Re: [patch, fortran] Always return malloc(1) for empty arrays in the library
On Wed, Jun 29, 2011 at 00:50, Thomas Koenig wrote: > Hello world, > > looking at PR 49479 and other functions in the library made me realize > there are lots of places where we don't malloc one byte for empty > arrays. I'd prefer to add the zero check to runtime/memory.c (internal_malloc_size), i.e. change if (size == 0) return NULL; to if (size == 0) size = 1; Or perhaps even better would be to add that check to get_mem() instead (as a side effect making internal_malloc_size redundant), thus matching how memory allocation works in the frontend? Looking at get_mem() we actually have a small bug here; if size == 0 and malloc happens to return NULL we treat it as an error. -- Janne Blomqvist
Re: [RFC] Fix full memory barrier on SPARC-V8
> Linux doesn't ever run the cpu in the RMO memory model any more. All > sparc64 chips run only in TSO now. > > All of the Niagara chips implement an even stricter than TSO memory > model, and the membars we used to have all over the kernel to handle > that properly were just wasted I-cache space. So I just moved > unilaterally to TSO everywhere and killed off the membars necessitated > by RMO. OK, thanks for the clarification. That's also fine from GCC's viewpoint. -- Eric Botcazou
Re: [PATCH, i386] Enable -mprefer-avx128 by default for Bulldozer
> * config/i386/i386.opt (mprefer-avx128): Redefine the flag as a Mask > option. > > * config/i386/i386.h (ix86_tune_indices): Add X86_TUNE_AVX128_OPTIMAL > entry. > (TARGET_AVX128_OPTIMAL): New definition. > > * config/i386/i386.c (initial_ix86_tune_features): Initialize > X86_TUNE_AVX128_OPTIMAL entry. > (ix86_option_override_internal): Enable the generation > of the 128-bit instructions when TARGET_AVX128_OPTIMAL is set. > (ix86_preferred_simd_mode): Use TARGET_PREFER_AVX128. > (ix86_autovectorize_vector_sizes): Use TARGET_PREFER_AVX128. OK for mainline. For 4.6 it is RM's call. Thanks, Honza
Re: [PATCH, i386] Enable -mprefer-avx128 by default for Bulldozer
On Wed, Jun 29, 2011 at 09:49:52AM +0200, Jan Hubicka wrote: > > * config/i386/i386.opt (mprefer-avx128): Redefine the flag as a Mask > > option. > > > > * config/i386/i386.h (ix86_tune_indices): Add X86_TUNE_AVX128_OPTIMAL > > entry. > > (TARGET_AVX128_OPTIMAL): New definition. > > > > * config/i386/i386.c (initial_ix86_tune_features): Initialize > > X86_TUNE_AVX128_OPTIMAL entry. > > (ix86_option_override_internal): Enable the generation > > of the 128-bit instructions when TARGET_AVX128_OPTIMAL is set. > > (ix86_preferred_simd_mode): Use TARGET_PREFER_AVX128. > > (ix86_autovectorize_vector_sizes): Use TARGET_PREFER_AVX128. > > OK for mainline. For 4.6 it is RM's call. For 4.6 it is fine as well. Jakub
Re: Ping #1: [Patch, AVR]: Fix PR34734
Denis Chertykov wrote: > 2011/6/28 Georg-Johann Lay : >> http://gcc.gnu.org/ml/gcc-patches/2011-06/msg01462.html >> >> Georg-Johann Lay wrote: >>> PR34734 produces annoying, false warnings if __attribute__((progmem)) >>> is used in conjunction with C++. DECL_INITIAL is not yet set up in >>> avr_handle_progmem_attribute. >>> >>> Johann >>> >>> PR target/34734 >>> * config/avr/avr.c (avr_handle_progmem_attribute): Move warning >>> about uninitialized data attributed 'progmem' from here... >>> (avr_encode_section_info): ...to this new function. >>> (TARGET_ENCODE_SECTION_INFO): New define. >>> (avr_section_type_flags): For data in ".progmem.data", remove >>> section flag SECTION_WRITE. >> avr_encode_section_info is good place to emit the warning: >> DECL_INITIAL has stabilized for C++, the warning will appear even for >> unused variables that will eventually be thrown away, and the warning >> appears only once (new_decl_p). > > Approved. > > Denis. Is this a patch that should be backported? 4.6? 4.5? It's not fix for "bug or doc" but very annoying, false warning. Johann
Re: Unreviewed build/libgcc, mudflap patches
> [ada, build] Link gnattools with $(LIBINTL) > http://gcc.gnu.org/ml/gcc-patches/2011-06/msg01443.html > > This needs a build and/or Ada maintainer. I guess this is obvious, > though. Can you clarify how LIBINTL is set, and to what value? Will it be empty when not needed? Arno
Re: Unreviewed build/libgcc, mudflap patches
Arnaud,
>> [ada, build] Link gnattools with $(LIBINTL)
>> http://gcc.gnu.org/ml/gcc-patches/2011-06/msg01443.html
>>
>> This needs a build and/or Ada maintainer. I guess this is obvious,
>> though.
>
> Can you clarify how LIBINTL is set, and to what value? Will it be empty
> when not needed?
It's substituted in gcc/ada/gcc-interface/Makefile.in:
LIBINTL = @LIBINTL@
On Solaris 10, it becomes ${top_builddir}/../intl/libintl.a, on Solaris
11, it is empty.
Rainer
--
-
Rainer Orth, Center for Biotechnology, Bielefeld University
[committed] Fix ICE in mem_loc_descriptor (PR debug/49567)
Hi!
My assumption that SIGN/ZERO_EXTEND will always have MODE_INT class mode
was provably wrong, as can be seen on the attached testcase where mode
is integer vector. Fixed thusly, bootstrapped/regtested on x86_64-linux
and i686-linux, committed as obvious.
2011-06-29 Jakub Jelinek
PR debug/49567
* dwarf2out.c (mem_loc_descriptor) : Give up
for non-MODE_INT modes instead of asserting the mode has MODE_INT
class.
* gcc.target/i386/pr49567.c: New test.
--- gcc/dwarf2out.c.jj 2011-06-23 10:13:58.0 +0200
+++ gcc/dwarf2out.c 2011-06-28 16:12:08.0 +0200
@@ -14762,7 +14762,8 @@ mem_loc_descriptor (rtx rtl, enum machin
case SIGN_EXTEND:
case ZERO_EXTEND:
- gcc_assert (GET_MODE_CLASS (mode) == MODE_INT);
+ if (GET_MODE_CLASS (mode) != MODE_INT)
+ break;
op0 = mem_loc_descriptor (XEXP (rtl, 0), GET_MODE (XEXP (rtl, 0)),
mem_mode, VAR_INIT_STATUS_INITIALIZED);
if (op0 == 0)
--- gcc/testsuite/gcc.target/i386/pr49567.c.jj 2011-06-28 19:02:11.0
+0200
+++ gcc/testsuite/gcc.target/i386/pr49567.c 2011-06-28 19:01:41.0
+0200
@@ -0,0 +1,13 @@
+/* PR debug/49567 */
+/* { dg-do compile } */
+/* { dg-options "-g -O2 -msse4" } */
+
+#include
+
+__m128
+foo (__m128i x)
+{
+ __m128i y;
+ y = _mm_cvtepi16_epi32 (x);
+ return _mm_cvtepi32_ps (y);
+}
Jakub
Re: Unreviewed build/libgcc, mudflap patches
> >>[ada, build] Link gnattools with $(LIBINTL)
> >> http://gcc.gnu.org/ml/gcc-patches/2011-06/msg01443.html
> >>
> >>
> >> This needs a build and/or Ada maintainer. I guess this is obvious,
> >> though.
> >
> > Can you clarify how LIBINTL is set, and to what value? Will it be empty
> > when not needed?
>
> It's substituted in gcc/ada/gcc-interface/Makefile.in:
>
> LIBINTL = @LIBINTL@
>
> On Solaris 10, it becomes ${top_builddir}/../intl/libintl.a, on Solaris
> 11, it is empty.
The patch is OK then, thanks.
[testsuite, libffi] XFAIL libffi.call/cls_{,long}double_va.c on IRIX 6.5 (PR libffi/46660)
A fresh look at the remaining two libffi testsuite failures on IRIX 6.5
revealed why they cannot currently work:
FAIL: libffi.call/cls_double_va.c -O0 -W -Wall output pattern test, is 0.0
FAIL: libffi.call/cls_longdouble_va.c -O0 -W -Wall output pattern test, is 0.0
Those tests pass floating point arguments to varargs functions, which
libffi currently (i.e. before 3.0.11) doesn't support. Unfortunately,
the MIPS N32 and N64 ABIs pass floating point args in integer registers
in this case, as described in the MIPSpro N32 ABI Handbook, p. 7:
http://techpubs.sgi.com/library/tpl/cgi-bin/getdoc.cgi?coll=0650&db=bks&srch=&fname=/SGI_Developer/Mpro_n32_ABI/sgi_html/ch02.html
while for the O32 ABI there's no difference between varargs and
non-varargs functions.
The following patch handles this. On mainline and the 4.6 branch, it
xfails the tests on IRIX 6. The 4.5 branch is a bit different: since
the O32 ABI is still supported there, the difference between them needs
to be taken into account. Unfortunately, dg-output doesn't take the
input-opts and exclude-opts parameters dg-xfail-if and dg-skip-if do
(and is taken from upstream DejaGnu, not overridden/implemented in GCC),
so I'm forced to use dg-skip-if. Even dg-xfail-if is not enough since
the execution tests would XPASS and the output test FAIL.
Unfortunately, libffi has its own implementation of dg-xfail-if and is
missing dg-skip-if completely. To deal with this without duplication, I
include the implementation from gcc/testsuite/lib, which shouldn't be an
issue since other .exp files are already included from there.
Tested on mips-sgi-irix6.5 with the appropriate runtest invocations.
For the 4.5 branch, I've rerun the whole libffi testsuite for all 3 ABIs
to make sure nothing broke.
Installed on mainline, 4.6 and 4.5 branches.
Rainer
Mainline/4.6 branch version:
2011-06-29 Rainer Orth
PR libffi/46660
* testsuite/libffi.call/cls_double_va.c: xfail dg-output on
mips-sgi-irix6*.
* testsuite/libffi.call/cls_longdouble_va.c: Likewise.
Index: libffi/testsuite/libffi.call/cls_double_va.c
===
--- libffi/testsuite/libffi.call/cls_double_va.c(revision 175617)
+++ libffi/testsuite/libffi.call/cls_double_va.c(working copy)
@@ -6,6 +6,7 @@
/* { dg-do run { xfail strongarm*-*-* xscale*-*-* } } */
/* { dg-output "" { xfail avr32*-*-* } } */
+/* { dg-output "PR libffi/46660" { xfail mips-sgi-irix6* } } */
/* { dg-skip-if "" arm*-*-* { "-mfloat-abi=hard" } { "" } } */
#include "ffitest.h"
Index: libffi/testsuite/libffi.call/cls_longdouble_va.c
===
--- libffi/testsuite/libffi.call/cls_longdouble_va.c(revision 175617)
+++ libffi/testsuite/libffi.call/cls_longdouble_va.c(working copy)
@@ -6,6 +6,7 @@
/* { dg-do run { xfail strongarm*-*-* xscale*-*-* } } */
/* { dg-output "" { xfail avr32*-*-* x86_64-*-mingw* } } */
+/* { dg-output "PR libffi/46660" { xfail mips-sgi-irix6* } } */
/* { dg-skip-if "" arm*-*-* { "-mfloat-abi=hard" } { "" } } */
#include "ffitest.h"
4.5 branch version:
2011-06-29 Rainer Orth
PR libffi/46660
* testsuite/lib/libffi-dg.exp: Load target-supports.exp,
target-supports-dg.exp.
(dg-xfail-if): Remove.
* testsuite/libffi.call/cls_double_va.c: Skip on mips-sgi-irix6*
unless -mabi=32.
* testsuite/libffi.call/cls_longdouble_va.c: Likewise.
Index: libffi/testsuite/libffi.call/cls_double_va.c
===
--- libffi/testsuite/libffi.call/cls_double_va.c(revision 175619)
+++ libffi/testsuite/libffi.call/cls_double_va.c(working copy)
@@ -6,6 +6,7 @@
/* { dg-do run { xfail strongarm*-*-* xscale*-*-* } } */
/* { dg-output "" { xfail avr32*-*-* } } */
+/* { dg-skip-if "PR libffi/46660" { mips-sgi-irix6* } { "*" } { "-mabi=32" } }
*/
#include "ffitest.h"
static void
Index: libffi/testsuite/libffi.call/cls_longdouble_va.c
===
--- libffi/testsuite/libffi.call/cls_longdouble_va.c(revision 175619)
+++ libffi/testsuite/libffi.call/cls_longdouble_va.c(working copy)
@@ -6,6 +6,7 @@
/* { dg-do run { xfail strongarm*-*-* xscale*-*-* } } */
/* { dg-output "" { xfail avr32*-*-* x86_64-*-mingw* } } */
+/* { dg-skip-if "PR libffi/46660" { mips-sgi-irix6* } { "*" } { "-mabi=32" } }
*/
#include "ffitest.h"
static void
Index: libffi/testsuite/lib/libffi-dg.exp
===
--- libffi/testsuite/lib/libffi-dg.exp (revision 175619)
+++ libffi/testsuite/lib/libffi-dg.exp (working copy)
@@ -1,4 +1,5 @@
-# Copyright (C) 2003, 2005, 2008, 2009, 2010 Free Software Foundation, Inc.
+# Copyright (C) 2003, 2005, 2008, 2009, 2010, 2011
+# Free Software Foundation, Inc.
# This program is free s
Re: PR 49169: testing the alignment of a function
"H.J. Lu" writes:
> This caused:
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49545
Sorry for the breakage. I should obviously have tested on x86_64 as well.
To recap, there are (at least) two concepts of what "the address of X"
can mean. It can mean (a) the address at which X is actually located
or (b) the value to which symbol X resolves. The problem is that these
two are different for things like Thumb and MIPS16 functions.
The DECL_ALIGN on a function should be (a). For example, if we have:
void __attribute__((aligned(16))) foo (void) { ... }
then foo() really ought to start on a 16-byte boundary, even for Thumb.
However, most parts of GCC want (b). In particular, if we want to know
the alignment of an ADDR_EXPR, or the alignment of a tree or RTL memory
reference, the address is going to come from symbol resolution.
As far as I can tell, all uses of get_object_alignment* want (b).
In particular, get_object_alignment* isn't used for __alignof__,
which might have been one case where (a) made more sense.
(Any thoughts on what __alignof__ should mean here? Either way,
it's a separate patch.)
So DECL_ALIGN gives (a) rather than (b). But as the PR shows,
there are cases where we know (and need to know) that (b) >= 2 bytes,
namely whenever TARGET_PTRMEMFUNC_VBIT_LOCATION == ptrmemfunc_vbit_in_pfn.
This patch uses that to decide whether &function has 1 or 2 bytes of
alignment.
Note that this isn't just about ARM and MIPS. Some targets define
FUNCTION_BOUNDARY in a way that depends on optimisation flags,
whereas get_object_alignment previously treated it as an ABI property.
It's definitely arguable that those targets are buggy and should be
using align_functions instead, but the docs aren't really clear.
That's why I'm still not going for a target hook at this stage.
It's just too hard to say in general whether a port's current
FUNCTION_BOUNDARY is guaranteed by the ABI or not, so I think
the default definition of any hook would still be the condition
that I've used here. I think it makes sense to leave it like
this until someone is motivated to guarantee greater alignment.
Tested on x86_64-linux-gnu and arm-linux-gnueabi. OK to install?
Richard
gcc/
PR tree-optimization/49545
* builtins.c (get_object_alignment_1): Update function comment.
Do not use DECL_ALIGN for functions, but test
TARGET_PTRMEMFUNC_VBIT_LOCATION instead.
* fold-const.c (get_pointer_modulus_and_residue): Don't check
for functions here.
* tree-ssa-ccp.c (get_value_from_alignment): Likewise.
gcc/testsuite/
* gcc.dg/torture/pr49169.c: Restrict to ARM and MIPS targets.
Index: gcc/builtins.c
===
--- gcc/builtins.c 2011-06-28 14:06:19.0 +0100
+++ gcc/builtins.c 2011-06-29 09:07:00.0 +0100
@@ -264,8 +264,15 @@ called_as_built_in (tree node)
return is_builtin_name (name);
}
-/* Return the alignment in bits of EXP, an object.
- Don't return more than MAX_ALIGN no matter what. */
+/* Compute values M and N such that M divides (address of EXP - N) and
+ such that N < M. Store N in *BITPOSP and return M.
+
+ Note that the address (and thus the alignment) computed here is based
+ on the address to which a symbol resolves, whereas DECL_ALIGN is based
+ on the address at which an object is actually located. These two
+ addresses are not always the same. For example, on ARM targets,
+ the address &foo of a Thumb function foo() has the lowest bit set,
+ whereas foo() itself starts on an even address. */
unsigned int
get_object_alignment_1 (tree exp, unsigned HOST_WIDE_INT *bitposp)
@@ -287,7 +294,21 @@ get_object_alignment_1 (tree exp, unsign
exp = DECL_INITIAL (exp);
if (DECL_P (exp)
&& TREE_CODE (exp) != LABEL_DECL)
-align = DECL_ALIGN (exp);
+{
+ if (TREE_CODE (exp) == FUNCTION_DECL)
+ {
+ /* Function addresses can encode extra information besides their
+alignment. However, if TARGET_PTRMEMFUNC_VBIT_LOCATION
+allows the low bit to be used as a virtual bit, we know
+that the address itself must be 2-byte aligned. */
+ if (TARGET_PTRMEMFUNC_VBIT_LOCATION == ptrmemfunc_vbit_in_pfn)
+ align = 2 * BITS_PER_UNIT;
+ else
+ align = BITS_PER_UNIT;
+ }
+ else
+ align = DECL_ALIGN (exp);
+}
else if (CONSTANT_CLASS_P (exp))
{
align = TYPE_ALIGN (TREE_TYPE (exp));
Index: gcc/fold-const.c
===
--- gcc/fold-const.c2011-06-28 14:06:19.0 +0100
+++ gcc/fold-const.c2011-06-29 08:59:02.0 +0100
@@ -9216,8 +9216,7 @@ get_pointer_modulus_and_residue (tree ex
*residue = 0;
code = TREE_CODE (expr);
- if (code == ADDR_EXPR
- && TREE_CODE (TREE_OPERAND (expr, 0)) != FUNCTION_DECL)
+ if (code == ADDR_EXPR)
{
u
Re: [testsuite] Remove dg-extra-errors in gcc.dg/inline_[12].c etc.
David, > Your fix works ok for me (on x86-64/linux) too. based on that, I've installed the patch. Thanks. Rainer -- - Rainer Orth, Center for Biotechnology, Bielefeld University
Re: [testsuite, objc] Don't XFAIL objc.dg/torture/forward-1.m
On 28 Jun 2011, at 18:01, Iain Sandoe wrote:
On 28 Jun 2011, at 17:47, Rainer Orth wrote:
objc.dg/torture/forward-1.m now seems to XPASS everywhere, creating
an
annoying amount of testsuite noise. Dominique provided the following
patch in PR libobjc/Bug 36610.
Tested with the appropriate runtest invocations on i386-pc-
solaris2.10
(both multilibs), sparc-sun-solaris2.10 (both multilibs),
alpha-dec-osf5.1b, mips-sgi-irix6.5 (both multilibs),
powerpc-apple-darwin9.8.0 (32-bit only).
Ok for mainline?
Thanks.
Rainer
2011-06-28 Dominique d'Humieres
* objc.dg/torture/forward-1.m: Remove dg-xfail-run-if, dg-skip-if.
Index: gcc/testsuite/objc.dg/torture/forward-1.m
===
--- gcc/testsuite/objc.dg/torture/forward-1.m (revision 175589)
+++ gcc/testsuite/objc.dg/torture/forward-1.m (working copy)
@@ -1,7 +1,5 @@
/* { dg-do run } */
/* See if -forward:: is able to work. */
-/* { dg-xfail-run-if "PR36610" { ! { { i?86-*-* x86_64-*-* } &&
ilp32 } } { "-fgnu-runtime" } { "" } } */
-/* { dg-skip-if "Needs OBJC2 Implementation" { *-*-darwin* &&
{ lp64 } } { "-fnext-runtime" } { "" } } */
actually, looking at this, it should likely read (untested):
/* { dg-skip-if "Needs OBJC2 Implementation" { *-*-darwin8* &&
{ lp64 && { ! objc2 } } } { "-fnext-runtime" } { "" } } */
and should stay in place to protect the test-cases for m64 on *-*-
darwin8*
(not that there's ever likely to be an m64 objc2 on darwin 8.. but)
Just FTR, this works for me on powerpc-apple-darwin9
Iain
Index: gcc/testsuite/objc.dg/torture/forward-1.m
===
--- gcc/testsuite/objc.dg/torture/forward-1.m (revision 175578)
+++ gcc/testsuite/objc.dg/torture/forward-1.m (working copy)
@@ -1,7 +1,6 @@
/* { dg-do run } */
/* See if -forward:: is able to work. */
-/* { dg-xfail-run-if "PR36610" { ! { { i?86-*-* x86_64-*-* } &&
ilp32 } } { "-fgnu-runtime" } { "" } } */
-/* { dg-skip-if "Needs OBJC2 Implementation" { *-*-darwin* &&
{ lp64 } } { "-fnext-runtime" } { "" } } */
+/* { dg-skip-if "Needs OBJC2 Implementation" { *-*-darwin8* && { lp64
&& { ! objc2 } } } { "-fnext-runtime" } { "" } } */
#include
#include
Re: Unreviewed build/libgcc, mudflap patches
On 06/27/2011 01:32 PM, Rainer Orth wrote: Support libmudflap on Solaris (PR libmudflap/38738) http://gcc.gnu.org/ml/gcc-patches/2011-06/msg01446.html This needs the mudflap maintainer. Build changes are okay, the rest is pretty obvious. For pass-stratcliff I'd rather use autoconf tests as you suggested. There's nothing in the toplevel to support feature tests that enable/disable modules, unfortunately. It's been done in a very ad-hoc way so far, perhaps you can just use "case $target in" (or is it "case $host in"? would it be possible to cross-compile from Solaris 10 to Solaris 11 using Sun tools?). Paolo
Re: [build] Move MD_UNWIND_SUPPORT to toplevel libgcc
On 06/21/2011 12:04 PM, Rainer Orth wrote: For md_unwind_header on the other hand, you'd have almost as many cases as in the general case. I fear it's hard to have the configuration split over too many places. So I'd suggest to split the affected cases into Linux and non-Linux ones, with the slight duplication necessary for extra_parts and tmake_file. I agree. Thomas, are you going to do that? Paolo
Re: PATCH [2/n]: Prepare x32: Convert pointer to TLS symbol if needed
"H.J. Lu" writes:
> @@ -706,7 +706,13 @@ precompute_register_parameters (int num_actuals, struct
> arg_data *args,
> pseudo now. TLS symbols sometimes need a call to resolve. */
> if (CONSTANT_P (args[i].value)
> && !targetm.legitimate_constant_p (args[i].mode, args[i].value))
> - args[i].value = force_reg (args[i].mode, args[i].value);
> + {
> + if (GET_MODE (args[i].value) != args[i].mode)
> + args[i].value = convert_to_mode (args[i].mode,
> +args[i].value,
> +args[i].unsignedp);
> + args[i].value = force_reg (args[i].mode, args[i].value);
> + }
But if GET_MODE (args[i].value) != args[i].mode, then the call to
targetm.legitimate_constant_p looks wrong. The mode passed in the
first argument is supposed to the mode of the second argument.
Is there any reason why this and the following:
/* If we are to promote the function arg to a wider mode,
do it now. */
if (args[i].mode != TYPE_MODE (TREE_TYPE (args[i].tree_value)))
args[i].value
= convert_modes (args[i].mode,
TYPE_MODE (TREE_TYPE (args[i].tree_value)),
args[i].value, args[i].unsignedp);
need to be done in the current order? I can't think of any off-hand.
If not, would swapping them also fix the bug?
(I can't review this either way, of course.)
Richard
Re: Unreviewed build/libgcc, mudflap patches
Paolo, > On 06/27/2011 01:32 PM, Rainer Orth wrote: >> Support libmudflap on Solaris (PR libmudflap/38738) >> http://gcc.gnu.org/ml/gcc-patches/2011-06/msg01446.html >> >> This needs the mudflap maintainer. > > Build changes are okay, the rest is pretty obvious. > > For pass-stratcliff I'd rather use autoconf tests as you suggested. will do in the next step when dealing with remaining libmudflap failures on Solaris. > There's nothing in the toplevel to support feature tests that > enable/disable modules, unfortunately. It's been done in a very ad-hoc way > so far, perhaps you can just use "case $target in" (or is it "case $host > in"? would it be possible to cross-compile from Solaris 10 to Solaris 11 > using Sun tools?). It would, but I'd rather move this out of toplevel completely, having the modules enable/disable themselves depending on either static configuration or configure test outcome. I think Joseph had ideas for something like this. Rainer -- - Rainer Orth, Center for Biotechnology, Bielefeld University
[PATCH, ARM] [1/3] iWMMXt intrinsics maintenance and pipeline description
Hi, This patch maintains iWMMXt intrinsics code, and adds WMMX pipeline description. *config/arm/arm.c (arm_option_override): Enable iWMMt with VFP (enum arm_builtins): Add/fix iWMMXT/iWMMXT2 intrinsics. (builtin_description bdesc_2arg): Same. (builtin_description bdesc_1arg): Same. (arm_init_iwmmxt_builtins): Same. (arm_expand_binop_builtin): Same. (arm_expand_builtin): Same. (arm_output_iwmmxt_shift_immediate): New function. Same. (arm_output_iwmmxt_tinsr): New function. Same. *config/arm/iterators.md: Same. *config/arm/mmintrin.h: Same. *config/arm/arm.md: Same. Add wtype. *config/arm/iwmmxt.md: Same. *config/arm/iwmmxt2.md: New file. Same. *config/arm/arm-protos.h: Add new functions protos. *config/arm/marvell-f-iwmmxt.md: New file. Add Marvell WMMX pipeline description. Thanks, Xinyu iwmmxt_intrinsic_pipeline.patch.gz Description: iwmmxt_intrinsic_pipeline.patch.gz
[PATCH, ARM] [2/3] iWMMXt intrinsics testsuite
Hi, This patch adds some simple iWMMXt intrinsic test cases. gcc.target/arm/iwmmxt/i_shift_error.c: New. gcc.target/arm/iwmmxt/iwmmxt.exp: New. gcc.target/arm/iwmmxt/i_wshufh_error.c: New. gcc.target/arm/iwmmxt/i_merge_si64_error.c: New. gcc.target/arm/iwmmxt/i_waligni_error.c: New. gcc.target/arm/iwmmxt/i_extract_error.c: New. Thanks, Xinyu iwmmxt_testsuite.patch Description: iwmmxt_testsuite.patch
[PATCH, ARM] [3/3] iWMMXt intrinsics documentation
Hi, This patch adds iWMMXt intrinsics documentation into gcc doc. doc/extend.texi: Update iwmmxt intrinsics doc. doc/arm-iwmmxt-intrinsics.texi: New. Thanks, Xinyu iwmmxt_doc.patch Description: iwmmxt_doc.patch
Re: [patch] Fix PR tree-optimization/49539
On Wed, Jun 29, 2011 at 12:17 AM, Eric Botcazou wrote: > Hi, > > this is an ICE building the gnattools on ARM, a regression present on the > mainline (and reproducible on x86/Linux by switching to SJLJ exceptions). > > For the reduced testcase compiled at -O: > > Unable to coalesce ssa_names 2 and 174 which are marked as MUST COALESCE. > comp_last_2(ab) and comp_last_174(ab) > +===GNAT BUG DETECTED==+ > | 4.7.0 20110626 (experimental) [trunk revision 175408] (i586-suse-linux-gnu) > GCC error:| > | SSA corruption | > | Error detected around p.adb:3:4 > > The SSA names (or rather 2 related ones) have overlapping lifetimes. The > problem is created by forwprop1. Before: > > : > # comp_last_1(ab) = PHI > > [...] > > comp_last_174(ab) = comp_last_1(ab) + 1; > D.2425_175 = args.P_BOUNDS; > D.2426_176 = D.2425_175->LB0; > if (D.2426_176 > comp_last_174(ab)) > goto ; > else > goto ; > > : > D.2425_177 = args.P_BOUNDS; > D.2427_178 = D.2425_177->UB0; > if (D.2427_178 < comp_last_174(ab)) > goto ; > else > goto ; > > [...] > > comp_last_185(ab) = comp_last_174(ab) + 1; > D.2425_186 = args.P_BOUNDS; > D.2426_187 = D.2425_186->LB0; > if (D.2426_187 > comp_last_185(ab)) > goto ; > else > goto ; > > > After: > > comp_last_185(ab) = comp_last_1(ab) + 2; > D.2425_186 = args.P_BOUNDS; > D.2426_187 = D.2425_186->LB0; > if (D.2426_187 > comp_last_185(ab)) > goto ; > else > goto ; > > > The pass already contains a check for this situation in can_propagate_from but > it isn't applied in this case. > > Tested on x86_64-suse-linux, OK for the mainline? Ok. This is also latent on the 4.6 branch, so feel free to backport it. Thanks, Richard. > > 2011-06-28 Eric Botcazou > > PR tree-optimization/49539 > * tree-ssa-forwprop.c (can_propagate_from): Check for abnormal SSA > by means of stmt_references_abnormal_ssa_name. > (associate_plusminus): Call can_propagate_from before propagating > from definition statements. > (ssa_forward_propagate_and_combine): Remove superfluous newline. > > > -- > Eric Botcazou >
Re: [patch] Fix oversight in tuplification of DOM
On Tue, Jun 28, 2011 at 10:36 PM, Eric Botcazou wrote: > Hi, > > the attached testcase triggers an ICE when compiled at -O or above, on all the > open branches. This is a regression introduced with the tuplification. The > problem is that 2 ARRAY_RANGE_REFs are recognized as equivalent, although they > don't have the same number of elements. This is so because their type isn't > taken into account by the hash equality function as it simply isn't recorded > in initialize_hash_element (GIMPLE_SINGLE_RHS case). Now in all the other > cases it is recorded so this very likely is an oversight. > > Tested on x86_64-suse-linux, OK for all branches? Oops. This would also make two MEM-REFs equivalent which access different amount of memory. Ok. Thanks, Richard. > > 2011-06-28 Eric Botcazou > > * tree-ssa-dom.c (initialize_hash_element): Fix oversight. > > > 2011-06-28 Eric Botcazou > > * gnat.dg/opt17.ad[sb]: New test. > > > -- > Eric Botcazou >
Re: PR 49169: testing the alignment of a function
On Wed, Jun 29, 2011 at 10:28 AM, Richard Sandiford
wrote:
> "H.J. Lu" writes:
>> This caused:
>>
>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49545
>
> Sorry for the breakage. I should obviously have tested on x86_64 as well.
>
> To recap, there are (at least) two concepts of what "the address of X"
> can mean. It can mean (a) the address at which X is actually located
> or (b) the value to which symbol X resolves. The problem is that these
> two are different for things like Thumb and MIPS16 functions.
>
> The DECL_ALIGN on a function should be (a). For example, if we have:
>
> void __attribute__((aligned(16))) foo (void) { ... }
>
> then foo() really ought to start on a 16-byte boundary, even for Thumb.
> However, most parts of GCC want (b). In particular, if we want to know
> the alignment of an ADDR_EXPR, or the alignment of a tree or RTL memory
> reference, the address is going to come from symbol resolution.
>
> As far as I can tell, all uses of get_object_alignment* want (b).
> In particular, get_object_alignment* isn't used for __alignof__,
> which might have been one case where (a) made more sense.
> (Any thoughts on what __alignof__ should mean here? Either way,
> it's a separate patch.)
>
> So DECL_ALIGN gives (a) rather than (b). But as the PR shows,
> there are cases where we know (and need to know) that (b) >= 2 bytes,
> namely whenever TARGET_PTRMEMFUNC_VBIT_LOCATION == ptrmemfunc_vbit_in_pfn.
> This patch uses that to decide whether &function has 1 or 2 bytes of
> alignment.
>
> Note that this isn't just about ARM and MIPS. Some targets define
> FUNCTION_BOUNDARY in a way that depends on optimisation flags,
> whereas get_object_alignment previously treated it as an ABI property.
> It's definitely arguable that those targets are buggy and should be
> using align_functions instead, but the docs aren't really clear.
>
> That's why I'm still not going for a target hook at this stage.
> It's just too hard to say in general whether a port's current
> FUNCTION_BOUNDARY is guaranteed by the ABI or not, so I think
> the default definition of any hook would still be the condition
> that I've used here. I think it makes sense to leave it like
> this until someone is motivated to guarantee greater alignment.
>
> Tested on x86_64-linux-gnu and arm-linux-gnueabi. OK to install?
Ok.
Thanks,
Richard.
> Richard
>
>
> gcc/
> PR tree-optimization/49545
> * builtins.c (get_object_alignment_1): Update function comment.
> Do not use DECL_ALIGN for functions, but test
> TARGET_PTRMEMFUNC_VBIT_LOCATION instead.
> * fold-const.c (get_pointer_modulus_and_residue): Don't check
> for functions here.
> * tree-ssa-ccp.c (get_value_from_alignment): Likewise.
>
> gcc/testsuite/
> * gcc.dg/torture/pr49169.c: Restrict to ARM and MIPS targets.
>
> Index: gcc/builtins.c
> ===
> --- gcc/builtins.c 2011-06-28 14:06:19.0 +0100
> +++ gcc/builtins.c 2011-06-29 09:07:00.0 +0100
> @@ -264,8 +264,15 @@ called_as_built_in (tree node)
> return is_builtin_name (name);
> }
>
> -/* Return the alignment in bits of EXP, an object.
> - Don't return more than MAX_ALIGN no matter what. */
> +/* Compute values M and N such that M divides (address of EXP - N) and
> + such that N < M. Store N in *BITPOSP and return M.
> +
> + Note that the address (and thus the alignment) computed here is based
> + on the address to which a symbol resolves, whereas DECL_ALIGN is based
> + on the address at which an object is actually located. These two
> + addresses are not always the same. For example, on ARM targets,
> + the address &foo of a Thumb function foo() has the lowest bit set,
> + whereas foo() itself starts on an even address. */
>
> unsigned int
> get_object_alignment_1 (tree exp, unsigned HOST_WIDE_INT *bitposp)
> @@ -287,7 +294,21 @@ get_object_alignment_1 (tree exp, unsign
> exp = DECL_INITIAL (exp);
> if (DECL_P (exp)
> && TREE_CODE (exp) != LABEL_DECL)
> - align = DECL_ALIGN (exp);
> + {
> + if (TREE_CODE (exp) == FUNCTION_DECL)
> + {
> + /* Function addresses can encode extra information besides their
> + alignment. However, if TARGET_PTRMEMFUNC_VBIT_LOCATION
> + allows the low bit to be used as a virtual bit, we know
> + that the address itself must be 2-byte aligned. */
> + if (TARGET_PTRMEMFUNC_VBIT_LOCATION == ptrmemfunc_vbit_in_pfn)
> + align = 2 * BITS_PER_UNIT;
> + else
> + align = BITS_PER_UNIT;
> + }
> + else
> + align = DECL_ALIGN (exp);
> + }
> else if (CONSTANT_CLASS_P (exp))
> {
> align = TYPE_ALIGN (TREE_TYPE (exp));
> Index: gcc/fold-const.c
> ===
> --- gcc/fold-const.c 2011-06-28 14:06:19.0 +0100
> +++ gcc/fold-const.c 2011-06
Re: [build] Move unwinder to toplevel libgcc
On 06/20/2011 02:24 PM, Rainer Orth wrote:
* The only unwinder-related macro I haven't moved is
LIBGCC2_UNWIND_ATTRIBUTE. It is only defined gcc/config/mips/mips.h.
I suppose we would need a libgcc equivalent of tm.h for that,
something I didn't want to attack at this point.
Something like the attached?
Paolo
Index: configure.ac
===
--- configure.ac (revision 169877)
+++ configure.ac (working copy)
@@ -267,6 +267,16 @@ done
tmake_file="${tmake_file_}"
AC_SUBST(tmake_file)
+# Likewise export definitions for libgcc_tm.h
+tm_file_=
+for f in ${tm_file}
+do
+ tm_file_="${tm_file_} \$(srcdir)/config/$f"
+done
+tm_file="${tm_file_}"
+AC_SUBST(tm_file)
+AC_SUBST(tm_defines)
+
# Substitute configuration variables
AC_SUBST(extra_parts)
AC_SUBST(asm_hidden_op)
Index: Makefile.in
===
--- Makefile.in (revision 169877)
+++ Makefile.in (working copy)
@@ -253,6 +253,16 @@ objects = $(filter %$(objext),$^)
tmake_file = @tmake_file@
include $(srcdir)/empty.mk $(tmake_file)
+# Collect target defines and headers from config.host.
+libgcc_tm_defines = @tm_defines@
+libgcc_tm_file = @tm_file@
+libgcc_tm.h: libgcc_tm.stamp; @true
+libgcc_tm.stamp:
+ DEFINES='$(libgcc_tm_defines)' HEADERS='$(libgcc_tm_file)' \
+ $(srcdir)/mkheader.sh > tmp-libgcc_tm.h
+ $(SHELL) $(srcdir)/../move-if-change tmp-libgcc_tm.h libgcc_tm.h
+ echo timestamp > $@
+
# Only handle shared libraries if both:
# - the user requested them
# - we know how to build them
Index: config.host
===
--- config.host (revision 169877)
+++ config.host (working copy)
@@ -57,6 +57,8 @@
asm_hidden_op=.hidden
extra_parts=
tmake_file=
+tm_defines=
+tm_file=
# Set default cpu_type so it can be updated in each machine entry.
cpu_type=`echo ${host} | sed 's/-.*$//'`
Index: mkheader.sh
===
--- mkheader.sh (revision 0)
+++ mkheader.sh (revision 0)
@@ -0,0 +1,41 @@
+#! /bin/sh
+
+# Copyright (C) 2001, 2002, 2006, 2007, 2010, 2011 Free Software Foundation, Inc.
+# This file is part of GCC.
+
+# GCC is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3, or (at your option)
+# any later version.
+
+# GCC is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+
+# You should have received a copy of the GNU General Public License
+# along with GCC; see the file COPYING3. If not see
+#
Re: [build] Move unwinder to toplevel libgcc
Paolo Bonzini writes: > On 06/20/2011 02:24 PM, Rainer Orth wrote: >> * The only unwinder-related macro I haven't moved is >>LIBGCC2_UNWIND_ATTRIBUTE. It is only defined gcc/config/mips/mips.h. >>I suppose we would need a libgcc equivalent of tm.h for that, >>something I didn't want to attack at this point. > > Something like the attached? Yep, that should do it. Thanks. Rainer -- - Rainer Orth, Center for Biotechnology, Bielefeld University
[testsuite, ada] Cope with HP-UX sh in run_acats (PR ada/49511)
As described in the PR, HP-UX /bin/sh type -p ignores the -p, so we need
to extract the last field of the output just as for regular type. The
following patch does this, tested by Dave and bootstrapped on
sparc-sun-solaris2.11.
Ok for mainline, 4.6 and 4.5 branches?
Thanks.
Rainer
2011-06-29 Rainer Orth
PR ada/49511
* ada/acats/run_acats (which): Use last field of type -p output.
diff --git a/gcc/testsuite/ada/acats/run_acats
b/gcc/testsuite/ada/acats/run_acats
--- a/gcc/testsuite/ada/acats/run_acats
+++ b/gcc/testsuite/ada/acats/run_acats
@@ -10,11 +10,11 @@ fi
# type -p is missing from Solaris 2 /bin/sh and /bin/ksh (ksh88), but both
# ksh93 and bash have it.
# type output format differs between ksh88 and ksh93, so avoid it if
-# type -p is present.
+# type -p is present. Unfortunately, HP-UX /bin/sh ignores -p with type.
# Fall back to whence which ksh88 and ksh93 provide, but bash does not.
which () {
-path=`type -p $* 2>/dev/null` && { echo $path; return 0; }
+path=`type -p $* 2>/dev/null | awk '{print $NF}'` && { echo $path; return
0; }
path=`type $* 2>/dev/null | awk '{print $NF}'` && { echo $path; return 0; }
path=`whence $* 2>/dev/null` && { echo $path; return 0; }
return 1
--
-
Rainer Orth, Center for Biotechnology, Bielefeld University
Re: [PATCH, ARM] [1/3] iWMMXt intrinsics maintenance and pipeline description
On 29/06/11 09:48, Xinyu Qi wrote: > Hi, > > This patch maintains iWMMXt intrinsics code, and adds WMMX pipeline > description. > > *config/arm/arm.c (arm_option_override): Enable iWMMt with VFP > (enum arm_builtins): Add/fix iWMMXT/iWMMXT2 intrinsics. > (builtin_description bdesc_2arg): Same. > (builtin_description bdesc_1arg): Same. > (arm_init_iwmmxt_builtins): Same. > (arm_expand_binop_builtin): Same. > (arm_expand_builtin): Same. > (arm_output_iwmmxt_shift_immediate): New function. Same. > (arm_output_iwmmxt_tinsr): New function. Same. > *config/arm/iterators.md: Same. > *config/arm/mmintrin.h: Same. > *config/arm/arm.md: Same. Add wtype. > *config/arm/iwmmxt.md: Same. > *config/arm/iwmmxt2.md: New file. Same. > *config/arm/arm-protos.h: Add new functions protos. > *config/arm/marvell-f-iwmmxt.md: New file. > Add Marvell WMMX pipeline description. > > Thanks, > Xinyu This is still too big to review (if a patch is not a mechanical change and you need to compress it before posting, then it's almost certainly too big). I suggest you subdivide it along the following lines. 1) The new pipeline description 2) Changes to generic ARM code 3) New additions for iwmmxt I think you also need to expand on your change log entry (it doesn't help me understand the patch's contents). Please also use 'diff -up -F "^(define"' to generate your diff files (so that there's some function context on each hunk). Finally, I note that you've made some changes along the lines of (define_insn "ssaddv8qi3" [(set (match_operand:V8QI 0 "register_operand" "=y") -(ss_plus:V8QI (match_operand:V8QI 1 "register_operand" "y") +(ss_plus:V8QI (match_operand:V8QI 1 "register_operand" "%y") (match_operand:V8QI 2 "register_operand" "y")))] Marking such patterns as commutative doesn't give any better results, even though the operation is commutative. Why? well all marking the pattern like this is that it tells reload to try swapping the order of the commutative operands if reloading is needed (to see if the result will need fewer reloads); but you've only got a single register class for both alternatives and swapping them won't make any difference to that answer -- so the result is simply that the compiler will spend more time working out that there are still the same number of reloads to deal with. R.
Re: Ping #1: [Patch, AVR]: Fix PR34734
2011/6/29 Georg-Johann Lay : > Denis Chertykov wrote: >> 2011/6/28 Georg-Johann Lay : >>> http://gcc.gnu.org/ml/gcc-patches/2011-06/msg01462.html >>> >>> Georg-Johann Lay wrote: PR34734 produces annoying, false warnings if __attribute__((progmem)) is used in conjunction with C++. DECL_INITIAL is not yet set up in avr_handle_progmem_attribute. Johann PR target/34734 * config/avr/avr.c (avr_handle_progmem_attribute): Move warning about uninitialized data attributed 'progmem' from here... (avr_encode_section_info): ...to this new function. (TARGET_ENCODE_SECTION_INFO): New define. (avr_section_type_flags): For data in ".progmem.data", remove section flag SECTION_WRITE. >>> avr_encode_section_info is good place to emit the warning: >>> DECL_INITIAL has stabilized for C++, the warning will appear even for >>> unused variables that will eventually be thrown away, and the warning >>> appears only once (new_decl_p). >> >> Approved. >> >> Denis. > > Is this a patch that should be backported? > 4.6? > 4.5? > > It's not fix for "bug or doc" but very annoying, false warning. You can backport it if you want. I'm usually didn't backport such patches. Denis.
Re: Richard Sandiford appointed RTL maintainer
Gerald Pfeifer writes: > It's my pleasure to announce that, also based on the recommendation of > Eric Botcazou as the current maintainer in that area¹, the steering > committee is appointing Richard Sandiford as an additional RTL optimizers > maintainer. > > Please adjust the MAINTAINERS file accordingly, and Happy Hacking, > Richard! Thanks, and thanks to Eric. Here's what I've installed. This appointment seems to have been rather controversial, so FWIW, I'd be happy to voluntarily downgrade to something else, if there's a clear idea what that something else should be. Richard * MAINTAINERS (RTL optimizers): Add self. Index: MAINTAINERS === --- MAINTAINERS 2011-06-29 09:33:39.0 +0100 +++ MAINTAINERS 2011-06-29 11:01:46.0 +0100 @@ -231,6 +231,7 @@ reload Ulrich Weigand uweig...@de.ibm reload Bernd Schmidt ber...@codesourcery.com dfp.c, related Ben Ellistonb...@gnu.org RTL optimizers Eric Botcazou ebotca...@libertysurf.fr +RTL optimizers Richard Sandiford rdsandif...@googlemail.com auto-vectorizerRichard Guentherrguent...@suse.de auto-vectorizerZdenek Dvorak o...@ucw.cz auto-vectorizerIra Rosen i...@il.ibm.com
Re: [patch tree-optimization]: Do bitwise operator optimizations for X op !X patterns
On Tue, Jun 28, 2011 at 5:05 PM, Kai Tietz wrote: > Hello, > > this patch implements the X op !X patterns within tree-ssa-forwprop.c without > using here const-fold routines. Additionally it does some trivial folding > for X op X. Implementation > also looks through [(type)] X op [(type)] !X, if type of X is integral and > precision is suitable > for operation. > > ChangeLog gcc/ > > 2011-06-28 Kai Tietz > > * tree-ssa-forwprop.c (operand_precision_onep): New > function. > (find_possible_not_expr_argument): Likewise. > (simplify_bitwise_binary_1): Likewise. > (simplify_bitwise_binary): Use simplify_bitwise_binary_1 > for detecting various X op !X optimizations. > > ChangeLog gcc/testsuite > > 2011-06-28 Kai Tietz > > * gcc.dg/binop-notand1a.c: New test. > * gcc.dg/binop-notand2a.c: New test. > * gcc.dg/binop-notand3a.c: New test. > * gcc.dg/binop-notand4a.c: New test. > * gcc.dg/binop-notand5a.c: New test. > * gcc.dg/binop-notand6a.c: New test. > * gcc.dg/binop-notor1.c: New test. > * gcc.dg/binop-notor2.c: New test. > * gcc.dg/binop-notxor1.c: New test. > * gcc.dg/binop-notxor2.c: New test. > > Bootstrapped and regression tested for all languages plus Ada and Obj-C for > x86_64-pc-linux-gnu. Ok for apply? I can't follow the code in find_possible_not_expr_argument or its uses at all. Please try to produce patches that look more obvious in what they are doing - don't try to solve every testcase you can come up with in a single patch. Especially don't write functions like find_possible_not_expr_argument which seems to have evolved a lot after you wrote the overall function comment. Thanks, Richard. > Regards, > Kai >
Re: [patch, vectorizer] Handle pattern statements with multiple uses
On Wed, Jun 29, 2011 at 8:31 AM, Ira Rosen wrote: > Hi, > > This is a follow-up patch for > http://gcc.gnu.org/ml/gcc-patches/2011-06/msg01205.html. > The previous patch added a support for widen-mult in intermediate type, e.g.,: > > char a_t; > short a_it; > int a_T, prod_T, prod_T'; > > S1 a_t = ; > S3 a_T = (int) a_t; > '--> a_it = (short) a_t; > S5 prod_T = a_T * CONST; > '--> prod_T' = a_it w* CONST; > > > This way we create a patten stmt a_it = (short) a_t; (used in the > pattern) to replace the original a_T = (int) a_t;, which prevents > vectorization if there are other uses of a_T outside the pattern, > e.g., > > char a_t; > short a_it; > int a_T, prod_T, prod_T', x_T, y_T; > > S1 a_t = ; > S3 a_T = (int) a_t; > '--> a_it = (short) a_t; > S5 prod_T = a_T * CONST; > '--> prod_T' = a_it w* CONST; > S6 x_T = a_T + y_T; > > Here the addition (S6) should be performed on int and use the original > definition. > Currently the vectorization fails on the type check of this statement. > > This patch adds a support of such cases: we mark the original > statement as relevant for vectorization (and not only its related > pattern statement) and handle both the original and pattern statements > in the analysis and the transformation. > > Bootstrapped and tested on powerpc64-suse-linux. > Comments are welcome. Looks reasonable to me. Thanks, Richard. > Thanks, > Ira > > ChangeLog: > > * tree-vect-loop.c (vect_determine_vectorization_factor): Handle > both pattern and original statements if necessary. > (vect_transform_loop): Likewise. > * tree-vect-patterns.c (vect_pattern_recog): Update documentation. > * tree-vect-stmts.c (vect_mark_relevant): Add new argument. > Mark the pattern statement only if the original statement doesn't > have its own uses. > (process_use): Call vect_mark_relevant with additional parameter. > (vect_mark_stmts_to_be_vectorized): Likewise. > (vect_get_vec_def_for_operand): Use vectorized pattern statement. > (vect_analyze_stmt): Handle both pattern and original statements > if necessary. > (vect_transform_stmt): Don't store vectorized pattern statement > in the original statement. > (vect_is_simple_use_1): Use related pattern statement only if the > original statement is irrelevant. > * tree-vect-slp.c (vect_get_and_check_slp_defs): Likewise. > > testsuite/ChangeLog: > * gcc.dg/vect/slp-widen-mult-half.c: New test. > * gcc.dg/vect/vect-widen-mult-half.c: New test. >
Re: [build] Move unwinder to toplevel libgcc
On Wed, 29 Jun 2011, Paolo Bonzini wrote: > On 06/20/2011 02:24 PM, Rainer Orth wrote: > > * The only unwinder-related macro I haven't moved is > >LIBGCC2_UNWIND_ATTRIBUTE. It is only defined gcc/config/mips/mips.h. > >I suppose we would need a libgcc equivalent of tm.h for that, > >something I didn't want to attack at this point. > > Something like the attached? This patch appears to create a *second* file called libgcc_tm.h, so there is one created in gcc/ and one in libgcc/, which seems incredibly confusing. By all means move the libgcc_tm.h logic to libgcc/ (though tm.h will need to continue to include libgcc_tm.h for the target, until the transition away from tm.h target macros in target code is complete and code includes libgcc_tm.h directly instead), but don't have two headers with the same name in different build directories. I interpreted Rainer's comment as meaning that a libgcc/config/mips/mips-lib.h (or similar) file would need to be created and inserted in the right place in libgcc_tm_file for the large number of targets presently putting mips.h in tm_file. -- Joseph S. Myers jos...@codesourcery.com
[PATCH] Handle vectorization of invariant loads (PR46787)
The following patch makes us handle invariant loads during vectorization.
Dependence analysis currently isn't clever enough to disambiguate them
thus we insert versioning-for-alias checks. For the testcase hoisting
the load is still always possible though, and for a read-after-write
dependence it would be possible for the vectorized loop copy as the
may-aliasing write is varying by the scalar variable size.
The existing code for vectorizing invariant accesses looks very
suspicious - it generates a vector load at the scalar address
to then just extract the first vector element. Huh. IMHO this
can be simplified as done, by just re-using the scalar load result.
But maybe this code was supposed to deal with something entirely
different?
This patch gives a 33% speedup to the phoronix himeno testcase
if you bump the maximum alias versioning checks we want to insert.
I'm currently re-bootstrapping & testing this but an earlier version
was ok on x86_64-unknown-linux-gnu.
2011-06-29 Richard Guenther
PR tree-optimization/46787
* tree-data-ref.c (dr_address_invariant_p): Remove.
(find_data_references_in_stmt): Invariant accesses are ok now.
* tree-vect-stmts.c (vectorizable_load): Handle invariant
loads.
* tree-vect-data-refs.c (vect_analyze_data_ref_access): Allow
invariant loads.
* gcc.dg/vect/vect-121.c: New testcase.
Index: gcc/tree-data-ref.c
===
--- gcc/tree-data-ref.c (revision 175531)
+++ gcc/tree-data-ref.c (working copy)
@@ -919,21 +919,6 @@ dr_analyze_alias (struct data_reference
}
}
-/* Returns true if the address of DR is invariant. */
-
-static bool
-dr_address_invariant_p (struct data_reference *dr)
-{
- unsigned i;
- tree idx;
-
- FOR_EACH_VEC_ELT (tree, DR_ACCESS_FNS (dr), i, idx)
-if (tree_contains_chrecs (idx, NULL))
- return false;
-
- return true;
-}
-
/* Frees data reference DR. */
void
@@ -4228,19 +4213,6 @@ find_data_references_in_stmt (struct loo
dr = create_data_ref (nest, loop_containing_stmt (stmt),
*ref->pos, stmt, ref->is_read);
gcc_assert (dr != NULL);
-
- /* FIXME -- data dependence analysis does not work correctly for objects
- with invariant addresses in loop nests. Let us fail here until the
-problem is fixed. */
- if (dr_address_invariant_p (dr) && nest)
- {
- free_data_ref (dr);
- if (dump_file && (dump_flags & TDF_DETAILS))
- fprintf (dump_file, "\tFAILED as dr address is invariant\n");
- ret = false;
- break;
- }
-
VEC_safe_push (data_reference_p, heap, *datarefs, dr);
}
VEC_free (data_ref_loc, heap, references);
Index: gcc/tree-vect-stmts.c
===
--- gcc/tree-vect-stmts.c (revision 175531)
+++ gcc/tree-vect-stmts.c (working copy)
@@ -4076,7 +4076,8 @@ vectorizable_load (gimple stmt, gimple_s
&& code != COMPONENT_REF
&& code != IMAGPART_EXPR
&& code != REALPART_EXPR
- && code != MEM_REF)
+ && code != MEM_REF
+ && TREE_CODE_CLASS (code) != tcc_declaration)
return false;
if (!STMT_VINFO_DATA_REF (stmt_info))
@@ -4527,30 +4528,14 @@ vectorizable_load (gimple stmt, gimple_s
if (inv_p && !bb_vinfo)
{
gcc_assert (!strided_load);
- gcc_assert (nested_in_vect_loop_p (loop, stmt));
if (j == 0)
{
- int k;
- tree t = NULL_TREE;
- tree vec_inv, bitpos, bitsize = TYPE_SIZE (scalar_type);
-
- /* CHECKME: bitpos depends on endianess? */
- bitpos = bitsize_zero_node;
- vec_inv = build3 (BIT_FIELD_REF, scalar_type, new_temp,
- bitsize, bitpos);
- vec_dest = vect_create_destination_var (scalar_dest,
- NULL_TREE);
- new_stmt = gimple_build_assign (vec_dest, vec_inv);
- new_temp = make_ssa_name (vec_dest, new_stmt);
- gimple_assign_set_lhs (new_stmt, new_temp);
- vect_finish_stmt_generation (stmt, new_stmt, gsi);
-
- for (k = nunits - 1; k >= 0; --k)
- t = tree_cons (NULL_TREE, new_temp, t);
- /* FIXME: use build_constructor directly. */
- vec_inv = build_constructor_from_list (vectype, t);
+ tree vec_inv;
+ gimple_stmt_iterator gsi2 = *gsi;
+ gsi_next (&gsi2);
+ vec_inv = build_vector_from_val (vectype, scalar_dest);
new_temp = vect_init_vector (stmt, vec_inv,
-
[ARM] Deprecate -mwords-little-endian
ARM has an option called -mwords-little-endian that provides big-endian
compatibility with pre-2.8 compilers. When I asked Richard about it,
he seemed to think it had outlived its usefulness, so this patch
deprecates it. We can then remove it once 4.7 is out.
Tested on arm-linux-gnueabi. OK to install? If so, I'll do a patch
for the web page as well.
Richard
Index: gcc/config/arm/arm.c
===
--- gcc/config/arm/arm.c2011-06-29 09:33:37.0 +0100
+++ gcc/config/arm/arm.c2011-06-29 12:20:35.0 +0100
@@ -1483,6 +1483,10 @@ arm_option_override (void)
if (TARGET_APCS_FLOAT)
warning (0, "passing floating point arguments in fp regs not yet
supported");
+ if (TARGET_LITTLE_WORDS)
+warning (OPT_Wdeprecated, "% is deprecated and "
+"will be removed in a future release");
+
/* Initialize boolean versions of the flags, for use in the arm.md file. */
arm_arch3m = (insn_flags & FL_ARCH3M) != 0;
arm_arch4 = (insn_flags & FL_ARCH4) != 0;
Index: gcc/doc/invoke.texi
===
--- gcc/doc/invoke.texi 2011-06-22 16:45:29.0 +0100
+++ gcc/doc/invoke.texi 2011-06-29 12:17:40.0 +0100
@@ -10237,7 +10237,7 @@ Generate code for a little-endian word o
order. That is, a byte order of the form @samp{32107654}. Note: this
option should only be used if you require compatibility with code for
big-endian ARM processors generated by versions of the compiler prior to
-2.8.
+2.8. This option is now deprecated.
@item -mcpu=@var{name}
@opindex mcpu
Re: [build] Move unwinder to toplevel libgcc
On 06/29/2011 12:41 PM, Joseph S. Myers wrote: On Wed, 29 Jun 2011, Paolo Bonzini wrote: On 06/20/2011 02:24 PM, Rainer Orth wrote: * The only unwinder-related macro I haven't moved is LIBGCC2_UNWIND_ATTRIBUTE. It is only defined gcc/config/mips/mips.h. I suppose we would need a libgcc equivalent of tm.h for that, something I didn't want to attack at this point. Something like the attached? This patch appears to create a *second* file called libgcc_tm.h, so there is one created in gcc/ and one in libgcc/, which seems incredibly confusing. By all means move the libgcc_tm.h logic to libgcc/ (though tm.h will need to continue to include libgcc_tm.h for the target, until the transition away from tm.h target macros in target code is complete and code includes libgcc_tm.h directly instead), but don't have two headers with the same name in different build directories. Yes, the patch was not complete and only meant to be the plumbing for what you describe. I think Rainer missed the existence of libgcc_tm_file. :) Paolo
Re: Ping #1: [Patch, AVR]: Fix PR34734
Denis Chertykov wrote:
> 2011/6/29 Georg-Johann Lay :
>> Denis Chertykov wrote:
>>> 2011/6/28 Georg-Johann Lay :
http://gcc.gnu.org/ml/gcc-patches/2011-06/msg01462.html
Georg-Johann Lay wrote:
> PR34734 produces annoying, false warnings if __attribute__((progmem))
> is used in conjunction with C++. DECL_INITIAL is not yet set up in
> avr_handle_progmem_attribute.
>
> Johann
>
> PR target/34734
> * config/avr/avr.c (avr_handle_progmem_attribute): Move warning
> about uninitialized data attributed 'progmem' from here...
> (avr_encode_section_info): ...to this new function.
> (TARGET_ENCODE_SECTION_INFO): New define.
> (avr_section_type_flags): For data in ".progmem.data", remove
> section flag SECTION_WRITE.
avr_encode_section_info is good place to emit the warning:
DECL_INITIAL has stabilized for C++, the warning will appear even for
unused variables that will eventually be thrown away, and the warning
appears only once (new_decl_p).
>>> Approved.
>>>
>>> Denis.
>> Is this a patch that should be backported?
>> 4.6?
>> 4.5?
>>
>> It's not fix for "bug or doc" but very annoying, false warning.
>
> You can backport it if you want.
>
> I'm usually didn't backport such patches.
>
> Denis.
Ok, maybe Eric or Anatoly have some preference for 4.5/4.6.
Applied to 4.7 together with following corrigendum:
Index: config/avr/avr.c
===
--- config/avr/avr.c(revision 175628)
+++ config/avr/avr.c(working copy)
@@ -5161,7 +5161,7 @@ avr_section_type_flags (tree decl, const
/* Implement `TARGET_ENCODE_SECTION_INFO'. */
static void
-avr_encode_section_info (tree decl, rtx rtl ATTRIBUTE_UNUSED,
+avr_encode_section_info (tree decl, rtx rtl,
int new_decl_p)
{
/* In avr_handle_progmem_attribute, DECL_INITIAL is not yet
@@ -5177,6 +5177,8 @@ avr_encode_section_info (tree decl, rtx
"uninitialized variable %q+D put into "
"program memory area", decl);
}
+
+ default_encode_section_info (decl, rtl, new_decl_p);
}
Re: [build] Move unwinder to toplevel libgcc
"Joseph S. Myers" writes: > I interpreted Rainer's comment as meaning that a > libgcc/config/mips/mips-lib.h (or similar) file would need to be created > and inserted in the right place in libgcc_tm_file for the large number of > targets presently putting mips.h in tm_file. Indeed, that's why I refrained from attacking this by other means. Rainer -- - Rainer Orth, Center for Biotechnology, Bielefeld University
Re: [build] Move unwinder to toplevel libgcc
Paolo Bonzini writes: > I think Rainer missed the existence of libgcc_tm_file. :) I didn't, but it lives in gcc/config.gcc, which is contrary to the move from gcc to libgcc. Rainer -- - Rainer Orth, Center for Biotechnology, Bielefeld University
Re: [patch tree-optimization]: Do bitwise operator optimizations for X op !X patterns
- Original Message -
From: "Richard Guenther"
To: "Kai Tietz"
Cc: gcc-patches@gcc.gnu.org
Sent: Wednesday, June 29, 2011 12:14:10 PM
Subject: Re: [patch tree-optimization]: Do bitwise operator optimizations for X
op !X patterns
On Tue, Jun 28, 2011 at 5:05 PM, Kai Tietz wrote:
> Hello,
>
> this patch implements the X op !X patterns within tree-ssa-forwprop.c without
> using here const-fold routines. Additionally it does some trivial folding
> for X op X. Implementation
> also looks through [(type)] X op [(type)] !X, if type of X is integral and
> precision is suitable
> for operation.
>
> ChangeLog gcc/
>
> 2011-06-28 Kai Tietz
>
> * tree-ssa-forwprop.c (operand_precision_onep): New
> function.
> (find_possible_not_expr_argument): Likewise.
> (simplify_bitwise_binary_1): Likewise.
> (simplify_bitwise_binary): Use simplify_bitwise_binary_1
> for detecting various X op !X optimizations.
>
> ChangeLog gcc/testsuite
>
> 2011-06-28 Kai Tietz
>
> * gcc.dg/binop-notand1a.c: New test.
> * gcc.dg/binop-notand2a.c: New test.
> * gcc.dg/binop-notand3a.c: New test.
> * gcc.dg/binop-notand4a.c: New test.
> * gcc.dg/binop-notand5a.c: New test.
> * gcc.dg/binop-notand6a.c: New test.
> * gcc.dg/binop-notor1.c: New test.
> * gcc.dg/binop-notor2.c: New test.
> * gcc.dg/binop-notxor1.c: New test.
> * gcc.dg/binop-notxor2.c: New test.
>
> Bootstrapped and regression tested for all languages plus Ada and Obj-C for
> x86_64-pc-linux-gnu. Ok for apply?
I can't follow the code in find_possible_not_expr_argument or its uses
at all. Please try to produce patches that look more obvious in what
they are doing - don't try to solve every testcase you can come up with
in a single patch. Especially don't write functions like
find_possible_not_expr_argument which seems to have evolved a lot
after you wrote the overall function comment.
Thanks,
Richard.
> Regards,
> Kai
>
Well, I added some comments to these functions and renamed the
find_possible_not_expr_argument function to detect_not_expr_operand, which hits
its use better.
The cause for this function is, that there are more then one variant of
expressing a logical-not and all of them are used.
This routine simply tries to detect different variants used for not. Eg ~X ==
!X and (X ^ 1) == !X for integral type of X with precision one. For X with
integral type, (X == 0) == !X.
The folding for the three different bitwise-operations is pretty easy and it
makes sense to implement them at once. I see here no good point to separate
them into different patches. To separate them might even lead to questions
about abstracting some code-pieces out of the main function.
I didn't added testcases for all variants I am aware now. Just those, which are
now handled.
So hope you can read and understand logic of patch better by updated patch.
Regards,
Kai
Index: gcc-head/gcc/tree-ssa-forwprop.c
===
--- gcc-head.orig/gcc/tree-ssa-forwprop.c
+++ gcc-head/gcc/tree-ssa-forwprop.c
@@ -1674,6 +1674,228 @@ simplify_builtin_call (gimple_stmt_itera
return false;
}
+/* Checks if expression has type of one-bit precision, or is result of
+ a known boolean expression. */
+static bool
+operand_precision_onep (tree expr)
+{
+ enum tree_code code;
+
+ do
+{
+ code = TREE_CODE (expr);
+ if (!INTEGRAL_TYPE_P (TREE_TYPE (expr)))
+ return false;
+ if (TYPE_PRECISION (TREE_TYPE (expr)) == 1)
+ return true;
+ if (code == SSA_NAME)
+ {
+ gimple def_stmt = SSA_NAME_DEF_STMT (expr);
+ if (!def_stmt || !is_gimple_assign (def_stmt))
+ break;
+ code = gimple_assign_rhs_code (def_stmt);
+ if (!CONVERT_EXPR_CODE_P (code))
+ break;
+ expr = gimple_assign_rhs1 (def_stmt);
+ }
+ else
+ break;
+}
+ while (CONVERT_EXPR_CODE_P (code));
+ if (code == TRUTH_NOT_EXPR || TREE_CODE_CLASS (code) == tcc_comparison)
+return true;
+ return false;
+}
+
+/* Helper routine for simplify_bitwise_binary_1 function.
+ Ignored integral type cast is counted in CNT_CASTS. If a not expression
+ is found, the operand of the not-expression is stored in NEXPR.
+ This function returns the either the expression after the first
+ integral cast expression, or NAME.
+ Detected not-patterns are !X or X == 0 for X with integral type, and
+ X ^ 1 or ~X for X with integral type with precision of one.
+ The value of CNT_CASTS is either zero, or one. */
+static tree
+detect_not_expr_operand (tree name, int *cnt_casts, tree *nexpr)
+{
+ enum tree_code code = ERROR_MARK;
+ *cnt_casts = 0;
+ *nexpr = NULL_TREE;
+
+ while (1)
+{
+ tree op1, op2;
+ gimple def_stmt;
+ code = TREE_CODE (name);
+ /* If name has none-intergal type, or isn't a SSA_NAME, then
+stop search. */
Re: [PATCH] Handle vectorization of invariant loads (PR46787)
Richard Guenther wrote on 29/06/2011 02:19:40 PM:
>
> The following patch makes us handle invariant loads during vectorization.
> Dependence analysis currently isn't clever enough to disambiguate them
> thus we insert versioning-for-alias checks. For the testcase hoisting
> the load is still always possible though, and for a read-after-write
> dependence it would be possible for the vectorized loop copy as the
> may-aliasing write is varying by the scalar variable size.
>
> The existing code for vectorizing invariant accesses looks very
> suspicious - it generates a vector load at the scalar address
> to then just extract the first vector element. Huh. IMHO this
> can be simplified as done, by just re-using the scalar load result.
> But maybe this code was supposed to deal with something entirely
> different?
This code was added to support outer loop vectorization:
http://gcc.gnu.org/ml/gcc-patches/2007-08/msg00729.html (with an intention
to be improved...). I think your version is fine for the outer loops as
well. But with this patch an unused vector load is still created (it will
probably be removed later though).
Ira
>
> This patch gives a 33% speedup to the phoronix himeno testcase
> if you bump the maximum alias versioning checks we want to insert.
>
> I'm currently re-bootstrapping & testing this but an earlier version
> was ok on x86_64-unknown-linux-gnu.
>
> 2011-06-29 Richard Guenther
>
>PR tree-optimization/46787
>* tree-data-ref.c (dr_address_invariant_p): Remove.
>(find_data_references_in_stmt): Invariant accesses are ok now.
>* tree-vect-stmts.c (vectorizable_load): Handle invariant
>loads.
>* tree-vect-data-refs.c (vect_analyze_data_ref_access): Allow
>invariant loads.
>
>* gcc.dg/vect/vect-121.c: New testcase.
>
> Index: gcc/tree-data-ref.c
> ===
> --- gcc/tree-data-ref.c (revision 175531)
> +++ gcc/tree-data-ref.c (working copy)
> @@ -919,21 +919,6 @@ dr_analyze_alias (struct data_reference
> }
> }
>
> -/* Returns true if the address of DR is invariant. */
> -
> -static bool
> -dr_address_invariant_p (struct data_reference *dr)
> -{
> - unsigned i;
> - tree idx;
> -
> - FOR_EACH_VEC_ELT (tree, DR_ACCESS_FNS (dr), i, idx)
> -if (tree_contains_chrecs (idx, NULL))
> - return false;
> -
> - return true;
> -}
> -
> /* Frees data reference DR. */
>
> void
> @@ -4228,19 +4213,6 @@ find_data_references_in_stmt (struct loo
>dr = create_data_ref (nest, loop_containing_stmt (stmt),
> *ref->pos, stmt, ref->is_read);
>gcc_assert (dr != NULL);
> -
> - /* FIXME -- data dependence analysis does not work correctly
> for objects
> - with invariant addresses in loop nests. Let us fail here until
the
> -problem is fixed. */
> - if (dr_address_invariant_p (dr) && nest)
> - {
> - free_data_ref (dr);
> - if (dump_file && (dump_flags & TDF_DETAILS))
> - fprintf (dump_file, "\tFAILED as dr address is invariant\n");
> - ret = false;
> - break;
> - }
> -
>VEC_safe_push (data_reference_p, heap, *datarefs, dr);
> }
>VEC_free (data_ref_loc, heap, references);
> Index: gcc/tree-vect-stmts.c
> ===
> --- gcc/tree-vect-stmts.c (revision 175531)
> +++ gcc/tree-vect-stmts.c (working copy)
> @@ -4076,7 +4076,8 @@ vectorizable_load (gimple stmt, gimple_s
>&& code != COMPONENT_REF
>&& code != IMAGPART_EXPR
>&& code != REALPART_EXPR
> - && code != MEM_REF)
> + && code != MEM_REF
> + && TREE_CODE_CLASS (code) != tcc_declaration)
> return false;
>
>if (!STMT_VINFO_DATA_REF (stmt_info))
> @@ -4527,30 +4528,14 @@ vectorizable_load (gimple stmt, gimple_s
> if (inv_p && !bb_vinfo)
>{
> gcc_assert (!strided_load);
> -gcc_assert (nested_in_vect_loop_p (loop, stmt));
> if (j == 0)
>{
> -int k;
> -tree t = NULL_TREE;
> -tree vec_inv, bitpos, bitsize = TYPE_SIZE (scalar_type);
> -
> -/* CHECKME: bitpos depends on endianess? */
> -bitpos = bitsize_zero_node;
> -vec_inv = build3 (BIT_FIELD_REF, scalar_type, new_temp,
> - bitsize, bitpos);
> -vec_dest = vect_create_destination_var (scalar_dest,
> - NULL_TREE);
> -new_stmt = gimple_build_assign (vec_dest, vec_inv);
> -new_temp = make_ssa_name (vec_dest, new_stmt);
> -gimple_assign_set_lhs (new_stmt, new_temp);
> -vect_finish_stmt_generation (stmt, new_stmt, gsi);
> -
> -for (k = nunits - 1; k >= 0; --k)
> - t = tree_cons (NULL_TREE, new_temp, t);
> -/* FIXME: use build_constructor directly. */
> -vec_inv = build_constructor_from_list (vectype, t);
> +tree vec_inv;
> +
Re: PATCH [10/n]: Prepare x32: PR rtl-optimization/49114: Reload failed to handle (set reg:X (plus:X (subreg:X (reg:Y) 0) (const
H.J. Lu wrote: > >>* reload.c (struct replacement): Remove SUBREG_LOC member. > >>(push_reload): Do not set it. > >>(push_replacement): Likewise. > >>(subst_reload): Remove dead code. > >>(copy_replacements): Remove assertion. > >>(copy_replacements_1): Do not handle SUBREG_LOC. > >>(move_replacements): Likewise. > >>(find_replacement): Remove dead code. Use > >> reload_adjust_reg_for_mode. > >>Detect subregs via recursive descent instead of via SUBREG_LOC. > >> > > > > It works much better. I am testing it now. > > > > It works. There are no regressions on Linux/ia32 nor Linux/x86-64. > Can you check it in and mention PR rtl-optimization/49114 ChangeLog? OK, I've checked the patch in now. Thanks for testing! Bye, Ulrich -- Dr. Ulrich Weigand GNU Toolchain for Linux on System z and Cell BE ulrich.weig...@de.ibm.com
Re: [testsuite, libffi] XFAIL libffi.call/cls_{,long}double_va.c on IRIX 6.5 (PR libffi/46660)
On Wed, Jun 29, 2011 at 10:26 AM, Rainer Orth
wrote:
> A fresh look at the remaining two libffi testsuite failures on IRIX 6.5
> revealed why they cannot currently work:
>
> FAIL: libffi.call/cls_double_va.c -O0 -W -Wall output pattern test, is 0.0
> FAIL: libffi.call/cls_longdouble_va.c -O0 -W -Wall output pattern test, is 0.0
>
> Those tests pass floating point arguments to varargs functions, which
> libffi currently (i.e. before 3.0.11) doesn't support. Unfortunately,
> the MIPS N32 and N64 ABIs pass floating point args in integer registers
> in this case, as described in the MIPSpro N32 ABI Handbook, p. 7:
>
> http://techpubs.sgi.com/library/tpl/cgi-bin/getdoc.cgi?coll=0650&db=bks&srch=&fname=/SGI_Developer/Mpro_n32_ABI/sgi_html/ch02.html
>
> while for the O32 ABI there's no difference between varargs and
> non-varargs functions.
>
> The following patch handles this. On mainline and the 4.6 branch, it
> xfails the tests on IRIX 6. The 4.5 branch is a bit different: since
> the O32 ABI is still supported there, the difference between them needs
> to be taken into account. Unfortunately, dg-output doesn't take the
> input-opts and exclude-opts parameters dg-xfail-if and dg-skip-if do
> (and is taken from upstream DejaGnu, not overridden/implemented in GCC),
> so I'm forced to use dg-skip-if. Even dg-xfail-if is not enough since
> the execution tests would XPASS and the output test FAIL.
> Unfortunately, libffi has its own implementation of dg-xfail-if and is
> missing dg-skip-if completely. To deal with this without duplication, I
> include the implementation from gcc/testsuite/lib, which shouldn't be an
> issue since other .exp files are already included from there.
>
> Tested on mips-sgi-irix6.5 with the appropriate runtest invocations.
> For the 4.5 branch, I've rerun the whole libffi testsuite for all 3 ABIs
> to make sure nothing broke.
>
> Installed on mainline, 4.6 and 4.5 branches.
The tests now fail on x86_64-linux and i?86-linux like
FAIL: libffi.call/cls_double_va.c -O0 -W -Wall output pattern test, is 7.0
FAIL: libffi.call/cls_longdouble_va.c -O0 -W -Wall output pattern test, is 7.0
FAIL: libffi.call/cls_double_va.c -O2 output pattern test, is 7.0
FAIL: libffi.call/cls_longdouble_va.c -O2 output pattern test, is 7.0
FAIL: libffi.call/cls_double_va.c -O3 output pattern test, is 7.0
FAIL: libffi.call/cls_longdouble_va.c -O3 output pattern test, is 7.0
FAIL: libffi.call/cls_double_va.c -Os output pattern test, is 7.0
FAIL: libffi.call/cls_longdouble_va.c -Os output pattern test, is 7.0
FAIL: libffi.call/cls_double_va.c -O2 -fomit-frame-pointer output pattern test,
is 7.0
FAIL: libffi.call/cls_longdouble_va.c -O2 -fomit-frame-pointer output pattern te
st, is 7.0
spawn [open ...]^M
7.0
res: 4
7.0
res: 4
PASS: libffi.call/cls_double_va.c -O0 -W -Wall execution test
FAIL: libffi.call/cls_double_va.c -O0 -W -Wall output pattern test, is 7.0
res: 4
7.0
res: 4
, should match PR libffi/466607.0^M?
res: 4^M?
7.0^M?
res: 4
I believe your dg-output first arguments are bogus.
Richard.
> Rainer
>
>
> Mainline/4.6 branch version:
>
> 2011-06-29 Rainer Orth
>
> PR libffi/46660
> * testsuite/libffi.call/cls_double_va.c: xfail dg-output on
> mips-sgi-irix6*.
> * testsuite/libffi.call/cls_longdouble_va.c: Likewise.
>
> Index: libffi/testsuite/libffi.call/cls_double_va.c
> ===
> --- libffi/testsuite/libffi.call/cls_double_va.c (revision 175617)
> +++ libffi/testsuite/libffi.call/cls_double_va.c (working copy)
> @@ -6,6 +6,7 @@
>
> /* { dg-do run { xfail strongarm*-*-* xscale*-*-* } } */
> /* { dg-output "" { xfail avr32*-*-* } } */
> +/* { dg-output "PR libffi/46660" { xfail mips-sgi-irix6* } } */
> /* { dg-skip-if "" arm*-*-* { "-mfloat-abi=hard" } { "" } } */
>
> #include "ffitest.h"
> Index: libffi/testsuite/libffi.call/cls_longdouble_va.c
> ===
> --- libffi/testsuite/libffi.call/cls_longdouble_va.c (revision 175617)
> +++ libffi/testsuite/libffi.call/cls_longdouble_va.c (working copy)
> @@ -6,6 +6,7 @@
>
> /* { dg-do run { xfail strongarm*-*-* xscale*-*-* } } */
> /* { dg-output "" { xfail avr32*-*-* x86_64-*-mingw* } } */
> +/* { dg-output "PR libffi/46660" { xfail mips-sgi-irix6* } } */
> /* { dg-skip-if "" arm*-*-* { "-mfloat-abi=hard" } { "" } } */
>
> #include "ffitest.h"
>
> 4.5 branch version:
>
> 2011-06-29 Rainer Orth
>
> PR libffi/46660
> * testsuite/lib/libffi-dg.exp: Load target-supports.exp,
> target-supports-dg.exp.
> (dg-xfail-if): Remove.
> * testsuite/libffi.call/cls_double_va.c: Skip on mips-sgi-irix6*
> unless -mabi=32.
> * testsuite/libffi.call/cls_longdouble_va.c: Likewise.
>
> Index: libffi/testsuite/libffi.call/cls_double_va.c
> ===
> --- libffi/testsuite/libffi.call/cls
Re: [PATCH, SRA] Total scalarization and padding
On Tue, 28 Jun 2011, Martin Jambor wrote:
> Hi,
>
> On Tue, Jun 28, 2011 at 03:01:17PM +0200, Richard Guenther wrote:
> > On Tue, Jun 28, 2011 at 2:50 PM, Martin Jambor wrote:
> > > Hi,
> > >
> > > at the moment SRA can get confused by alignment padding and think that
> > > it actually contains some data for which there is no planned
> > > replacement and thus might leave some loads and stores in place
> > > instead of removing them. This is perhaps the biggest problem when we
> > > attempt total scalarization of simple structures exactly in order to
> > > get rid of these and of the variables altogether.
> > >
> > > I've pondered for quite a while how to best deal with them. One
> > > option was to make just the total scalarization stronger. I have also
> > > contemplated creating phantom accesses for padding I could detect
> > > (i.e. in simple structures) which would be more general, but this
> > > would complicate the parts of SRA which are already quite convoluted
> > > and I was not really sure it was worth it.
> > >
> > > Eventually I decided for the total scalarization option. This patch
> > > changes it such that the flag is propagated down the access tree but
> > > also, if it does not work out, is reset on the way up. If the flag
> > > survives, the access tree is considered "covered" by scalar
> > > replacements and thus it is known not to contain unscalarized data.
> > >
> > > While changing function analyze_access_subtree I have simplified the
> > > way we compute the hole flag and also fixed one comparison which we
> > > currently have the wrong way round but it fortunately does not matter
> > > because if there is a hole, the covered_to will never add up to the
> > > total size. I'll probably post a separate patch against 4.6 just in
> > > case someone attempts to read the source.
> > >
> > > Bootstrapped and tested on x86_64-linux, OK for trunk?
> >
> > So, what will it do for the testcase?
> >
> > The following is what I _think_ it should do:
> >
> > :
> > l = *p_1(D);
> > l$i_6 = p_1(D)->i;
> > D.2700_2 = l$i_6;
> > D.2701_3 = D.2700_2 + 1;
> > l$i_12 = D.2701_3;
> > *p_1(D) = l;
> > p_1(D)->i = l$i_12;
> >
> > and let FRE/DSE do their job (which they don't do, unfortunately).
> > So does your patch then remove the load/store from/to l but keep
> > the elementwise loads/stores (which are probably cleaned up by FRE)?
> >
>
> Well, that is what would happen if no total scalarization was going
> on. Total scalarization is a poor-man's aggregate copy-propagation by
> splitting up small structures to individual fields whenever we can get
> rid of them this way (i.e. if they are never used in a non-assignment)
> which I introduced to fix PR 42585 - but unfortunately the padding
> problem did not occur to me until this winter.
>
> Currently, SRA performs very badly on the testcase, creating:
>
> :
> l = *p_1(D);
> l$i_6 = p_1(D)->i;
> l$f1_8 = p_1(D)->f1;
> l$f2_9 = p_1(D)->f2;
> l$f3_10 = p_1(D)->f3;
> l$f4_11 = p_1(D)->f4;
> D.1966_2 = l$i_6;
> D.1967_3 = D.1966_2 + 1;
> l$i_12 = D.1967_3;
> *p_1(D) = l; <-- this should not be here
> p_1(D)->i = l$i_12;
> p_1(D)->f1 = l$f1_8;
> p_1(D)->f2 = l$f2_9;
> p_1(D)->f3 = l$f3_10;
> p_1(D)->f4 = l$f4_11;
> return;
>
> Unfortunately, this basically survives all the way to the "optimized"
> dump. With the patch, the assignment *p_1(D) = l; is removed and
> copyprop1 and cddce1 turn this into:
>
> :
> l$i_6 = p_1(D)->i;
> D.1967_3 = l$i_6 + 1;
> p_1(D)->i = D.1967_3;
> return;
>
> which is then the "optimized" gimple, already before IPA and at -O1.
Ok, that's certainly better. Can you file a bugreport for the FRE/DSE
issue please?
> For the record, without total scalarization, the "optimized" gimple
> would be:
>
> :
> l = *p_1(D);
> l$i_6 = p_1(D)->i;
> D.1967_3 = l$i_6 + 1;
> *p_1(D) = l;
> p_1(D)->i = D.1967_3;
> return;
>
> So at the moment FRE/DSE certainly does not help. Eventually we
> should do something like that or a real aggregate copy propagation but
> until then we probably need to live with the total scalarization
> thingy - I have learned in the PR mentioned above and a few others,
> there are people who really want at least this functionality now - and
> it should not perform this badly on unaligned structures.
The only thing I'm worried about is
struct X { int i; short s; pad p; int j; };
struct Y { int i; short s; int j; };
int main()
{
struct X x = { 1, 2, 3, 4 };
struct Y y = *(struct Y)&x;
y.s++;
*(struct Y)&x = y;
if (x->pad != 3)
abort ();
return 0;
}
untested, but any variant needs to make sure the padding value is not
lost by means of assigning x to y and back. The above should translate
to y = MEM[&x + 0]; ... MEM[&x + 0] = y;
So if the result for sth like the above looks ok the patch is ok.
Thanks,
Richard.
> Martin
>
>
>
>
> > Richard.
> >
> >
> > > Thanks,
> > >
> > > Martin
> > >
> > >
> >
Re: [testsuite, libffi] XFAIL libffi.call/cls_{,long}double_va.c on IRIX 6.5 (PR libffi/46660)
Richard Guenther writes:
> The tests now fail on x86_64-linux and i?86-linux like
>
> FAIL: libffi.call/cls_double_va.c -O0 -W -Wall output pattern test, is 7.0
> FAIL: libffi.call/cls_longdouble_va.c -O0 -W -Wall output pattern test, is 7.0
> FAIL: libffi.call/cls_double_va.c -O2 output pattern test, is 7.0
> FAIL: libffi.call/cls_longdouble_va.c -O2 output pattern test, is 7.0
> FAIL: libffi.call/cls_double_va.c -O3 output pattern test, is 7.0
> FAIL: libffi.call/cls_longdouble_va.c -O3 output pattern test, is 7.0
> FAIL: libffi.call/cls_double_va.c -Os output pattern test, is 7.0
> FAIL: libffi.call/cls_longdouble_va.c -Os output pattern test, is 7.0
> FAIL: libffi.call/cls_double_va.c -O2 -fomit-frame-pointer output pattern
> test,
> is 7.0
> FAIL: libffi.call/cls_longdouble_va.c -O2 -fomit-frame-pointer output pattern
> te
> st, is 7.0
>
> spawn [open ...]^M
> 7.0
> res: 4
> 7.0
> res: 4
> PASS: libffi.call/cls_double_va.c -O0 -W -Wall execution test
> FAIL: libffi.call/cls_double_va.c -O0 -W -Wall output pattern test, is 7.0
> res: 4
> 7.0
> res: 4
> , should match PR libffi/466607.0^M?
> res: 4^M?
> 7.0^M?
> res: 4
>
> I believe your dg-output first arguments are bogus.
You're right, I'm an idiot ;-) I should have tested on a non-xfailed
target, too.
Fixed as follows, tested on i386-pc-solaris2.11 and mips-sgi-irix6.5,
applied to mainline and 4.6 branch. The 4.5 branch is unaffected.
Sorry.
Rainer
2011-06-29 Rainer Orth
* testsuite/libffi.call/cls_double_va.c: Move PR number to comment.
* testsuite/libffi.call/cls_longdouble_va.c: Likewise.
Index: libffi/testsuite/libffi.call/cls_double_va.c
===
--- libffi/testsuite/libffi.call/cls_double_va.c(revision 175618)
+++ libffi/testsuite/libffi.call/cls_double_va.c(working copy)
@@ -6,7 +6,7 @@
/* { dg-do run { xfail strongarm*-*-* xscale*-*-* } } */
/* { dg-output "" { xfail avr32*-*-* } } */
-/* { dg-output "PR libffi/46660" { xfail mips-sgi-irix6* } } */
+/* { dg-output "" { xfail mips-sgi-irix6* } } PR libffi/46660 */
/* { dg-skip-if "" arm*-*-* { "-mfloat-abi=hard" } { "" } } */
#include "ffitest.h"
Index: libffi/testsuite/libffi.call/cls_longdouble_va.c
===
--- libffi/testsuite/libffi.call/cls_longdouble_va.c(revision 175618)
+++ libffi/testsuite/libffi.call/cls_longdouble_va.c(working copy)
@@ -6,7 +6,7 @@
/* { dg-do run { xfail strongarm*-*-* xscale*-*-* } } */
/* { dg-output "" { xfail avr32*-*-* x86_64-*-mingw* } } */
-/* { dg-output "PR libffi/46660" { xfail mips-sgi-irix6* } } */
+/* { dg-output "" { xfail mips-sgi-irix6* } } PR libffi/46660 */
/* { dg-skip-if "" arm*-*-* { "-mfloat-abi=hard" } { "" } } */
#include "ffitest.h"
--
-
Rainer Orth, Center for Biotechnology, Bielefeld University
Re: [patch tree-optimization]: Do bitwise operator optimizations for X op !X patterns
- Original Message -
From: "Kai Tietz"
To: "Richard Guenther"
Cc: gcc-patches@gcc.gnu.org
Sent: Wednesday, June 29, 2011 1:33:30 PM
Subject: Re: [patch tree-optimization]: Do bitwise operator optimizations for X
op !X patterns
- Original Message -
From: "Richard Guenther"
To: "Kai Tietz"
Cc: gcc-patches@gcc.gnu.org
Sent: Wednesday, June 29, 2011 12:14:10 PM
Subject: Re: [patch tree-optimization]: Do bitwise operator optimizations for X
op !X patterns
On Tue, Jun 28, 2011 at 5:05 PM, Kai Tietz wrote:
> Hello,
>
> this patch implements the X op !X patterns within tree-ssa-forwprop.c without
> using here const-fold routines. Additionally it does some trivial folding
> for X op X. Implementation
> also looks through [(type)] X op [(type)] !X, if type of X is integral and
> precision is suitable
> for operation.
>
> ChangeLog gcc/
>
> 2011-06-28 Kai Tietz
>
> * tree-ssa-forwprop.c (operand_precision_onep): New
> function.
> (find_possible_not_expr_argument): Likewise.
> (simplify_bitwise_binary_1): Likewise.
> (simplify_bitwise_binary): Use simplify_bitwise_binary_1
> for detecting various X op !X optimizations.
>
> ChangeLog gcc/testsuite
>
> 2011-06-28 Kai Tietz
>
> * gcc.dg/binop-notand1a.c: New test.
> * gcc.dg/binop-notand2a.c: New test.
> * gcc.dg/binop-notand3a.c: New test.
> * gcc.dg/binop-notand4a.c: New test.
> * gcc.dg/binop-notand5a.c: New test.
> * gcc.dg/binop-notand6a.c: New test.
> * gcc.dg/binop-notor1.c: New test.
> * gcc.dg/binop-notor2.c: New test.
> * gcc.dg/binop-notxor1.c: New test.
> * gcc.dg/binop-notxor2.c: New test.
>
> Bootstrapped and regression tested for all languages plus Ada and Obj-C for
> x86_64-pc-linux-gnu. Ok for apply?
I can't follow the code in find_possible_not_expr_argument or its uses
at all. Please try to produce patches that look more obvious in what
they are doing - don't try to solve every testcase you can come up with
in a single patch. Especially don't write functions like
find_possible_not_expr_argument which seems to have evolved a lot
after you wrote the overall function comment.
Thanks,
Richard.
> Regards,
> Kai
>
Well, I added some comments to these functions and renamed the
find_possible_not_expr_argument function to detect_not_expr_operand, which hits
its use better.
The cause for this function is, that there are more then one variant of
expressing a logical-not and all of them are used.
This routine simply tries to detect different variants used for not. Eg ~X ==
!X and (X ^ 1) == !X for integral type of X with precision one. For X with
integral type, (X == 0) == !X.
The folding for the three different bitwise-operations is pretty easy and it
makes sense to implement them at once. I see here no good point to separate
them into different patches. To separate them might even lead to questions
about abstracting some code-pieces out of the main function.
I didn't added testcases for all variants I am aware now. Just those, which are
now handled.
So hope you can read and understand logic of patch better by updated patch.
Regards,
Kai
I found that in version I've sent there is an unclosed comment. So here is
updated patch, which additionally simplify some code to ease reading.
Regards,
Kai
Index: gcc-head/gcc/tree-ssa-forwprop.c
===
--- gcc-head.orig/gcc/tree-ssa-forwprop.c
+++ gcc-head/gcc/tree-ssa-forwprop.c
@@ -1674,6 +1674,223 @@ simplify_builtin_call (gimple_stmt_itera
return false;
}
+/* Checks if expression has type of one-bit precision, or is result of
+ a known boolean expression. */
+static bool
+operand_precision_onep (tree expr)
+{
+ enum tree_code code;
+ gimple def_stmt;
+
+ do
+{
+ code = TREE_CODE (expr);
+ if (!INTEGRAL_TYPE_P (TREE_TYPE (expr)))
+ return false;
+ if (TYPE_PRECISION (TREE_TYPE (expr)) == 1)
+ return true;
+ if (code != SSA_NAME)
+ break;
+ def_stmt = SSA_NAME_DEF_STMT (expr);
+ if (!def_stmt || !is_gimple_assign (def_stmt))
+ break;
+ code = gimple_assign_rhs_code (def_stmt);
+ if (!CONVERT_EXPR_CODE_P (code))
+ break;
+ expr = gimple_assign_rhs1 (def_stmt);
+}
+ while (CONVERT_EXPR_CODE_P (code));
+
+ if (code == TRUTH_NOT_EXPR || TREE_CODE_CLASS (code) == tcc_comparison)
+return true;
+ return false;
+}
+
+/* Helper routine for simplify_bitwise_binary_1 function.
+ If a NOT-expression is found, the operand of the NOT-expression is
+ stored in NEXPR.
+ This function returns either the expression after the first
+ integral cast expression, or NAME.
+ Detected not-patterns are !X or X == 0 for X with integral type, and
+ X ^ 1 or ~X for X with integral type with precision of one.
+ The value of CNT_CASTS is either zero, or one. */
+static tree
+detect_not_expr_operand (tree name,
Re: [ARM] Deprecate -mwords-little-endian
On 29/06/11 12:28, Richard Sandiford wrote:
> ARM has an option called -mwords-little-endian that provides big-endian
> compatibility with pre-2.8 compilers. When I asked Richard about it,
> he seemed to think it had outlived its usefulness, so this patch
> deprecates it. We can then remove it once 4.7 is out.
>
> Tested on arm-linux-gnueabi. OK to install? If so, I'll do a patch
> for the web page as well.
>
Please also update the in-line help text in arm.opt. OK with that change.
R.
> Richard
>
>
> Index: gcc/config/arm/arm.c
> ===
> --- gcc/config/arm/arm.c 2011-06-29 09:33:37.0 +0100
> +++ gcc/config/arm/arm.c 2011-06-29 12:20:35.0 +0100
> @@ -1483,6 +1483,10 @@ arm_option_override (void)
>if (TARGET_APCS_FLOAT)
> warning (0, "passing floating point arguments in fp regs not yet
> supported");
>
> + if (TARGET_LITTLE_WORDS)
> +warning (OPT_Wdeprecated, "% is deprecated and "
> + "will be removed in a future release");
> +
>/* Initialize boolean versions of the flags, for use in the arm.md file.
> */
>arm_arch3m = (insn_flags & FL_ARCH3M) != 0;
>arm_arch4 = (insn_flags & FL_ARCH4) != 0;
> Index: gcc/doc/invoke.texi
> ===
> --- gcc/doc/invoke.texi 2011-06-22 16:45:29.0 +0100
> +++ gcc/doc/invoke.texi 2011-06-29 12:17:40.0 +0100
> @@ -10237,7 +10237,7 @@ Generate code for a little-endian word o
> order. That is, a byte order of the form @samp{32107654}. Note: this
> option should only be used if you require compatibility with code for
> big-endian ARM processors generated by versions of the compiler prior to
> -2.8.
> +2.8. This option is now deprecated.
>
> @item -mcpu=@var{name}
> @opindex mcpu
>
Re: [testsuite] ARM wmul tests: require arm_dsp_multiply
On 23/06/11 22:38, Janis Johnson wrote:
> Tests wmul-[1234].c and mla-2.c in gcc.target/arm require support that
> the arm backend identifies as TARGET_DSP_MULTIPLY. The tests all
> specify a -march option with that support, but it is overridden by
> multilib flags.
>
> This patch adds a new effective target, arm_dsp_multiply, and requires
> it for those tests instead of having them specify a -march value. This
> means that the tests will be skipped for older targets and test coverage
> relies on testing for some newer multilibs.
>
> The same effective target is needed for tests smlaltb-1.c, smlaltt-1.c,
> smlatb-1.c, and smlatt-1.c, but those also need to be renamed so the
> scans don't pass just because the file name is in the assembly file.
>
> OK for trunk, and later for 4.6?
>
> (btw, I'm currently testing ARM compile-only tests with 43 sets of
> multilib flags)
>
I've recently approved a patch from James Greenhalgh
(http://gcc.gnu.org/ml/gcc-patches/2011-06/msg01852.html) that defines
__ARM_DSP_MULTIPLY when these features are available. That should
simplify your target-supports change and also serve as a check that we
aren't erroneously defining that macro.
R.
>
> gcc-20110623-2
>
>
> 2011-06-23 Janis Johnson
>
> * lib/target-supports.exp (check_effective_target_arm_dsp_multiply):
> New.
> * gcc.target/arm/wmul-1.c: Require arm_dsp_multiply, don't
> supply -march.
> * gcc.target/arm/wmul-2.c: Likewise.
> * gcc.target/arm/wmul-3.c: Likewise.
> * gcc.target/arm/wmul-4.c: Likewise.
> * gcc.target/arm/mla-2.c: Likewise.
>
> Index: lib/target-supports.exp
> ===
> --- lib/target-supports.exp (revision 175313)
> +++ lib/target-supports.exp (working copy)
> @@ -1902,6 +1902,33 @@
> }
> }
>
> +# Return 1 if this is an ARM target that supports DSP multiply with
> +# current multilib flags.
> +
> +proc check_effective_target_arm_dsp_multiply { } {
> +return [check_no_compiler_messages arm_dsp_multiply assembly {
> + #if defined(__ARM_ARCH_2__) || defined(__ARM_ARCH_2A__) \
> + || defined(__ARM_ARCH_3__) || defined(__ARM_ARCH_3M__) \
> + || defined(__ARM_ARCH_4__) || defined(__ARM_ARCH_4T__) \
> + || defined(__ARM_ARCH_5T__)
> + # error NOT_SUPPORTED
> + #elif defined(__thumb__) || defined(__thumb2__)
> + # if defined(__ARM_ARCH_5TE__) || defined(__ARM_ARCH_6__) \
> + || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6M__) \
> + || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) \
> + || defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7M__) \
> + || defined(__ARM_ARCH_IWMMXT__) || defined(__ARM_ARCH_IWMMXT2__)
> + #error NOT_SUPPORTED
> + # endif
> + #else
> + # if defined(__ARM_ARCH_5__) || defined(__ARM_ARCH_EP9312__)
> + #error NOT_SUPPORTED
> + # endif
> + #endif
> + int i;
> +}]
> +}
> +
> # Add the options needed for NEON. We need either -mfloat-abi=softfp
> # or -mfloat-abi=hard, but if one is already specified by the
> # multilib, use it. Similarly, if a -mfpu option already enables
> Index: gcc.target/arm/wmul-1.c
> ===
> --- gcc.target/arm/wmul-1.c (revision 175313)
> +++ gcc.target/arm/wmul-1.c (working copy)
> @@ -1,5 +1,6 @@
> /* { dg-do compile } */
> -/* { dg-options "-O2 -march=armv6t2" } */
> +/* { dg-require-effective-target arm_dsp_multiply } */
> +/* { dg-options "-O2" } */
>
> int mac(const short *a, const short *b, int sqr, int *sum)
> {
> Index: gcc.target/arm/wmul-2.c
> ===
> --- gcc.target/arm/wmul-2.c (revision 175313)
> +++ gcc.target/arm/wmul-2.c (working copy)
> @@ -1,5 +1,6 @@
> /* { dg-do compile } */
> -/* { dg-options "-O2 -march=armv6t2" } */
> +/* { dg-require-effective-target arm_dsp_multiply } */
> +/* { dg-options "-O2" } */
>
> void vec_mpy(int y[], const short x[], short scaler)
> {
> Index: gcc.target/arm/wmul-3.c
> ===
> --- gcc.target/arm/wmul-3.c (revision 175313)
> +++ gcc.target/arm/wmul-3.c (working copy)
> @@ -1,5 +1,6 @@
> /* { dg-do compile } */
> -/* { dg-options "-O2 -march=armv6t2" } */
> +/* { dg-require-effective-target arm_dsp_multiply } */
> +/* { dg-options "-O2" } */
>
> int mac(const short *a, const short *b, int sqr, int *sum)
> {
> Index: gcc.target/arm/wmul-4.c
> ===
> --- gcc.target/arm/wmul-4.c (revision 175313)
> +++ gcc.target/arm/wmul-4.c (working copy)
> @@ -1,5 +1,6 @@
> /* { dg-do compile } */
> -/* { dg-options "-O2 -march=armv6t2" } */
> +/* { dg-require-effective-target arm_dsp_multiply } */
> +/* { dg-options "-O2" } */
>
> int mac(const int *a, const int *b, long long sqr
RE: Ping #1: [Patch, AVR]: Fix PR34734
> -Original Message- > From: Georg-Johann Lay [mailto:a...@gjlay.de] > Sent: Wednesday, June 29, 2011 5:26 AM > To: Denis Chertykov > Cc: gcc-patches@gcc.gnu.org; Weddington, Eric; Anatoly Sokolov > Subject: Re: Ping #1: [Patch, AVR]: Fix PR34734 > > > > > You can backport it if you want. > > > > I'm usually didn't backport such patches. > > > > Denis. > > Ok, maybe Eric or Anatoly have some preference for 4.5/4.6. > I would like it for the 4.6 series, if possible. I think that the next releases of the various avr toolchain distributions will very probably be moving up to 4.6.x. Eric
Re: [ARM] fix C++ EH interoperability
On 23/05/11 16:52, Nathan Sidwell wrote: > This patch fixes an interoperability issue with code generated by ARM's > EABI compiler. > > Unlike the generic C++ ABI, which always catches pointers by value, > ARM's ABI only catches pointers by value when there's the possibility of > derived->base conversion happening. The ARM __cxa_type_matcher can > return one of 3 values to indicate the three possibilities. GCC was > returning the success value when a pointer matched exactly, but > returning the pointer by value. This is incompatible with EABI > compliant code, which expects a reference to the thrown pointer in such > cases. > > However, we cannot simply change the GCC type matcher to return a > reference in such cases, as then it will be incompatible with existing > GCC-compiled code. GCC compiled code expects the pointer value at the > landing pad, ARM-compiled code expects a reference to the pointer > value. Hence ARM-compiled code (usually) inserts an additional > dereference. > > Note that the compiler generating the catching code, specifies the > personality routine (unwinder), but the code doing the type-matching > could come from any compliant library. > > In discussion with ARM, I developed the attached patch, which changes > __cxa_type_match to return the succeeded_with_ptr_to_base enum value > whenever a pointer is caught (rather than only when a pointer-to-derived > is converted to a pointer-to-base). This value indicates to the (ARM) > unwinding machinery that the pointer value is being returned. And makes > the type matcher work with the ARM unwinder. We do not change the > conditions under which _cxa_type_match performs the pointer indirection. > > At the other side of the interface, we have to make GCC's implementation > of the EABI unwinder work with an ARM-provided type matcher. Here, the > unwinder notes the return value from _cxa_match_type, and inserts an > additional indirection by using a spare field in the barrier_cache > structure. Fortunately GCC-generated code doesn't use the common > unwinder, but uses a gnu personality routine, which is unchanged. > > This patch results has been tested for arm-linux, and independently > tested by ARM with mixed RVCT-generated code confirming the defect has > been fixed. > > ok? > > nathan > > > arm-eh.patch > > > 2011-05-23 Nathan Sidwell > > gcc/ > * config/arm/unwind-arm.c (enum __cxa_type_match_result): New. > (cxa_type_match): Correct declaration. > (__gnu_unwind_pr_common): Reconstruct > additional indirection when __cxa_type_match returns > succeeded_with_ptr_to_base. > > libstdc++/ > * libsupc++/eh_arm.c (__cxa_type_match): Construct address of > thrown object here. Return succeded_with_ptr_to_base for all > pointer cases. > OK. R.
Re: [testsuite, libffi] XFAIL libffi.call/cls_{,long}double_va.c on IRIX 6.5 (PR libffi/46660)
On Wed, Jun 29, 2011 at 5:52 AM, Rainer Orth
wrote:
> Richard Guenther writes:
>
>> The tests now fail on x86_64-linux and i?86-linux like
>>
>> FAIL: libffi.call/cls_double_va.c -O0 -W -Wall output pattern test, is 7.0
>> FAIL: libffi.call/cls_longdouble_va.c -O0 -W -Wall output pattern test, is
>> 7.0
>> FAIL: libffi.call/cls_double_va.c -O2 output pattern test, is 7.0
>> FAIL: libffi.call/cls_longdouble_va.c -O2 output pattern test, is 7.0
>> FAIL: libffi.call/cls_double_va.c -O3 output pattern test, is 7.0
>> FAIL: libffi.call/cls_longdouble_va.c -O3 output pattern test, is 7.0
>> FAIL: libffi.call/cls_double_va.c -Os output pattern test, is 7.0
>> FAIL: libffi.call/cls_longdouble_va.c -Os output pattern test, is 7.0
>> FAIL: libffi.call/cls_double_va.c -O2 -fomit-frame-pointer output pattern
>> test,
>> is 7.0
>> FAIL: libffi.call/cls_longdouble_va.c -O2 -fomit-frame-pointer output
>> pattern te
>> st, is 7.0
>>
>> spawn [open ...]^M
>> 7.0
>> res: 4
>> 7.0
>> res: 4
>> PASS: libffi.call/cls_double_va.c -O0 -W -Wall execution test
>> FAIL: libffi.call/cls_double_va.c -O0 -W -Wall output pattern test, is 7.0
>> res: 4
>> 7.0
>> res: 4
>> , should match PR libffi/466607.0^M?
>> res: 4^M?
>> 7.0^M?
>> res: 4
>>
>> I believe your dg-output first arguments are bogus.
>
> You're right, I'm an idiot ;-) I should have tested on a non-xfailed
> target, too.
>
> Fixed as follows, tested on i386-pc-solaris2.11 and mips-sgi-irix6.5,
> applied to mainline and 4.6 branch. The 4.5 branch is unaffected.
>
> Sorry.
> Rainer
>
>
> 2011-06-29 Rainer Orth
>
> * testsuite/libffi.call/cls_double_va.c: Move PR number to comment.
> * testsuite/libffi.call/cls_longdouble_va.c: Likewise.
>
> Index: libffi/testsuite/libffi.call/cls_double_va.c
> ===
> --- libffi/testsuite/libffi.call/cls_double_va.c (revision 175618)
> +++ libffi/testsuite/libffi.call/cls_double_va.c (working copy)
> @@ -6,7 +6,7 @@
>
> /* { dg-do run { xfail strongarm*-*-* xscale*-*-* } } */
> /* { dg-output "" { xfail avr32*-*-* } } */
> -/* { dg-output "PR libffi/46660" { xfail mips-sgi-irix6* } } */
> +/* { dg-output "" { xfail mips-sgi-irix6* } } PR libffi/46660 */
> /* { dg-skip-if "" arm*-*-* { "-mfloat-abi=hard" } { "" } } */
>
Why not just add mips-sgi-irix6* to avr32*-*-*?
--
H.J.
Re: [testsuite, libffi] XFAIL libffi.call/cls_{,long}double_va.c on IRIX 6.5 (PR libffi/46660)
"H.J. Lu" writes:
>> Index: libffi/testsuite/libffi.call/cls_double_va.c
>> ===
>> --- libffi/testsuite/libffi.call/cls_double_va.c (revision 175618)
>> +++ libffi/testsuite/libffi.call/cls_double_va.c (working copy)
>> @@ -6,7 +6,7 @@
>>
>> /* { dg-do run { xfail strongarm*-*-* xscale*-*-* } } */
>> /* { dg-output "" { xfail avr32*-*-* } } */
>> -/* { dg-output "PR libffi/46660" { xfail mips-sgi-irix6* } } */
>> +/* { dg-output "" { xfail mips-sgi-irix6* } } PR libffi/46660 */
>> /* { dg-skip-if "" arm*-*-* { "-mfloat-abi=hard" } { "" } } */
>>
>
> Why not just add mips-sgi-irix6* to avr32*-*-*?
So the corresponding PR can be noted. If you list it without
attribution, you'd have to search all of them to find which one is for
IRIX.
Rainer
--
-
Rainer Orth, Center for Biotechnology, Bielefeld University
Re: [PATCH, ARM] [2/3] iWMMXt intrinsics testsuite
On 29/06/11 09:50, Xinyu Qi wrote:
> Hi,
>
> This patch adds some simple iWMMXt intrinsic test cases.
>
> gcc.target/arm/iwmmxt/i_shift_error.c: New.
> gcc.target/arm/iwmmxt/iwmmxt.exp: New.
> gcc.target/arm/iwmmxt/i_wshufh_error.c: New.
> gcc.target/arm/iwmmxt/i_merge_si64_error.c: New.
> gcc.target/arm/iwmmxt/i_waligni_error.c: New.
> gcc.target/arm/iwmmxt/i_extract_error.c: New.
>
Please use "New test" when adding new tests.
> Thanks,
> Xinyu
>
>
> iwmmxt_testsuite.patch
>
>
> Index: gcc/testsuite/gcc.target/arm/iwmmxt/i_shift_error.c
> ===
> --- gcc/testsuite/gcc.target/arm/iwmmxt/i_shift_error.c (revision 0)
> +++ gcc/testsuite/gcc.target/arm/iwmmxt/i_shift_error.c (revision 0)
> @@ -0,0 +1,39 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -march=iwmmxt2" } */
No, this will interact badly with multi-lib testing. You need to add a
test along the lines of arm_iwmmxt_ok from lib/target-supports.exp
R.
Re: [testsuite, libffi] XFAIL libffi.call/cls_{,long}double_va.c on IRIX 6.5 (PR libffi/46660)
On 29 June 2011 14:43, Rainer Orth wrote:
> -/* { dg-output "PR libffi/46660" { xfail mips-sgi-irix6* } } */
> +/* { dg-output "" { xfail mips-sgi-irix6* } } PR libffi/46660 */
Do you fancy adding the appropriate MIPS fix on top of the libffi varargs patch
I posted a few months back - then it could actually pass the test!
Dave
Re: [testsuite, libffi] XFAIL libffi.call/cls_{,long}double_va.c on IRIX 6.5 (PR libffi/46660)
David Gilbert writes:
> On 29 June 2011 14:43, Rainer Orth wrote:
>> -/* { dg-output "PR libffi/46660" { xfail mips-sgi-irix6* } } */
>> +/* { dg-output "" { xfail mips-sgi-irix6* } } PR libffi/46660 */
>
> Do you fancy adding the appropriate MIPS fix on top of the libffi varargs
> patch
> I posted a few months back - then it could actually pass the test!
I plan to do so once libffi 3.0.11 with the varargs patch lands in the
gcc tree, unless someone beats me to it.
Rainer
--
-
Rainer Orth, Center for Biotechnology, Bielefeld University
Re: [testsuite] skip ARM tests if no thumb2 support
On 15/06/11 17:38, Janis Johnson wrote:
> On 06/15/2011 01:13 AM, Ramana Radhakrishnan wrote:
>> On 14 June 2011 21:35, Janis Johnson wrote:
>>> These tests apparently require thumb2 support (I don't yet know much
>>> about ARM). OK for trunk, and later 4.6?
>>
>> OK - The -march=armv7-a is redundant in these tests. You should be
>> able to lose them if arm_thumb2_ok returns true.
>
> That brings up an important point. Without any changes, this test
> passes scan-assembler when run with multilibs that don't override the
> -march= option, but fails the scan when -march from multilibs
> overrides the one from dg-options and doesn't support thumb2. With
> "dg-require-effective-target arm_thumb2_ok" the test is skipped for
> multilibs that don't support thumb2, including those that default to an
> arch that doesn't support it.
>
> If we want this and similar tests to continue to be run for default
> multilibs with default support for older values of -march then instead
> of using ""dg-require-effective-target arm_thumb2_ok" the test can skip
> multilibs that use -march that don't support thumb2, e.g.
>
> /* { dg-skip-if "need thumb2" { arm*-*-* } { "-march=*" } { "-march=armv6t2"
> "-march=armv[7-9]*" "-march=armv[1-9][0-9]*" } } */
>
> In summary, the choice for a test like this is:
>
> 1. Provide options needed to test particular functionality, and skip the
>test if multilib flags that override those options would not test the
>relevant functionality.
> 2. Skip the test if multilib options on their own do not provide the
>functionality that the test covers.
>
I'd posit a third option:
3. Add a new dg directive (perhaps dg-ignore-multilib) that instructs
the framework to ignore the multilib options entirely (only supported
for compile/assembly tests).
There are a lot of target-specific tests that are not really testing
execution, just that the compiler generates the right instruction under
specific compilation options. It seems silly to force all this into the
multi-lib framework.
R.
> Option 2 assumes that there is enough test coverage for later arch
> versions, either with default support or multilib options. This choice
> applies to several tests in gcc.target/arm.
>
> Janis
>
>
>
Re: PATCH [2/n]: Prepare x32: Convert pointer to TLS symbol if needed
On Wed, Jun 29, 2011 at 1:45 AM, Richard Sandiford
wrote:
> "H.J. Lu" writes:
>> @@ -706,7 +706,13 @@ precompute_register_parameters (int num_actuals, struct
>> arg_data *args,
>> pseudo now. TLS symbols sometimes need a call to resolve. */
>> if (CONSTANT_P (args[i].value)
>> && !targetm.legitimate_constant_p (args[i].mode, args[i].value))
>> - args[i].value = force_reg (args[i].mode, args[i].value);
>> + {
>> + if (GET_MODE (args[i].value) != args[i].mode)
>> + args[i].value = convert_to_mode (args[i].mode,
>> + args[i].value,
>> + args[i].unsignedp);
>> + args[i].value = force_reg (args[i].mode, args[i].value);
>> + }
>
> But if GET_MODE (args[i].value) != args[i].mode, then the call to
> targetm.legitimate_constant_p looks wrong. The mode passed in the
> first argument is supposed to the mode of the second argument.
>
> Is there any reason why this and the following:
>
> /* If we are to promote the function arg to a wider mode,
> do it now. */
>
> if (args[i].mode != TYPE_MODE (TREE_TYPE (args[i].tree_value)))
> args[i].value
> = convert_modes (args[i].mode,
> TYPE_MODE (TREE_TYPE (args[i].tree_value)),
> args[i].value, args[i].unsignedp);
>
> need to be done in the current order? I can't think of any off-hand.
> If not, would swapping them also fix the bug?
>
> (I can't review this either way, of course.)
It works on the testcase. I will do a full test.
Thanks.
--
H.J.
C++ PATCHes to list-value-initialization
The first patch implements the resolution of DR 990, which clarifies
that {} means value-initialization if the type has a default constructor.
The second patch fixes a bug in the standard I noticed while looking at
related issues: it says that if a class has any user-provided
constructor, we just call the default constructor. This wording should
have been adjusted when we added defaulted functions; I've raised the
issue with the committee, but am also applying the obvious fix to the
compiler, namely to only consider the user-providedness of the default
constructor when deciding whether or not to zero-initialize first.
Tested x86_64-pc-linux-gnu, applying to trunk.
commit 2bd88a2546673c5f10bedefff0dcfa3565d5b6fa
Author: Jason Merrill
Date: Tue Jun 28 10:20:50 2011 -0400
DR 990
* call.c (convert_like_real) [ck_user]: Handle value-initialization.
(build_new_method_call_1): Likewise.
* init.c (expand_default_init): Handle direct list-initialization
of aggregates.
diff --git a/gcc/cp/call.c b/gcc/cp/call.c
index cfaef7d..e2d455a 100644
--- a/gcc/cp/call.c
+++ b/gcc/cp/call.c
@@ -5592,6 +5592,18 @@ convert_like_real (conversion *convs, tree expr, tree fn, int argnum,
tree convfn = cand->fn;
unsigned i;
+ /* If we're initializing from {}, it's value-initialization. */
+ if (BRACE_ENCLOSED_INITIALIZER_P (expr)
+ && CONSTRUCTOR_NELTS (expr) == 0
+ && TYPE_HAS_DEFAULT_CONSTRUCTOR (totype))
+ {
+ expr = build_value_init (totype, complain);
+ expr = get_target_expr_sfinae (expr, complain);
+ if (expr != error_mark_node)
+ TARGET_EXPR_LIST_INIT_P (expr) = true;
+ return expr;
+ }
+
expr = mark_rvalue_use (expr);
/* When converting from an init list we consider explicit
@@ -5634,7 +5646,7 @@ convert_like_real (conversion *convs, tree expr, tree fn, int argnum,
{
int nelts = CONSTRUCTOR_NELTS (expr);
if (nelts == 0)
- expr = build_value_init (totype, tf_warning_or_error);
+ expr = build_value_init (totype, complain);
else if (nelts == 1)
expr = CONSTRUCTOR_ELT (expr, 0)->value;
else
@@ -7138,10 +7150,29 @@ build_new_method_call_1 (tree instance, tree fns, VEC(tree,gc) **args,
&& BRACE_ENCLOSED_INITIALIZER_P (VEC_index (tree, *args, 0))
&& CONSTRUCTOR_IS_DIRECT_INIT (VEC_index (tree, *args, 0)))
{
+ tree init_list = VEC_index (tree, *args, 0);
+
gcc_assert (VEC_length (tree, *args) == 1
&& !(flags & LOOKUP_ONLYCONVERTING));
- add_list_candidates (fns, first_mem_arg, VEC_index (tree, *args, 0),
+ /* If the initializer list has no elements and T is a class type with
+ a default constructor, the object is value-initialized. Handle
+ this here so we don't need to handle it wherever we use
+ build_special_member_call. */
+ if (CONSTRUCTOR_NELTS (init_list) == 0
+ && TYPE_HAS_DEFAULT_CONSTRUCTOR (basetype)
+ && !processing_template_decl)
+ {
+ tree ob, init = build_value_init (basetype, complain);
+ if (integer_zerop (instance_ptr))
+ return get_target_expr_sfinae (init, complain);
+ ob = build_fold_indirect_ref (instance_ptr);
+ init = build2 (INIT_EXPR, TREE_TYPE (ob), ob, init);
+ TREE_SIDE_EFFECTS (init) = true;
+ return init;
+ }
+
+ add_list_candidates (fns, first_mem_arg, init_list,
basetype, explicit_targs, template_only,
conversion_path, access_binfo, flags, &candidates);
}
@@ -8365,7 +8396,7 @@ perform_implicit_conversion (tree type, tree expr, tsubst_flags_t complain)
permitted. If the conversion is valid, the converted expression is
returned. Otherwise, NULL_TREE is returned, except in the case
that TYPE is a class type; in that case, an error is issued. If
- C_CAST_P is true, then this direction initialization is taking
+ C_CAST_P is true, then this direct-initialization is taking
place as part of a static_cast being attempted as part of a C-style
cast. */
diff --git a/gcc/cp/init.c b/gcc/cp/init.c
index 3ceed90..1719339 100644
--- a/gcc/cp/init.c
+++ b/gcc/cp/init.c
@@ -1443,6 +1443,17 @@ expand_default_init (tree binfo, tree true_exp, tree exp, tree init, int flags,
tree rval;
VEC(tree,gc) *parms;
+ /* If we have direct-initialization from an initializer list, pull
+ it out of the TREE_LIST so the code below can see it. */
+ if (init && TREE_CODE (init) == TREE_LIST
+ && BRACE_ENCLOSED_INITIALIZER_P (TREE_VALUE (init))
+ && CONSTRUCTOR_IS_DIRECT_INIT (TREE_VALUE (init)))
+{
+ gcc_checking_assert ((flags & LOOKUP_ONLYCONVERTING) == 0
+ && TREE_CHAIN (init) == NULL_TREE);
+ init = TREE_VALUE (init);
+}
+
if (init && BRACE_ENCLOSED_INITIALIZER_P (init)
&& CP_AGGREGATE_TYPE_P (type))
{
diff --git a/gcc/testsuite/g++.dg/cpp0x/constexpr-initlist4.C b/gcc/testsuite/g++.dg/cpp0x/constexpr-initlist4.C
new file mode 100644
index 000..8151857
--- /dev/null
+++ b/gcc/testsuite/g++.dg/cpp0x/constexpr-initlist4.C
[PATCH] Split out SCEV dumping
I'm finally annoyed enough about this to do the patch splitting out
SCEV details dumping to a TDF_SCEV (-scev) flag.
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk.
Richard.
2011-06-29 Richard Guenther
* doc/invoke.texi: Document -scev dump modifier.
* tree-pass.h (TDF_SCEV): New dump flag.
* tree-dump.c (dump_option_value_in): Add scev.
* tree-chrec.c: Replace all TDF_DETAILS checks with TDF_SCEV.
* tree-scalar-evolution.c: Likewise.
* gcc.dg/tree-ssa/loop-17.c: Adjust.
Index: gcc/doc/invoke.texi
===
*** gcc/doc/invoke.texi (revision 175630)
--- gcc/doc/invoke.texi (working copy)
*** Enable showing the unique ID (@code{DECL
*** 5559,5564
--- 5559,5566
Enable showing the tree dump for each statement.
@item eh
Enable showing the EH region number holding each statement.
+ @item scev
+ Enable showing scalar evolution analysis details.
@item all
Turn on all options, except @option{raw}, @option{slim}, @option{verbose}
and @option{lineno}.
Index: gcc/tree-pass.h
===
*** gcc/tree-pass.h (revision 175630)
--- gcc/tree-pass.h (working copy)
*** enum tree_dump_index
*** 83,88
--- 83,89
#define TDF_ALIAS (1 << 21) /* display alias information */
#define TDF_ENUMERATE_LOCALS (1 << 22)/* Enumerate locals by uid. */
#define TDF_CSELIB(1 << 23) /* Dump cselib details. */
+ #define TDF_SCEV (1 << 24) /* Dump SCEV details. */
/* In tree-dump.c */
Index: gcc/tree-dump.c
===
*** gcc/tree-dump.c (revision 175630)
--- gcc/tree-dump.c (working copy)
*** static const struct dump_option_value_in
*** 823,831
{"alias", TDF_ALIAS},
{"nouid", TDF_NOUID},
{"enumerate_locals", TDF_ENUMERATE_LOCALS},
{"all", ~(TDF_RAW | TDF_SLIM | TDF_LINENO | TDF_TREE | TDF_RTL | TDF_IPA
| TDF_STMTADDR | TDF_GRAPH | TDF_DIAGNOSTIC | TDF_VERBOSE
! | TDF_RHS_ONLY | TDF_NOUID | TDF_ENUMERATE_LOCALS)},
{NULL, 0}
};
--- 823,832
{"alias", TDF_ALIAS},
{"nouid", TDF_NOUID},
{"enumerate_locals", TDF_ENUMERATE_LOCALS},
+ {"scev", TDF_SCEV},
{"all", ~(TDF_RAW | TDF_SLIM | TDF_LINENO | TDF_TREE | TDF_RTL | TDF_IPA
| TDF_STMTADDR | TDF_GRAPH | TDF_DIAGNOSTIC | TDF_VERBOSE
! | TDF_RHS_ONLY | TDF_NOUID | TDF_ENUMERATE_LOCALS | TDF_SCEV)},
{NULL, 0}
};
Index: gcc/tree-chrec.c
===
*** gcc/tree-chrec.c(revision 175630)
--- gcc/tree-chrec.c(working copy)
*** chrec_apply (unsigned var,
*** 587,593
|| chrec_contains_symbols_defined_in_loop (chrec, var))
return chrec_dont_know;
! if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "(chrec_apply \n");
if (TREE_CODE (x) == INTEGER_CST && SCALAR_FLOAT_TYPE_P (type))
--- 587,593
|| chrec_contains_symbols_defined_in_loop (chrec, var))
return chrec_dont_know;
! if (dump_file && (dump_flags & TDF_SCEV))
fprintf (dump_file, "(chrec_apply \n");
if (TREE_CODE (x) == INTEGER_CST && SCALAR_FLOAT_TYPE_P (type))
*** chrec_apply (unsigned var,
*** 628,634
break;
}
! if (dump_file && (dump_flags & TDF_DETAILS))
{
fprintf (dump_file, " (varying_loop = %d\n", var);
fprintf (dump_file, ")\n (chrec = ");
--- 628,634
break;
}
! if (dump_file && (dump_flags & TDF_SCEV))
{
fprintf (dump_file, " (varying_loop = %d\n", var);
fprintf (dump_file, ")\n (chrec = ");
Index: gcc/tree-scalar-evolution.c
===
*** gcc/tree-scalar-evolution.c (revision 175630)
--- gcc/tree-scalar-evolution.c (working copy)
*** set_scalar_evolution (basic_block instan
*** 572,578
if (dump_file)
{
! if (dump_flags & TDF_DETAILS)
{
fprintf (dump_file, "(set_scalar_evolution \n");
fprintf (dump_file, " instantiated_below = %d \n",
--- 572,578
if (dump_file)
{
! if (dump_flags & TDF_SCEV)
{
fprintf (dump_file, "(set_scalar_evolution \n");
fprintf (dump_file, " instantiated_below = %d \n",
*** get_scalar_evolution (basic_block instan
*** 600,606
if (dump_file)
{
! if (dump_flags & TDF_DETAILS)
{
fprintf (dump_file, "(get_scalar_evolution \n");
fprintf (dump_file, " (scalar = ");
--- 600,606
if (dump_file)
{
! if (dump_flags & TDF_SCEV)
{
fprintf (dump_file, "(get_scalar_evolution \n");
fprintf (dump_file
C++ PATCHes for c++/45923 (constexpr diagnostics)
In C++0x when an expression fails to satisfy the requirements of a
constant expression it can be hard to figure out why; previously the
compiler would just say that a particular class or function was not
literal or constexpr without explaining why. This patch improves
diagnostics by having the compiler recursively explain why classes are
not literal or functions not constexpr.
The second patch updates the rules for what classes can be literal,
adding aggregates. I've left in trivial default constructors for now
even though it was dropped from the FDIS.
The third patch is necessary to avoid some testsuite failures from
complaints about non-literal temporaries if we do eventually drop
trivial default constructors.
The last patch tweaks the similar function maybe_explain_implicit_delete
to use a pointer set instead of a hash table, since we were only using
it as a set anyway.
Tested x86_64-pc-linux-gnu, applying to trunk.
commit fe193797db267abf26a2632fed0a36828505d0c9
Author: Jason Merrill
Date: Mon Jun 27 17:02:23 2011 -0400
PR c++/45923
* class.c (explain_non_literal_class): New.
(finalize_literal_type_property): Call it.
* cp-tree.h: Declare it.
* semantics.c (ensure_literal_type_for_constexpr_object): Call it.
(is_valid_constexpr_fn): Likewise.
(massage_constexpr_body): Split out from...
(register_constexpr_fundef): ...here.
(is_instantiation_of_constexpr): New.
(expand_or_defer_fn_1): Leave DECL_SAVED_TREE alone in that case.
(explain_invalid_constexpr_fn): New.
(cxx_eval_call_expression): Call it.
(potential_constant_expression_1): Likewise. Avoid redundant errors.
* method.c (process_subob_fn): Diagnose non-constexpr.
(walk_field_subobs): Likewise.
(synthesized_method_walk): Don't shortcut if we want diagnostics.
(explain_implicit_non_constexpr): New.
(defaulted_late_check): Use it.
* call.c (build_cxx_call): Remember location.
diff --git a/gcc/cp/call.c b/gcc/cp/call.c
index e2d455a..56f3408 100644
--- a/gcc/cp/call.c
+++ b/gcc/cp/call.c
@@ -6721,7 +6721,10 @@ build_cxx_call (tree fn, int nargs, tree *argarray)
{
tree fndecl;
+ /* Remember roughly where this call is. */
+ location_t loc = EXPR_LOC_OR_HERE (fn);
fn = build_call_a (fn, nargs, argarray);
+ SET_EXPR_LOCATION (fn, loc);
/* If this call might throw an exception, note that fact. */
fndecl = get_callee_fndecl (fn);
diff --git a/gcc/cp/class.c b/gcc/cp/class.c
index 9054b5c..6aefd68 100644
--- a/gcc/cp/class.c
+++ b/gcc/cp/class.c
@@ -37,6 +37,7 @@ along with GCC; see the file COPYING3. If not see
#include "cgraph.h"
#include "tree-dump.h"
#include "splay-tree.h"
+#include "pointer-set.h"
/* The number of nested classes being processed. If we are not in the
scope of any class, this is zero. */
@@ -4582,10 +4583,73 @@ finalize_literal_type_property (tree t)
{
DECL_DECLARED_CONSTEXPR_P (fn) = false;
if (!DECL_TEMPLATE_INFO (fn))
- error ("enclosing class of %q+#D is not a literal type", fn);
+ {
+ error ("enclosing class of constexpr non-static member "
+ "function %q+#D is not a literal type", fn);
+ explain_non_literal_class (t);
+ }
}
}
+/* T is a non-literal type used in a context which requires a constant
+ expression. Explain why it isn't literal. */
+
+void
+explain_non_literal_class (tree t)
+{
+ static struct pointer_set_t *diagnosed;
+
+ if (!CLASS_TYPE_P (t))
+return;
+ t = TYPE_MAIN_VARIANT (t);
+
+ if (diagnosed == NULL)
+diagnosed = pointer_set_create ();
+ if (pointer_set_insert (diagnosed, t) != 0)
+/* Already explained. */
+return;
+
+ inform (0, "%q+T is not literal because:", t);
+ if (TYPE_HAS_NONTRIVIAL_DESTRUCTOR (t))
+inform (0, " %q+T has a non-trivial destructor", t);
+ else if (CLASSTYPE_NON_AGGREGATE (t)
+ && !TYPE_HAS_TRIVIAL_DFLT (t)
+ && !TYPE_HAS_CONSTEXPR_CTOR (t))
+inform (0, " %q+T is not an aggregate, does not have a trivial "
+ "default constructor, and has no constexpr constructor that "
+ "is not a copy or move constructor", t);
+ else
+{
+ tree binfo, base_binfo, field; int i;
+ for (binfo = TYPE_BINFO (t), i = 0;
+ BINFO_BASE_ITERATE (binfo, i, base_binfo); i++)
+ {
+ tree basetype = TREE_TYPE (base_binfo);
+ if (!CLASSTYPE_LITERAL_P (basetype))
+ {
+ inform (0, " base class %qT of %q+T is non-literal",
+ basetype, t);
+ explain_non_literal_class (basetype);
+ return;
+ }
+ }
+ for (field = TYPE_FIELDS (t); field; field = TREE_CHAIN (field))
+ {
+ tree ftype;
+ if (TREE_CODE (field) != FIELD_DECL)
+ continue;
+ ftype = TREE_TYPE (field);
+ if (!literal_type_p (ftype))
+ {
+ inform (0, " non-static data member %q+D has "
+ "non-literal type", field);
+ if (CLASS_TYPE_P (ftype))
+ explain_non_literal_class (ftype);
+ }
+ }
+}
+}
+
/* Check the validity
Re: PATCH [5/n]: Prepare x32: PR middle-end/48016: Inconsistency in non-local goto save area
Ping.
On Sat, Jun 25, 2011 at 9:06 AM, H.J. Lu wrote:
> On Thu, Jun 16, 2011 at 10:18 AM, H.J. Lu wrote:
>> On Thu, Jun 16, 2011 at 12:56 AM, Richard Guenther
>> wrote:
>>> On Wed, Jun 15, 2011 at 9:55 PM, H.J. Lu wrote:
On Wed, Jun 15, 2011 at 8:16 AM, Michael Matz wrote:
> Hi,
>
> On Wed, 15 Jun 2011, H.J. Lu wrote:
>
>> >> + /* FIXME: update_nonlocal_goto_save_area may pass SA in the wrong
>> >> mode. */
>> >> + if (GET_MODE (sa) != mode)
>> >> + {
>> >> + gcc_assert (ptr_mode != Pmode
>> >> + && GET_MODE (sa) == ptr_mode
>> >> + && mode == Pmode);
>> >> + sa = adjust_address (sa, mode, 0);
>> >> + }
>> >
>> > That may be appropriate for a branch, but trunk shouldn't contain
>> > FIXMEs
>> > that explain how something should be fixed, instead that something
>> > should
>> > be carried out. I.e. just fix update_nonlocal_goto_save_area.
>> >
>>
>> I don't know update_nonlocal_goto_save_area enough to fix it
>> without breaking other targets. This patch is the lest invasive.
>> Any suggestions how to properly fix it is appreciated.
>
> Well, the most obvious variant would be to move the above code right
> before the call of emit_stack_save in update_nonlocal_goto_save_area
> (using r_save and STACK_SAVEAREA_MODE (SAVE_NONLOCAL)). All other callers
> of emit_stack_save already make sure to pass an object of correct mode, so
> this one should too.
>
> But I think it's better to just produce a correct array_ref from the
> start. get_nl_goto_field creates an array_type for the
> nonlocal_goto_save_area of correct type (ptr_type_node or
> lang_hooks.types.type_for_mode (Pmode, 1)), and we should use that.
>
> So something like this in update_nonlocal_goto_save_area:
> t_save = build4 (ARRAY_REF,
> TREE_TYPE (TREE_TYPE (cfun->nonlocal_goto_save_area)),
> cfun->nonlocal_goto_save_area,
> integer_one_node, NULL_TREE, NULL_TREE);
>
> instead of the current building of t_save. Then r_save also should get
> the correct mode automatically.
>
Here is the updated patch. OK for trunk?
>>>
>>> The explow.c change is ok. For the function.c change I wonder why
>>> convert_memory_address doesn't do the right thing - from it's documentation
>>> it definitely should, so it should be fixed instead of being replaced by
>>> adjust_address with a zero offset.
>>>
>>
>> convert_memory_address may return a pseudo register converted
>> to Pmode. But here what we want is the same memory address
>> adjusted for Pmode. I don't think the usage of convert_memory_address
>>
>
> Here is the code in question:
>
> r_save = convert_memory_address (Pmode, r_save);
>
> emit_move_insn (r_save, targetm.builtin_setjmp_frame_value ());
>
> R_SAVE must be lvalue. But return from convert_memory_address
> isn't. I am re-posting my patch here. OK for trunk?
>
> Thanks.
>
> --
> H.J.
> ---
> 2011-06-15 H.J. Lu
>
> PR middle-end/48016
> * explow.c (update_nonlocal_goto_save_area): Use proper mode
> for stack save area.
>
> * function.c (expand_function_start): Properly store frame
> pointer for non-local goto.
>
> diff --git a/gcc/explow.c b/gcc/explow.c
> index c7d8183..efe6c7e 100644
> --- a/gcc/explow.c
> +++ b/gcc/explow.c
> @@ -1102,7 +1097,9 @@ update_nonlocal_goto_save_area (void)
> first one is used for the frame pointer save; the rest are sized by
> STACK_SAVEAREA_MODE. Create a reference to array index 1, the first
> of the stack save area slots. */
> - t_save = build4 (ARRAY_REF, ptr_type_node, cfun->nonlocal_goto_save_area,
> + t_save = build4 (ARRAY_REF,
> + TREE_TYPE (TREE_TYPE (cfun->nonlocal_goto_save_area)),
> + cfun->nonlocal_goto_save_area,
> integer_one_node, NULL_TREE, NULL_TREE);
> r_save = expand_expr (t_save, NULL_RTX, VOIDmode, EXPAND_WRITE);
>
> diff --git a/gcc/function.c b/gcc/function.c
> index 81c4d39..131bc09 100644
> --- a/gcc/function.c
> +++ b/gcc/function.c
> @@ -4780,7 +4780,7 @@ expand_function_start (tree subr)
> cfun->nonlocal_goto_save_area,
> integer_zero_node, NULL_TREE, NULL_TREE);
> r_save = expand_expr (t_save, NULL_RTX, VOIDmode, EXPAND_WRITE);
> - r_save = convert_memory_address (Pmode, r_save);
> + r_save = adjust_address (r_save, Pmode, 0);
>
> emit_move_insn (r_save, targetm.builtin_setjmp_frame_value ());
> update_nonlocal_goto_save_area ();
>
--
H.J.
Re: [PATCH] Un-obsolete Interix
An update to the IA-32/x86-64 section seems the right place. I confess to being ignorant about how to update this html page. Please advise. --Doug On 12 Jun 2011, Gerald Pfeifer wrote: And I assume you'll be updating the release notes at http://gcc.gnu.org/gcc-4.7/changes.html . ;-)
Re: [testsuite] skip ARM tests if no thumb2 support
On 06/29/2011 06:55 AM, Richard Earnshaw wrote:
> On 15/06/11 17:38, Janis Johnson wrote:
>> On 06/15/2011 01:13 AM, Ramana Radhakrishnan wrote:
>>> On 14 June 2011 21:35, Janis Johnson wrote:
These tests apparently require thumb2 support (I don't yet know much
about ARM). OK for trunk, and later 4.6?
>>>
>>> OK - The -march=armv7-a is redundant in these tests. You should be
>>> able to lose them if arm_thumb2_ok returns true.
>>
>> That brings up an important point. Without any changes, this test
>> passes scan-assembler when run with multilibs that don't override the
>> -march= option, but fails the scan when -march from multilibs
>> overrides the one from dg-options and doesn't support thumb2. With
>> "dg-require-effective-target arm_thumb2_ok" the test is skipped for
>> multilibs that don't support thumb2, including those that default to an
>> arch that doesn't support it.
>>
>> If we want this and similar tests to continue to be run for default
>> multilibs with default support for older values of -march then instead
>> of using ""dg-require-effective-target arm_thumb2_ok" the test can skip
>> multilibs that use -march that don't support thumb2, e.g.
>>
>> /* { dg-skip-if "need thumb2" { arm*-*-* } { "-march=*" } { "-march=armv6t2"
>> "-march=armv[7-9]*" "-march=armv[1-9][0-9]*" } } */
>>
>> In summary, the choice for a test like this is:
>>
>> 1. Provide options needed to test particular functionality, and skip the
>>test if multilib flags that override those options would not test the
>>relevant functionality.
>> 2. Skip the test if multilib options on their own do not provide the
>>functionality that the test covers.
>>
>
> I'd posit a third option:
>
> 3. Add a new dg directive (perhaps dg-ignore-multilib) that instructs
> the framework to ignore the multilib options entirely (only supported
> for compile/assembly tests).
>
> There are a lot of target-specific tests that are not really testing
> execution, just that the compiler generates the right instruction under
> specific compilation options. It seems silly to force all this into the
> multi-lib framework.
>
> R.
The MIPS tests have support for something like that that, but it seems
to me that if tests are being run for lots of multilibs then it's
better to test things like code generation for a variety of relevant
options instead of using the same options every time.
Having said that, I'm happy that you and other ARM maintainers and
developers are now thinking about how multilibs and dg-options interact
and how you want the tests to be run. My primary goal is to get rid of
spurious failures when we run tests internally with 10 or 12 multilibs.
So far I've been stumbling about trying to skip tests when multilibs
don't support them, but I'll follow whatever approach you guys prefer.
Janis
>> Option 2 assumes that there is enough test coverage for later arch
>> versions, either with default support or multilib options. This choice
>> applies to several tests in gcc.target/arm.
>>
>> Janis
>>
>
[Patch, AVR]: Cleanup avr.c a bit
This patch is mostly mechanical:
Add some comment, break long lines, use CONST_INT_P instead of
CONST_INT == GET_CODE().
The targetm initializer is moved down in the file similar to m32c,
s390, alpha, etc. back ends.
That way a hook can be defined on the fly:
static foo_t
foo_hook (foo_args)
{
...
}
#undef FOO_HOOK
#define FOO_HOOK foo_hook
targetm = ...
instead of the redundant prototyping
static foo_t foo_hook (foo_args)
#undef FOO_HOOK
#define FOO_HOOK foo_hook
targetm = ...
foo_t
foo_hook (foo_args)
{
...
}
This reduced redundancy in avr.c and subsequent patches that
add/remove hooks will need less lines to be changed.
reg_class_tab is now local static and use smallest register class.
Tested without regressions.
Johann
* config/avr/avr.c (targetm): Move down.
(TARGET_ASM_ALIGNED_HI_OP): Move down.
(TARGET_ASM_ALIGNED_SI_OP): Move down.
(TARGET_ASM_UNALIGNED_HI_OP): Move down.
(TARGET_ASM_UNALIGNED_SI_OP): Move down.
(TARGET_ASM_INTEGER): Move down.
(TARGET_ASM_FILE_START): Move down.
(TARGET_ASM_FILE_END): Move down.
(TARGET_ASM_FUNCTION_END_PROLOGUE): Move down.
(TARGET_ASM_FUNCTION_BEGIN_EPILOGUE): Move down.
(TARGET_FUNCTION_VALUE): Move down.
(TARGET_LIBCALL_VALUE): Move down.
(TARGET_FUNCTION_VALUE_REGNO_P): Move down.
(TARGET_ATTRIBUTE_TABLE): Move down.
(TARGET_ASM_FUNCTION_RODATA_SECTION): Move down.
(TARGET_INSERT_ATTRIBUTES): Move down.
(TARGET_SECTION_TYPE_FLAGS): Move down.
(TARGET_ASM_INIT_SECTIONS): Move down.
(TARGET_ENCODE_SECTION_INFO): Move down.
(TARGET_REGISTER_MOVE_COST): Move down.
(TARGET_MEMORY_MOVE_COST): Move down.
(TARGET_RTX_COSTS): Move down.
(TARGET_ADDRESS_COST): Move down.
(TARGET_MACHINE_DEPENDENT_REORG): Move down.
(TARGET_FUNCTION_ARG): Move down.
(TARGET_FUNCTION_ARG_ADVANCE): Move down.
(TARGET_LEGITIMIZE_ADDRESS): Move down.
(TARGET_RETURN_IN_MEMORY): Move down.
(TARGET_STRICT_ARGUMENT_NAMING): Move down.
(TARGET_BUILTIN_SETJMP_FRAME_VALUE): Move down.
(TARGET_HARD_REGNO_SCRATCH_OK): Move down.
(TARGET_CASE_VALUES_THRESHOLD): Move down.
(TARGET_LEGITIMATE_ADDRESS_P): Move down.
(TARGET_FRAME_POINTER_REQUIRED): Move down.
(TARGET_CAN_ELIMINATE): Move down.
(TARGET_CLASS_LIKELY_SPILLED_P): Move down.
(TARGET_OPTION_OVERRIDE): Move down.
(TARGET_CANNOT_MODIFY_JUMPS_P): Move down.
(TARGET_FUNCTION_OK_FOR_SIBCALL): Move down.
(TARGET_INIT_BUILTINS): Move down.
(TARGET_EXPAND_BUILTIN): Move down.
(avr_operand_rtx_cost): Move down.
(avr_init_machine_status): Move up.
(avr_legitimize_address): Define on the fly.
(avr_legitimate_address_p): Define on the fly.
(avr_hard_regno_scratch_ok): Define on the fly.
(avr_builtin_setjmp_frame_value): Define on the fly.
(avr_case_values_threshold): Define on the fly.
(avr_can_eliminate): Define on the fly.
(avr_asm_named_section): Define on the fly.
(avr_frame_pointer_required_p): Define on the fly.
(avr_regnames): Remove unused variable.
(reg_class_tab): Use SIMPLE_LD_REGS for r16-r23.
Use NO_LD_REGS for r0-r15.
(avr_regno_reg_class): Incorporate reg_class_tab.
(out_movqi_r_mr): Use CONST_INT_P.
(out_movqi_mr_r): Use CONST_INT_P.
(output_movsisf): Use CONST_INT_P.
(out_shift_with_cnt): Use CONST_INT_P.
(ashlqi3_out): Use CONST_INT_P.
(ashlhi3_out): Use CONST_INT_P.
(ashlsi3_out): Use CONST_INT_P.
(ashrqi3_out): Use CONST_INT_P.
(ashrhi3_out): Use CONST_INT_P.
(ashrsi3_out): Use CONST_INT_P.
(lshrqi3_out): Use CONST_INT_P.
(lshrhi3_out): Use CONST_INT_P.
(lshrsi3_out): Use CONST_INT_P.
(adjust_insn_length): Use CONST_INT_P.
(avr_rtx_costs): Use CONST_INT_P.
(avr_address_cost): Use CONST_INT_P.
(extra_constraint_Q): Use CONST_INT_P.
(avr_reorg): Use CONST_INT_P.
(output_reload_inhi): Use CONST_INT_P.
(output_reload_insisf): Use CONST_INT_P.
(avr_out_sbxx_branch): Use CONST_INT_P.
(print_operand_address): Use CONST_INT_P.
(print_operand): Use CONST_INT_P.
(notice_update_cc): Use CONST_INT_P.
(avr_simplify_comparison_p): Use CONST_INT_P.
(output_movqi): Use CONST_INT_P.
(output_movhi): Use CONST_INT_P.
Fix long lines.
Index: config/avr/avr.c
===
--- config/avr/avr.c (revision 175629)
+++ config/avr/avr.c (working copy)
@@ -51,7 +51,6 @@
/* Maximal allowed offset for an address in the LD command */
#define MAX_LD_OFFSET(MODE) (64 - (signed)GET_MODE_SIZE (MODE))
-static void avr_option_override (
Re: [Patch, AVR]: Cleanup avr.c a bit
Georg-Johann Lay writes:
> @@ -1257,17 +1129,19 @@ print_operand_address (FILE *file, rtx a
> rtx x = addr;
> if (GET_CODE (x) == CONST)
> x = XEXP (x, 0);
> - if (GET_CODE (x) == PLUS && GET_CODE (XEXP (x,1)) == CONST_INT)
> + if (GET_CODE (x) == PLUS
> + && CONST_INT_P (XEXP (x,1)))
Please also add a space after the comma.
> @@ -5034,9 +4933,11 @@ avr_insert_attributes (tree node, tree *
> {
>static const char dsec[] = ".progmem.data";
>
> - *attributes = tree_cons (get_identifier ("section"),
> - build_tree_list (NULL, build_string
> (strlen (dsec), dsec)),
> - *attributes);
> + *attributes =
> +tree_cons (get_identifier ("section"),
> + build_tree_list (NULL,
> +build_string (strlen (dsec), dsec)),
> + *attributes);
GNU style is to break before the operator.
Andreas.
--
Andreas Schwab, sch...@redhat.com
GPG Key fingerprint = D4E8 DBE3 3813 BB5D FA84 5EC7 45C6 250E 6F00 984E
"And now for something completely different."
Backport patches that are correctness or performance bug fixes for latest AMD/Intel hardware.
Is it ok to commit backported patch from trunk below to gcc 4.6 as long as bootstrap and tests pass (ongoing)? This is one of the patches that is significant enough a bug for recent AMD and Intel hardware. 2011-06-29 Harsha Jagasia Backport from mainline 2011-05-31 Alexandre Oliva * config/i386/i386.c (ix86_rtx_costs): Drop NEG from sub for FMA. * config/i386/sse.md: Add n to negated FMA pattern names. Index: config/i386/sse.md === --- config/i386/sse.md (revision 175646) +++ config/i386/sse.md (working copy) @@ -2130,7 +2130,7 @@ (define_insn "*fma_fmsub_" [(set_attr "type" "ssemuladd") (set_attr "mode" "")]) -(define_insn "*fma_fmadd_" +(define_insn "*fma_fnmadd_" [(set (match_operand:FMAMODE 0 "register_operand" "=x,x,x") (fma:FMAMODE (neg:FMAMODE @@ -2145,7 +2145,7 @@ (define_insn "*fma_fmadd_" [(set_attr "type" "ssemuladd") (set_attr "mode" "")]) -(define_insn "*fma_fmsub_" +(define_insn "*fma_fnmsub_" [(set (match_operand:FMAMODE 0 "register_operand" "=x,x,x") (fma:FMAMODE (neg:FMAMODE Index: config/i386/i386.c === --- config/i386/i386.c (revision 175646) +++ config/i386/i386.c (working copy) @@ -29081,12 +29081,12 @@ ix86_rtx_costs (rtx x, int code, int out /* Negate in op0 or op2 is free: FMS, FNMA, FNMS. */ sub = XEXP (x, 0); if (GET_CODE (sub) == NEG) - sub = XEXP (x, 0); + sub = XEXP (sub, 0); *total += rtx_cost (sub, FMA, speed); sub = XEXP (x, 2); if (GET_CODE (sub) == NEG) - sub = XEXP (x, 0); + sub = XEXP (sub, 0); *total += rtx_cost (sub, FMA, speed); return true; }
Re: [PATCH, PR 49089] Don't split AVX256 unaligned loads by default on bdver1 and generic
> Thanks, Note that there is no "i386" component in Bugzilla, only a "target" so this should have been PR target/49089. The end result is that there are no xrefs in the PR, which is still open btw. So please add the xrefs to the commits in the PR manually and close it if you are done with it. -- Eric Botcazou
Re: [testsuite] skip ARM tests if no thumb2 support
Janis Johnson writes: > On 06/29/2011 06:55 AM, Richard Earnshaw wrote: >> I'd posit a third option: >> >> 3. Add a new dg directive (perhaps dg-ignore-multilib) that instructs >> the framework to ignore the multilib options entirely (only supported >> for compile/assembly tests). >> >> There are a lot of target-specific tests that are not really testing >> execution, just that the compiler generates the right instruction under >> specific compilation options. It seems silly to force all this into the >> multi-lib framework. >> >> R. > > The MIPS tests have support for something like that that, but it seems > to me that if tests are being run for lots of multilibs then it's > better to test things like code generation for a variety of relevant > options instead of using the same options every time. I might be misunderstanding what you mean, but the MIPS version tries to keep as many of the original options as possible, only overriding those that are known to be incompatible with the test. So e.g. a test for a MIPS IV feature will run normally on a MIPS IV-compatible target, but will override the architecture on other targets. That said, I wouldn't recommend the mips.exp approach to anyone. It's been useful, and it seems to do what it's meant to, but it's a very... heavy-weight approach. Richard
Re: [PATCH] Split out SCEV dumping
On Wed, Jun 29, 2011 at 09:27, Richard Guenther wrote: > > I'm finally annoyed enough about this to do the patch splitting out > SCEV details dumping to a TDF_SCEV (-scev) flag. > > Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk. Looks very good. Thanks for this cleanup. Sebastian
Re: PATCH [5/n]: Prepare x32: PR middle-end/48016: Inconsistency in non-local goto save area
Hi, On Wed, 29 Jun 2011, H.J. Lu wrote: > > diff --git a/gcc/function.c b/gcc/function.c > > index 81c4d39..131bc09 100644 > > --- a/gcc/function.c > > +++ b/gcc/function.c > > @@ -4780,7 +4780,7 @@ expand_function_start (tree subr) > > cfun->nonlocal_goto_save_area, > > integer_zero_node, NULL_TREE, NULL_TREE); > > r_save = expand_expr (t_save, NULL_RTX, VOIDmode, EXPAND_WRITE); > > - r_save = convert_memory_address (Pmode, r_save); > > + r_save = adjust_address (r_save, Pmode, 0); This is actually the same problem as in explow.c. t_save is built with ptr_type_node, where it should have been using TREE_TYPE (TREE_TYPE (cfun->nonlocal_goto_save_area)) Then r_save should have the correct mode already, possibly this could be asserted. You are right that r_save needs to correspond to the nonlocal_goto_save_area[0] array-ref, hence pseudos aren't okay, therefore convert_memory_address isn't. Actually I think we might even want to assert that indeed the expanded r_save is of Pmode already. Ciao, Michael.
Re: Backport patches that are correctness or performance bug fixes for latest AMD/Intel hardware.
On Wed, Jun 29, 2011 at 11:03:55AM -0500, harsha.jaga...@amd.com wrote: > Is it ok to commit backported patch from trunk below to gcc 4.6 as long as > bootstrap and tests pass (ongoing)? This is one of the patches that is > significant enough a bug for recent AMD and Intel hardware. Yeah, this is ok for 4.6. > 2011-06-29 Harsha Jagasia > Backport from mainline > > 2011-05-31 Alexandre Oliva > > * config/i386/i386.c (ix86_rtx_costs): Drop NEG from sub for FMA. > * config/i386/sse.md: Add n to negated FMA pattern names. Jakub
[commit] Fix -Werror build break (Re: PATCH [10/n]: Prepare x32: PR rtl-optimization/49114)
> H.J. Lu wrote:
> > >>* reload.c (struct replacement): Remove SUBREG_LOC member.
> > >>(push_reload): Do not set it.
> > >>(push_replacement): Likewise.
> > >>(subst_reload): Remove dead code.
> > >>(copy_replacements): Remove assertion.
> > >>(copy_replacements_1): Do not handle SUBREG_LOC.
> > >>(move_replacements): Likewise.
> > >>(find_replacement): Remove dead code. Use
> > >> reload_adjust_reg_for_mode.
> > >>Detect subregs via recursive descent instead of via SUBREG_LOC.
> > >>
> > >
> > > It works much better. I am testing it now.
> > >
> >
> > It works. There are no regressions on Linux/ia32 nor Linux/x86-64.
> > Can you check it in and mention PR rtl-optimization/49114 ChangeLog?
>
> OK, I've checked the patch in now. Thanks for testing!
Pat points out that this breaks the build on platforms that do not define
LIMIT_RELOAD_CLASS due to a -Werror unused variable warning:
/home/gccbuild/gcc_trunk_anonsvn/gcc/gcc/reload.c: In function 'push_reload':
/home/gccbuild/gcc_trunk_anonsvn/gcc/gcc/reload.c:926:28: error: variable
'out_subreg_loc' set but not used [-Werror=unused-but-set-variable]
/home/gccbuild/gcc_trunk_anonsvn/gcc/gcc/reload.c:926:8: error: variable
'in_subreg_loc' set but not used [-Werror=unused-but-set-variable]
Fixed by placing the variable under #ifdef LIMIT_RELOAD_CLASS as well.
Committed to mainline.
Bye,
Ulrich
ChangeLog:
PR rtl-optimization/49114
* reload.c (push_reload): Define in_subreg_loc and out_subreg_loc
only if LIMIT_RELOAD_CLASS to avoid -Werror build breaks.
Index: gcc/reload.c
===
*** gcc/reload.c(revision 175631)
--- gcc/reload.c(working copy)
*** push_reload (rtx in, rtx out, rtx *inloc
*** 923,929
--- 923,931
int i;
int dont_share = 0;
int dont_remove_subreg = 0;
+ #ifdef LIMIT_RELOAD_CLASS
rtx *in_subreg_loc = 0, *out_subreg_loc = 0;
+ #endif
int secondary_in_reload = -1, secondary_out_reload = -1;
enum insn_code secondary_in_icode = CODE_FOR_nothing;
enum insn_code secondary_out_icode = CODE_FOR_nothing;
*** push_reload (rtx in, rtx out, rtx *inloc
*** 1068,1074
--- 1070,1078
#endif
))
{
+ #ifdef LIMIT_RELOAD_CLASS
in_subreg_loc = inloc;
+ #endif
inloc = &SUBREG_REG (in);
in = *inloc;
#if ! defined (LOAD_EXTEND_OP) && ! defined (WORD_REGISTER_OPERATIONS)
*** push_reload (rtx in, rtx out, rtx *inloc
*** 1163,1169
--- 1167,1175
#endif
))
{
+ #ifdef LIMIT_RELOAD_CLASS
out_subreg_loc = outloc;
+ #endif
outloc = &SUBREG_REG (out);
out = *outloc;
#if ! defined (LOAD_EXTEND_OP) && ! defined (WORD_REGISTER_OPERATIONS)
--
Dr. Ulrich Weigand
GNU Toolchain for Linux on System z and Cell BE
ulrich.weig...@de.ibm.com
Re: PATCH [2/n]: Prepare x32: Convert pointer to TLS symbol if needed
On Wed, Jun 29, 2011 at 7:06 AM, H.J. Lu wrote:
> On Wed, Jun 29, 2011 at 1:45 AM, Richard Sandiford
> wrote:
>> "H.J. Lu" writes:
>>> @@ -706,7 +706,13 @@ precompute_register_parameters (int num_actuals,
>>> struct arg_data *args,
>>> pseudo now. TLS symbols sometimes need a call to resolve. */
>>> if (CONSTANT_P (args[i].value)
>>> && !targetm.legitimate_constant_p (args[i].mode, args[i].value))
>>> - args[i].value = force_reg (args[i].mode, args[i].value);
>>> + {
>>> + if (GET_MODE (args[i].value) != args[i].mode)
>>> + args[i].value = convert_to_mode (args[i].mode,
>>> + args[i].value,
>>> + args[i].unsignedp);
>>> + args[i].value = force_reg (args[i].mode, args[i].value);
>>> + }
>>
>> But if GET_MODE (args[i].value) != args[i].mode, then the call to
>> targetm.legitimate_constant_p looks wrong. The mode passed in the
>> first argument is supposed to the mode of the second argument.
>>
>> Is there any reason why this and the following:
>>
>> /* If we are to promote the function arg to a wider mode,
>> do it now. */
>>
>> if (args[i].mode != TYPE_MODE (TREE_TYPE (args[i].tree_value)))
>> args[i].value
>> = convert_modes (args[i].mode,
>> TYPE_MODE (TREE_TYPE (args[i].tree_value)),
>> args[i].value, args[i].unsignedp);
>>
>> need to be done in the current order? I can't think of any off-hand.
>> If not, would swapping them also fix the bug?
>>
>> (I can't review this either way, of course.)
>
> It works on the testcase. I will do a full test.
>
It works. There are no regressions on Linux/x86-64.
OK for trunk?
Thanks.
--
H.J.
2011-06-29 H.J. Lu
PR middle-end/47715
* calls.c (precompute_register_parameters): Promote the function
argument before checking non-legitimate constant.
2011-06-29 H.J. Lu
PR middle-end/47715
* calls.c (precompute_register_parameters): Promote the function
argument before checking non-legitimate constant.
diff --git a/gcc/calls.c b/gcc/calls.c
index bba477c..7538e4e 100644
--- a/gcc/calls.c
+++ b/gcc/calls.c
@@ -702,12 +702,6 @@ precompute_register_parameters (int num_actuals, struct
arg_data *args,
pop_temp_slots ();
}
- /* If the value is a non-legitimate constant, force it into a
- pseudo now. TLS symbols sometimes need a call to resolve. */
- if (CONSTANT_P (args[i].value)
- && !targetm.legitimate_constant_p (args[i].mode, args[i].value))
- args[i].value = force_reg (args[i].mode, args[i].value);
-
/* If we are to promote the function arg to a wider mode,
do it now. */
@@ -717,6 +711,12 @@ precompute_register_parameters (int num_actuals, struct
arg_data *args,
TYPE_MODE (TREE_TYPE (args[i].tree_value)),
args[i].value, args[i].unsignedp);
+ /* If the value is a non-legitimate constant, force it into a
+ pseudo now. TLS symbols sometimes need a call to resolve. */
+ if (CONSTANT_P (args[i].value)
+ && !targetm.legitimate_constant_p (args[i].mode, args[i].value))
+ args[i].value = force_reg (args[i].mode, args[i].value);
+
/* If we're going to have to load the value by parts, pull the
parts into pseudos. The part extraction process can involve
non-trivial computation. */
Re: [cxx-mem-model] __sync_mem_load
On 06/23/2011 03:50 PM, Andrew MacLeod wrote:
> * doc/extend.texi (__sync_mem_load): Document.
> * c-family/c-common.c (resolve_overloaded_builtin): Add
> BUILT_IN_SYNC_MEM_LOAD_N.
> * optabs.c (expand_sync_mem_load): New.
> * optabs.h (enum direct_optab_index): Add DOI_sync_mem_load.
> (sync_mem_load_optab): Define.
> * genopinit.c: Add entry for sync_mem_load.
> * builtins.c (expand_builtin_sync_mem_load): New.
> (expand_builtin): Handle BUILT_IN_SYNC_MEM_LOAD_*
> * sync-bultins.def: Add entries for BUILT_IN_SYNC_MEM_LOAD_*.
> * testsuite/gcc.dg/sync-mem-invalid.c: Add invalid load tests.
> * testsuite/gcc.dg/sync-mem.h: Add load executable tests.
> * builtin-types.def (BT_FN_I{1,2,4,8,16}_VPTR_INT): New.
> * expr.h (expand_sync_mem_load): Declare.
> * fortran/types.def (BT_FN_I{1,2,4,8,16}_VPTR_INT): New.
> * config/i386/sync.md (sync_mem_load): New pattern.
Looks good.
> + (define_expand "sync_mem_load"
> + [(match_operand:SWI 0 "register_operand" "") ;; output
> +(match_operand:SWI 1 "memory_operand" "");; memory
> +(match_operand:SI 2 "const_int_operand" "")];; memory model
> +""
> + {
> + if (INTVAL (operands[2]) == MEMMODEL_ACQUIRE ||
> + INTVAL (operands[2]) == MEMMODEL_SEQ_CST)
> + expand_builtin_sync_synchronize ();
> + ix86_expand_move (mode, operands);
> + DONE;
> + })
Formatting error (|| operator on next line).
Invoke gen_memory_barrier directly instead of a
call back into expand_builtin_sync_synchronize.
Isn't there a MEMMODEL setting that would call for the
use of LFENCE instead of MFENCE?
r~
C++ PATCH for c++/49520 (bogus error with using in constexpr)
A CLEANUP_POINT_EXPR wraps each statement, so we need to strip it at a
lower level.
Tested x86_64-pc-linux-gnu, applied to trunk.
commit 505b4baa3619375d81b409f9ff2bde95cce0f50a
Author: Jason Merrill
Date: Wed Jun 29 12:43:19 2011 -0400
PR c++/49520
* semantics.c (constexpr_fn_retval): Handle CLEANUP_POINT_EXPR here.
(massage_constexpr_body): Not here.
diff --git a/gcc/cp/cp-tree.def b/gcc/cp/cp-tree.def
index 12c01cb..bb1b753 100644
--- a/gcc/cp/cp-tree.def
+++ b/gcc/cp/cp-tree.def
@@ -207,7 +207,7 @@ DEFTREECODE (UNBOUND_CLASS_TEMPLATE, "unbound_class_template", tcc_type, 0)
DEFTREECODE (USING_DECL, "using_decl", tcc_declaration, 0)
/* A using directive. The operand is USING_STMT_NAMESPACE. */
-DEFTREECODE (USING_STMT, "using_directive", tcc_statement, 1)
+DEFTREECODE (USING_STMT, "using_stmt", tcc_statement, 1)
/* An un-parsed default argument. Holds a vector of input tokens and
a vector of places where the argument was instantiated before
diff --git a/gcc/cp/semantics.c b/gcc/cp/semantics.c
index fb984d4..ad68a01 100644
--- a/gcc/cp/semantics.c
+++ b/gcc/cp/semantics.c
@@ -5657,6 +5657,9 @@ constexpr_fn_retval (tree body)
return NULL_TREE;
return error_mark_node;
+case CLEANUP_POINT_EXPR:
+ return constexpr_fn_retval (TREE_OPERAND (body, 0));
+
case USING_STMT:
return NULL_TREE;
@@ -5683,8 +5686,6 @@ massage_constexpr_body (tree fun, tree body)
body = EH_SPEC_STMTS (body);
if (TREE_CODE (body) == MUST_NOT_THROW_EXPR)
body = TREE_OPERAND (body, 0);
- if (TREE_CODE (body) == CLEANUP_POINT_EXPR)
-body = TREE_OPERAND (body, 0);
body = constexpr_fn_retval (body);
}
return body;
diff --git a/gcc/testsuite/g++.dg/cpp0x/constexpr-using2.C b/gcc/testsuite/g++.dg/cpp0x/constexpr-using2.C
new file mode 100644
index 000..6b28281
--- /dev/null
+++ b/gcc/testsuite/g++.dg/cpp0x/constexpr-using2.C
@@ -0,0 +1,18 @@
+// PR c++/49520
+// { dg-options -std=c++0x }
+
+namespace x { void foo(); }
+
+template
+struct traits
+{
+static constexpr bool f() { return true; }
+
+static constexpr bool g()
+{
+using x::foo;
+return f() && noexcept(foo());
+}
+};
+
+template struct traits;
C++ PATCH for c++/49554 (bogus error with lambda in template)
The failure in this testcase was happening because we were trying to
evaluate the capture decltype again when instantiating the lambda
operator(), but it only works when we're still in the enclosing function
context. Fixed by basically waiting to copy the type over at
instantiation time rather than copy the dependent type at template
definition time.
Tested x86_64-pc-linux-gnu, applied to trunk.
commit 7493431b3a7c540254aa92cd0e8ea873eace94f8
Author: Jason Merrill
Date: Wed Jun 29 11:55:58 2011 -0400
PR c++/49554
* semantics.c (lambda_proxy_type): New.
(build_capture_proxy): Use it.
* cp-tree.h (DECLTYPE_FOR_LAMBDA_PROXY): New.
* pt.c (tsubst) [DECLTYPE_TYPE]: Use them.
diff --git a/gcc/cp/cp-tree.h b/gcc/cp/cp-tree.h
index 7244cc8..55c88e3 100644
--- a/gcc/cp/cp-tree.h
+++ b/gcc/cp/cp-tree.h
@@ -3418,11 +3418,13 @@ more_aggr_init_expr_args_p (const aggr_init_expr_arg_iterator *iter)
/* These flags indicate that we want different semantics from normal
decltype: lambda capture just drops references, lambda return also does
- type decay. */
+ type decay, lambda proxies look through implicit dereference. */
#define DECLTYPE_FOR_LAMBDA_CAPTURE(NODE) \
TREE_LANG_FLAG_0 (DECLTYPE_TYPE_CHECK (NODE))
#define DECLTYPE_FOR_LAMBDA_RETURN(NODE) \
TREE_LANG_FLAG_1 (DECLTYPE_TYPE_CHECK (NODE))
+#define DECLTYPE_FOR_LAMBDA_PROXY(NODE) \
+ TREE_LANG_FLAG_2 (DECLTYPE_TYPE_CHECK (NODE))
/* Nonzero for VAR_DECL and FUNCTION_DECL node means that `extern' was
specified in its declaration. This can also be set for an
@@ -5455,6 +5457,7 @@ extern tree build_lambda_object (tree);
extern tree begin_lambda_type (tree);
extern tree lambda_capture_field_type (tree);
extern tree lambda_return_type (tree);
+extern tree lambda_proxy_type (tree);
extern tree lambda_function (tree);
extern void apply_lambda_return_type(tree, tree);
extern tree add_capture (tree, tree, tree, bool, bool);
diff --git a/gcc/cp/pt.c b/gcc/cp/pt.c
index b3dd85f..d1d8336 100644
--- a/gcc/cp/pt.c
+++ b/gcc/cp/pt.c
@@ -11108,6 +11108,8 @@ tsubst (tree t, tree args, tsubst_flags_t complain, tree in_decl)
type = lambda_capture_field_type (type);
else if (DECLTYPE_FOR_LAMBDA_RETURN (t))
type = lambda_return_type (type);
+ else if (DECLTYPE_FOR_LAMBDA_PROXY (t))
+ type = lambda_proxy_type (type);
else
type = finish_decltype_type
(type, DECLTYPE_TYPE_ID_EXPR_OR_MEMBER_ACCESS_P (t), complain);
diff --git a/gcc/cp/semantics.c b/gcc/cp/semantics.c
index 4581729..fb984d4 100644
--- a/gcc/cp/semantics.c
+++ b/gcc/cp/semantics.c
@@ -8489,6 +8489,27 @@ insert_pending_capture_proxies (void)
LAMBDA_EXPR_PENDING_PROXIES (lam) = NULL;
}
+/* Given REF, a COMPONENT_REF designating a field in the lambda closure,
+ return the type we want the proxy to have: the type of the field itself,
+ with added const-qualification if the lambda isn't mutable and the
+ capture is by value. */
+
+tree
+lambda_proxy_type (tree ref)
+{
+ tree type;
+ if (REFERENCE_REF_P (ref))
+ref = TREE_OPERAND (ref, 0);
+ type = TREE_TYPE (ref);
+ if (!dependent_type_p (type))
+return type;
+ type = cxx_make_type (DECLTYPE_TYPE);
+ DECLTYPE_TYPE_EXPR (type) = ref;
+ DECLTYPE_FOR_LAMBDA_PROXY (type) = true;
+ SET_TYPE_STRUCTURAL_EQUALITY (type);
+ return type;
+}
+
/* MEMBER is a capture field in a lambda closure class. Now that we're
inside the operator(), build a placeholder var for future lookups and
debugging. */
@@ -8496,7 +8517,7 @@ insert_pending_capture_proxies (void)
tree
build_capture_proxy (tree member)
{
- tree var, object, fn, closure, name, lam;
+ tree var, object, fn, closure, name, lam, type;
closure = DECL_CONTEXT (member);
fn = lambda_function (closure);
@@ -8511,7 +8532,8 @@ build_capture_proxy (tree member)
/* Remove the __ inserted by add_capture. */
name = get_identifier (IDENTIFIER_POINTER (DECL_NAME (member)) + 2);
- var = build_decl (input_location, VAR_DECL, name, TREE_TYPE (object));
+ type = lambda_proxy_type (object);
+ var = build_decl (input_location, VAR_DECL, name, type);
SET_DECL_VALUE_EXPR (var, object);
DECL_HAS_VALUE_EXPR_P (var) = 1;
DECL_ARTIFICIAL (var) = 1;
diff --git a/gcc/testsuite/g++.dg/cpp0x/lambda/lambda-template3.C b/gcc/testsuite/g++.dg/cpp0x/lambda/lambda-template3.C
new file mode 100644
index 000..fd6f1d3
--- /dev/null
+++ b/gcc/testsuite/g++.dg/cpp0x/lambda/lambda-template3.C
@@ -0,0 +1,33 @@
+// PR c++/49554
+// { dg-options -std=c++0x }
+
+template
+ struct base
+ {
+struct iterator { };
+
+iterator begin();
+ };
+
+template
+class flist : public base
+{
+ typedef base Base;
+
+ typedef typename Base::iterator Base_iterator;
+public:
+
+ void
+ resize()
+ {
+Base_iterator b = Base::begin();
+
+[b](int i) { return i; };
+ }
+};
+
+void test01()
+{
+ flist fl;
+ fl.resize();
+}
[PATCH 1/6] Correct typo.
2011-06-29 Sebastian Pop * graphite-clast-to-gimple.c (clast_name_to_index): Add missing space. --- gcc/ChangeLog |4 gcc/graphite-clast-to-gimple.c |2 +- 2 files changed, 5 insertions(+), 1 deletions(-) diff --git a/gcc/ChangeLog b/gcc/ChangeLog index e37d823..8be5adb 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,7 @@ +2011-06-29 Sebastian Pop + + * graphite-clast-to-gimple.c (clast_name_to_index): Add missing space. + 2011-06-29 Eric Botcazou PR tree-optimization/49539 diff --git a/gcc/graphite-clast-to-gimple.c b/gcc/graphite-clast-to-gimple.c index c8356d3..4a4c3d2 100644 --- a/gcc/graphite-clast-to-gimple.c +++ b/gcc/graphite-clast-to-gimple.c @@ -88,7 +88,7 @@ clast_name_to_index (clast_name_p name, htab_t index_table) #ifdef CLOOG_ORG gcc_assert (name->type == clast_expr_name); - tmp.name = ((const struct clast_name*) name)->name; + tmp.name = ((const struct clast_name *) name)->name; #else tmp.name = name; #endif -- 1.7.4.1
[PATCH 2/6] Correct computation of max.
2011-06-29 Sebastian Pop
* graphite-ppl.h (value_max): Correct computation of max.
---
gcc/ChangeLog |4
gcc/graphite-ppl.h |3 ++-
2 files changed, 6 insertions(+), 1 deletions(-)
diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 8be5adb..b0e3173 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,5 +1,9 @@
2011-06-29 Sebastian Pop
+ * graphite-ppl.h (value_max): Correct computation of max.
+
+2011-06-29 Sebastian Pop
+
* graphite-clast-to-gimple.c (clast_name_to_index): Add missing space.
2011-06-29 Eric Botcazou
diff --git a/gcc/graphite-ppl.h b/gcc/graphite-ppl.h
index 695d01f..49bde61 100644
--- a/gcc/graphite-ppl.h
+++ b/gcc/graphite-ppl.h
@@ -131,7 +131,8 @@ value_max (mpz_t res, mpz_t v1, mpz_t v2)
{
if (mpz_cmp (v1, v2) < 0)
mpz_set (res, v2);
- mpz_set (res, v1);
+ else
+mpz_set (res, v1);
}
/* Builds a new identity map for dimension DIM. */
--
1.7.4.1
[PATCH 0/6] Fix PR47654
Hi, the following patch set fixes PR47654: Correct typo. Correct computation of max. Fix PR47654: Loop blocking should strip-mine at least two loops. Fix computation of precision. Compute the type of the IV based only on the CLAST bounds. Fix PR47654: Compute LB and UB of a CLAST expression. First, "Loop blocking should strip-mine at least two loops" disables loop blocking when the strip mine is not applied to at least two loops. In the testcase of this PR we have the first loop that does not contain enough iterations to be strip mined: for (i = 0; i < 40; i++) for (j = 0; j < 128; j++) a[j][i] = 4; The second interesting patch "Fix computation of precision" uses the mpz_sizeinbase function instead of computing the log of the gmp value. "Compute the type of the IV based only on the CLAST bounds" removes the computation of the type of the induction variable based on the polyhedral representation: this is a redundant information at the code generation level, as cloog has already integrated this info in the CLAST. Finally, "Compute LB and UB of a CLAST expression" fixes the root cause of PR47654: it reimplements the type computation based on the low and up bounds of CLAST expressions. This patch set passed bootstrap and test c,c++,fortran on amd64-linux. Full bootstrap and test in progress on amd64-linux. Ok for trunk? Thanks, Sebastian gcc/ChangeLog | 53 +++ gcc/graphite-blocking.c| 59 ++-- gcc/graphite-clast-to-gimple.c | 417 ++-- gcc/graphite-interchange.c | 21 +- gcc/graphite-poly.h|4 +- gcc/graphite-ppl.h | 14 +- gcc/testsuite/ChangeLog| 10 + gcc/testsuite/gcc.dg/graphite/block-pr47654.c | 25 ++ gcc/testsuite/gcc.dg/graphite/run-id-pr47654.c | 24 ++ 9 files changed, 343 insertions(+), 284 deletions(-) create mode 100644 gcc/testsuite/gcc.dg/graphite/block-pr47654.c create mode 100644 gcc/testsuite/gcc.dg/graphite/run-id-pr47654.c -- 1.7.4.1
[PATCH 5/6] Compute the type of the IV based only on the CLAST bounds.
2011-06-29 Sebastian Pop
* graphite-clast-to-gimple.c (compute_bounds_for_level): Removed.
(compute_type_for_level): Removed.
(clast_get_body_of_loop): Removed.
(gcc_type_for_iv_of_clast_loop): Removed.
(graphite_create_new_loop): Use max_precision_type. Compute the type
of the IV based only on the CLAST bounds.
(translate_clast_for_loop): Do not pass level to
graphite_create_new_loop.
---
gcc/ChangeLog | 11 +
gcc/graphite-clast-to-gimple.c | 95 +--
2 files changed, 14 insertions(+), 92 deletions(-)
diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 0616b10..3117f23 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,5 +1,16 @@
2011-06-29 Sebastian Pop
+ * graphite-clast-to-gimple.c (compute_bounds_for_level): Removed.
+ (compute_type_for_level): Removed.
+ (clast_get_body_of_loop): Removed.
+ (gcc_type_for_iv_of_clast_loop): Removed.
+ (graphite_create_new_loop): Use max_precision_type. Compute the type
+ of the IV based only on the CLAST bounds.
+ (translate_clast_for_loop): Do not pass level to
+ graphite_create_new_loop.
+
+2011-06-29 Sebastian Pop
+
* graphite-clast-to-gimple.c (precision_for_value): Removed.
(precision_for_interval): Removed.
(gcc_type_for_interval): Use mpz_sizeinbase.
diff --git a/gcc/graphite-clast-to-gimple.c b/gcc/graphite-clast-to-gimple.c
index 70031a0..c8d76c1 100644
--- a/gcc/graphite-clast-to-gimple.c
+++ b/gcc/graphite-clast-to-gimple.c
@@ -603,94 +603,6 @@ graphite_create_new_guard (sese region, edge entry_edge,
return exit_edge;
}
-/* Compute the lower bound LOW and upper bound UP for the induction
- variable at LEVEL for the statement PBB, based on the transformed
- scattering of PBB: T|I|G|Cst, with T the scattering transform, I
- the iteration domain, and G the context parameters. */
-
-static void
-compute_bounds_for_level (poly_bb_p pbb, int level, mpz_t low, mpz_t up)
-{
- ppl_Pointset_Powerset_C_Polyhedron_t ps;
- ppl_Linear_Expression_t le;
-
- combine_context_id_scat (&ps, pbb, false);
-
- /* Prepare the linear expression corresponding to the level that we
- want to maximize/minimize. */
- {
-ppl_dimension_type dim = pbb_nb_scattering_transform (pbb)
- + pbb_dim_iter_domain (pbb) + pbb_nb_params (pbb);
-
-ppl_new_Linear_Expression_with_dimension (&le, dim);
-ppl_set_coef (le, 2 * level + 1, 1);
- }
-
- ppl_max_for_le_pointset (ps, le, up);
- ppl_min_for_le_pointset (ps, le, low);
- ppl_delete_Linear_Expression (le);
- ppl_delete_Pointset_Powerset_C_Polyhedron (ps);
-}
-
-/* Compute the type for the induction variable at LEVEL for the
- statement PBB, based on the transformed schedule of PBB. */
-
-static tree
-compute_type_for_level (poly_bb_p pbb, int level)
-{
- mpz_t low, up;
- tree type;
-
- mpz_init (low);
- mpz_init (up);
-
- compute_bounds_for_level (pbb, level, low, up);
- type = gcc_type_for_interval (low, up);
-
- mpz_clear (low);
- mpz_clear (up);
- return type;
-}
-
-/* Walks a CLAST and returns the first statement in the body of a
- loop. */
-
-static struct clast_user_stmt *
-clast_get_body_of_loop (struct clast_stmt *stmt)
-{
- if (!stmt
- || CLAST_STMT_IS_A (stmt, stmt_user))
-return (struct clast_user_stmt *) stmt;
-
- if (CLAST_STMT_IS_A (stmt, stmt_for))
-return clast_get_body_of_loop (((struct clast_for *) stmt)->body);
-
- if (CLAST_STMT_IS_A (stmt, stmt_guard))
-return clast_get_body_of_loop (((struct clast_guard *) stmt)->then);
-
- if (CLAST_STMT_IS_A (stmt, stmt_block))
-return clast_get_body_of_loop (((struct clast_block *) stmt)->body);
-
- gcc_unreachable ();
-}
-
-/* Returns the type for the induction variable for the loop translated
- from STMT_FOR. */
-
-static tree
-gcc_type_for_iv_of_clast_loop (struct clast_for *stmt_for, int level,
- tree lb_type, tree ub_type)
-{
- struct clast_stmt *stmt = (struct clast_stmt *) stmt_for;
- struct clast_user_stmt *body = clast_get_body_of_loop (stmt);
- CloogStatement *cs = body->statement;
- poly_bb_p pbb = (poly_bb_p) cloog_statement_usr (cs);
-
- return max_signed_precision_type (lb_type, max_precision_type
- (ub_type, compute_type_for_level
-(pbb, level - 1)));
-}
-
/* Creates a new LOOP corresponding to Cloog's STMT. Inserts an
induction variable for the new LOOP. New LOOP is attached to CFG
starting at ENTRY_EDGE. LOOP is inserted into the loop tree and
@@ -703,13 +615,13 @@ static struct loop *
graphite_create_new_loop (sese region, edge entry_edge,
struct clast_for *stmt,
loop_p outer, VEC (tree, heap) **newivs,
- htab_t newivs_index, htab_t params_index, int level)
+ htab_t
[PATCH 3/6] Fix PR47654: Loop blocking should strip-mine at least two loops.
2011-06-29 Sebastian Pop
PR tree-optimization/47654
* graphite-blocking.c (pbb_strip_mine_time_depth): Do not return bool.
(lst_do_strip_mine_loop): Return an int.
(lst_do_strip_mine): Same.
(scop_do_strip_mine): Same.
(scop_do_block): Loop blocking should strip-mine at least two loops.
* graphite-interchange.c (lst_interchange_select_outer): Return an int.
(scop_do_interchange): Same.
* graphite-poly.h (scop_do_interchange): Update declaration.
(scop_do_strip_mine): Same.
* gcc.dg/graphite/block-pr47654.c: New.
---
gcc/ChangeLog | 13 +
gcc/graphite-blocking.c | 59 +++--
gcc/graphite-interchange.c| 21 +
gcc/graphite-poly.h |4 +-
gcc/testsuite/ChangeLog |5 ++
gcc/testsuite/gcc.dg/graphite/block-pr47654.c | 25 ++
6 files changed, 82 insertions(+), 45 deletions(-)
create mode 100644 gcc/testsuite/gcc.dg/graphite/block-pr47654.c
diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index b0e3173..828559a 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,5 +1,18 @@
2011-06-29 Sebastian Pop
+ PR tree-optimization/47654
+ * graphite-blocking.c (pbb_strip_mine_time_depth): Do not return bool.
+ (lst_do_strip_mine_loop): Return an int.
+ (lst_do_strip_mine): Same.
+ (scop_do_strip_mine): Same.
+ (scop_do_block): Loop blocking should strip-mine at least two loops.
+ * graphite-interchange.c (lst_interchange_select_outer): Return an int.
+ (scop_do_interchange): Same.
+ * graphite-poly.h (scop_do_interchange): Update declaration.
+ (scop_do_strip_mine): Same.
+
+2011-06-29 Sebastian Pop
+
* graphite-ppl.h (value_max): Correct computation of max.
2011-06-29 Sebastian Pop
diff --git a/gcc/graphite-blocking.c b/gcc/graphite-blocking.c
index bcd077a..967de9d 100644
--- a/gcc/graphite-blocking.c
+++ b/gcc/graphite-blocking.c
@@ -89,7 +89,7 @@ along with GCC; see the file COPYING3. If not see
# }
*/
-static bool
+static void
pbb_strip_mine_time_depth (poly_bb_p pbb, int time_depth, int stride)
{
ppl_dimension_type iter, dim, strip;
@@ -151,8 +151,6 @@ pbb_strip_mine_time_depth (poly_bb_p pbb, int time_depth,
int stride)
ppl_Polyhedron_add_constraint (res, new_cstr);
ppl_delete_Constraint (new_cstr);
}
-
- return true;
}
/* Returns true when strip mining with STRIDE of the loop LST is
@@ -177,10 +175,10 @@ lst_strip_mine_profitable_p (lst_p lst, int stride)
return res;
}
-/* Strip-mines all the loops of LST with STRIDE. Return true if it
- did strip-mined some loops. */
+/* Strip-mines all the loops of LST with STRIDE. Return the number of
+ loops strip-mined. */
-static bool
+static int
lst_do_strip_mine_loop (lst_p lst, int depth, int stride)
{
int i;
@@ -188,26 +186,26 @@ lst_do_strip_mine_loop (lst_p lst, int depth, int stride)
poly_bb_p pbb;
if (!lst)
-return false;
+return 0;
if (LST_LOOP_P (lst))
{
- bool res = false;
+ int res = 0;
FOR_EACH_VEC_ELT (lst_p, LST_SEQ (lst), i, l)
- res |= lst_do_strip_mine_loop (l, depth, stride);
+ res += lst_do_strip_mine_loop (l, depth, stride);
return res;
}
pbb = LST_PBB (lst);
- return pbb_strip_mine_time_depth (pbb, psct_dynamic_dim (pbb, depth),
- stride);
+ pbb_strip_mine_time_depth (pbb, psct_dynamic_dim (pbb, depth), stride);
+ return 1;
}
/* Strip-mines all the loops of LST with STRIDE. When STRIDE is zero,
- read the stride from the PARAM_LOOP_BLOCK_TILE_SIZE. Return true
- if it did strip-mined some loops.
+ read the stride from the PARAM_LOOP_BLOCK_TILE_SIZE. Return the
+ number of strip-mined loops.
Strip mining transforms a loop
@@ -221,12 +219,12 @@ lst_do_strip_mine_loop (lst_p lst, int depth, int stride)
| S (i = k + j);
*/
-static bool
+static int
lst_do_strip_mine (lst_p lst, int stride)
{
int i;
lst_p l;
- bool res = false;
+ int res = 0;
int depth;
if (!stride)
@@ -237,23 +235,23 @@ lst_do_strip_mine (lst_p lst, int stride)
return false;
FOR_EACH_VEC_ELT (lst_p, LST_SEQ (lst), i, l)
-res |= lst_do_strip_mine (l, stride);
+res += lst_do_strip_mine (l, stride);
depth = lst_depth (lst);
if (depth >= 0
&& lst_strip_mine_profitable_p (lst, stride))
{
- res |= lst_do_strip_mine_loop (lst, lst_depth (lst), stride);
+ res += lst_do_strip_mine_loop (lst, lst_depth (lst), stride);
lst_add_loop_under_loop (lst);
}
return res;
}
-/* Strip mines all the loops in SCOP. Returns true when some loops
- have been strip-mined. */
+/* Strip mines all the loops in SCOP. Returns the number of
+ strip-mined loops. */
-bool
+int
scop_do_strip_min
[PATCH 6/6] Fix PR47654: Compute LB and UB of a CLAST expression.
2011-06-29 Sebastian Pop
PR tree-optimization/47654
* graphite-clast-to-gimple.c (gcc_type_for_value): Removed.
(gcc_type_for_clast_term): Removed.
(gcc_type_for_clast_red): Removed.
(gcc_type_for_clast_bin): Removed.
(lb_ub_for_expr_name): New.
(lb_ub_for_term): New.
(lb_ub_for_expr): New.
(lb_ub_for_red): New.
(lb_ub_for_bin): New.
(gcc_type_for_clast_expr): Reimplemented.
* graphite-ppl.h (value_min): New.
* gcc.dg/graphite/run-id-pr47654.c: New.
---
gcc/ChangeLog | 15 ++
gcc/graphite-clast-to-gimple.c | 281
gcc/graphite-ppl.h | 11 +
gcc/testsuite/ChangeLog|5 +
gcc/testsuite/gcc.dg/graphite/run-id-pr47654.c | 24 ++
5 files changed, 240 insertions(+), 96 deletions(-)
create mode 100644 gcc/testsuite/gcc.dg/graphite/run-id-pr47654.c
diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 3117f23..f69f7f8 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,5 +1,20 @@
2011-06-29 Sebastian Pop
+ PR tree-optimization/47654
+ * graphite-clast-to-gimple.c (gcc_type_for_value): Removed.
+ (gcc_type_for_clast_term): Removed.
+ (gcc_type_for_clast_red): Removed.
+ (gcc_type_for_clast_bin): Removed.
+ (lb_ub_for_expr_name): New.
+ (lb_ub_for_term): New.
+ (lb_ub_for_expr): New.
+ (lb_ub_for_red): New.
+ (lb_ub_for_bin): New.
+ (gcc_type_for_clast_expr): Reimplemented.
+ * graphite-ppl.h (value_min): New.
+
+2011-06-29 Sebastian Pop
+
* graphite-clast-to-gimple.c (compute_bounds_for_level): Removed.
(compute_type_for_level): Removed.
(clast_get_body_of_loop): Removed.
diff --git a/gcc/graphite-clast-to-gimple.c b/gcc/graphite-clast-to-gimple.c
index c8d76c1..686c921 100644
--- a/gcc/graphite-clast-to-gimple.c
+++ b/gcc/graphite-clast-to-gimple.c
@@ -379,147 +379,236 @@ clast_to_gcc_expression (tree type, struct clast_expr
*e,
return NULL_TREE;
}
-/* Return a type that could represent the values between LOW and UP.
- The value of LOW can be bigger than UP. */
+/* Return the lower bound LB and upper bound UB of the clast_name N. */
-static tree
-gcc_type_for_interval (mpz_t low, mpz_t up)
+static void
+lb_ub_for_name (clast_name_p n, sese region, VEC (tree, heap) *newivs,
+ htab_t newivs_index, htab_t params_index, mpz_t lb, mpz_t ub)
{
- bool unsigned_p;
- tree type;
- enum machine_mode mode;
- int precision = MAX (mpz_sizeinbase (low, 2),
- mpz_sizeinbase (up, 2));
-
- if (precision > BITS_PER_WORD)
-{
- gloog_error = true;
- return integer_type_node;
-}
+ tree l, u;
+ tree type = TREE_TYPE (clast_name_to_gcc (n, region, newivs,
+ newivs_index, params_index));
- if (mpz_cmp (low, up) <= 0)
-unsigned_p = (mpz_sgn (low) >= 0);
+ if (POINTER_TYPE_P (type) || !TYPE_MIN_VALUE (type))
+l = lower_bound_in_type (type, type);
else
-unsigned_p = (mpz_sgn (up) >= 0);
+l = TYPE_MIN_VALUE (type);
- mode = smallest_mode_for_size (precision, MODE_INT);
- precision = GET_MODE_PRECISION (mode);
- type = build_nonstandard_integer_type (precision, unsigned_p);
-
- if (!type)
-{
- gloog_error = true;
- return integer_type_node;
-}
+ if (POINTER_TYPE_P (type) || !TYPE_MAX_VALUE (type))
+u = upper_bound_in_type (type, type);
+ else
+u = TYPE_MAX_VALUE (type);
- return type;
+ tree_int_to_gmp (l, lb);
+ tree_int_to_gmp (u, ub);
}
-/* Return a type that could represent the integer value VAL, or
- otherwise return NULL_TREE. */
-
-static tree
-gcc_type_for_value (mpz_t val)
-{
- return gcc_type_for_interval (val, val);
-}
+/* Return the lower bound LB and upper bound UB of the clast_term T. */
-/* Return the type for the clast_term T used in STMT. */
-
-static tree
-gcc_type_for_clast_term (struct clast_term *t,
-sese region, VEC (tree, heap) *newivs,
-htab_t newivs_index, htab_t params_index)
+static void
+lb_ub_for_term (struct clast_term *t, sese region,
+ VEC (tree, heap) *newivs, htab_t newivs_index,
+ htab_t params_index, mpz_t lb, mpz_t ub)
{
gcc_assert (t->expr.type == clast_expr_term);
- if (!t->var)
-return gcc_type_for_value (t->val);
-
- return TREE_TYPE (clast_name_to_gcc (t->var, region, newivs,
- newivs_index, params_index));
+ if (t->var)
+{
+ mpz_t v;
+ lb_ub_for_name ((clast_name_p) (t->var), region, newivs, newivs_index,
+ params_index, lb, ub);
+ mpz_init (v);
+ mpz_abs (v, t->val);
+ mpz_mul (lb, lb, v);
+ mpz_mul (ub, ub, v);
+ mpz_clear (v);
+}
+ else
+{
+ mpz_set (lb, t->val);
+ mpz_se
Re: [cxx-mem-model] __sync_mem_load
On 06/23/2011 03:50 PM, Andrew MacLeod wrote:
> + (define_expand "sync_mem_load"
> + [(match_operand:SWI 0 "register_operand" "") ;; output
> +(match_operand:SWI 1 "memory_operand" "");; memory
> +(match_operand:SI 2 "const_int_operand" "")];; memory model
> +""
> + {
> + if (INTVAL (operands[2]) == MEMMODEL_ACQUIRE ||
> + INTVAL (operands[2]) == MEMMODEL_SEQ_CST)
Oh, and I suspect all of these will be easier to debug with
enum memmodel mm = (enum memmodel) INTVAL (operands[2]);
if (mm == ...)
r~
[PATCH 4/6] Fix computation of precision.
2011-06-29 Sebastian Pop
* graphite-clast-to-gimple.c (precision_for_value): Removed.
(precision_for_interval): Removed.
(gcc_type_for_interval): Use mpz_sizeinbase.
---
gcc/ChangeLog |6 +++
gcc/graphite-clast-to-gimple.c | 77 +---
2 files changed, 15 insertions(+), 68 deletions(-)
diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 828559a..0616b10 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,5 +1,11 @@
2011-06-29 Sebastian Pop
+ * graphite-clast-to-gimple.c (precision_for_value): Removed.
+ (precision_for_interval): Removed.
+ (gcc_type_for_interval): Use mpz_sizeinbase.
+
+2011-06-29 Sebastian Pop
+
PR tree-optimization/47654
* graphite-blocking.c (pbb_strip_mine_time_depth): Do not return bool.
(lst_do_strip_mine_loop): Return an int.
diff --git a/gcc/graphite-clast-to-gimple.c b/gcc/graphite-clast-to-gimple.c
index 4a4c3d2..70031a0 100644
--- a/gcc/graphite-clast-to-gimple.c
+++ b/gcc/graphite-clast-to-gimple.c
@@ -379,72 +379,17 @@ clast_to_gcc_expression (tree type, struct clast_expr *e,
return NULL_TREE;
}
-/* Return the precision needed to represent the value VAL. */
-
-static int
-precision_for_value (mpz_t val)
-{
- mpz_t x, y, two;
- int precision;
-
- mpz_init (x);
- mpz_init (y);
- mpz_init (two);
- mpz_set_si (x, 2);
- mpz_set (y, val);
- mpz_set_si (two, 2);
- precision = 1;
-
- if (mpz_sgn (y) < 0)
-mpz_neg (y, y);
-
- while (mpz_cmp (y, x) >= 0)
-{
- mpz_mul (x, x, two);
- precision++;
-}
-
- mpz_clear (x);
- mpz_clear (y);
- mpz_clear (two);
-
- return precision;
-}
-
-/* Return the precision needed to represent the values between LOW and
- UP. */
-
-static int
-precision_for_interval (mpz_t low, mpz_t up)
-{
- mpz_t diff;
- int precision;
-
- gcc_assert (mpz_cmp (low, up) <= 0);
-
- mpz_init (diff);
- mpz_sub (diff, up, low);
- precision = precision_for_value (diff);
- mpz_clear (diff);
-
- return precision;
-}
-
-/* Return a type that could represent the integer value VAL. */
+/* Return a type that could represent the values between LOW and UP.
+ The value of LOW can be bigger than UP. */
static tree
gcc_type_for_interval (mpz_t low, mpz_t up)
{
- bool unsigned_p = true;
- int precision, prec_up, prec_int;
+ bool unsigned_p;
tree type;
enum machine_mode mode;
-
- gcc_assert (mpz_cmp (low, up) <= 0);
-
- prec_up = precision_for_value (up);
- prec_int = precision_for_interval (low, up);
- precision = MAX (prec_up, prec_int);
+ int precision = MAX (mpz_sizeinbase (low, 2),
+ mpz_sizeinbase (up, 2));
if (precision > BITS_PER_WORD)
{
@@ -452,14 +397,10 @@ gcc_type_for_interval (mpz_t low, mpz_t up)
return integer_type_node;
}
- if (mpz_sgn (low) <= 0)
-unsigned_p = false;
-
- else if (precision < BITS_PER_WORD)
-{
- unsigned_p = false;
- precision++;
-}
+ if (mpz_cmp (low, up) <= 0)
+unsigned_p = (mpz_sgn (low) >= 0);
+ else
+unsigned_p = (mpz_sgn (up) >= 0);
mode = smallest_mode_for_size (precision, MODE_INT);
precision = GET_MODE_PRECISION (mode);
--
1.7.4.1
Re: [testsuite, ada] Cope with HP-UX sh in run_acats (PR ada/49511)
> As described in the PR, HP-UX /bin/sh type -p ignores the -p, so we need > to extract the last field of the output just as for regular type. The > following patch does this, tested by Dave and bootstrapped on > sparc-sun-solaris2.11. > > Ok for mainline, 4.6 and 4.5 branches? OK
[Patch] Fix objc/48109
The bug arises because of the use, by the ObjC FE, of two old target
macros that emit efficient representations of class definitions and
references.
This 'works fine' (however wrong it might be conceptually), until LTO
is engaged, whereupon the definitions vanish without trace (since no
corresponding real variable is ever created).
---
The patch creates appropriate variables in the FE and tags them as
ObjC meta-data (in the same manner as is done for other objc meta-data).
We then intercept them in varasm.c with a hook that allows the target
to declare that it has completely handled the output of a variable -
allowing us to handle them in the required special manner.
FAOD, It is necessary to preserve this mechanism for emitting the
definitions and references to permit linkage with existing system
libraries.
If the patch is acceptable, then I would expect to follow up with a
patch to remove ASM_DECLARE_CLASS_REFERENCE and
ASM_DECLARE_UNRESOLVED_REFERENCE from the tree - since this is their
only use.
bootstrapped on i686-linux, i686-darwin9, x86_64-darwin10, (checked
to do the Right Thing on darwin).
Mike has already given this a 'seems reasonable' in the PR thread,
however, I need an approver for the varasm.c and target hook changes.
OK for trunk & 4.6?
Iain
===
gcc/
* target.def (handled_assemble_variable_p): New hook.
* varasm.c (assemble_variable): Allow target to handle variable output
in some special manner.
* doc/tm.texi: Regenerate.
* config/darwin.c (darwin_objc1_section): Handle class defs/refs.
(darwin_handled_assemble_variable_p): New.
* config/darwin-protos.h (darwin_handled_assemble_variable_p): New.
* config/darwin.h (TARGET_ASM_HANDLED_ASSEMBLE_VARIABLE_P): New.
gcc/objc/
*objc-next-runtime-abi-01.c (handle_next_class_ref): Don't emit lazy
refs. for
cases where the class is local. Declare a real, meta-data item
tagged as
a class reference.
(handle_next_impent): Declare a real, meta-data item tagged as a
class def.
===
Index: gcc/target.def
===
--- gcc/target.def (revision 175628)
+++ gcc/target.def (working copy)
@@ -449,6 +449,13 @@ DEFHOOK
bool, (FILE *file, rtx x),
default_asm_output_addr_const_extra)
+DEFHOOK
+(handled_assemble_variable_p,
+ "Returns @code{true} iff the target has handled the assembly of the\
+ variable @var{var_decl}",
+ bool, (tree var_decl),
+ hook_bool_tree_false)
+
/* ??? The TARGET_PRINT_OPERAND* hooks are part of the asm_out struct,
even though that is not reflected in the macro name to override
their
initializers. */
Index: gcc/objc/objc-next-runtime-abi-01.c
===
--- gcc/objc/objc-next-runtime-abi-01.c (revision 175628)
+++ gcc/objc/objc-next-runtime-abi-01.c (working copy)
@@ -2267,27 +2267,51 @@ generate_objc_symtab_decl (void)
init_objc_symtab (TREE_TYPE (UOBJC_SYMBOLS_decl)));
}
-
static void
handle_next_class_ref (tree chain)
{
+ tree decl, exp;
+ struct imp_entry *impent;
const char *name = IDENTIFIER_POINTER (TREE_VALUE (chain));
char *string = (char *) alloca (strlen (name) + 30);
+ for (impent = imp_list; impent; impent = impent->next)
+if (TREE_CODE (impent->imp_context) == CLASS_IMPLEMENTATION_TYPE
+&& IDENTIFIER_POINTER (CLASS_NAME (impent->imp_context))
+ == IDENTIFIER_POINTER (TREE_VALUE (chain)))
+ return; /* we declare this, no need for a lazy ref. */
+
sprintf (string, ".objc_class_name_%s", name);
-#ifdef ASM_DECLARE_UNRESOLVED_REFERENCE
- ASM_DECLARE_UNRESOLVED_REFERENCE (asm_out_file, string);
-#else
- return ; /* NULL build for targets other than Darwin. */
-#endif
+ decl = build_decl (UNKNOWN_LOCATION,
+VAR_DECL, get_identifier (string), char_type_node);
+ TREE_PUBLIC (decl) = 1;
+ DECL_EXTERNAL (decl) = 1;
+ DECL_CONTEXT (decl) = NULL_TREE;
+ finish_var_decl (decl, NULL);
+
+ /* We build a variable to signal the reference. This will be
intercepted
+ and output as a lazy reference. */
+ sprintf (string, "_OBJC_class_ref_%s", name);
+ exp = build1 (ADDR_EXPR, string_type_node, decl);
+ decl = build_decl (input_location,
+VAR_DECL, get_identifier (string), string_type_node);
+ TREE_STATIC (decl) = 1;
+ DECL_ARTIFICIAL (decl) = 1;
+ DECL_INITIAL (decl) = error_mark_node;
+
+ /* We must force the reference. */
+ DECL_PRESERVE_P (decl) = 1;
+ OBJCMETA (decl, objc_meta, get_identifier ("V1_CREF"));
+ DECL_CONTEXT (decl) = NULL_TREE;
+ finish_var_decl (decl, exp);
}
static void
handle_next_impent (struct imp_entry *impent)
{
char buf[BUFSIZE];
-
+ tree decl;
switch (TREE_CODE (impent->imp_context))
{
case CLASS_IMPLEMENTATION_TYPE:
@@ -2303,11 +2327,16 @@ handle_next_impent (struct imp_entry *
Re: [testsuite] skip ARM tests if no thumb2 support
On 06/29/2011 06:55 AM, Richard Earnshaw wrote:
> 3. Add a new dg directive (perhaps dg-ignore-multilib) that instructs
> the framework to ignore the multilib options entirely (only supported
> for compile/assembly tests).
>
> There are a lot of target-specific tests that are not really testing
> execution, just that the compiler generates the right instruction under
> specific compilation options. It seems silly to force all this into the
> multi-lib framework.
If it's true that most testsuite runs on ARM include a default multilib,
then some tests can be skipped for everything other than the default
multilib and specify all the flags that are needed for the test, e.g.
with
/* { dg-require-effective-target default_multilib } */
That's much simpler than ignoring multilib flags, and avoids running
the same test multiple times with exactly the same options.
Janis
Re: [Design notes, RFC] Address-lowering prototype design (PR46556)
On Tue, 2011-06-14 at 15:39 +0200, Richard Guenther wrote:
> On Fri, Jun 10, 2011 at 5:11 PM, William J. Schmidt
> wrote:
> > On Tue, 2011-06-07 at 16:49 +0200, Richard Guenther wrote:
> >> On Tue, Jun 7, 2011 at 4:14 PM, William J. Schmidt
> >> wrote:
> >
> >
> >
> >> >> > Loss of aliasing information
> >> >> >
> >> >> > The most serious problem I've run into is degraded performance due to
> >> >> > poorer
> >> >> > instruction scheduling choices. I tracked this down to
> >> >> > alias.c:nonoverlapping_component_refs_p.
> >> >> >
> >> >> > This code proves that two memory accesses don't overlap by attempting
> >> >> > to prove
> >> >> > that they access different fields of the same structure. This is done
> >> >> > using
> >> >> > the MEM_EXPRs of the two rtx's, which record the expression trees
> >> >> > that were
> >> >> > translated into the rtx's during expand. When address lowering is not
> >> >> > present, a simple COMPONENT_REF will appear in the MEM_EXPR: x.a, for
> >> >> > example. However, address lowering changes the simple COMPONENT_REF
> >> >> > into a
> >> >> > [TARGET_]MEM_REF that is no longer necessarily identifiable as a field
> >> >> > reference. Thus the aliasing machinery can no longer prove that two
> >> >> > such
> >> >> > field references are disjoint.
> >> >> >
> >> >> > This has severe consequences for performance, and has to be dealt
> >> >> > with if
> >> >> > address lowering is to be successful.
> >> >> >
> >> >> > I've worked around this with an admittedly fragile solution; I'll
> >> >> > discuss the
> >> >> > drawbacks below. The idea is to construct a mapping from replacement
> >> >> > mem_refs
> >> >> > to the original expressions that they replaced. When a MEM_EXPR is
> >> >> > being set
> >> >> > during expand, we first look up the mem_ref in the mapping. If
> >> >> > present, the
> >> >> > MEM_EXPR is set to the original expression, rather than to the
> >> >> > mem_ref. This
> >> >> > essentially duplicates the behavior in the absence of address
> >> >> > lowering.
> >> >>
> >> >> Ick. We had this in the past via TMR_ORIGINAL which caused all sorts
> >> >> of problems. Removing it didn't cause much degradation because we now
> >> >> preserve points-to information.
> >> >>
> >> >> Originally I played with lowering all memory accesses to MEM_REFs
> >> >> (see the old mem-ref branch), and the loss of type-based alias
> >> >> disambiguation was indeed an issue.
> >> >>
> >> >> But - I definitely do not like the idea of preserving something similar
> >> >> to TMR_ORIGINAL. Instead we can try preserving some information
> >> >> we derive from it. We keep the original access type that we can use
> >> >> for TBAA but do not retain knowledge on whether the type of the
> >> >> MEM_REF is valid for TBAA or if it is view-converted.
> >> >
> >> > Yes, I really don't like what I have at the moment, either. I put it in
> >> > place as a stopgap to let me proceed to look for other performance
> >> > problems.
> >> >
> >> > The question is how we can infer useful information for TBAA from the
> >> > MEM_REFs and TMRs. I poked at trying to identify types and offsets from
> >> > the MEM_EXPRs, but this ended up being useless; I had to constrain too
> >> > many cases to maintain correctness, and couldn't prove the type
> >> > information for the important cases in SPEC I was trying to address.
> >> >
> >> > Unfortunately, the whole design goes down the drain if we can't find a
> >> > way to solve the TBAA issue. The performance degradations are too
> >> > costly.
> >>
> >> If you look at what basic TBAA the alias oracle performs then it boils
> >> down to the fact that get_alias_set for a.b.c might end up using the
> >> alias-set of the type of C but for MEM[&a + 4] it will use the alias set
> >> of the type of a. The tree alias-oracle extracts both alias sets, that
> >> of the outermost valid type and that of the innermost as both are
> >> equally useful. But the MEM_REF (or TARGET_MEM_REF) tree
> >> only have storage for one such alias-set. Thus my idea at some point
> >> was to store the other one as well in some form. It will not be
> >> the full information (after all, the complete access path does provide
> >> some extra information - see aliasing_component_refs_p).
> >
> > This is what concerns me. TBAA information for the outer and inner
> > components doesn't seem sufficient to provide what
> > nonoverlapping_component_refs_p is currently able to prove. The latter
> > searches for a common RECORD_TYPE somewhere along the two access paths,
> > and then disambiguates if the two associated referenced fields differ.
> > For a simple case like "struct x { int a; int b; };", a and b have the
> > same type and alias-set, so the alias-set information doesn't add
> > anything. It isn't sufficient alone for the disambiguation of x1.a =
> > MEM_REF[&x1, 0] and x2.b = MEM_REF[&x2, 4].
> >
> > Obviously the offset is sufficient to disambiguate for
Re: [PATCH, MELT] correct meltgc_read_from_val without location
Hello,
here is an improvment to
http://gcc.gnu.org/ml/gcc-patches/2011-06/msg01888.html.
The function meltgc_read_from_val (in melt-runtime.c) takes two
arguments, a string value and a second one which is a location. In the
comments, it is written that we can pass a NULL pointer if we have no
location (it is a direct string). However, this conduct MELT to crash
because it doesn't handle correctly the absence of file.
In the first patch, I modified makesexpr to create a location with a
'virtual' file. This works however, as this function is used recursively
this is not very elegant.
This patch is more elegant, it adds a boolean field to the struct
reading_st to declare if there is no file location.
meltgc_read_* functions are modified to test if there is a given
location, and if no, it create a location with a virtual file and set
the boolean to false, conducting to avoid crash.
I have been able to build MELT with this pass and test it without problem.
On 24/06/2011 18:13, Pierre Vittet wrote:
Hello,
The function meltgc_read_from_val (in melt-runtime.c) takes two
arguments, a string value and a second one which is a location.
In the comments, it is written that we can pass a NULL pointer if we
have no location (it is a direct string). However, this conduct MELT to
crash because it doesn't handle correctly the absence of file.
This patch correct this, if there is no file, it create a "virtual" one
which is named "stringBuffer".
Pierre Vittet
correct_read_from_val_without_location-175348.diff
Index: gcc/melt-runtime.c
===
--- gcc/melt-runtime.c (revision 175348)
+++ gcc/melt-runtime.c (working copy)
@@ -6292,6 +6292,7 @@ struct reading_st
int rcol;/* current column */
source_location rsrcloc; /* current source location */
melt_ptr_t *rpfilnam;/* pointer to location of file name string */
+ bool has_file_location; /* precise if the string comes from a file */
};
#define MELT_READ_TABULATION_FACTOR 8
@@ -6326,7 +6327,7 @@ melt_linemap_compute_current_location (struct read
{
int colnum = 1;
int cix = 0;
- if (!rd || !rd->rcurlin)
+ if (!rd || !rd->rcurlin || !rd->has_file_location)
return;
for (cix=0; cixrcol; cix++) {
char c = rd->rcurlin[cix];
@@ -8314,6 +8315,7 @@ meltgc_read_file (const char *filnam, const char *
rd = &rds;
locnamv = meltgc_new_stringdup ((meltobject_ptr_t) MELT_PREDEF
(DISCR_STRING), locnam);
rds.rpfilnam = (melt_ptr_t *) & locnamv;
+ rds.has_file_location = true;
seqv = meltgc_new_list ((meltobject_ptr_t) MELT_PREDEF (DISCR_LIST));
while (!rdeof ())
{
@@ -8371,7 +8373,16 @@ meltgc_read_from_rawstring (const char *rawstr, co
rds.rsrcloc = loch;
rd = &rds;
if (locnam)
-locnamv = meltgc_new_stringdup ((meltobject_ptr_t) MELT_PREDEF
(DISCR_STRING), locnam);
+{
+ rds.has_file_location = true;
+ locnamv = meltgc_new_stringdup ((meltobject_ptr_t) MELT_PREDEF
(DISCR_STRING), locnam);
+}
+ else
+{
+ rds.has_file_location = false;
+ locnamv = meltgc_new_string ((meltobject_ptr_t)
MELT_PREDEF(DISCR_STRING),
+ "stringBuffer");
+}
seqv = meltgc_new_list ((meltobject_ptr_t) MELT_PREDEF (DISCR_LIST));
rds.rpfilnam = (melt_ptr_t *) & locnamv;
while (rdcurc ())
@@ -8415,6 +8426,7 @@ meltgc_read_from_val (melt_ptr_t strv_p, melt_ptr_
locnamv = locnam_p;
rbuf = 0;
strmagic = melt_magic_discr ((melt_ptr_t) strv);
+ seqv = meltgc_new_list ((meltobject_ptr_t) MELT_PREDEF (DISCR_LIST));
switch (strmagic)
{
case MELTOBMAG_STRING:
@@ -8441,7 +8453,14 @@ meltgc_read_from_val (melt_ptr_t strv_p, melt_ptr_
rds.rpath = 0;
rds.rlineno = 0;
rds.rcurlin = rbuf;
+ rds.has_file_location = true;
rd = &rds;
+ if (locnamv == NULL){
+rds.has_file_location = false;
+locnamv = meltgc_new_string ((meltobject_ptr_t) MELT_PREDEF(DISCR_STRING),
+ "stringBuffer");
+rd->rpfilnam = (melt_ptr_t *) &locnamv;
+ }
rds.rpfilnam = (melt_ptr_t *) & locnamv;
while (rdcurc ())
{
2011-06-29 Pierre Vittet
* melt-runtime.c (struct reading_st): add a boolean has_file_location
field.
(melt_linemap_compute_current_location): return immediately if no file
location.
(meltgc_read_file, meltgc_read_from_rawstring, meltgc_read_from_val):
set has_file_location accordingly
(meltgc_read_from_val): create seqv list (it was used without being
created).
Re: [PATCH, MELT] correct meltgc_read_from_val without location
On Wed, 29 Jun 2011 20:32:08 +0200
Pierre wrote:
> Hello,
>
> here is an improvment to
> http://gcc.gnu.org/ml/gcc-patches/2011-06/msg01888.html.
>
Thanks. I applied it with minor changes on the MELT branch.
{spelling mistakes & indentation mostly}.
Perhaps a future improvement might be to set the location (i.e. locnamv
in meltgc_read_from_val) to something which depends upon the string
rbuf. I have no idea if that would be useful.
Regards.
--
Basile STARYNKEVITCH http://starynkevitch.net/Basile/
email: basilestarynkevitchnet mobile: +33 6 8501 2359
8, rue de la Faiencerie, 92340 Bourg La Reine, France
*** opinions {are only mine, sont seulement les miennes} ***
Re: [gcc patch] Re: C++ member function template id not matching linkage name (PR debug/49408)
On Mon, 27 Jun 2011 20:00:24 +0200, Jason Merrill wrote:
> They should be supressed whenever the function appears in an
> expression context, either as a pointer to member function (i.e. the
> operand of '&')
Done, therefore it is no longer restricted only to templates as before.
> or as the function being called in a call expression.
I implemented it in the patch below but I do not agree + understand it.
The call expression is in libiberty/testsuite/demangle-expected modified by
this patch as:
# decltype/fn call test
--format=gnu-v3
_Z4add3IidEDTclL_Z1gEfp_fp0_EET_T0_
-decltype (g({parm#1}, {parm#2})) add3(int, double)
+decltype (g) add3(int, double)
I agree it is sufficient to determine the return type just from the function
type as return type cannot be overloaded by the function parameters. But it
no longer matches a valid C++ source code now:
char g (int x, double y) { return 0; }
template
decltype (g((T) 0, (U) 0)) add3 (T x, U y) { return 'z'; }
// error: ‘add3’ declared as function returning a function
// decltype (g) add3 (T x, U y) { return 'z'; }
int main () { add3 (1, 2.0); }
g++ -Wall -g -std=c++0x
g++ (GCC) 4.7.0 20110629 (experimental)
(Regression testing underway.)
Thanks,
Jan
libiberty/
2011-06-29 Jan Kratochvil
* cp-demangle.c (d_print_comp): Suppress argument list for function
references by the '&' unary operator. Keep also already processed
variant without the argument list. Suppress argument list also for
function call used in an expression.
* testsuite/demangle-expected: Remove parameters from function call
expressions of 6 testcases. Create 3 new testcases for function
references by the '&' unary operator..
--- a/libiberty/cp-demangle.c
+++ b/libiberty/cp-demangle.c
@@ -4139,7 +4169,46 @@ d_print_comp (struct d_print_info *dpi, int options,
return;
case DEMANGLE_COMPONENT_UNARY:
- if (d_left (dc)->type != DEMANGLE_COMPONENT_CAST)
+ if (d_left (dc)->type == DEMANGLE_COMPONENT_OPERATOR
+ && d_left (dc)->u.s_operator.op->len == 1
+ && d_left (dc)->u.s_operator.op->name[0] == '&'
+ && d_right (dc)->type == DEMANGLE_COMPONENT_TYPED_NAME
+ && d_left (d_right (dc))->type == DEMANGLE_COMPONENT_QUAL_NAME
+ && d_right (d_right (dc))->type == DEMANGLE_COMPONENT_FUNCTION_TYPE)
+ {
+ /* Address of a function (therefore in an expression context) must
+have its argument list suppressed.
+
+unary operator ... dc
+ operator & ... d_left (dc)
+ typed name ... d_right (dc)
+qualified name ... d_left (d_right (dc))
+
+function type ... d_right (d_right (dc))
+ argument list
+ */
+
+ d_print_expr_op (dpi, options, d_left (dc));
+ d_print_comp (dpi, options, d_left (d_right (dc)));
+ return;
+ }
+ else if (d_left (dc)->type == DEMANGLE_COMPONENT_OPERATOR
+ && d_left (dc)->u.s_operator.op->len == 1
+ && d_left (dc)->u.s_operator.op->name[0] == '&'
+ && d_right (dc)->type == DEMANGLE_COMPONENT_QUAL_NAME)
+ {
+ /* Keep also already processed variant without the argument list.
+
+unary operator ... dc
+ operator & ... d_left (dc)
+ qualified name ... d_right (dc)
+ */
+
+ d_print_expr_op (dpi, options, d_left (dc));
+ d_print_comp (dpi, options, d_right (dc));
+ return;
+ }
+ else if (d_left (dc)->type != DEMANGLE_COMPONENT_CAST)
d_print_expr_op (dpi, options, d_left (dc));
else
{
@@ -4172,10 +4241,11 @@ d_print_comp (struct d_print_info *dpi, int options,
d_print_comp (dpi, options, d_right (d_right (dc)));
d_append_char (dpi, ']');
}
- else
+ /* Function call used in an expression should not have the argument list
+printed. */
+ else if (strcmp (d_left (dc)->u.s_operator.op->code, "cl") != 0)
{
- if (strcmp (d_left (dc)->u.s_operator.op->code, "cl") != 0)
- d_print_expr_op (dpi, options, d_left (dc));
+ d_print_expr_op (dpi, options, d_left (dc));
d_print_subexpr (dpi, options, d_right (d_right (dc)));
}
diff --git a/libiberty/testsuite/demangle-expected
b/libiberty/testsuite/demangle-expected
index bbd418c..da87282 100644
--- a/libiberty/testsuite/demangle-expected
+++ b/libiberty/testsuite/demangle-expected
@@ -3904,7 +3904,7 @@ decltype ({parm#1}+{parm#2}) add(int, double)
# decltype/fn call test
--format=gnu-v3
_Z4add3
Re: cleanup patch
Attached patch applied
2011-06-29 François Dumont
* include/debug/set.h, unordered_map, multiset.h, forward_list,
unordered_set, vector, deque, string, list, multimap.h: Remove base
class default constructor calls.
* include/debug/map.h: Likewise and cleanup several redefinition of
base iterator typedef.
Regards
On 06/29/2011 10:34 AM, Jonathan Wakely wrote:
On 28 June 2011 21:36, François Dumont wrote:
Here is a small patch to cleanup debug code.
2011-06-28 François Dumont
* include/debug/set.h, unordered_map, multiset.h, forward_list,
unordered_set, vector, deque, string, list, multimap.h: Remove base
class default constructor calls.
* include/debug/map.h: Likewise and cleanup several redefinition of
base iterator typedef.
Tested under linux x86_64.
(Remember to CC gcc-patches for all patches)
This is OK to check in, thanks.
Index: include/debug/set.h
===
--- include/debug/set.h (revision 175553)
+++ include/debug/set.h (working copy)
@@ -46,7 +46,6 @@
public __gnu_debug::_Safe_sequence >
{
typedef _GLIBCXX_STD_C::set<_Key, _Compare, _Allocator> _Base;
- typedef __gnu_debug::_Safe_sequence _Safe_base;
typedef typename _Base::const_iterator _Base_const_iterator;
typedef typename _Base::iterator _Base_iterator;
@@ -88,21 +87,21 @@
__comp, __a) { }
set(const set& __x)
- : _Base(__x), _Safe_base() { }
+ : _Base(__x) { }
set(const _Base& __x)
- : _Base(__x), _Safe_base() { }
+ : _Base(__x) { }
#ifdef __GXX_EXPERIMENTAL_CXX0X__
set(set&& __x)
noexcept(is_nothrow_copy_constructible<_Compare>::value)
- : _Base(std::move(__x)), _Safe_base()
+ : _Base(std::move(__x))
{ this->_M_swap(__x); }
set(initializer_list __l,
const _Compare& __comp = _Compare(),
const allocator_type& __a = allocator_type())
- : _Base(__l, __comp, __a), _Safe_base() { }
+ : _Base(__l, __comp, __a) { }
#endif
~set() _GLIBCXX_NOEXCEPT { }
@@ -206,7 +205,6 @@
std::pair
insert(value_type&& __x)
{
- typedef typename _Base::iterator _Base_iterator;
std::pair<_Base_iterator, bool> __res
= _Base::insert(std::move(__x));
return std::pair(iterator(__res.first, this),
Index: include/debug/unordered_map
===
--- include/debug/unordered_map (revision 175553)
+++ include/debug/unordered_map (working copy)
@@ -89,25 +89,25 @@
: _Base(__gnu_debug::__base(__gnu_debug::__check_valid_range(__first,
__last)),
__gnu_debug::__base(__last), __n,
- __hf, __eql, __a), _Safe_base() { }
+ __hf, __eql, __a) { }
unordered_map(const unordered_map& __x)
- : _Base(__x), _Safe_base() { }
+ : _Base(__x) { }
unordered_map(const _Base& __x)
- : _Base(__x), _Safe_base() { }
+ : _Base(__x) { }
unordered_map(unordered_map&& __x)
noexcept(__and_,
is_nothrow_copy_constructible<_Pred>>::value)
- : _Base(std::move(__x)), _Safe_base() { }
+ : _Base(std::move(__x)) { }
unordered_map(initializer_list __l,
size_type __n = 0,
const hasher& __hf = hasher(),
const key_equal& __eql = key_equal(),
const allocator_type& __a = allocator_type())
- : _Base(__l, __n, __hf, __eql, __a), _Safe_base() { }
+ : _Base(__l, __n, __hf, __eql, __a) { }
~unordered_map() noexcept { }
@@ -381,25 +381,25 @@
: _Base(__gnu_debug::__base(__gnu_debug::__check_valid_range(__first,
__last)),
__gnu_debug::__base(__last), __n,
- __hf, __eql, __a), _Safe_base() { }
+ __hf, __eql, __a) { }
unordered_multimap(const unordered_multimap& __x)
- : _Base(__x), _Safe_base() { }
+ : _Base(__x) { }
unordered_multimap(const _Base& __x)
- : _Base(__x), _Safe_base() { }
+ : _Base(__x) { }
unordered_multimap(unordered_multimap&& __x)
noexcept(__and_,
is_nothrow_copy_constructible<_Pred>>::value)
- : _Base(std::move(__x)), _Safe_base() { }
+ : _Base(std::move(__x)) { }
unordered_multimap(initializer_list __l,
size_type __n = 0,
const hasher& __hf = hasher(),
const key_equal& __eql = key_equal(),
const allocator_type& __a = allocator_type())
- : _Base(__l, __n, __hf, __eql, __a), _Safe_base() { }
+ : _Base(__l, __n, __hf, __eql, __a) { }
~unordered_multimap() noexcept { }
Index: include/debug/multiset.h
===
--- include/debug/multiset.h (revision 175553)
+++ include/debug/multiset.h (working copy)
@@ -46,7 +46,6 @@
public __gnu_debug::_Safe_sequence >
{
typedef _GLIBCXX_STD_C::multiset<_Key, _Compare, _Allocator> _Base;
- typedef __gnu_debug::
[pph] Fix ICE during gimplification (issue4636074)
This patch fixes an ICE during gimplification. When reading a
STATEMENT_LIST, we were calling append_to_statement_list, but at
that point some elements in the list do not have side-effects
(the final VAR_DECL in a stmt expression), so we were dropping it
out.
Fixed by force-adding every statetement we read. This fixes
c1attr-warn-unused-result.cc.
Tested on x86_64. Committed to branch.
* pph-streamer-in.c (pph_read_tree): Call
append_to_statement_list_force instead of append_to_statement_list.
testsuite/ChangeLog.pph
* g++.dg/pph/c1attr-warn-unused-result.cc: Remove xfail marker.
diff --git a/gcc/cp/ChangeLog.pph b/gcc/cp/ChangeLog.pph
index c2679e6..a62dd66 100644
--- a/gcc/cp/ChangeLog.pph
+++ b/gcc/cp/ChangeLog.pph
@@ -1,3 +1,8 @@
+2011-06-29 Diego Novillo
+
+ * pph-streamer-in.c (pph_read_tree): Call
+ append_to_statement_list_force instead of append_to_statement_list.
+
2011-06-28 Diego Novillo
* pph-streamer-in.c (pph_in_ld_fn): Instantiate
diff --git a/gcc/cp/pph-streamer-in.c b/gcc/cp/pph-streamer-in.c
index 1dabcf1..1a59640 100644
--- a/gcc/cp/pph-streamer-in.c
+++ b/gcc/cp/pph-streamer-in.c
@@ -1488,7 +1488,7 @@ pph_read_tree (struct lto_input_block *ib
ATTRIBUTE_UNUSED,
for (i = 0; i < num_trees; i++)
{
tree stmt = pph_in_tree (stream);
- append_to_statement_list (stmt, &expr);
+ append_to_statement_list_force (stmt, &expr);
}
}
break;
diff --git a/gcc/testsuite/ChangeLog.pph b/gcc/testsuite/ChangeLog.pph
index 9c33875..aeede51 100644
--- a/gcc/testsuite/ChangeLog.pph
+++ b/gcc/testsuite/ChangeLog.pph
@@ -1,3 +1,7 @@
+2011-06-29 Diego Novillo
+
+ * g++.dg/pph/c1attr-warn-unused-result.cc: Remove xfail marker.
+
2011-06-28 Diego Novillo
* g++.dg/pph/c1attr-warn-unused-result.cc: Expect an ICE.
diff --git a/gcc/testsuite/g++.dg/pph/c1attr-warn-unused-result.cc
b/gcc/testsuite/g++.dg/pph/c1attr-warn-unused-result.cc
index 4633106..da75561 100644
--- a/gcc/testsuite/g++.dg/pph/c1attr-warn-unused-result.cc
+++ b/gcc/testsuite/g++.dg/pph/c1attr-warn-unused-result.cc
@@ -1,6 +1,3 @@
-/* { dg-xfail-if "ICE" { "*-*-*" } { "-fpph-map=pph.map" } } */
-// { dg-bogus "internal compiler error: Segmentation fault" "" { xfail *-*-* }
0 }
-// { dg-prune-output "In file included from" }
/* { dg-options "-w" } */
// pph asm xdiff
#include "c1attr-warn-unused-result.h"
--
This patch is available for review at http://codereview.appspot.com/4636074
[pph] Allow DECL_LANG_SPECIFIC to be shared (issue4667049)
This is a patch that Lawrence and I were working on yesterday
that I need for my next patch. We were not expecting
DECL_LANG_SPECIFIC to be shared, but in fact there are several
occasions (thunks, aliases) where it is shared.
By itself, this does not fix any tests, but it fixes a problem
exposed by the next patch.
Tested on x86_64. Committed to the branch.
2011-06-29 Diego Novillo
Lawrence Crowl
* pph-streamer-in.c (pph_in_lang_specific): Expect shared
DECL_LANG_SPECIFIC fields.
diff --git a/gcc/cp/ChangeLog.pph b/gcc/cp/ChangeLog.pph
index a62dd66..9465484 100644
--- a/gcc/cp/ChangeLog.pph
+++ b/gcc/cp/ChangeLog.pph
@@ -1,4 +1,10 @@
2011-06-29 Diego Novillo
+Lawrence Crowl
+
+ * pph-streamer-in.c (pph_in_lang_specific): Expect shared
+ DECL_LANG_SPECIFIC fields.
+
+2011-06-29 Diego Novillo
* pph-streamer-in.c (pph_read_tree): Call
append_to_statement_list_force instead of append_to_statement_list.
diff --git a/gcc/cp/pph-streamer-in.c b/gcc/cp/pph-streamer-in.c
index 1a59640..1c71c30 100644
--- a/gcc/cp/pph-streamer-in.c
+++ b/gcc/cp/pph-streamer-in.c
@@ -881,18 +881,23 @@ pph_in_lang_specific (pph_stream *stream, tree decl)
marker = pph_in_start_record (stream, &ix);
if (marker == PPH_RECORD_END)
return;
-
- /* Since lang_decl is embedded in every decl, LD cannot
- be shared. */
- gcc_assert (marker != PPH_RECORD_SHARED);
+ else if (marker == PPH_RECORD_SHARED)
+{
+ DECL_LANG_SPECIFIC (decl) =
+ (struct lang_decl *) pph_in_shared_data (stream, ix);
+ return;
+}
/* Allocate a lang_decl structure for DECL. */
retrofit_lang_decl (decl);
-
ld = DECL_LANG_SPECIFIC (decl);
- ldb = &ld->u.base;
+
+ /* Now register it. We would normally use ALLOC_AND_REGISTER,
+ but retrofit_lang_decl does not return a pointer. */
+ pph_register_shared_data (stream, ld, ix);
/* Read all the fields in lang_decl_base. */
+ ldb = &ld->u.base;
pph_in_ld_base (stream, ldb);
if (ldb->selector == 0)
--
This patch is available for review at http://codereview.appspot.com/4667049
[pph] Fix ICE during function expansion (issue4662067)
This patch fixes a segmentation fault during function expansion
by delaying all expansions to happen after the PPH file has been
read.
The patch fixes one test (x1template.cc), but it exposes a common
ICE in resume_scope in other two tests. Lawrence, this is the
ICE that you are now looking at, I think. Your patch should fix
these two cases now.
Tested on x86_64. Committed to the branch.
* pph-streamer-in.c (pph_read_file_contents): Call
expand_or_defer_fn on every function in stream->fns_to_expand.
(pph_in_function_decl): If FNDECL has a body, add it to
stream->fns_to_expand.
(pph_read_tree): Tidy argument list.
* pph-streamer.h (struct pph_stream): Add field fns_to_expand.
testsuite/ChangeLog.pph
* g++.dg/pph/x1dynarray1.cc: Adjust xfail.
* g++.dg/pph/x1namespace.cc: Likewise.
* g++.dg/pph/x1template.cc: Mark fixed.
diff --git a/gcc/cp/ChangeLog.pph b/gcc/cp/ChangeLog.pph
index 9465484..eaf7121 100644
--- a/gcc/cp/ChangeLog.pph
+++ b/gcc/cp/ChangeLog.pph
@@ -1,4 +1,13 @@
2011-06-29 Diego Novillo
+
+ * pph-streamer-in.c (pph_read_file_contents): Call
+ expand_or_defer_fn on every function in stream->fns_to_expand.
+ (pph_in_function_decl): If FNDECL has a body, add it to
+ stream->fns_to_expand.
+ (pph_read_tree): Tidy argument list.
+ * pph-streamer.h (struct pph_stream): Add field fns_to_expand.
+
+2011-06-29 Diego Novillo
Lawrence Crowl
* pph-streamer-in.c (pph_in_lang_specific): Expect shared
diff --git a/gcc/cp/pph-streamer-in.c b/gcc/cp/pph-streamer-in.c
index 1c71c30..3ac5243 100644
--- a/gcc/cp/pph-streamer-in.c
+++ b/gcc/cp/pph-streamer-in.c
@@ -1311,6 +1311,8 @@ pph_read_file_contents (pph_stream *stream)
cpp_ident_use *bad_use;
const char *cur_def;
cpp_idents_used idents_used;
+ tree fndecl;
+ unsigned i;
pph_in_identifiers (stream, &idents_used);
@@ -1333,6 +1335,23 @@ pph_read_file_contents (pph_stream *stream)
/* FIXME pph: This call replaces the tinfo, we should merge instead.
See pph_in_tree_vec. */
unemitted_tinfo_decls = pph_in_tree_vec (stream);
+
+ /* Expand all the functions with bodies that we read from STREAM. */
+ for (i = 0; VEC_iterate (tree, stream->fns_to_expand, i, fndecl); i++)
+{
+ /* FIXME pph - This is somewhat gross. When we generated the
+PPH image, the parser called expand_or_defer_fn on FNDECL,
+which marked it DECL_EXTERNAL (see expand_or_defer_fn_1 for
+details).
+
+However, this is not really an extern definition, so it was
+also marked not-really-extern (yes, I know...). If this
+happens, we need to unmark it, otherwise the code generator
+will toss it out. */
+ if (DECL_NOT_REALLY_EXTERN (fndecl))
+ DECL_EXTERNAL (fndecl) = 0;
+ expand_or_defer_fn (fndecl);
+}
}
@@ -1372,20 +1391,7 @@ pph_in_function_decl (pph_stream *stream, tree fndecl)
DECL_STRUCT_FUNCTION (fndecl) = pph_in_struct_function (stream, fndecl);
DECL_CHAIN (fndecl) = pph_in_tree (stream);
if (DECL_SAVED_TREE (fndecl))
-{
- /* FIXME pph - This is somewhat gross. When we generated the
-PPH image, the parser called expand_or_defer_fn on FNDECL,
-which marked it DECL_EXTERNAL (see expand_or_defer_fn_1 for
-details).
-
-However, this is not really an extern definition, so it was
-also marked not-really-extern (yes, I know...). If this
-happens, we need to unmark it, otherwise the code generator
-will toss it out. */
- if (DECL_NOT_REALLY_EXTERN (fndecl))
- DECL_EXTERNAL (fndecl) = 0;
- expand_or_defer_fn (fndecl);
-}
+VEC_safe_push (tree, gc, stream->fns_to_expand, fndecl);
}
@@ -1396,7 +1402,7 @@ pph_in_function_decl (pph_stream *stream, tree fndecl)
void
pph_read_tree (struct lto_input_block *ib ATTRIBUTE_UNUSED,
- struct data_in *data_in, tree expr)
+ struct data_in *data_in, tree expr)
{
pph_stream *stream = (pph_stream *) data_in->sdata;
diff --git a/gcc/cp/pph-streamer.h b/gcc/cp/pph-streamer.h
index b36d622..b899501 100644
--- a/gcc/cp/pph-streamer.h
+++ b/gcc/cp/pph-streamer.h
@@ -116,6 +116,10 @@ typedef struct pph_stream {
/* Nonzero if the stream was opened for writing. */
unsigned int write_p : 1;
+
+ /* List of functions with bodies that need to be expanded after
+ reading the PPH file. */
+ VEC(tree,gc) *fns_to_expand;
} pph_stream;
/* Filter values for pph_out_chain_filtered. */
diff --git a/gcc/testsuite/ChangeLog.pph b/gcc/testsuite/ChangeLog.pph
index aeede51..c5e1634 100644
--- a/gcc/testsuite/ChangeLog.pph
+++ b/gcc/testsuite/ChangeLog.pph
@@ -1,5 +1,11 @@
2011-06-29 Diego Novillo
+ * g++.dg/pph/x1dynarray1.cc: Adjust xfail.
+ * g++.dg/pph/x1namespace.cc: Likewise.
+ * g++.dg/pph/x1template.cc: Mark fixed.
+
+2011-06-29 Di
Re: [gcc patch] Re: C++ member function template id not matching linkage name (PR debug/49408)
On 06/29/2011 04:00 PM, Jan Kratochvil wrote:
> On Mon, 27 Jun 2011 20:00:24 +0200, Jason Merrill wrote:
>> They should be supressed whenever the function appears in an
>> expression context, either as a pointer to member function (i.e. the
>> operand of '&')
>
> Done, therefore it is no longer restricted only to templates as before.
>
>> or as the function being called in a call expression.
>
> I implemented it in the patch below but I do not agree + understand it.
>
> The call expression is in libiberty/testsuite/demangle-expected modified by
> this patch as:
>
> # decltype/fn call test
> --format=gnu-v3
> _Z4add3IidEDTclL_Z1gEfp_fp0_EET_T0_
> -decltype (g({parm#1}, {parm#2})) add3(int, double)
> +decltype (g) add3(int, double)
Here you're suppressing the arguments to a call, which we want to keep;
we only want to suppress printing the parameter types (which are not
part of the source expression).
Jason
Re: [gcc patch] Re: C++ member function template id not matching linkage name (PR debug/49408)
On Wed, 29 Jun 2011 22:56:26 +0200, Jason Merrill wrote:
> On 06/29/2011 04:00 PM, Jan Kratochvil wrote:
> > On Mon, 27 Jun 2011 20:00:24 +0200, Jason Merrill wrote:
> > # decltype/fn call test
> > --format=gnu-v3
> > _Z4add3IidEDTclL_Z1gEfp_fp0_EET_T0_
> > -decltype (g({parm#1}, {parm#2})) add3(int, double)
> > +decltype (g) add3(int, double)
>
> Here you're suppressing the arguments to a call, which we want to keep;
> we only want to suppress printing the parameter types (which are not
> part of the source expression).
Sorry but what is therefore the expect output in this case?
Thanks,
Jan
Re: [trans-mem] Beginning of refactoring
Patch contains more clean-up: the trycommit, rollback and
registerThrownObject functions in the ABI are unnecessary for how we
support exception handling.
Ok for branch?
commit 79ae722913f40715882b4c7b3fb798306f1ad3ae
Author: Torvald Riegel
Date: Wed Jun 29 23:05:28 2011 +0200
Removed unnecessary trycommit, rollback, and registerThrownObject ABI
functions.
diff --git a/libitm/beginend.cc b/libitm/beginend.cc
index 7ed5073..9b16973 100644
--- a/libitm/beginend.cc
+++ b/libitm/beginend.cc
@@ -184,25 +184,12 @@ GTM::gtm_transaction::rollback ()
}
void ITM_REGPARM
-_ITM_rollbackTransaction (void)
-{
- gtm_transaction *tx = gtm_tx();
-
- assert ((tx->prop & pr_hasNoAbort) == 0);
- assert ((tx->state & gtm_transaction::STATE_ABORTING) == 0);
-
- tx->rollback ();
- tx->state |= gtm_transaction::STATE_ABORTING;
-}
-
-void ITM_REGPARM
_ITM_abortTransaction (_ITM_abortReason reason)
{
gtm_transaction *tx = gtm_tx();
assert (reason == userAbort);
assert ((tx->prop & pr_hasNoAbort) == 0);
- assert ((tx->state & gtm_transaction::STATE_ABORTING) == 0);
if (tx->state & gtm_transaction::STATE_IRREVOCABLE)
abort ();
@@ -238,7 +225,7 @@ GTM::gtm_transaction::trycommit ()
bool
GTM::gtm_transaction::trycommit_and_finalize ()
{
- if ((this->state & gtm_transaction::STATE_ABORTING) || trycommit ())
+ if (trycommit ())
{
abi_disp()->fini ();
set_gtm_tx (this->prev);
@@ -252,14 +239,6 @@ GTM::gtm_transaction::trycommit_and_finalize ()
return false;
}
-bool ITM_REGPARM
-_ITM_tryCommitTransaction (void)
-{
- gtm_transaction *tx = gtm_tx();
- assert ((tx->state & gtm_transaction::STATE_ABORTING) == 0);
- return tx->trycommit ();
-}
-
void ITM_NORETURN
GTM::gtm_transaction::restart (gtm_restart_reason r)
{
diff --git a/libitm/libitm.h b/libitm/libitm.h
index df89d33..abd4274 100644
--- a/libitm/libitm.h
+++ b/libitm/libitm.h
@@ -45,7 +45,7 @@ extern "C" {
#define ITM_NORETURN __attribute__((noreturn))
#define ITM_PURE __attribute__((transaction_pure))
-
+
/* The following are externally visible definitions and functions, though
only very few of these should be called by user code. */
@@ -91,6 +91,7 @@ typedef enum
pr_RaRBarriersOmitted = 0x0200,
pr_undoLogCode = 0x0400,
pr_preferUninstrumented = 0x0800,
+ /* Exception blocks are not used nor supported. */
pr_exceptionBlock = 0x1000,
pr_readOnly = 0x4000,
pr_hasElse = 0x20,
@@ -138,12 +139,8 @@ extern _ITM_transactionId_t _ITM_getTransactionId(void)
ITM_REGPARM;
extern uint32_t _ITM_beginTransaction(uint32_t, ...) ITM_REGPARM;
extern void _ITM_abortTransaction(_ITM_abortReason) ITM_REGPARM ITM_NORETURN;
-extern void _ITM_rollbackTransaction (void) ITM_REGPARM;
extern void _ITM_commitTransaction (void) ITM_REGPARM;
-extern bool _ITM_tryCommitTransaction(void) ITM_REGPARM;
-
-extern void _ITM_registerThrownObject (const void *, size_t) ITM_REGPARM;
extern void _ITM_changeTransactionMode (_ITM_transactionState) ITM_REGPARM;
diff --git a/libitm/libitm.map b/libitm/libitm.map
index 0d52a7c..49d0b1b 100644
--- a/libitm/libitm.map
+++ b/libitm/libitm.map
@@ -12,9 +12,6 @@ LIBITM_1.0 {
_ITM_getTransactionId;
_ITM_inTransaction;
_ITM_libraryVersion;
- _ITM_registerThrownObject;
- _ITM_rollbackTransaction;
- _ITM_tryCommitTransaction;
_ITM_versionCompatible;
_ITM_registerTMCloneTable;
diff --git a/libitm/libitm.texi b/libitm/libitm.texi
index 8d8de8c..046b0bd 100644
--- a/libitm/libitm.texi
+++ b/libitm/libitm.texi
@@ -221,25 +221,34 @@ reported for the dynamic scope as well, not just for the
lexical scope
(@code{hasNoAbort}). Without this, a library cannot exploit this together
with flat nesting.
-@strong{TODO} @code{exceptionBlock?}
+@code{exceptionBlock} is not supported because exception blocks are not used.
@subsubsection [No changes] Windows exception state
@subsubsection [No changes] Other machine state
-@subsubsection Results from beginTransaction
-@strong{TODO} @code{abortTransaction} supported?
+@subsubsection [No changes] Results from beginTransaction
@subsection Aborting a transaction
-@strong{TODO} make consistent with EH.
-@subsection Committing a transaction
-@strong{TODO} make consistent with EH.
+@code{_ITM_rollbackTransaction} is not supported. @code{_ITM_abortTransaction}
+is supported but the abort reason @code{exceptionBlockAbort} is not (and there
+are no exception blocks in general, so the related cases also do not have to
+be considered).
-@subsection Exception handling support
+@subsection Committing a transaction
-@strong{TODO} Document wrappers. Document code generated for commit, perhaps
-with examples similar to those in the specification. What can be removed from
-the ABI in turn? Document requirements on libstdc++ (@code{_cxa_tm_cleanup()}).
+The exception handling (EH) scheme is di
C++ PATCH for c++/49003 (DR 1207, use of 'this' in trailing return type)
This patch adds support for use of 'this' (implicitly or explicitly) in
the trailing-return-type of a member function.
Tested x86_64-pc-linux-gnu, applying to trunk.
commit 8154a9854f1d5ff3407af1707d112899eb39f013
Author: Jason Merrill
Date: Wed Jun 29 15:19:46 2011 -0400
DR 1207
PR c++/49003
* cp-tree.h (struct saved_scope): Add x_current_class_ptr,
x_current_class_ref.
(current_class_ptr, current_class_ref): Use them.
* decl.c (build_this_parm): Handle getting the class type.
* parser.c (cp_parser_late_return_type_opt): Set up 'this'
for use within the trailing return type.
diff --git a/gcc/cp/cp-tree.h b/gcc/cp/cp-tree.h
index 55c88e3..ef25c97 100644
--- a/gcc/cp/cp-tree.h
+++ b/gcc/cp/cp-tree.h
@@ -953,6 +953,10 @@ struct GTY(()) saved_scope {
struct cp_binding_level *x_previous_class_level;
tree x_saved_tree;
+ /* Only used for uses of this in trailing return type. */
+ tree x_current_class_ptr;
+ tree x_current_class_ref;
+
int x_processing_template_decl;
int x_processing_specialization;
BOOL_BITFIELD x_processing_explicit_instantiation : 1;
@@ -1070,12 +1074,14 @@ struct GTY(()) language_function {
PARM_DECL for the `this' pointer. The current_class_ref is an
expression for `*this'. */
-#define current_class_ptr \
- (cfun && cp_function_chain \
- ? cp_function_chain->x_current_class_ptr : NULL_TREE)
-#define current_class_ref \
- ((cfun && cp_function_chain) \
- ? cp_function_chain->x_current_class_ref : NULL_TREE)
+#define current_class_ptr \
+ (*(cfun && cp_function_chain \
+ ? &cp_function_chain->x_current_class_ptr \
+ : &scope_chain->x_current_class_ptr))
+#define current_class_ref \
+ (*(cfun && cp_function_chain \
+ ? &cp_function_chain->x_current_class_ref \
+ : &scope_chain->x_current_class_ref))
/* The EH_SPEC_BLOCK for the exception-specifiers for the current
function, if any. */
diff --git a/gcc/cp/decl.c b/gcc/cp/decl.c
index b8435a6..94d686d 100644
--- a/gcc/cp/decl.c
+++ b/gcc/cp/decl.c
@@ -7001,7 +7001,14 @@ build_this_parm (tree type, cp_cv_quals quals)
tree parm;
cp_cv_quals this_quals;
- this_type = type_of_this_parm (type);
+ if (CLASS_TYPE_P (type))
+{
+ this_type
+ = cp_build_qualified_type (type, quals & ~TYPE_QUAL_RESTRICT);
+ this_type = build_pointer_type (this_type);
+}
+ else
+this_type = type_of_this_parm (type);
/* The `this' parameter is implicitly `const'; it cannot be
assigned to. */
this_quals = (quals & TYPE_QUAL_RESTRICT) | TYPE_QUAL_CONST;
@@ -12675,6 +12682,7 @@ start_preparsed_function (tree decl1, tree attrs, int flags)
cp_function_chain->x_current_class_ref
= cp_build_indirect_ref (t, RO_NULL, tf_warning_or_error);
+ /* Set this second to avoid shortcut in cp_build_indirect_ref. */
cp_function_chain->x_current_class_ptr = t;
/* Constructors and destructors need to know whether they're "in
diff --git a/gcc/cp/parser.c b/gcc/cp/parser.c
index f1b7976..d79326d 100644
--- a/gcc/cp/parser.c
+++ b/gcc/cp/parser.c
@@ -1696,7 +1696,7 @@ static cp_cv_quals cp_parser_cv_qualifier_seq_opt
static cp_virt_specifiers cp_parser_virt_specifier_seq_opt
(cp_parser *);
static tree cp_parser_late_return_type_opt
- (cp_parser *);
+ (cp_parser *, cp_cv_quals);
static tree cp_parser_declarator_id
(cp_parser *, bool);
static tree cp_parser_type_id
@@ -14968,7 +14968,7 @@ cp_parser_direct_declarator (cp_parser* parser,
virt_specifiers = cp_parser_virt_specifier_seq_opt (parser);
late_return
- = cp_parser_late_return_type_opt (parser);
+ = cp_parser_late_return_type_opt (parser, cv_quals);
/* Create the function-declarator. */
declarator = make_call_declarator (declarator,
@@ -15537,9 +15537,10 @@ cp_parser_virt_specifier_seq_opt (cp_parser* parser)
Returns the type indicated by the type-id. */
static tree
-cp_parser_late_return_type_opt (cp_parser* parser)
+cp_parser_late_return_type_opt (cp_parser* parser, cp_cv_quals quals)
{
cp_token *token;
+ tree type;
/* Peek at the next token. */
token = cp_lexer_peek_token (parser->lexer);
@@ -15550,7 +15551,23 @@ cp_parser_late_return_type_opt (cp_parser* parser)
/* Consume the ->. */
cp_lexer_consume_token (parser->lexer);
- return cp_parser_trailing_type_id (parser);
+ if (current_class_type)
+{
+ /* DR 1207: 'this' is in scope in the trailing return type. */
+ tree this_parm = build_this_parm (current_class_type, quals);
+ gcc_assert (current_class_ptr == NULL_TREE);
+ current_class_ref
+ = cp_build_indirect_ref (this_parm, RO_NULL, tf_warning_or_error);
+ /* Set this second to avoid shortcut in cp_build_indirect_ref. */
+ current_class_ptr = this_parm;
+}
+
+ type = cp_parser_trailing_type_id (parser);
+
+ if (current_class_type)
+current_class_ptr = current_class_ref = NULL_TREE;
Re: C++ PATCH for c++/49554 (bogus error with lambda in template)
While looking at this issue, I also noticed that we were substituting
into DECL_INITIAL more than we need to.
Tested x86_64-pc-linux-gnu, applying to trunk.
commit 327bbebe3243b0423932862d3d12f92928853e9c
Author: Jason Merrill
Date: Wed Jun 29 14:21:17 2011 -0400
* pt.c (tsubst_decl) [VAR_DECL]: In unevaluated operand,
don't tsubst DECL_INITIAL unless our type use auto.
diff --git a/gcc/cp/pt.c b/gcc/cp/pt.c
index d1d8336..dc6cd50 100644
--- a/gcc/cp/pt.c
+++ b/gcc/cp/pt.c
@@ -10138,12 +10138,11 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain)
scope, such as for a lambda return type. Don't add it to
local_specializations, do perform auto deduction. */
tree auto_node = type_uses_auto (type);
- tree init
- = tsubst_expr (DECL_INITIAL (t), args, complain, in_decl,
- /*constant_expression_p=*/false);
-
- if (auto_node && init)
+ if (auto_node)
{
+ tree init
+ = tsubst_expr (DECL_INITIAL (t), args, complain, in_decl,
+ /*constant_expression_p=*/false);
init = resolve_nondeduced_context (init);
TREE_TYPE (r) = type
= do_auto_deduction (type, init, auto_node);
[PATCH] Sign extend before converting constants to GMP values.
Hi,
This patch fixes PR47653 by sign extending the double int constants
before converting them to a GMP value. There still are some places
where we should not sign extend the values converted: upper bounds of
unsigned types should for example not be sign extended.
The patch passed make -k check RUNTESTFLAGS=graphite.exp on c,c++,fortran.
Regstrap with all languages in progress on amd64-linux. Ok for trunk?
Thanks,
Sebastian
2011-06-29 Sebastian Pop
PR tree-optimization/47653
* graphite-ppl.h (tree_int_to_gmp): Sign extend before converting
constants to GMP values. Add a sext parameter.
(ppl_set_inhomogeneous_tree): Add sext parameter.
(ppl_set_coef_tree): Removed.
* graphite-sese-to-poly.c (scan_tree_for_params_right_scev): Adjust
call to tree_int_to_gmp.
(scan_tree_for_params_int): Use tree_int_to_gmp.
(scan_tree_for_params): Adjust call to tree_int_to_gmp.
(build_loop_iteration_domains): Adjust call to
ppl_set_inhomogeneous_tree.
(add_param_constraints): Same.
(pdr_add_data_dimensions): Same.
* gcc.dg/graphite/run-id-pr47653.c: New.
---
gcc/ChangeLog | 16
gcc/graphite-ppl.h | 30 +--
gcc/graphite-sese-to-poly.c| 27 ++---
gcc/testsuite/ChangeLog|5
gcc/testsuite/gcc.dg/graphite/run-id-pr47653.c | 17 +
5 files changed, 63 insertions(+), 32 deletions(-)
create mode 100644 gcc/testsuite/gcc.dg/graphite/run-id-pr47653.c
diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index e37d823..bed0070 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,19 @@
+2011-06-29 Sebastian Pop
+
+ PR tree-optimization/47653
+ * graphite-ppl.h (tree_int_to_gmp): Sign extend before converting
+ constants to GMP values. Add a sext parameter.
+ (ppl_set_inhomogeneous_tree): Add sext parameter.
+ (ppl_set_coef_tree): Removed.
+ * graphite-sese-to-poly.c (scan_tree_for_params_right_scev): Adjust
+ call to tree_int_to_gmp.
+ (scan_tree_for_params_int): Use tree_int_to_gmp.
+ (scan_tree_for_params): Adjust call to tree_int_to_gmp.
+ (build_loop_iteration_domains): Adjust call to
+ ppl_set_inhomogeneous_tree.
+ (add_param_constraints): Same.
+ (pdr_add_data_dimensions): Same.
+
2011-06-29 Eric Botcazou
PR tree-optimization/49539
diff --git a/gcc/graphite-ppl.h b/gcc/graphite-ppl.h
index 695d01f..4ae9f63 100644
--- a/gcc/graphite-ppl.h
+++ b/gcc/graphite-ppl.h
@@ -50,13 +50,18 @@ void debug_gmp_value (mpz_t);
bool ppl_powerset_is_empty (ppl_Pointset_Powerset_C_Polyhedron_t);
-/* Assigns to RES the value of the INTEGER_CST T. */
+/* Assigns to RES the value of the INTEGER_CST T. When SEXT is true,
+ sign extend the value of T to not get "-1 = 2^n - 1". */
static inline void
-tree_int_to_gmp (tree t, mpz_t res)
+tree_int_to_gmp (tree t, mpz_t res, bool sext)
{
+ tree type = TREE_TYPE (t);
double_int di = tree_to_double_int (t);
- mpz_set_double_int (res, di, TYPE_UNSIGNED (TREE_TYPE (t)));
+
+ if (sext)
+di = double_int_sext (di, TYPE_PRECISION (type));
+ mpz_set_double_int (res, di, false);
}
/* Converts a GMP constant VAL to a tree and returns it. */
@@ -88,14 +93,15 @@ ppl_set_inhomogeneous (ppl_Linear_Expression_t e, int x)
mpz_clear (v);
}
-/* Set the inhomogeneous term of E to the tree X. */
+/* Set the inhomogeneous term of E to the tree X. When SEXT is true,
+ sign extend the value of X. */
static inline void
-ppl_set_inhomogeneous_tree (ppl_Linear_Expression_t e, tree x)
+ppl_set_inhomogeneous_tree (ppl_Linear_Expression_t e, tree x, bool sext)
{
mpz_t v;
mpz_init (v);
- tree_int_to_gmp (x, v);
+ tree_int_to_gmp (x, v, sext);
ppl_set_inhomogeneous_gmp (e, v);
mpz_clear (v);
}
@@ -112,18 +118,6 @@ ppl_set_coef (ppl_Linear_Expression_t e,
ppl_dimension_type i, int x)
mpz_clear (v);
}
-/* Set E[I] to tree X. */
-
-static inline void
-ppl_set_coef_tree (ppl_Linear_Expression_t e, ppl_dimension_type i, tree x)
-{
- mpz_t v;
- mpz_init (v);
- tree_int_to_gmp (x, v);
- ppl_set_coef_gmp (e, i, v);
- mpz_clear (v);
-}
-
/* Sets RES to the max of V1 and V2. */
static inline void
diff --git a/gcc/graphite-sese-to-poly.c b/gcc/graphite-sese-to-poly.c
index 7e23c9d..5f8188b 100644
--- a/gcc/graphite-sese-to-poly.c
+++ b/gcc/graphite-sese-to-poly.c
@@ -633,7 +633,7 @@ scan_tree_for_params_right_scev (sese s, tree e, int var,
gcc_assert (TREE_CODE (e) == INTEGER_CST);
mpz_init (val);
- tree_int_to_gmp (e, val);
+ tree_int_to_gmp (e, val, true);
add_value_to_dim (l, expr, val);
mpz_clear (val);
}
@@ -647,14 +647,9 @@ scan_tree_for_params_int (tree cst,
ppl_Linear_Expression_t expr, mpz_t k)
{
mpz_t val;
ppl_Coefficien
