[Bug c++/44320] New: ADL names are looked up in namespaces that are only referenced by template arguments
I believe that the following is a bug. it is accepted by MSVC and comeau.
GCC seems to ADL-lookup names in namespaces that are only used in the template
arguments of the function arguments, for example:
Aotherns::type a;
adl_func(a); //otherns::adl_func is called
this can cause unintended conflicts with non-function names in otherns, e.g.
if the ADL function is called apply and the argument type happens to be a
template instantiated with a type from namespace boost::mpl, only the fact that
a header containing the type boost::mpl::apply is included can cause a compile
error.
here's a simplified test case that causes a conflict:
namespace mpl{
class apply{};
class vector{};
}
templateclass T
void apply(T const ){}
templateclass T
class A{};
int main(){
Ampl::vector a;
apply(a);
}
--
Summary: ADL names are looked up in namespaces that are only
referenced by template arguments
Product: gcc
Version: 4.3.2
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: sstrasser at systemhaus-gruppe dot de
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44320
[Bug c++/44320] ADL names are looked up in namespaces that are only referenced by template arguments
--- Comment #1 from sstrasser at systemhaus-gruppe dot de 2010-05-29 07:45 --- a workaround (other than renaming the ADL function) would be appreciated -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44320
[Bug bootstrap/43170] gcc 4.5 20100218 bootstrap compare fails on os x 10.6
--- Comment #31 from iains at gcc dot gnu dot org 2010-05-29 07:47 --- 4.5-branch (as of r160013) has an error in config.gcc (which I just fixed on trunk yesterday, with r159979) in which several t-make fragments are included twice on x86_64-*-darwin*. I wonder if this could be the origin of the problem? If anyone is interested to try it out I could make you a patch for 4.5-branch to do the same fix. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43170
[Bug bootstrap/44304] Building gcc 4.5.0 under Snow Leopard : stages 2 3 differ
--- Comment #4 from iains at gcc dot gnu dot org 2010-05-29 07:49 --- 4.5-branch (as of r160013) has an error in config.gcc (which I just fixed on trunk yesterday, with r159979) in which several t-make fragments are included twice on x86_64-*-darwin*. If anyone is interested to try it out I could make you a patch for 4.5-branch to do the same fix. still, first we need a repeatable phenomenon - and to be sure that it's not tool-related. -- iains at gcc dot gnu dot org changed: What|Removed |Added CC||iains at gcc dot gnu dot org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44304
[Bug tree-optimization/44137] [4.6 Regression]: objc.dg/torture/tls/thr-init-2.m and thr-init.m
--- Comment #3 from iains at gcc dot gnu dot org 2010-05-29 08:01 --- (In reply to comment #2) Fixed after 159920 but before or including 159930. At closer inspection, it has to be r159929. :) On the other hand, from the patch message it seems it's just intended to be a stop-gap measure, so I'll leave it to Iain to close this PR. Indeed, a stop-gap measure; whilst I believe that this PR is part a duplicate of PR44140 and part a duplicate of 44132, I think to leave it open for now - and try to resolve once PR44140 is fixed. Thus I am going to mark it as depending on PR44140. -- iains at gcc dot gnu dot org changed: What|Removed |Added BugsThisDependsOn||44140 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44137
[Bug c++/41233] Templated conversion operator produces symbol name that won't demangle
--- Comment #4 from gcc at magfr dot user dot lysator dot liu dot se 2010-05-29 08:17 --- I have run head-first into this problem as well. I have found that _ZN1tcvT_IiEEv ( t::operator intint() ) works but _ZN1tcvKT_IiEEv ( t::operator int constint() ) fails so it seems the problem is that the template isn't detected, and looking at the code in libiberty/cp-demangle.c (r159908) we have line 4689 that tries to detect if the cast target is a template type but only handles the case where the type is unadorned by anything and then there is the comment on lines 4696-4699 that explains the problem. -- gcc at magfr dot user dot lysator dot liu dot se changed: What|Removed |Added CC||gcc at magfr dot user dot ||lysator dot liu dot se http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41233
[Bug c++/44320] ADL names are looked up in namespaces that are only referenced by template arguments
--- Comment #2 from paolo dot carlini at oracle dot com 2010-05-29 09:30 --- Works fine in 4_5-branch and mainline, the behavior in 4_3-branch and 4_4-branch isn't a regression in any way, as fas as I can see. -- paolo dot carlini at oracle dot com changed: What|Removed |Added Status|UNCONFIRMED |RESOLVED Known to work||4.5.0 4.5.1 4.6.0 Resolution||WORKSFORME http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44320
[Bug middle-end/44321] New: attribute warn_unused_result fails under inlining.
In the (to-be) attached testcase, GCC warns about an unused result when the
return value of the function is ignored, or cast to void, but it fails to warn
when the cast-to-void is hidden inside an inline function, like so:
$ cat -n wur.c
1
2 static inline void ignore_value (int i) { (void) i; }
3
4 extern int foo (void) __attribute__ ((warn_unused_result));
5
6 void bar1 (void)
7 {
8foo ();
9 }
10
11 void bar2 (void)
12 {
13(void) foo ();
14 }
15
16 void bar3 (void)
17 {
18ignore_value (foo ());
19 }
$ gcc-4 -c wur.c -o wur.o --save-temps -W -Wall -Wextra
wur.c: In function 'bar2':
wur.c:13:3: warning: ignoring return value of 'foo', declared with attribute
war
n_unused_result [-Wunused-result]
wur.c: In function 'bar1':
wur.c:8:7: warning: ignoring return value of 'foo', declared with attribute
warn
_unused_result [-Wunused-result]
This happens at all -O levels. I haven't filled in the host/target/build
triplets because I don't suppose it is in any way specific to any of them.
BTW, it's interesting that the warnings come out in reverse order, is that
supposed to happen?
--
Summary: attribute warn_unused_result fails under inlining.
Product: gcc
Version: 4.6.0
Status: UNCONFIRMED
Keywords: diagnostic
Severity: normal
Priority: P3
Component: middle-end
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: davek at gcc dot gnu dot org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44321
[Bug middle-end/44321] attribute warn_unused_result fails under inlining.
--- Comment #1 from davek at gcc dot gnu dot org 2010-05-29 11:33 --- Created an attachment (id=20771) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=20771action=view) testcase as per initial comment. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44321
[Bug tree-optimization/44290] [4.5 Regression] arm linux kernel crahes when built with -fipa-sra
--- Comment #8 from mikpe at it dot uu dot se 2010-05-29 11:35 --- (In reply to comment #6) I note that copypage-xscale.c:xscale_mc_copy_user_highpage() calls a __naked function to do the bulk copy. Converting that to a plain inline function (changing 'pc' to 'lr' in the final instruction that restores the scrach regs), does not prevent the crash. So I suspect a plain C code miscompilation. Actually that conversion away from __naked may have been flawed. What I'm seeing is that r148981 causes gcc to clone the __naked function and change its calling conventions in ways that don't match the proper function call ABI. This breaks the body of the __naked function which is just a big asm() statement. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290
[Bug middle-end/44321] attribute warn_unused_result fails under inlining.
--- Comment #2 from rguenth at gcc dot gnu dot org 2010-05-29 11:40 ---
You can add
void bar4 (void)
{
int dummy;
dummy = foo ();
}
so I'm not sure the ignore_value () function call isn't a use. In fact
if you externalize that function it is at least a possible use.
Which also raises the question whether (void) is really more ignoring
the result than the assignment to a dummy variable.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44321
[Bug target/44290] [4.5 Regression] arm linux kernel crahes when built with -fipa-sra, __naked attribute is broken
--- Comment #9 from rguenth at gcc dot gnu dot org 2010-05-29 11:42 --- (In reply to comment #8) (In reply to comment #6) I note that copypage-xscale.c:xscale_mc_copy_user_highpage() calls a __naked function to do the bulk copy. Converting that to a plain inline function (changing 'pc' to 'lr' in the final instruction that restores the scrach regs), does not prevent the crash. So I suspect a plain C code miscompilation. Actually that conversion away from __naked may have been flawed. What I'm seeing is that r148981 causes gcc to clone the __naked function and change its calling conventions in ways that don't match the proper function call ABI. This breaks the body of the __naked function which is just a big asm() statement. Well. The arm backend needs to mark the function as used in non-visible ways then. Thus this is a target bug. -- rguenth at gcc dot gnu dot org changed: What|Removed |Added Status|UNCONFIRMED |NEW Component|tree-optimization |target Ever Confirmed|0 |1 Last reconfirmed|-00-00 00:00:00 |2010-05-29 11:42:59 date|| Summary|[4.5 Regression] arm linux |[4.5 Regression] arm linux |kernel crahes when built|kernel crahes when built |with -fipa-sra |with -fipa-sra, __naked ||attribute is broken http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290
[Bug target/44290] [4.5 Regression] arm linux kernel crahes when built with -fipa-sra, __naked attribute is broken
--- Comment #10 from rguenth at gcc dot gnu dot org 2010-05-29 11:47
---
Or rather, if you have
void __attribute__((naked)) foo (int i)
{
asm(use i);
}
without any inputs refering to i that is invalid. Like I see in the
attached preprocessed source:
static void __attribute__((naked)) __attribute__((no_instrument_function))
v4wb_copy_user_page(void *kto, const void *kfrom)
{
asm( stmfd sp!, {r4, lr} @ 2\n mov r2, %0
@ 1\n ldmia r1!, {r3, r4, ip, lr} @ 4\n1: mcr
p15, 0, r0, c7, c6, 1 @ 1 invalidate D line\n stmia r0!,
{r3, r4, ip, lr} @ 4\n ldmia r1!, {r3, r4, ip, lr} @
4+1\n stmia r0!, {r3, r4, ip, lr} @ 4\n ldmia r1!, {r3, r4, ip,
lr} @ 4\n mcr p15, 0, r0, c7, c6, 1 @ 1 invalidate
D line\n stmia r0!, {r3, r4, ip, lr} @ 4\n ldmia r1!,
{r3, r4, ip, lr} @ 4\n subsr2, r2, #1 @
1\n stmia r0!, {r3, r4, ip, lr} @ 4\n ldmneia r1!, {r3, r4, ip,
lr} @ 4\n bne 1b @ 1\n mcr
p15, 0, r1, c7, c10, 4 @ 1 drain WB\nldmfdsp!, {r4, pc}
@ 3
# 46 /home/kraj/work/linux-2.6.34/arch/arm/mm/copypage-v4wb.c
:
: I (((1UL) 12) / 64));
}
kto and kfrom are unused.
--
rguenth at gcc dot gnu dot org changed:
What|Removed |Added
Status|NEW |RESOLVED
Resolution||INVALID
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290
[Bug target/44290] [4.5 Regression] arm linux kernel crahes when built with -fipa-sra, __naked attribute is broken
--- Comment #11 from rguenth at gcc dot gnu dot org 2010-05-29 11:50 --- (it seems quite stupid to have naked functions with only an asm inside in the first place - you can equally well use plain assembly) -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290
[Bug c/44322] New: Bogus warning when assigning pointer-to-array with both const and restrict
The following code is valid according to C99:
void * restrict const a[2];
void * restrict const (*p2)[2];
void foo(void) {
p2 = a;
}
However GCC complains with:
a.c: In function 'foo':
a.c:14:7: warning: assignment from incompatible pointer type
If either const or restrict (or both) is omitted from the declaration of
a and p, no warning is generated.
Despite the warning, code generation and optimization behave properly.
FYI, Clang 1.0.2 accepts this code without warning.
--
Summary: Bogus warning when assigning pointer-to-array with both
const and restrict
Product: gcc
Version: 4.5.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: r dot c dot poss at uva dot nl
GCC build triplet: x86_64-apple-darwin10
GCC host triplet: x86_64-apple-darwin10
GCC target triplet: x86_64-apple-darwin10
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44322
Re: [Bug target/44290] [4.5 Regression] arm linux kernel crahes when built with -fipa-sra, __naked attribute is broken
The naked attribute should cause two things noinline and noclone. Sent from my iPhone On May 29, 2010, at 4:50 AM, rguenth at gcc dot gnu dot org gcc-bugzi...@gcc.gnu.org wrote: --- Comment #11 from rguenth at gcc dot gnu dot org 2010-05-29 11:50 --- (it seems quite stupid to have naked functions with only an asm inside in the first place - you can equally well use plain assembly) -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290
[Bug target/44290] [4.5 Regression] arm linux kernel crahes when built with -fipa-sra, __naked attribute is broken
--- Comment #12 from pinskia at gmail dot com 2010-05-29 13:30 --- Subject: Re: [4.5 Regression] arm linux kernel crahes when built with -fipa-sra, __naked attribute is broken The naked attribute should cause two things noinline and noclone. Sent from my iPhone On May 29, 2010, at 4:50 AM, rguenth at gcc dot gnu dot org gcc-bugzi...@gcc.gnu.org wrote: --- Comment #11 from rguenth at gcc dot gnu dot org 2010-05-29 11:50 --- (it seems quite stupid to have naked functions with only an asm inside in the first place - you can equally well use plain assembly) -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290 -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290
[Bug target/44261] Multiplying -1 by NaN is not valid.
--- Comment #4 from danglin at gcc dot gnu dot org 2010-05-29 14:16 --- Subject: Bug 44261 Author: danglin Date: Sat May 29 14:16:11 2010 New Revision: 160027 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=160027 Log: PR target/44261 config/pa/pa.md (negdf2_slow, negsf2_slow): New patterns. (negdf2): Adjust expander pattern and use negdf2_slow. (negsf2): Likewise. Modified: trunk/gcc/ChangeLog trunk/gcc/config/pa/pa.md -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44261
[Bug target/44266] stack frame lacks parameter save area
--- Comment #5 from bergner at gcc dot gnu dot org 2010-05-29 14:17 --- Subject: Bug 44266 Author: bergner Date: Sat May 29 14:17:26 2010 New Revision: 160028 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=160028 Log: Backport from mainline: 2010-05-28 Alan Modra amo...@gmail.com PR target/44266 * config/rs6000/rs6000.c (rs6000_legitimize_tls_address): Use emit_library_call machinery to set up __tls_get_addr calls. Modified: branches/ibm/gcc-4_4-branch/gcc/ChangeLog.ibm branches/ibm/gcc-4_4-branch/gcc/config/rs6000/rs6000.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44266
[Bug target/44290] [4.5 Regression] arm linux kernel crahes when built with -fipa-sra, __naked attribute is broken
--- Comment #13 from mikpe at it dot uu dot se 2010-05-29 14:27 ---
(In reply to comment #10)
Or rather, if you have
void __attribute__((naked)) foo (int i)
{
asm(use i);
}
without any inputs refering to i that is invalid.
Not according to gcc/doc/extend.texi:
@item naked
@cindex function without a prologue/epilogue code
Use this attribute on the ARM, AVR, IP2K, RX and SPU ports to indicate that
the specified function does not need prologue/epilogue sequences generated by
the compiler. It is up to the programmer to provide these sequences. The
only statements that can be safely included in naked functions are
@code{asm} statements that do not have operands.
Note: do not have operands. Thus the only way such an asm() can refer to
parameters is by assuming a standard function call sequence and hardcoding
corresponding register numbers or stack frame offsets.
However, even if the asm() refers to those parameters via r(...) inputs,
gcc-4.5 changes the register assignment to not agree with the standard call
sequence, I'll attach a small test case showing that in a moment.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290
[Bug target/44261] Multiplying -1 by NaN is not valid.
--- Comment #5 from danglin at gcc dot gnu dot org 2010-05-29 14:33 --- Fixed on trunk. -- danglin at gcc dot gnu dot org changed: What|Removed |Added Status|UNCONFIRMED |RESOLVED Resolution||FIXED http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44261
[Bug target/44290] [4.5 Regression] arm linux kernel crahes when built with -fipa-sra, __naked attribute is broken
--- Comment #14 from mikpe at it dot uu dot se 2010-05-29 14:39 --- (In reply to comment #11) (it seems quite stupid to have naked functions with only an asm inside in the first place - you can equally well use plain assembly) Except that with plain asm() for an entire function definition you'd also have to include boring preamble/postamble stuff like .align/.type/.size if you want it to appear as a proper function, and you still have to declarate a prototype. And the reason for making it a separate function rather than an inline asm() is probably related to register assignment: a separate function can (could) make assumptions about parameter registers and scratch registers. With inline asm() you have to be much more elaborate, esp. if you have constraints that gcc cannot express, like even/odd register pairs on ARM. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290
[Bug target/44290] [4.5 Regression] __naked attribute is broken
--- Comment #15 from rguenth at gcc dot gnu dot org 2010-05-29 14:45
---
(In reply to comment #13)
(In reply to comment #10)
Or rather, if you have
void __attribute__((naked)) foo (int i)
{
asm(use i);
}
without any inputs refering to i that is invalid.
Not according to gcc/doc/extend.texi:
@item naked
@cindex function without a prologue/epilogue code
Use this attribute on the ARM, AVR, IP2K, RX and SPU ports to indicate that
the specified function does not need prologue/epilogue sequences generated
by
the compiler. It is up to the programmer to provide these sequences. The
only statements that can be safely included in naked functions are
@code{asm} statements that do not have operands.
Note: do not have operands. Thus the only way such an asm() can refer to
parameters is by assuming a standard function call sequence and hardcoding
corresponding register numbers or stack frame offsets.
Then the target has to properly communicate this to the middle-end.
However, even if the asm() refers to those parameters via r(...) inputs,
gcc-4.5 changes the register assignment to not agree with the standard call
sequence, I'll attach a small test case showing that in a moment.
I'd have required dummy inputs like g (kto), g (kfrom) not used by
the actual assembly.
For now re-open as a target bug.
--
rguenth at gcc dot gnu dot org changed:
What|Removed |Added
Status|RESOLVED|REOPENED
Resolution|INVALID |
Summary|[4.5 Regression] arm linux |[4.5 Regression] __naked
|kernel crahes when built|attribute is broken
|with -fipa-sra, __naked |
|attribute is broken |
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290
[Bug target/44290] [4.5 Regression] __naked attribute is broken
--- Comment #16 from rguenth at gcc dot gnu dot org 2010-05-29 14:46 --- And CC a arm maintainer. There might be more target-specific attributes that need adjustment. -- rguenth at gcc dot gnu dot org changed: What|Removed |Added CC||rearnsha at gcc dot gnu dot ||org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290
[Bug target/44290] [4.5 Regression] __naked attribute is broken
--- Comment #17 from mikpe at it dot uu dot se 2010-05-29 14:47 --- Created an attachment (id=20772) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=20772action=view) test case from copypage-xscale.c This is distilled from the kernel's copypage-xscale.c file and illustrates the issue. With gcc-4.4 the __naked__ function foo() is called with the standard call sequence register assignment, so the asm() body of foo() works. With gcc-4.5 foo() is cloned and gets its second parameter `to' in r0 (not r1 as expected), and the body of foo() is modified to set up the actual first parameter (fie[0]) in r1 (not r0 as expected). Obviously the asm() then breaks. Compiling with -fno-ipa-cp avoids this problem, as does adding __noclone__ and __noinline__ to foo()'s function definition. I don't immediately see how to enforce __noclone__ and __noinline__ in the ARM backend when it sees __naked__. Any ideas? -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290
[Bug bootstrap/44315] [4.6 Regression] Circular build/gencondmd.o - insn-flags.h dependency dropped
--- Comment #2 from hjl at gcc dot gnu dot org 2010-05-29 15:05 --- Subject: Bug 44315 Author: hjl Date: Sat May 29 15:05:30 2010 New Revision: 160029 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=160029 Log: Add a missing `\'. 2010-05-29 H.J. Lu hongjiu...@intel.com PR bootstrap/44315 * Makefile.in (build/gencondmd.o): Add a missing `\'. Modified: trunk/gcc/ChangeLog trunk/gcc/Makefile.in -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44315
[Bug libstdc++/44268] abi docs say that hppa-linux defaults to libgcc_s.so.2
--- Comment #5 from dave at hiauly1 dot hia dot nrc dot ca 2010-05-29 15:11 --- Subject: Re: abi docs say that hppa-linux defaults to libgcc_s.so.2 Dave, when did the hppa-linux so version change from 2 to 4? I'd like to document that, rather than just say it's always 1 or 4 Oops, I have messed up. After some discussion, it was decided to revert the change. It was supposed to have been reverted on 2010-01-31, but the change apparently didn't work at least on the trunk. The original change was made on 2010-01-02. On further review, I see that I didn't mess up the revert in January! The sjlj and elf shared library versions were set to 3 and 4, respectively, in revision 109894 on Jan. 18, 2006 on the trunk. The version was previously set to 2 in revision 77920 on Feb. 16, 2004. Prior to that, the version was 1. I would have to dig some more to determine the version numbers for specific releases. Dave -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44268
[Bug rtl-optimization/44323] New: IRA/reload moves asm statement
On Linux/x86-64, IRA/reload moves asm statement by
adding movss at -O2:
[...@gnu-6 vzeroupper-1]$ cat foo.c
extern void bar2 (void);
float
foo (float y)
{
asm volatile (nop);
bar2 ();
return y;
}
[...@gnu-6 vzeroupper-1]$ gcc -S -O foo.c
[...@gnu-6 vzeroupper-1]$ cat foo.s
.file foo.c
.text
.globl foo
.type foo, @function
foo:
.LFB0:
.cfi_startproc
subq$24, %rsp
.cfi_def_cfa_offset 32
movss %xmm0, 12(%rsp)
#APP
# 5 foo.c 1
nop
# 0 2
#NO_APP
callbar2
movss 12(%rsp), %xmm0
addq$24, %rsp
.cfi_def_cfa_offset 8
ret
.cfi_endproc
.LFE0:
.size foo, .-foo
.ident GCC: (GNU) 4.4.3 20100127 (Red Hat 4.4.3-4)
.section.note.GNU-stack,,@progbits
[...@gnu-6 vzeroupper-1]$ gcc -S -O2 foo.c
[...@gnu-6 vzeroupper-1]$ cat foo.s
.file foo.c
.text
.p2align 4,,15
.globl foo
.type foo, @function
foo:
.LFB0:
.cfi_startproc
subq$24, %rsp
.cfi_def_cfa_offset 32
#APP
# 5 foo.c 1
nop
# 0 2
#NO_APP
movss %xmm0, (%rsp)
callbar2
movss (%rsp), %xmm0
addq$24, %rsp
.cfi_def_cfa_offset 8
ret
.cfi_endproc
.LFE0:
.size foo, .-foo
.ident GCC: (GNU) 4.4.3 20100127 (Red Hat 4.4.3-4)
.section.note.GNU-stack,,@progbits
[...@gnu-6 vzeroupper-1]$
--
Summary: IRA/reload moves asm statement
Product: gcc
Version: 4.6.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: rtl-optimization
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: hjl dot tools at gmail dot com
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44323
Re: [Bug rtl-optimization/44323] New: IRA/reload moves asm statement
I don't think this is valid. You cannot depend on where the spill will
happen around a function call. It is spilling to save the volatile
register. With -O, we don't use volatile registers to keep variables
across functions. While at -O2 we do so it saves it right before the
function call.
Sent from my iPhone
On May 29, 2010, at 8:25 AM, hjl dot tools at gmail dot com gcc-bugzi...@gcc.gnu.org
wrote:
On Linux/x86-64, IRA/reload moves asm statement by
adding movss at -O2:
[...@gnu-6 vzeroupper-1]$ cat foo.c
extern void bar2 (void);
float
foo (float y)
{
asm volatile (nop);
bar2 ();
return y;
}
[...@gnu-6 vzeroupper-1]$ gcc -S -O foo.c
[...@gnu-6 vzeroupper-1]$ cat foo.s
.file foo.c
.text
.globl foo
.type foo, @function
foo:
.LFB0:
.cfi_startproc
subq$24, %rsp
.cfi_def_cfa_offset 32
movss %xmm0, 12(%rsp)
#APP
# 5 foo.c 1
nop
# 0 2
#NO_APP
callbar2
movss 12(%rsp), %xmm0
addq$24, %rsp
.cfi_def_cfa_offset 8
ret
.cfi_endproc
.LFE0:
.size foo, .-foo
.ident GCC: (GNU) 4.4.3 20100127 (Red Hat 4.4.3-4)
.section.note.GNU-stack,,@progbits
[...@gnu-6 vzeroupper-1]$ gcc -S -O2 foo.c
[...@gnu-6 vzeroupper-1]$ cat foo.s
.file foo.c
.text
.p2align 4,,15
.globl foo
.type foo, @function
foo:
.LFB0:
.cfi_startproc
subq$24, %rsp
.cfi_def_cfa_offset 32
#APP
# 5 foo.c 1
nop
# 0 2
#NO_APP
movss %xmm0, (%rsp)
callbar2
movss (%rsp), %xmm0
addq$24, %rsp
.cfi_def_cfa_offset 8
ret
.cfi_endproc
.LFE0:
.size foo, .-foo
.ident GCC: (GNU) 4.4.3 20100127 (Red Hat 4.4.3-4)
.section.note.GNU-stack,,@progbits
[...@gnu-6 vzeroupper-1]$
--
Summary: IRA/reload moves asm statement
Product: gcc
Version: 4.6.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: rtl-optimization
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: hjl dot tools at gmail dot com
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44323
[Bug rtl-optimization/44323] IRA/reload moves asm statement
--- Comment #1 from pinskia at gmail dot com 2010-05-29 15:39 ---
Subject: Re: New: IRA/reload moves asm statement
I don't think this is valid. You cannot depend on where the spill will
happen around a function call. It is spilling to save the volatile
register. With -O, we don't use volatile registers to keep variables
across functions. While at -O2 we do so it saves it right before the
function call.
Sent from my iPhone
On May 29, 2010, at 8:25 AM, hjl dot tools at gmail dot com
gcc-bugzi...@gcc.gnu.org
wrote:
On Linux/x86-64, IRA/reload moves asm statement by
adding movss at -O2:
[...@gnu-6 vzeroupper-1]$ cat foo.c
extern void bar2 (void);
float
foo (float y)
{
asm volatile (nop);
bar2 ();
return y;
}
[...@gnu-6 vzeroupper-1]$ gcc -S -O foo.c
[...@gnu-6 vzeroupper-1]$ cat foo.s
.file foo.c
.text
.globl foo
.type foo, @function
foo:
.LFB0:
.cfi_startproc
subq$24, %rsp
.cfi_def_cfa_offset 32
movss %xmm0, 12(%rsp)
#APP
# 5 foo.c 1
nop
# 0 2
#NO_APP
callbar2
movss 12(%rsp), %xmm0
addq$24, %rsp
.cfi_def_cfa_offset 8
ret
.cfi_endproc
.LFE0:
.size foo, .-foo
.ident GCC: (GNU) 4.4.3 20100127 (Red Hat 4.4.3-4)
.section.note.GNU-stack,,@progbits
[...@gnu-6 vzeroupper-1]$ gcc -S -O2 foo.c
[...@gnu-6 vzeroupper-1]$ cat foo.s
.file foo.c
.text
.p2align 4,,15
.globl foo
.type foo, @function
foo:
.LFB0:
.cfi_startproc
subq$24, %rsp
.cfi_def_cfa_offset 32
#APP
# 5 foo.c 1
nop
# 0 2
#NO_APP
movss %xmm0, (%rsp)
callbar2
movss (%rsp), %xmm0
addq$24, %rsp
.cfi_def_cfa_offset 8
ret
.cfi_endproc
.LFE0:
.size foo, .-foo
.ident GCC: (GNU) 4.4.3 20100127 (Red Hat 4.4.3-4)
.section.note.GNU-stack,,@progbits
[...@gnu-6 vzeroupper-1]$
--
Summary: IRA/reload moves asm statement
Product: gcc
Version: 4.6.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: rtl-optimization
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: hjl dot tools at gmail dot com
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44323
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44323
[Bug rtl-optimization/44323] IRA/reload moves asm statement
--- Comment #2 from pinskia at gcc dot gnu dot org 2010-05-29 15:56 --- There is no bug here since the asm does not say it touches any registers so xmm0 is live across the asm. Reload/IRA can insert the spilling/saving before or after the asm. Both places are valid. -- pinskia at gcc dot gnu dot org changed: What|Removed |Added Status|UNCONFIRMED |RESOLVED Resolution||INVALID http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44323
[Bug bootstrap/44315] [4.6 Regression] Circular build/gencondmd.o - insn-flags.h dependency dropped
--- Comment #3 from hjl dot tools at gmail dot com 2010-05-29 16:23 --- A patch is posted at http://gcc.gnu.org/ml/gcc-patches/2010-05/msg02325.html -- hjl dot tools at gmail dot com changed: What|Removed |Added URL||http://gcc.gnu.org/ml/gcc- ||patches/2010- ||05/msg02325.html http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44315
[Bug middle-end/44324] New: gcc.dg/tree-ssa/ipa-cp-1.c failed
On Linux/ia32, gcc.dg/tree-ssa/ipa-cp-1.c failed: FAIL: gcc.dg/tree-ssa/ipa-cp-1.c scan-tree-dump-times optimized very_long_function.constprop.0 \(\) 3 -- Summary: gcc.dg/tree-ssa/ipa-cp-1.c failed Product: gcc Version: 4.6.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: middle-end AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: hjl dot tools at gmail dot com http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44324
[Bug middle-end/44324] [4.6 Regression] gcc.dg/tree-ssa/ipa-cp-1.c failed
--- Comment #1 from hjl dot tools at gmail dot com 2010-05-29 17:00 --- It is caused by revision 160016: http://gcc.gnu.org/ml/gcc-cvs/2010-05/msg01075.html -- hjl dot tools at gmail dot com changed: What|Removed |Added Summary|gcc.dg/tree-ssa/ipa-cp-1.c |[4.6 Regression] |failed |gcc.dg/tree-ssa/ipa-cp-1.c ||failed Target Milestone|--- |4.6.0 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44324
[Bug middle-end/44325] New: [4.6 regression] FAIL: gcc.dg/vect/vect-109.c
On Linux/ia64, revision 159926 gave: FAIL: gcc.dg/vect/vect-109.c scan-tree-dump-times vect vectorized 1 loops 2 Revision 159909 is OK. -- Summary: [4.6 regression] FAIL: gcc.dg/vect/vect-109.c Product: gcc Version: 4.6.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: middle-end AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: hjl dot tools at gmail dot com http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44325
[Bug middle-end/44325] [4.6 regression] FAIL: gcc.dg/vect/vect-109.c
--- Comment #1 from hjl dot tools at gmail dot com 2010-05-29 17:05 --- It may be caused by revision 159920: http://gcc.gnu.org/ml/gcc-cvs/2010-05/msg00977.html -- hjl dot tools at gmail dot com changed: What|Removed |Added CC||rguenth at gcc dot gnu dot ||org Target Milestone|--- |4.6.0 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44325
[Bug tree-optimization/44306] [4.6 Regression] 464.h264ref fails to build.
--- Comment #5 from spop at gcc dot gnu dot org 2010-05-29 17:14 --- Subject: Bug 44306 Author: spop Date: Sat May 29 17:14:31 2010 New Revision: 160031 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=160031 Log: Don't use unshare_expr when not necessary. 2010-05-29 Sebastian Pop sebastian@amd.com PR middle-end/44306 * gcc.dg/tree-ssa/pr44306.c: New. * tree-if-conv.c (is_true_predicate): New. (is_predicated): Use is_true_predicate. (add_to_predicate_list): Same. Do not use unshare_expr. (add_to_dst_predicate_list): Same. Added: trunk/gcc/testsuite/gcc.dg/tree-ssa/pr44306.c Modified: trunk/gcc/ChangeLog trunk/gcc/testsuite/ChangeLog trunk/gcc/tree-if-conv.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44306
[Bug tree-optimization/44306] [4.6 Regression] 464.h264ref fails to build.
--- Comment #6 from spop at gcc dot gnu dot org 2010-05-29 17:15 --- Fixed. -- spop at gcc dot gnu dot org changed: What|Removed |Added Status|ASSIGNED|RESOLVED Resolution||FIXED http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44306
[Bug middle-end/44321] attribute warn_unused_result fails under inlining.
--- Comment #3 from bonzini at gnu dot org 2010-05-29 17:32 --- I don't think this bug is of any use. Unlike nonnull, unused return values do not trigger undesirable optimizations and (as far as I can tell) cannot possibly result in miscompilation. This bug is indeed about a loophole, and one that could indeed cause some bugs (even security bugs). But, fixing it would also make it impossible for many projects to use -Werror, since they have to deal with the fact that __wur was used inappropriately by glibc. Adding another attribute and making __wur overridable BTW would not be a solution since glibc would certainly switch to the new __really_wur attribute and we'd be in the same situation. :-) -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44321
[Bug middle-end/44321] attribute warn_unused_result fails under inlining.
--- Comment #4 from rguenth at gcc dot gnu dot org 2010-05-29 17:46 --- (In reply to comment #3) I don't think this bug is of any use. Unlike nonnull, unused return values do not trigger undesirable optimizations and (as far as I can tell) cannot possibly result in miscompilation. This bug is indeed about a loophole, and one that could indeed cause some bugs (even security bugs). But, fixing it would also make it impossible for many projects to use -Werror, since they have to deal with the fact that __wur was used inappropriately by glibc. Adding another attribute and making __wur overridable BTW would not be a solution since glibc would certainly switch to the new __really_wur attribute and we'd be in the same situation. :-) True, but if int dummy = foo() works then we can also make (void) foo () work. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44321
[Bug bootstrap/44315] [4.6 Regression] Circular build/gencondmd.o - insn-flags.h dependency dropped
--- Comment #4 from mrs at gcc dot gnu dot org 2010-05-29 17:58 --- I believe this has already been fixed by r160020. -- mrs at gcc dot gnu dot org changed: What|Removed |Added CC||mikestump at comcast dot net Status|NEW |RESOLVED Resolution||FIXED http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44315
[Bug target/44290] [4.5 Regression] __naked attribute is broken
--- Comment #18 from mikpe at it dot uu dot se 2010-05-29 18:00 --- Created an attachment (id=20773) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=20773action=view) linux kernel workaround for attribute naked breakage This patch makes the Linux kernel add noinline and noclone attributes to functions declared __naked. This allows gcc-4.5 to build a working 2.6.34 Linux kernel for my mach-iop32x/n2100 ARM box. Khem: can you check if this kernel-side workaround fixes your problem? Eventually I'd like the kernel to not use __naked, but that is non-trivial. This fix should work now and be easily backportable to older kernels. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290
[Bug middle-end/44324] [4.6 Regression] gcc.dg/tree-ssa/ipa-cp-1.c failed
--- Comment #2 from hubicka at gcc dot gnu dot org 2010-05-29 18:04 --- Subject: Bug 44324 Author: hubicka Date: Sat May 29 18:04:02 2010 New Revision: 160033 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=160033 Log: PR middle-end/44324 Modified: trunk/gcc/ChangeLog trunk/gcc/ipa-cp.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44324
[Bug middle-end/44295] [4.6 Regression] Failed to build 483.xalancbmk in SPEC CPU 2006
--- Comment #3 from hjl dot tools at gmail dot com 2010-05-29 18:35 --- Created an attachment (id=20774) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=20774action=view) A reduced testcase [...@gnu-32 delta]$ /export/gnu/import/rrs/159912/usr/bin/gcc -S -O3 pr44295.ii pr44295.ii: In member function \u2018xercesc_2_5::DOMNodeList* xercesc_2_5::DOMDocumentImpl::getElementsByTagName(const XMLCh*) const\u2019: pr44295.ii:157:68: error: edge points to wrong declaration: function_decl 0x7f3c21ea3600 _ZN11xercesc_2_519DOMDeepNodeListPoolINS_19DOMDeepNodeListImplEEC2Embm.clone.2 type method_type 0x7f3c220d8000 type void_type 0x7f3c2210de70 void type_6 VOID align 8 symtab 0 alias set -1 canonical type 0x7f3c2210de70 pointer_to_this pointer_type 0x7f3c2210df18 QI size integer_cst 0x7f3c220f8758 constant 8 unit size integer_cst 0x7f3c220f8780 constant 1 align 8 symtab 0 alias set -1 canonical type 0x7f3c220879d8 method basetype record_type 0x7f3c22082a80 DOMDeepNodeListPool arg-types tree_list 0x7f3c21e9f3e8 value pointer_type 0x7f3c22082b28 chain tree_list 0x7f3c2211d5f0 value void_type 0x7f3c2210de70 void pointer_to_this pointer_type 0x7f3c2208a5e8 asm_written used static decl_5 QI file pr44295.ii line 126 col 25 align 16 context record_type 0x7f3c22082a80 DOMDeepNodeListPool initial error_mark 0x7f3c22100ca8 abstract_origin function_decl 0x7f3c2207b800 DOMDeepNodeListPool arguments parm_decl 0x7f3c220b3a18 this type pointer_type 0x7f3c22082b28 type record_type 0x7f3c22082a80 DOMDeepNodeListPool public unsigned DI size integer_cst 0x7f3c220f8a50 constant 64 unit size integer_cst 0x7f3c220f8a78 constant 8 align 64 symtab 0 alias set 16 canonical type 0x7f3c22082b28 pointer_to_this pointer_type 0x7f3c21ecac78 readonly used unsigned DI file pr44295.ii line 126 col 239 size integer_cst 0x7f3c220f8a50 64 unit size integer_cst 0x7f3c220f8a78 8 align 64 context function_decl 0x7f3c21ea3600 _ZN11xercesc_2_519DOMDeepNodeListPoolINS_19DOMDeepNodeListImplEEC2Embm.clone.2 abstract_origin parm_decl 0x7f3c22088220 this (reg/f:DI 3 bx [orig:70 this ] [70]) arg-type pointer_type 0x7f3c22082b28 incoming-rtl (reg:DI 5 di [ this ]) result result_decl 0x7f3c220d3e80 D.2898 type void_type 0x7f3c2210de70 void used ignored VOID file pr44295.ii line 130 col 3 align 8 context function_decl 0x7f3c21ea3600 _ZN11xercesc_2_519DOMDeepNodeListPoolINS_19DOMDeepNodeListImplEEC2Embm.clone.2 abstract_origin result_decl 0x7f3c22080e80 D.2669 full-name xercesc_2_5::DOMDeepNodeListPoolTVal::DOMDeepNodeListPool(XMLSize_t, bool, XMLSize_t = 128) [with TVal = xercesc_2_5::DOMDeepNodeListImpl, XMLSize_t = long unsigned int] pending-inline-info 0x7f3c220a1310 template-info 0x7f3c220764a0 (mem:QI (symbol_ref:DI (_ZN11xercesc_2_519DOMDeepNodeListPoolINS_19DOMDeepNodeListImplEEC2Embm.clone.2) [flags 0x3] function_decl 0x7f3c21ea3600 _ZN11xercesc_2_519DOMDeepNodeListPoolINS_19DOMDeepNodeListImplEEC2Embm.clone.2) [0 S1 A8]) Instead of: function_decl 0x7f3c2207b700 __comp_ctor type method_type 0x7f3c22087930 type void_type 0x7f3c2210de70 void type_6 VOID align 8 symtab 0 alias set -1 canonical type 0x7f3c2210de70 pointer_to_this pointer_type 0x7f3c2210df18 QI size integer_cst 0x7f3c220f8758 constant 8 unit size integer_cst 0x7f3c220f8780 constant 1 align 8 symtab 0 alias set -1 canonical type 0x7f3c220879d8 method basetype record_type 0x7f3c22082a80 DOMDeepNodeListPool arg-types tree_list 0x7f3c22084870 value pointer_type 0x7f3c22082b28 chain tree_list 0x7f3c220847d0 value integer_type 0x7f3c2210d690 long unsigned int chain tree_list 0x7f3c220847a8 value boolean_type 0x7f3c2210d9d8 bool chain tree_list 0x7f3c22084780 purpose integer_cst 0x7f3c22031848 constant 128 value integer_type 0x7f3c2210d690 long unsigned int chain tree_list 0x7f3c2211d5f0 value void_type 0x7f3c2210de70 void pointer_to_this pointer_type 0x7f3c2208a5e8 addressable used public static weak decl_5 QI defer-output file pr44295.ii line 126 col 25 align 16 context record_type 0x7f3c22082a80 DOMDeepNodeListPool initial block 0x7f3c220a2580 abstract_origin function_decl 0x7f3c2207b800 DOMDeepNodeListPool arguments parm_decl 0x7f3c22088908 this type pointer_type 0x7f3c22082b28 type record_type 0x7f3c22082a80 DOMDeepNodeListPool public unsigned DI size integer_cst 0x7f3c220f8a50 constant 64 unit size integer_cst 0x7f3c220f8a78 constant 8 align 64 symtab 0 alias set 16 canonical type 0x7f3c22082b28 pointer_to_this pointer_type 0x7f3c21ecac78 readonly used unsigned DI file pr44295.ii line 126
[Bug bootstrap/44315] [4.6 Regression] Circular build/gencondmd.o - insn-flags.h dependency dropped
--- Comment #5 from hjl dot tools at gmail dot com 2010-05-29 18:40 --- (In reply to comment #4) I believe this has already been fixed by r160020. Revision 160020: http://gcc.gnu.org/ml/gcc-cvs/2010-05/msg01079.html doesn't fix this bug. In fact, it breaks bootstrap: http://gcc.gnu.org/ml/gcc-regression/2010-05/msg00395.html Please check out revision 160020 to see it yourselves. -- hjl dot tools at gmail dot com changed: What|Removed |Added Status|RESOLVED|REOPENED Resolution|FIXED | http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44315
[Bug target/44326] New: NONDEBUG_INSN_P should be used in implicit-zee.c
implicit-zee.c has
zeinsn_list = VEC_alloc (rtx, heap, 8);
FOR_EACH_BB (curr_block)
{
FOR_BB_INSNS (curr_block, curr_insn)
{
if (!INSN_P (curr_insn))
continue;
type = for_each_rtx (PATTERN (curr_insn),
is_set_with_extension_DI,
(void *)set_insn);
if (!type)
continue;
I believe NONDEBUG_INSN_P should be used here instead of INSN_P.
--
Summary: NONDEBUG_INSN_P should be used in implicit-zee.c
Product: gcc
Version: 4.6.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: hjl dot tools at gmail dot com
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44326
[Bug target/44326] NONDEBUG_INSN_P should be used in implicit-zee.c
-- hjl dot tools at gmail dot com changed: What|Removed |Added Target Milestone|--- |4.6.0 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44326
[Bug target/44165] 5/15/2010 snapshot build failure on hppa Linux
--- Comment #2 from danglin at gcc dot gnu dot org 2010-05-29 20:18 --- This was introduced by the following change: 2010-05-12 Jan Hubicka j...@suse.cz * cgraph.h (struct varpool_node): Add aux. * varasm.c (find_decl_and_mark_needed): Force output of varpool nodes. * varpool.c (varpool_remove_node): Do not remove initializer. (varpool_reset_queue): Export. (varpool_finalize_decl): Volatile vars are forced to be output. * lto-symtab.c (lto_varpool_replace_node): Clear out initializer of replaced decl. * ipa.c (enqueue_cgraph_node, enqueue_varpool_node, process_references, varpool_can_remove_if_no_refs): New functions. (cgraph_remove_unreachable_nodes): Handle variables too. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44165
[Bug bootstrap/44315] [4.6 Regression] Circular build/gencondmd.o - insn-flags.h dependency dropped
--- Comment #6 from mrs at gcc dot gnu dot org 2010-05-29 20:26 --- Subject: Bug 44315 Author: mrs Date: Sat May 29 20:26:12 2010 New Revision: 160035 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=160035 Log: 2010-05-29 H.J. Lu hongjiu...@intel.com PR bootstrap/44315 * Makefile.in (build/gencondmd.o): Remove TM_H := $(GTM_H). Filter out insn-flags.h. Modified: trunk/gcc/ChangeLog trunk/gcc/Makefile.in -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44315
[Bug bootstrap/44315] [4.6 Regression] Circular build/gencondmd.o - insn-flags.h dependency dropped
--- Comment #7 from mrs at gcc dot gnu dot org 2010-05-29 20:30 --- Ok, should be fixed now. -- mrs at gcc dot gnu dot org changed: What|Removed |Added Status|REOPENED|RESOLVED Resolution||FIXED http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44315
[Bug fortran/44292] [libgfortran ABI breakage] Increase internal size of RECL= of the OPEN statement
--- Comment #1 from tkoenig at gcc dot gnu dot org 2010-05-29 22:20 --- Confirmed. -- tkoenig at gcc dot gnu dot org changed: What|Removed |Added Status|UNCONFIRMED |NEW Ever Confirmed|0 |1 Last reconfirmed|-00-00 00:00:00 |2010-05-29 22:20:03 date|| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44292
[Bug target/44165] 5/15/2010 snapshot build failure on hppa Linux
--- Comment #3 from danglin at gcc dot gnu dot org 2010-05-29 22:25 --- Subject: Bug 44165 Author: danglin Date: Sat May 29 22:24:59 2010 New Revision: 160038 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=160038 Log: PR target/44165 * config/pa/pa32-linux.h (CTOR_LIST_BEGIN): Mark __CTOR_LIST__ as used. Modified: trunk/gcc/ChangeLog trunk/gcc/config/pa/pa32-linux.h -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44165
[Bug target/44165] 5/15/2010 snapshot build failure on hppa Linux
--- Comment #4 from danglin at gcc dot gnu dot org 2010-05-29 22:29 --- Fixed by patch. -- danglin at gcc dot gnu dot org changed: What|Removed |Added Status|NEW |RESOLVED Resolution||FIXED http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44165
[Bug java/44327] New: build_java_method_aliases should use cgraph_node-same_body aliases list
This blocks moving finalizing the compilation unit last in cp_write_global_declarations. -- Summary: build_java_method_aliases should use cgraph_node- same_body aliases list Product: gcc Version: 4.6.0 Status: UNCONFIRMED Severity: enhancement Priority: P3 Component: java AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: rguenth at gcc dot gnu dot org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44327
[Bug c++/44328] New: switch/case optimization produces an invalid jump table index
Notes:
1. the same issue occurs on x86 and x86_64 linux hosts, with a GCC
cross-compiler of the same version is built with the same options
2. the same issue occurs with GCC cross compiler 4.4.3 (same build options,
without the MPC host library)
I'm afraid I failed to understand something obvious, as this bug seems so
weird that I can't really believe it is an actual compiler bug. Please let me
know if/where I'm wrong.
I first encountered this bug while compiling the eCos 3.0+ kernel for an
ARM926 target.
The issue appends within the fopen.cxx source file (fopen() implementation):
the 'mode' parameter is converted from a char string - such as r - into a
fcntl enumerated value - O_RDONLY in this case, through two switch/case
statements
When built with -O0 or-01, the compiler does not emit a jump table for the
switch/case and the produced 'mode' integer value is the expected one.
When build with -O2 or -O3, the compiler does emit a jump table, but the jump
table is invalid: the net result is that for the '0' case value, the jump table
index points to an address that is outside the jump table, and the value
returned from the jump table is random.
When the -Wtype-limits option switch is used, the compiler does emit a warning,
but I failed to understand the reason for this warning.
I shrinked down the 'fopen.cxx' code to a single, simple function and the
assembly code contains the same invalid statement as it is in the original
file, the type-limits warning is also the same.
The source code does not rely on any other file (no header file).
//- C source --
#define O_RDONLY (10)
#define O_WRONLY (11)
#define O_RDWR (O_RDONLY|O_WRONLY)
#define O_CREAT (13)
#define O_TRUNC (16)
typedef enum {
OM_READ = 0,
OM_WRITE,
OM_READWRITE_NOCREATE,
OM_READWRITE_CREATE
} OpenMode;
extern C int open(const char *name, int mode);
void open_file(const char *filename, const OpenMode rw)
{
int mode = 0;
switch( rw )
{
case OM_WRITE:
mode = O_WRONLY|O_CREAT|O_TRUNC;
break;
case OM_READ:
mode = O_RDONLY;
break;
case OM_READWRITE_NOCREATE:
mode = O_RDWR;
break;
case OM_READWRITE_CREATE:
mode = O_RDWR|O_CREAT|O_TRUNC;
break;
}
open( filename, mode );
}
// - end of C code
The above C code is built with the following command:
arm-eabi-gcc -c -O2 -mcpu=arm926ej-s -fno-rtti -fno-exceptions -Wtype-limits \
-o fopen.o fopen.cxx
The compiler emits the following warning:
fopen.cxx: In function void open_file(const char*, OpenMode):
fopen.cxx:16:6: warning: comparison always true due to limited range of data
type
The disassemble code looks like the following:
//- ASM ---
fopen.o: file format elf32-littlearm
Disassembly of section .text:
_Z9open_filePKc8OpenMode:
0: e59f300cldr r3, [pc, #12] ; 14
_Z9open_filePKc8OpenMode+0x14
4: e2411001sub r1, r1, #1
8: e20110ffand r1, r1, #255; 0xff
c: e7931101ldr r1, [r3, r1, lsl #2]
10: eafeb 0 open
14: andeq r0, r0, r0
Disassembly of section .rodata:
CSWTCH.1:
0: 004aandeq r0, r0, r10, asr #32
4: 0003andeq r0, r0, r3
8: 004bandeq r0, r0, r11, asr #32
//- end of ASM
The jump table should contain 4 entries. It is only 3 entry wide.
There's no entry for 'case OM_READ'.
Nevertheless, the compiler shifts the index (sub r1, r1, #1), as if the
switch() value were starting from 1, rather than from 0.
When switch (OM_READ) is executed, r1 is decremented from 0 to 0x,
then masked to a single byte 0xff, following the ARM AAPCS variable-width
enumerated types.
This new index is used to access the jump table, but index 0xff is out of the
jump table.
Notes:
* If OpenMode enum is forced to start from 1 -rather than 0-, the emitted code
is valid (4-entry jump table):
OM_READ = 1
* If OpenMode enum is simply added another entry, the emitted code changes
and becomes valid (4-entry jump table):
OM_READWRITE_CREATE,
OM_DUMMY,
* If mode is not initialized to a default value, but mode is assigned the
same value (0) within a 'default', emitted code is valid:
int mode;
//
default: mode = 0; break;
The compiler has been built with the following options:
$ arm-eabi-gcc -c -O2 -mcpu=arm926ej-s -fno-rtti -fno-exceptions -o fopen.o
fopen.cxx -Wtype-limits -v
Using built-in specs.
COLLECT_GCC=arm-eabi-gcc
COLLECT_LTO_WRAPPER=/usr/local/homebrew/Cellar/gcc-arm-ecos/4.5.0/libexec/gcc/arm-eabi/4.5.0/lto-wrapper
Target: arm-eabi
Configured with: ../configure
[Bug middle-end/44329] New: vectorization loss in gas_dyn
Currently gcc trunk is showing significant loss of vectorization at r160038 for the gas_dyn.f90 benchmark... bash-3.2$ gfortran -O3 -ffast-math -ftree-vectorizer-verbose=2 gas_dyn.f90 gcc46.txt bash-3.2$ grep LOOP VECTORIZED gcc46.txt | wc 29 1161177 compared to... bash-3.2$ gfortran -O3 -ffast-math -ftree-vectorizer-verbose=2 gas_dyn.f90 gcc46.txt bash-3.2$ grep LOOP VECTORIZED gcc46.txt | wc 41 1641666 at r159941. -- Summary: vectorization loss in gas_dyn Product: gcc Version: 4.6.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: middle-end AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: howarth at nitro dot med dot uc dot edu GCC build triplet: x86_64-apple-darwin10 GCC host triplet: x86_64-apple-darwin10 GCC target triplet: x86_64-apple-darwin10 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44329
[Bug middle-end/44329] vectorization loss in gas_dyn
--- Comment #1 from howarth at nitro dot med dot uc dot edu 2010-05-30 00:34 --- Sorry for the noise. Wrong version of gcc was being executed. -- howarth at nitro dot med dot uc dot edu changed: What|Removed |Added Status|UNCONFIRMED |RESOLVED Resolution||WORKSFORME http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44329
[Bug objc/44140] ObjC lto/whopr fails
--- Comment #10 from iains at gcc dot gnu dot org 2010-05-30 00:56 --- Created an attachment (id=20775) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=20775action=view) with -lto the code @ comment # compiled with -flto -dA -fverbose-asm -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44140
[Bug objc/44140] ObjC lto/whopr fails
--- Comment #11 from iains at gcc dot gnu dot org 2010-05-30 00:57 --- Created an attachment (id=20776) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=20776action=view) without lto code from comment #5 compiled with -dA -fverbose-asm -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44140
[Bug objc/44140] ObjC lto/whopr fails
--- Comment #12 from iains at gcc dot gnu dot org 2010-05-30 01:00 --- OK, so I've patch obj-act.c to remove the two cases where ASM_ was called. The test code compiles and runs without -flto and produces the errors as per comment #9 when -flto is given. The required content is present, it's just that the elements have become disconnected - it looks like the reference has been updated, but not what it refers too. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44140
[Bug fortran/44292] [libgfortran ABI breakage] Increase internal size of RECL= of the OPEN statement
--- Comment #2 from jvdelisle at gcc dot gnu dot org 2010-05-30 05:38 --- I would like to work this one. -- jvdelisle at gcc dot gnu dot org changed: What|Removed |Added AssignedTo|unassigned at gcc dot gnu |jvdelisle at gcc dot gnu dot |dot org |org Status|NEW |ASSIGNED Last reconfirmed|2010-05-29 22:20:03 |2010-05-30 05:38:00 date|| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44292
[Bug target/44290] [4.5 Regression] __naked attribute is broken
--- Comment #19 from raj dot khem at gmail dot com 2010-05-30 05:58 --- (In reply to comment #18) Created an attachment (id=20773) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=20773action=view) [edit] linux kernel workaround for attribute naked breakage This patch makes the Linux kernel add noinline and noclone attributes to functions declared __naked. This allows gcc-4.5 to build a working 2.6.34 Linux kernel for my mach-iop32x/n2100 ARM box. Khem: can you check if this kernel-side workaround fixes your problem? I have tried using __noclone__ a couple of days ago to workaround the problem //static void __attribute__((__naked__, __noinline__, __noclone__, __no_instrument_function__)) static void __attribute__((__naked__, __no_instrument_function__)) v4wb_copy_user_page(void *kto, const void *kfrom) But my gcc seems to ignore it and its generating same code for both with Os and O2 for both above cases. Hence did not solve my issue. I am using snapshot of 4.5 branch from May 20 Eventually I'd like the kernel to not use __naked, but that is non-trivial. This fix should work now and be easily backportable to older kernels. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44290
