public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Christian Bruel <christian.bruel@st.com>
To: Bernd Schmidt <bschmidt@redhat.com>, "law@redhat.com" <law@redhat.com>
Cc: "gcc-patches@gcc.gnu.org" <gcc-patches@gcc.gnu.org>
Subject: refactoring TARGET_PTRMEMFUNC_VBIT_LOCATION checks
Date: Fri, 16 Oct 2015 08:03:00 -0000	[thread overview]
Message-ID: <5620AED8.4000805@st.com> (raw)
In-Reply-To: <561B973F.90302@redhat.com>

[-- Attachment #1: Type: text/plain, Size: 1425 bytes --]

Hello,

On 10/12/2015 01:19 PM, Bernd Schmidt wrote:
> On 10/12/2015 12:56 PM, Christian Bruel wrote:
>> yes I see, I was hoping to avoid a new hook, but as you said it seems
>> mandatory for the mere declaration case.
>>
>> Here is one proposal, it defaults to nothing and the ARM implementation
>> does not need to handle the vptr bit setting. so that simplifies a lot
>> the things.
>>
>> The hook is called from rest_of_decl_compilation for mere declarations
>> and allocate_struct_function for definitions.
>
> This looks good to me. I still think we also want your vptr patch.
>
 > Bernd
 >

Here is the "vptr patch".

2 changes to watch for the review :

  - I replaced the DECL_ALIGN checks by FUNCTION_BOUNDARY because there 
is no user align at that time of processing and it's consistent with the 
default TARGET_PTRMEMFUNC_VBIT_LOCATION macro implementation (or 
TARGET_PTRMEMFUNC_VBIT_LOCATION should also be function of DECL_ALIGN)

  - The second chunk from cp/lambda.c did not seem to be needed (?).
Thunks are FUNCTION_TYPE and I don't see why they would need to carry 
the vptr bit information (for METHOD_TYPE from my understanding). Or did 
I miss something here ?

bootstrapped on x86. no c/c++ new failures for
   x86 (ptrmemfunc_vbit_in_pfn, align 8)
   arm (ptrmemfunc_vbit_in_delta, align 16 or 32)
   sh4 (ptrmemfunc_vbit_in_pfn, align 16)

It's a code cleanup/refactoring. No new testcase.

thanks

[-- Attachment #2: align5.patch --]
[-- Type: text/x-patch, Size: 4934 bytes --]

2015-09-29  Christian Bruel  <christian.bruel@st.com>

	* function.h (MINIMUM_METHOD_BOUNDARY): New macro.
	* cp/decl.c (grokfndecl): Set DECL_ALIGN with MINIMUM_METHOD_BOUNDARY.
	* cp/method.c (implicitly_declare_fn): Likewise.
	* cp/lambda.c (maybe_add_lambda_conv_op): Likewise, remove wrong setting.
	* java/class.c (add_method_1): Likewise.



	  
Index: cp/decl.c
===================================================================
--- cp/decl.c	(revision 228756)
+++ cp/decl.c	(working copy)
@@ -7826,6 +7826,8 @@ grokfndecl (tree ctype,
       parm = build_this_parm (type, quals);
       DECL_CHAIN (parm) = parms;
       parms = parm;
+ 
+      /* Allocate space to hold the vptr bit if needed.  */
+      DECL_ALIGN (decl) = MINIMUM_METHOD_BOUNDARY;
     }
   DECL_ARGUMENTS (decl) = parms;
   for (t = parms; t; t = DECL_CHAIN (t))
@@ -7849,14 +7851,6 @@ grokfndecl (tree ctype,
       break;
     }
 
-  /* If pointers to member functions use the least significant bit to
-     indicate whether a function is virtual, ensure a pointer
-     to this function will have that bit clear.  */
-  if (TARGET_PTRMEMFUNC_VBIT_LOCATION == ptrmemfunc_vbit_in_pfn
-      && TREE_CODE (type) == METHOD_TYPE
-      && DECL_ALIGN (decl) < 2 * BITS_PER_UNIT)
-    DECL_ALIGN (decl) = 2 * BITS_PER_UNIT;
-
   if (friendp
       && TREE_CODE (orig_declarator) == TEMPLATE_ID_EXPR)
     {
Index: cp/lambda.c
===================================================================
--- cp/lambda.c	(revision 228756)
+++ cp/lambda.c	(working copy)
@@ -1006,11 +1006,8 @@ maybe_add_lambda_conv_op (tree type)
   tree convfn = build_lang_decl (FUNCTION_DECL, name, fntype);
   tree fn = convfn;
   DECL_SOURCE_LOCATION (fn) = DECL_SOURCE_LOCATION (callop);
-
-  if (TARGET_PTRMEMFUNC_VBIT_LOCATION == ptrmemfunc_vbit_in_pfn
-      && DECL_ALIGN (fn) < 2 * BITS_PER_UNIT)
-    DECL_ALIGN (fn) = 2 * BITS_PER_UNIT;
-
+  DECL_ALIGN (fn) = MINIMUM_METHOD_BOUNDARY;
   SET_OVERLOADED_OPERATOR_CODE (fn, TYPE_EXPR);
   grokclassfn (type, fn, NO_SPECIAL);
   set_linkage_according_to_type (type, fn);
@@ -1042,9 +1039,7 @@ maybe_add_lambda_conv_op (tree type)
   tree statfn = build_lang_decl (FUNCTION_DECL, name, stattype);
   fn = statfn;
   DECL_SOURCE_LOCATION (fn) = DECL_SOURCE_LOCATION (callop);
-  if (TARGET_PTRMEMFUNC_VBIT_LOCATION == ptrmemfunc_vbit_in_pfn
-      && DECL_ALIGN (fn) < 2 * BITS_PER_UNIT)
-    DECL_ALIGN (fn) = 2 * BITS_PER_UNIT;
   grokclassfn (type, fn, NO_SPECIAL);
   set_linkage_according_to_type (type, fn);
   rest_of_decl_compilation (fn, toplevel_bindings_p (), at_eof);
Index: cp/method.c
===================================================================
--- cp/method.c	(revision 228756)
+++ cp/method.c	(working copy)
@@ -1849,13 +1849,9 @@ implicitly_declare_fn (special_function_
       DECL_ASSIGNMENT_OPERATOR_P (fn) = 1;
       SET_OVERLOADED_OPERATOR_CODE (fn, NOP_EXPR);
     }
-  
-  /* If pointers to member functions use the least significant bit to
-     indicate whether a function is virtual, ensure a pointer
-     to this function will have that bit clear.  */
-  if (TARGET_PTRMEMFUNC_VBIT_LOCATION == ptrmemfunc_vbit_in_pfn
-      && DECL_ALIGN (fn) < 2 * BITS_PER_UNIT)
-    DECL_ALIGN (fn) = 2 * BITS_PER_UNIT;
+
+  DECL_ALIGN (fn) = MINIMUM_METHOD_BOUNDARY;
 
   /* Create the explicit arguments.  */
   if (rhs_parm_type)
Index: function.h
===================================================================
--- function.h	(revision 228756)
+++ function.h	(working copy)
@@ -537,6 +537,13 @@ do {								\
 #define ASLK_REDUCE_ALIGN 1
 #define ASLK_RECORD_PAD 2
 
+/* If pointers to member functions use the least significant bit to
+   indicate whether a function is virtual, ensure a pointer
+   to this function will have that bit clear.  */
+#define MINIMUM_METHOD_BOUNDARY \
+  ((TARGET_PTRMEMFUNC_VBIT_LOCATION == ptrmemfunc_vbit_in_pfn)	     \
+   ? MAX (FUNCTION_BOUNDARY, 2 * BITS_PER_UNIT) : FUNCTION_BOUNDARY)
+
 \f
 
 extern void push_function_context (void);
Index: java/class.c
===================================================================
--- java/class.c	(revision 228756)
+++ java/class.c	(working copy)
@@ -779,13 +779,8 @@ add_method_1 (tree this_class, int acces
   DECL_CHAIN (fndecl) = TYPE_METHODS (this_class);
   TYPE_METHODS (this_class) = fndecl;
 
-  /* If pointers to member functions use the least significant bit to
-     indicate whether a function is virtual, ensure a pointer
-     to this function will have that bit clear.  */
-  if (TARGET_PTRMEMFUNC_VBIT_LOCATION == ptrmemfunc_vbit_in_pfn
-      && !(access_flags & ACC_STATIC)
-      && DECL_ALIGN (fndecl) < 2 * BITS_PER_UNIT)
-    DECL_ALIGN (fndecl) = 2 * BITS_PER_UNIT;
+  if (!(access_flags & ACC_STATIC))
+    DECL_ALIGN (fndecl) = MINIMUM_METHOD_BOUNDARY;
 
   /* Notice that this is a finalizer and update the class type
      accordingly. This is used to optimize instance allocation. */

  parent reply	other threads:[~2015-10-16  8:03 UTC|newest]

Thread overview: 28+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-09-29 13:44 [PATCH ARM]: PR67745: Fix function alignment after __attribute__ 2/2 Christian Bruel
2015-09-29 13:59 ` Christian Bruel
2015-09-30 19:30 ` Jeff Law
2015-10-01  7:12   ` Christian Bruel
2015-10-01 16:11     ` Christian Bruel
2015-10-07  7:05       ` Christian Bruel
2015-10-07 10:18         ` Bernd Schmidt
2015-10-07 10:45           ` Christian Bruel
2015-10-07 17:37             ` Bernd Schmidt
2015-10-07 17:50               ` Bernd Schmidt
2015-10-08 13:14                 ` Christian Bruel
2015-10-08 13:20                   ` Bernd Schmidt
2015-10-08 13:51                     ` Christian Bruel
2015-10-08 14:30                       ` Bernd Schmidt
2015-10-12 10:56                         ` Christian Bruel
2015-10-12 11:19                           ` Bernd Schmidt
2015-10-12 11:26                             ` Christian Bruel
2015-10-16  8:03                             ` Christian Bruel [this message]
2015-10-16  9:47                               ` refactoring TARGET_PTRMEMFUNC_VBIT_LOCATION checks Bernd Schmidt
2015-10-16 10:50                                 ` Christian Bruel
2015-10-16 10:56                                   ` Christian Bruel
2015-10-16 10:56                                   ` Bernd Schmidt
2015-10-13  8:03                           ` [PATCH ARM]: PR67745: Fix function alignment after __attribute__ 2/2 Ramana Radhakrishnan
2015-10-13 10:18                             ` Christian Bruel
2015-10-16 14:18                             ` Christian Bruel
2015-10-08 14:01               ` Christian Bruel
2015-10-08 14:05                 ` Bernd Schmidt
2015-10-07 17:40         ` Jeff Law

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=5620AED8.4000805@st.com \
    --to=christian.bruel@st.com \
    --cc=bschmidt@redhat.com \
    --cc=gcc-patches@gcc.gnu.org \
    --cc=law@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).