[PATCH] c++: Fix ICE with __builtin_bit_cast [PR98469]

[PATCH] c++: Fix ICE with __builtin_bit_cast [PR98469]

[Jakub Jelinek]

Hi!

On the following testcase we ICE during constexpr evaluation (for warnings),
because the IL has ADDR_EXPR of BIT_CAST_EXPR and ADDR_EXPR case asserts
the result is not a CONSTRUCTOR.
I've tried to force a temporary for those in call.c next to:
        if (convs->need_temporary_p
            || TREE_CODE (expr) == CONSTRUCTOR
            || TREE_CODE (expr) == VA_ARG_EXPR)
but that resulted in a lot of ICEs, so this patch just punts on lval
evaluation of BIT_CAST_EXPR instead, normally __builtin_bit_cast is called
from std::bit_cast which is constexpr and therefore the BIT_CAST_EXPR
isn't evaluated there during parsing or tsubst and when evaluating the call
to std::bit_cast the NRV optimized return is assigned to some temporary or
variable and so BIT_CAST_EXPR is not evaluated as lval.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2020-12-30  Jakub Jelinek  <jakub@redhat.com>

	PR c++/98469
	* constexpr.c (cxx_eval_constant_expression) <case BIT_CAST_EXPR>:
	Punt if lval is true.

	* g++.dg/cpp2a/bit-cast8.C: New test.
	* g++.dg/cpp2a/bit-cast9.C: New test.

--- gcc/cp/constexpr.c.jj	2020-12-23 22:44:05.398093175 +0100
+++ gcc/cp/constexpr.c	2020-12-29 10:32:44.865030881 +0100
@@ -6900,6 +6900,15 @@ cxx_eval_constant_expression (const cons
       return t;
 
     case BIT_CAST_EXPR:
+      if (lval)
+	{
+	  if (!ctx->quiet)
+	    error_at (EXPR_LOCATION (t),
+		      "address of a call to %qs is not a constant expression",
+		      "__builtin_bit_cast");
+	  *non_constant_p = true;
+	  return t;
+	}
       r = cxx_eval_bit_cast (ctx, t, non_constant_p, overflow_p);
       break;
 
--- gcc/testsuite/g++.dg/cpp2a/bit-cast8.C.jj	2020-12-29 10:35:31.547140723 +0100
+++ gcc/testsuite/g++.dg/cpp2a/bit-cast8.C	2020-12-29 10:34:26.431879120 +0100
@@ -0,0 +1,11 @@
+// PR c++/98469
+// { dg-do compile { target c++20 } }
+// { dg-options "-Wall" }
+
+struct S { int s; };
+
+S
+foo ()
+{
+  return __builtin_bit_cast (S, 0);
+}
--- gcc/testsuite/g++.dg/cpp2a/bit-cast9.C.jj	2020-12-29 10:35:35.018101365 +0100
+++ gcc/testsuite/g++.dg/cpp2a/bit-cast9.C	2020-12-29 10:35:05.905431494 +0100
@@ -0,0 +1,15 @@
+// PR c++/98469
+// { dg-do compile { target c++20 } }
+// { dg-options "-Wall" }
+
+template<typename T, typename F>
+constexpr T
+bit_cast (const F &f) noexcept
+{
+  return __builtin_bit_cast (T, f);
+}
+struct S { int s; };
+constexpr int foo (const S &x) { return x.s; }
+constexpr int bar () { return foo (bit_cast<S> (0)); }
+constexpr int x = bar ();
+static_assert (!x);

	Jakub

Re: [PATCH] c++: Fix ICE with __builtin_bit_cast [PR98469]

[Jason Merrill]

On 12/30/20 4:13 AM, Jakub Jelinek wrote:
> Hi!
> 
> On the following testcase we ICE during constexpr evaluation (for warnings),
> because the IL has ADDR_EXPR of BIT_CAST_EXPR and ADDR_EXPR case asserts
> the result is not a CONSTRUCTOR.
> I've tried to force a temporary for those in call.c next to:
>          if (convs->need_temporary_p
>              || TREE_CODE (expr) == CONSTRUCTOR
>              || TREE_CODE (expr) == VA_ARG_EXPR)
> but that resulted in a lot of ICEs, so this patch just punts on lval
> evaluation of BIT_CAST_EXPR instead, normally __builtin_bit_cast is called
> from std::bit_cast which is constexpr and therefore the BIT_CAST_EXPR
> isn't evaluated there during parsing or tsubst and when evaluating the call
> to std::bit_cast the NRV optimized return is assigned to some temporary or
> variable and so BIT_CAST_EXPR is not evaluated as lval.

This change is OK, but part of the problem is that we're trying to do 
overload resolution for an S copy/move constructor, which we shouldn't 
be because bit_cast is a prvalue, so in C++17 and up we should use it to 
directly initialize the target without any implied constructor call.

It seems we're mishandling this because the code in 
build_special_member_call specifically looks for TARGET_EXPR or 
CONSTRUCTOR, and BIT_CAST_EXPR is neither of those.

Wrapping a BIT_CAST_EXPR of aggregate type in a TARGET_EXPR would 
address this, and any other places that expect a class prvalue to come 
in the form of a TARGET_EXPR.

> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
> 
> 2020-12-30  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR c++/98469
> 	* constexpr.c (cxx_eval_constant_expression) <case BIT_CAST_EXPR>:
> 	Punt if lval is true.
> 
> 	* g++.dg/cpp2a/bit-cast8.C: New test.
> 	* g++.dg/cpp2a/bit-cast9.C: New test.
> 
> --- gcc/cp/constexpr.c.jj	2020-12-23 22:44:05.398093175 +0100
> +++ gcc/cp/constexpr.c	2020-12-29 10:32:44.865030881 +0100
> @@ -6900,6 +6900,15 @@ cxx_eval_constant_expression (const cons
>         return t;
>   
>       case BIT_CAST_EXPR:
> +      if (lval)
> +	{
> +	  if (!ctx->quiet)
> +	    error_at (EXPR_LOCATION (t),
> +		      "address of a call to %qs is not a constant expression",
> +		      "__builtin_bit_cast");
> +	  *non_constant_p = true;
> +	  return t;
> +	}
>         r = cxx_eval_bit_cast (ctx, t, non_constant_p, overflow_p);
>         break;
>   
> --- gcc/testsuite/g++.dg/cpp2a/bit-cast8.C.jj	2020-12-29 10:35:31.547140723 +0100
> +++ gcc/testsuite/g++.dg/cpp2a/bit-cast8.C	2020-12-29 10:34:26.431879120 +0100
> @@ -0,0 +1,11 @@
> +// PR c++/98469
> +// { dg-do compile { target c++20 } }
> +// { dg-options "-Wall" }
> +
> +struct S { int s; };
> +
> +S
> +foo ()
> +{
> +  return __builtin_bit_cast (S, 0);
> +}
> --- gcc/testsuite/g++.dg/cpp2a/bit-cast9.C.jj	2020-12-29 10:35:35.018101365 +0100
> +++ gcc/testsuite/g++.dg/cpp2a/bit-cast9.C	2020-12-29 10:35:05.905431494 +0100
> @@ -0,0 +1,15 @@
> +// PR c++/98469
> +// { dg-do compile { target c++20 } }
> +// { dg-options "-Wall" }
> +
> +template<typename T, typename F>
> +constexpr T
> +bit_cast (const F &f) noexcept
> +{
> +  return __builtin_bit_cast (T, f);
> +}
> +struct S { int s; };
> +constexpr int foo (const S &x) { return x.s; }
> +constexpr int bar () { return foo (bit_cast<S> (0)); }
> +constexpr int x = bar ();
> +static_assert (!x);
> 
> 	Jakub
> 

Re: [PATCH] c++: Fix ICE with __builtin_bit_cast [PR98469]

[Jakub Jelinek]

On Mon, Jan 04, 2021 at 03:44:46PM -0500, Jason Merrill wrote:
> This change is OK, but part of the problem is that we're trying to do
> overload resolution for an S copy/move constructor, which we shouldn't be
> because bit_cast is a prvalue, so in C++17 and up we should use it to
> directly initialize the target without any implied constructor call.
> 
> It seems we're mishandling this because the code in
> build_special_member_call specifically looks for TARGET_EXPR or CONSTRUCTOR,
> and BIT_CAST_EXPR is neither of those.
> 
> Wrapping a BIT_CAST_EXPR of aggregate type in a TARGET_EXPR would address
> this, and any other places that expect a class prvalue to come in the form
> of a TARGET_EXPR.

I can try that tomorrow.  Won't that cause copying through extra temporary
in some cases though, or is that guaranteed to be optimized?

	Jakub

Re: [PATCH] c++: Fix ICE with __builtin_bit_cast [PR98469]

[Jason Merrill]

On 1/4/21 3:48 PM, Jakub Jelinek wrote:
> On Mon, Jan 04, 2021 at 03:44:46PM -0500, Jason Merrill wrote:
>> This change is OK, but part of the problem is that we're trying to do
>> overload resolution for an S copy/move constructor, which we shouldn't be
>> because bit_cast is a prvalue, so in C++17 and up we should use it to
>> directly initialize the target without any implied constructor call.
>>
>> It seems we're mishandling this because the code in
>> build_special_member_call specifically looks for TARGET_EXPR or CONSTRUCTOR,
>> and BIT_CAST_EXPR is neither of those.
>>
>> Wrapping a BIT_CAST_EXPR of aggregate type in a TARGET_EXPR would address
>> this, and any other places that expect a class prvalue to come in the form
>> of a TARGET_EXPR.
> 
> I can try that tomorrow.  Won't that cause copying through extra temporary
> in some cases though, or is that guaranteed to be optimized?

It won't cause any extra copying when it's used to initialize another 
object (like the return value of std::bit_cast).  Class prvalues are 
always expressed with a TARGET_EXPR in the front end; the TARGET_EXPR 
melts away when used as an initializer, it only creates a temporary when 
it's used in another way.

Jason

[PATCH] c++, v2: Fix ICE with __builtin_bit_cast [PR98469]

[Jakub Jelinek]

On Mon, Jan 04, 2021 at 04:01:25PM -0500, Jason Merrill via Gcc-patches wrote:
> On 1/4/21 3:48 PM, Jakub Jelinek wrote:
> > On Mon, Jan 04, 2021 at 03:44:46PM -0500, Jason Merrill wrote:
> > > This change is OK, but part of the problem is that we're trying to do
> > > overload resolution for an S copy/move constructor, which we shouldn't be
> > > because bit_cast is a prvalue, so in C++17 and up we should use it to
> > > directly initialize the target without any implied constructor call.
> > > 
> > > It seems we're mishandling this because the code in
> > > build_special_member_call specifically looks for TARGET_EXPR or CONSTRUCTOR,
> > > and BIT_CAST_EXPR is neither of those.
> > > 
> > > Wrapping a BIT_CAST_EXPR of aggregate type in a TARGET_EXPR would address
> > > this, and any other places that expect a class prvalue to come in the form
> > > of a TARGET_EXPR.
> > 
> > I can try that tomorrow.  Won't that cause copying through extra temporary
> > in some cases though, or is that guaranteed to be optimized?
> 
> It won't cause any extra copying when it's used to initialize another object
> (like the return value of std::bit_cast).  Class prvalues are always
> expressed with a TARGET_EXPR in the front end; the TARGET_EXPR melts away
> when used as an initializer, it only creates a temporary when it's used in
> another way.

Ok, this version wraps it into a TARGET_EXPR then, it alone fixes the bug,
but I've kept the constexpr.c change too.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2021-01-05  Jakub Jelinek  <jakub@redhat.com>

	PR c++/98469
	* constexpr.c (cxx_eval_constant_expression) <case BIT_CAST_EXPR>:
	Punt if lval is true.
	* semantics.c (cp_build_bit_cast): Call get_target_expr_sfinae on
	the result if it has a class type.

	* g++.dg/cpp2a/bit-cast8.C: New test.
	* g++.dg/cpp2a/bit-cast9.C: New test.

--- gcc/cp/constexpr.c.jj	2021-01-04 10:25:48.750121531 +0100
+++ gcc/cp/constexpr.c	2021-01-05 11:41:38.315032636 +0100
@@ -6900,6 +6900,15 @@ cxx_eval_constant_expression (const cons
       return t;
 
     case BIT_CAST_EXPR:
+      if (lval)
+	{
+	  if (!ctx->quiet)
+	    error_at (EXPR_LOCATION (t),
+		      "address of a call to %qs is not a constant expression",
+		      "__builtin_bit_cast");
+	  *non_constant_p = true;
+	  return t;
+	}
       r = cxx_eval_bit_cast (ctx, t, non_constant_p, overflow_p);
       break;
 
--- gcc/cp/semantics.c.jj	2021-01-04 10:25:48.489124486 +0100
+++ gcc/cp/semantics.c	2021-01-05 11:27:49.327372582 +0100
@@ -10761,6 +10761,10 @@ cp_build_bit_cast (location_t loc, tree
 
   tree ret = build_min (BIT_CAST_EXPR, type, arg);
   SET_EXPR_LOCATION (ret, loc);
+
+  if (!processing_template_decl && CLASS_TYPE_P (type))
+    ret = get_target_expr_sfinae (ret, complain);
+
   return ret;
 }
 
--- gcc/testsuite/g++.dg/cpp2a/bit-cast8.C.jj	2021-01-05 11:41:38.315032636 +0100
+++ gcc/testsuite/g++.dg/cpp2a/bit-cast8.C	2021-01-05 11:41:38.315032636 +0100
@@ -0,0 +1,11 @@
+// PR c++/98469
+// { dg-do compile { target c++20 } }
+// { dg-options "-Wall" }
+
+struct S { int s; };
+
+S
+foo ()
+{
+  return __builtin_bit_cast (S, 0);
+}
--- gcc/testsuite/g++.dg/cpp2a/bit-cast9.C.jj	2021-01-05 11:41:38.315032636 +0100
+++ gcc/testsuite/g++.dg/cpp2a/bit-cast9.C	2021-01-05 11:41:38.315032636 +0100
@@ -0,0 +1,15 @@
+// PR c++/98469
+// { dg-do compile { target c++20 } }
+// { dg-options "-Wall" }
+
+template<typename T, typename F>
+constexpr T
+bit_cast (const F &f) noexcept
+{
+  return __builtin_bit_cast (T, f);
+}
+struct S { int s; };
+constexpr int foo (const S &x) { return x.s; }
+constexpr int bar () { return foo (bit_cast<S> (0)); }
+constexpr int x = bar ();
+static_assert (!x);


	Jakub

Re: [PATCH] c++, v2: Fix ICE with __builtin_bit_cast [PR98469]

[Jason Merrill]

On 1/5/21 10:26 AM, Jakub Jelinek wrote:
> On Mon, Jan 04, 2021 at 04:01:25PM -0500, Jason Merrill via Gcc-patches wrote:
>> On 1/4/21 3:48 PM, Jakub Jelinek wrote:
>>> On Mon, Jan 04, 2021 at 03:44:46PM -0500, Jason Merrill wrote:
>>>> This change is OK, but part of the problem is that we're trying to do
>>>> overload resolution for an S copy/move constructor, which we shouldn't be
>>>> because bit_cast is a prvalue, so in C++17 and up we should use it to
>>>> directly initialize the target without any implied constructor call.
>>>>
>>>> It seems we're mishandling this because the code in
>>>> build_special_member_call specifically looks for TARGET_EXPR or CONSTRUCTOR,
>>>> and BIT_CAST_EXPR is neither of those.
>>>>
>>>> Wrapping a BIT_CAST_EXPR of aggregate type in a TARGET_EXPR would address
>>>> this, and any other places that expect a class prvalue to come in the form
>>>> of a TARGET_EXPR.
>>>
>>> I can try that tomorrow.  Won't that cause copying through extra temporary
>>> in some cases though, or is that guaranteed to be optimized?
>>
>> It won't cause any extra copying when it's used to initialize another object
>> (like the return value of std::bit_cast).  Class prvalues are always
>> expressed with a TARGET_EXPR in the front end; the TARGET_EXPR melts away
>> when used as an initializer, it only creates a temporary when it's used in
>> another way.
> 
> Ok, this version wraps it into a TARGET_EXPR then, it alone fixes the bug,
> but I've kept the constexpr.c change too.
> 
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

OK.

> 2021-01-05  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR c++/98469
> 	* constexpr.c (cxx_eval_constant_expression) <case BIT_CAST_EXPR>:
> 	Punt if lval is true.
> 	* semantics.c (cp_build_bit_cast): Call get_target_expr_sfinae on
> 	the result if it has a class type.
> 
> 	* g++.dg/cpp2a/bit-cast8.C: New test.
> 	* g++.dg/cpp2a/bit-cast9.C: New test.
> 
> --- gcc/cp/constexpr.c.jj	2021-01-04 10:25:48.750121531 +0100
> +++ gcc/cp/constexpr.c	2021-01-05 11:41:38.315032636 +0100
> @@ -6900,6 +6900,15 @@ cxx_eval_constant_expression (const cons
>         return t;
>   
>       case BIT_CAST_EXPR:
> +      if (lval)
> +	{
> +	  if (!ctx->quiet)
> +	    error_at (EXPR_LOCATION (t),
> +		      "address of a call to %qs is not a constant expression",
> +		      "__builtin_bit_cast");
> +	  *non_constant_p = true;
> +	  return t;
> +	}
>         r = cxx_eval_bit_cast (ctx, t, non_constant_p, overflow_p);
>         break;
>   
> --- gcc/cp/semantics.c.jj	2021-01-04 10:25:48.489124486 +0100
> +++ gcc/cp/semantics.c	2021-01-05 11:27:49.327372582 +0100
> @@ -10761,6 +10761,10 @@ cp_build_bit_cast (location_t loc, tree
>   
>     tree ret = build_min (BIT_CAST_EXPR, type, arg);
>     SET_EXPR_LOCATION (ret, loc);
> +
> +  if (!processing_template_decl && CLASS_TYPE_P (type))
> +    ret = get_target_expr_sfinae (ret, complain);
> +
>     return ret;
>   }
>   
> --- gcc/testsuite/g++.dg/cpp2a/bit-cast8.C.jj	2021-01-05 11:41:38.315032636 +0100
> +++ gcc/testsuite/g++.dg/cpp2a/bit-cast8.C	2021-01-05 11:41:38.315032636 +0100
> @@ -0,0 +1,11 @@
> +// PR c++/98469
> +// { dg-do compile { target c++20 } }
> +// { dg-options "-Wall" }
> +
> +struct S { int s; };
> +
> +S
> +foo ()
> +{
> +  return __builtin_bit_cast (S, 0);
> +}
> --- gcc/testsuite/g++.dg/cpp2a/bit-cast9.C.jj	2021-01-05 11:41:38.315032636 +0100
> +++ gcc/testsuite/g++.dg/cpp2a/bit-cast9.C	2021-01-05 11:41:38.315032636 +0100
> @@ -0,0 +1,15 @@
> +// PR c++/98469
> +// { dg-do compile { target c++20 } }
> +// { dg-options "-Wall" }
> +
> +template<typename T, typename F>
> +constexpr T
> +bit_cast (const F &f) noexcept
> +{
> +  return __builtin_bit_cast (T, f);
> +}
> +struct S { int s; };
> +constexpr int foo (const S &x) { return x.s; }
> +constexpr int bar () { return foo (bit_cast<S> (0)); }
> +constexpr int x = bar ();
> +static_assert (!x);
> 
> 
> 	Jakub
> 

Re: [PATCH] c++, v2: Fix ICE with __builtin_bit_cast [PR98469]

[Jason Merrill]

[-- Attachment #1: Type: text/plain, Size: 1647 bytes --]

On 1/5/21 10:26 AM, Jakub Jelinek wrote:
> On Mon, Jan 04, 2021 at 04:01:25PM -0500, Jason Merrill via Gcc-patches wrote:
>> On 1/4/21 3:48 PM, Jakub Jelinek wrote:
>>> On Mon, Jan 04, 2021 at 03:44:46PM -0500, Jason Merrill wrote:
>>>> This change is OK, but part of the problem is that we're trying to do
>>>> overload resolution for an S copy/move constructor, which we shouldn't be
>>>> because bit_cast is a prvalue, so in C++17 and up we should use it to
>>>> directly initialize the target without any implied constructor call.
>>>>
>>>> It seems we're mishandling this because the code in
>>>> build_special_member_call specifically looks for TARGET_EXPR or CONSTRUCTOR,
>>>> and BIT_CAST_EXPR is neither of those.
>>>>
>>>> Wrapping a BIT_CAST_EXPR of aggregate type in a TARGET_EXPR would address
>>>> this, and any other places that expect a class prvalue to come in the form
>>>> of a TARGET_EXPR.
>>>
>>> I can try that tomorrow.  Won't that cause copying through extra temporary
>>> in some cases though, or is that guaranteed to be optimized?
>>
>> It won't cause any extra copying when it's used to initialize another object
>> (like the return value of std::bit_cast).  Class prvalues are always
>> expressed with a TARGET_EXPR in the front end; the TARGET_EXPR melts away
>> when used as an initializer, it only creates a temporary when it's used in
>> another way.
> 
> Ok, this version wraps it into a TARGET_EXPR then, it alone fixes the bug,
> but I've kept the constexpr.c change too.

This patch corrects this and one other place to not be as dependent on 
TARGET_EXPR, but I think I'm going to save it for stage 1.

Jason

[-- Attachment #2: glvp.diff --]
[-- Type: text/x-patch, Size: 1952 bytes --]

commit 0d732b8c7fb3f8378dc1c894358bb5d766e6be5d
Author: Jason Merrill <jason@redhat.com>
Date:   Mon Jan 4 16:11:08 2021 -0500

    c++: Tweak prvalue test [PR98469]
    
    Discussing the 98469 patch and class prvalues with Jakub also inspired me to
    change the place that was mishandling BIT_CAST_EXPR and one other to use the
    lvalue_kind machinery to decide whether something is a prvalue, instead of
    looking specifically for a TARGET_EXPR.
    
    gcc/cp/ChangeLog:
    
            * call.c (build_special_member_call): Use !glvalue_p rather
            than specific tree codes to test for prvalue.
            (conv_is_prvalue): Likewise.
            (implicit_conversion): Check CLASS_TYPE_P first.

diff --git a/gcc/cp/call.c b/gcc/cp/call.c
index 218157088ef..e2d2b23e449 100644
--- a/gcc/cp/call.c
+++ b/gcc/cp/call.c
@@ -2118,8 +2118,8 @@ implicit_conversion (tree to, tree from, tree expr, bool c_cast_p,
 					    flags, complain);
   if (!conv || conv->bad_p)
     return conv;
-  if (conv_is_prvalue (conv)
-      && CLASS_TYPE_P (conv->type)
+  if (CLASS_TYPE_P (conv->type)
+      && conv_is_prvalue (conv)
       && CLASSTYPE_PURE_VIRTUALS (conv->type))
     conv->bad_p = true;
   return conv;
@@ -8500,8 +8500,7 @@ conv_is_prvalue (conversion *c)
     return true;
   if (c->kind == ck_user && !TYPE_REF_P (c->type))
     return true;
-  if (c->kind == ck_identity && c->u.expr
-      && TREE_CODE (c->u.expr) == TARGET_EXPR)
+  if (c->kind == ck_identity && c->u.expr && !glvalue_p (c->u.expr))
     return true;
 
   return false;
@@ -9950,8 +9949,7 @@ build_special_member_call (tree instance, tree name, vec<tree, va_gc> **args,
 	  && CONSTRUCTOR_NELTS (arg) == 1)
 	arg = CONSTRUCTOR_ELT (arg, 0)->value;
 
-      if ((TREE_CODE (arg) == TARGET_EXPR
-	   || TREE_CODE (arg) == CONSTRUCTOR)
+      if (!glvalue_p (arg)
 	  && (same_type_ignoring_top_level_qualifiers_p
 	      (class_type, TREE_TYPE (arg))))
 	{