the testcase is another example - in addition to recent PR 93516 - where
> the SRA access verifier is confused by the fact that get_ref_base_extent
> can return different sizes for the same type, depending whether they are
> COMPONENT_REF or not.  In the previous bug I decided to keep the
> verifier check for aggregate type even though it is not really important
> and instead avoid easily detectable type-within-the-same-type situation.
> This testcase is however a result of a fairly random looking type cast
> and so cannot be handled in the same way.
> 
> Because the check is not really important for aggregates, this patch
> simply disables it for non-register types.
> 
> Bootstrapped and tested on x86_64-linux.  OK for trunk?

OK.

Richard.

> Thanks,
> 
> Martin
> 
> 2020-02-20  Martin Jambor  <mjambor@suse.cz>
> 
> 	PR tree-optimization/93845
> 	* tree-sra.c (verify_sra_access_forest): Only test access size of
> 	scalar types.
> 
> 	testsuite/
> 	* g++.dg/tree-ssa/pr93845.C: New test.
> ---
>  gcc/ChangeLog                           |  6 +++++
>  gcc/testsuite/ChangeLog                 |  5 +++++
>  gcc/testsuite/g++.dg/tree-ssa/pr93845.C | 30 +++++++++++++++++++++++++
>  gcc/tree-sra.c                          |  3 ++-
>  4 files changed, 43 insertions(+), 1 deletion(-)
>  create mode 100644 gcc/testsuite/g++.dg/tree-ssa/pr93845.C
> 
> diff --git a/gcc/testsuite/g++.dg/tree-ssa/pr93845.C b/gcc/testsuite/g++.dg/tree-ssa/pr93845.C
> new file mode 100644
> index 00000000000..72e473fffcd
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/tree-ssa/pr93845.C
> @@ -0,0 +1,30 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O1" } */
> +
> +struct g;
> +struct h {
> +  g *operator->();
> +};
> +class i {
> +  void *a;
> +  int b;
> +
> +public:
> +  template <typename f> f j() { return *static_cast<f *>(this); }
> +};
> +struct k : i {};
> +struct l : k {};
> +struct m {
> +  i n();
> +  i o(l, l, int);
> +};
> +struct g {
> +  void m_fn4(k);
> +};
> +h a;
> +i b;
> +i m::n() {
> +  l c, d, e = o(d, c, 0).j<l>();
> +  a->m_fn4(e);
> +  return b;
> +}
> diff --git a/gcc/tree-sra.c b/gcc/tree-sra.c
> index 0cfac0a8192..1439f11f15a 100644
> --- a/gcc/tree-sra.c
> +++ b/gcc/tree-sra.c
> @@ -2339,7 +2339,8 @@ verify_sra_access_forest (struct access *root)
>        gcc_assert (offset == access->offset);
>        gcc_assert (access->grp_unscalarizable_region
>  		  || size == max_size);
> -      gcc_assert (max_size == access->size);
> +      gcc_assert (!is_gimple_reg_type (access->type)
> +		  || max_size == access->size);
>        gcc_assert (reverse == access->reverse);
>  
>        if (access->first_child)
> 

-- 
Richard Biener <rguenther@suse.de>
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imendörffer; HRB 36809 (AG Nuernberg)