[PATCH] store-merging: Allow enums during bswap recognition [PR94403]

[PATCH] store-merging: Allow enums during bswap recognition [PR94403]

[Jakub Jelinek]

Hi!

The following testcase is optimized with char/unsigned char/signed char,
but not with std::byte.  The following patch fixes that.  Didn't use
INTEGRAL_TYPE_P because bswapping bools is just too weird.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for GCC11 (or is
it simple enough that we want it in GCC10 already)?

2020-03-31  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/94403
	* gimple-ssa-store-merging.c (verify_symbolic_number_p): Allow also
	ENUMERAL_TYPE lhs_type.

	* g++.dg/tree-ssa/pr94403.C: New test.

--- gcc/gimple-ssa-store-merging.c.jj	2020-03-20 09:33:03.602112329 +0100
+++ gcc/gimple-ssa-store-merging.c	2020-03-30 18:30:31.723341519 +0200
@@ -315,7 +315,8 @@ verify_symbolic_number_p (struct symboli
 
   lhs_type = gimple_expr_type (stmt);
 
-  if (TREE_CODE (lhs_type) != INTEGER_TYPE)
+  if (TREE_CODE (lhs_type) != INTEGER_TYPE
+      && TREE_CODE (lhs_type) != ENUMERAL_TYPE)
     return false;
 
   if (TYPE_PRECISION (lhs_type) != TYPE_PRECISION (n->type))
--- gcc/testsuite/g++.dg/tree-ssa/pr94403.C.jj	2020-03-30 18:42:28.563719010 +0200
+++ gcc/testsuite/g++.dg/tree-ssa/pr94403.C	2020-03-30 18:51:21.135828389 +0200
@@ -0,0 +1,37 @@
+// PR tree-optimization/94403
+// Only test on some 64-bit targets which do have bswap{si,di}2 patterns and
+// are either big or little endian (not pdp endian).
+// { dg-do compile { target { lp64 && { i?86-*-* x86_64-*-* powerpc*-*-* aarch64*-*-* } } } }
+// { dg-require-effective-target store_merge }
+// { dg-options "-O2 -fdump-tree-store-merging -std=c++17" }
+
+namespace std {
+  template <typename T>
+  void swap (T& t1, T& t2) { T tmp (t1); t1 = t2; t2 = tmp; }
+  enum class byte : unsigned char {};
+}
+
+unsigned
+bswap32 (unsigned int x)
+{
+  unsigned int y = x;
+  std::byte *bytes = reinterpret_cast<std::byte*> (&y);
+  std::swap (bytes[0], bytes[3]);
+  std::swap (bytes[1], bytes[2]);
+  return y;
+}
+
+unsigned long long
+bswap64 (unsigned long long x)
+{
+  unsigned long long y = x;
+  std::byte *bytes = reinterpret_cast<std::byte*> (&y);
+  std::swap (bytes[0], bytes[7]);
+  std::swap (bytes[1], bytes[6]);
+  std::swap (bytes[2], bytes[5]);
+  std::swap (bytes[3], bytes[4]);
+  return y;
+}
+
+/* { dg-final { scan-tree-dump-times "__builtin_bswap64" 1 "store-merging" } } */
+/* { dg-final { scan-tree-dump-times "__builtin_bswap32" 1 "store-merging" } } */

	Jakub

Re: [PATCH] store-merging: Allow enums during bswap recognition [PR94403]

[Richard Biener]

On Tue, 31 Mar 2020, Jakub Jelinek wrote:

> Hi!
> 
> The following testcase is optimized with char/unsigned char/signed char,
> but not with std::byte.  The following patch fixes that.  Didn't use
> INTEGRAL_TYPE_P because bswapping bools is just too weird.

Is it?

> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for GCC11 (or is
> it simple enough that we want it in GCC10 already)?

OK for GCC 10.

Richard.

> 2020-03-31  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR tree-optimization/94403
> 	* gimple-ssa-store-merging.c (verify_symbolic_number_p): Allow also
> 	ENUMERAL_TYPE lhs_type.
> 
> 	* g++.dg/tree-ssa/pr94403.C: New test.
> 
> --- gcc/gimple-ssa-store-merging.c.jj	2020-03-20 09:33:03.602112329 +0100
> +++ gcc/gimple-ssa-store-merging.c	2020-03-30 18:30:31.723341519 +0200
> @@ -315,7 +315,8 @@ verify_symbolic_number_p (struct symboli
>  
>    lhs_type = gimple_expr_type (stmt);
>  
> -  if (TREE_CODE (lhs_type) != INTEGER_TYPE)
> +  if (TREE_CODE (lhs_type) != INTEGER_TYPE
> +      && TREE_CODE (lhs_type) != ENUMERAL_TYPE)
>      return false;
>  
>    if (TYPE_PRECISION (lhs_type) != TYPE_PRECISION (n->type))
> --- gcc/testsuite/g++.dg/tree-ssa/pr94403.C.jj	2020-03-30 18:42:28.563719010 +0200
> +++ gcc/testsuite/g++.dg/tree-ssa/pr94403.C	2020-03-30 18:51:21.135828389 +0200
> @@ -0,0 +1,37 @@
> +// PR tree-optimization/94403
> +// Only test on some 64-bit targets which do have bswap{si,di}2 patterns and
> +// are either big or little endian (not pdp endian).
> +// { dg-do compile { target { lp64 && { i?86-*-* x86_64-*-* powerpc*-*-* aarch64*-*-* } } } }
> +// { dg-require-effective-target store_merge }
> +// { dg-options "-O2 -fdump-tree-store-merging -std=c++17" }
> +
> +namespace std {
> +  template <typename T>
> +  void swap (T& t1, T& t2) { T tmp (t1); t1 = t2; t2 = tmp; }
> +  enum class byte : unsigned char {};
> +}
> +
> +unsigned
> +bswap32 (unsigned int x)
> +{
> +  unsigned int y = x;
> +  std::byte *bytes = reinterpret_cast<std::byte*> (&y);
> +  std::swap (bytes[0], bytes[3]);
> +  std::swap (bytes[1], bytes[2]);
> +  return y;
> +}
> +
> +unsigned long long
> +bswap64 (unsigned long long x)
> +{
> +  unsigned long long y = x;
> +  std::byte *bytes = reinterpret_cast<std::byte*> (&y);
> +  std::swap (bytes[0], bytes[7]);
> +  std::swap (bytes[1], bytes[6]);
> +  std::swap (bytes[2], bytes[5]);
> +  std::swap (bytes[3], bytes[4]);
> +  return y;
> +}
> +
> +/* { dg-final { scan-tree-dump-times "__builtin_bswap64" 1 "store-merging" } } */
> +/* { dg-final { scan-tree-dump-times "__builtin_bswap32" 1 "store-merging" } } */
> 
> 	Jakub
> 
> 

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