Easwaran Raman [Wed, 31 Oct 2012 23:28:45 +0000 (23:28 +0000)]
re PR target/54938 (sh libgcc_unpack_df.o fails to build: ../../../srcw/libgcc/fp-bit.h:221:19: internal compiler error: in emit_cmp_and_jump_insn_1, at optabs.c:4273)
2012-10-31 Easwaran Raman <eraman@google.com>
PR target/54938
PR middle-end/54957
* optabs.c (emit_cmp_and_jump_insn_1): Add REG_BR_PROB note
only if it doesn't already exist.
* stmt.c (get_outgoing_edge_probs): Return 0 if BB is NULL.
(emit_case_dispatch_table): Handle the case where STMT_BB is
NULL.
(expand_sjlj_dispatch_table): Pass BB containing before_case
to emit_case_dispatch_table.
Lawrence Crowl [Wed, 31 Oct 2012 23:15:10 +0000 (23:15 +0000)]
This patch implements generic type query and conversion functions,
and applies them to the use of cgraph_node, varpool_node, and symtab_node.
The functions are:
bool is_a <TYPE> (pointer)
Tests whether the pointer actually points to a more derived TYPE.
TYPE *as_a <TYPE> (pointer)
Converts pointer to a TYPE*.
TYPE *dyn_cast <TYPE> (pointer)
Converts pointer to TYPE* if and only if "is_a <TYPE> pointer".
Otherwise, returns NULL.
This function is essentially a checked down cast.
These functions reduce compile time and increase type safety when treating a
generic item as a more specific item. In essence, the code change is from
if (cgraph_node *cnode = dyn_cast <cgraph_node> (node))
{
....
}
The necessary conditional test defines a variable that holds a known good
pointer to the specific item and avoids subsequent conversion calls and
the assertion checks that may come with them.
When, the property test is embedded within a larger condition, the variable
declaration gets pulled out of the condition. (This leaves some room for
using the variable inappropriately.)
if (symtab_variable_p (node)
&& varpool (node)->finalized)
varpool_analyze_node (varpool (node));
Note that we have converted two sets of assertions in the calls to varpool
into safe and efficient use of a variable.
There are remaining calls to symtab_function_p and symtab_variable_p that
do not involve a pointer to a more specific type. These have been converted
to calls to a functions is_a <cgraph_node> and is_a <varpool_node>. The
original predicate functions have been removed.
The cgraph.h header defined both a struct and a function with the name
varpool_node. This name overloading can cause some unintuitive error messages
when, as is common in C++, one omits the struct keyword when using the type.
I have renamed the function to varpool_node_for_decl.
Tested on x86_64.
Index: gcc/ChangeLog
2012-10-31 Lawrence Crowl <crowl@google.com>
* is-a.h: New.
(is_a <T> (U*)): New. Test for is-a relationship.
(as_a <T> (U*)): New. Treat as a derived type.
(dyn_cast <T> (U*)): New. Conditionally cast based on is_a.
* cgraph.h (varpool_node): Rename to varpool_node_for_decl.
Adjust callers to match.
(is_a_helper <cgraph_node>::test (symtab_node_def *)): New.
(is_a_helper <varpool_node>::test (symtab_node_def *)): New.
(symtab_node_def::try_function): New. Change most calls to
symtab_function_p with calls to dyn_cast <cgraph_node> (p).
(symtab_node_def::try_variable): New. Change most calls to
symtab_variable_p with calls to dyn_cast <varpool_node> (p).
(symtab_function_p): Remove. Change callers to use
is_a <cgraph_node> (p) instead.
(symtab_variable_p): Remove. Change callers to use
is_a <varpool_node> (p) instead.
* cgraph.c (cgraph_node_for_asm): Remove redundant call to
symtab_node_for_asm.
* cgraphunit.c (symbol_finalized_and_needed): New.
(symbol_finalized): New.
(cgraph_analyze_functions): Split complicated conditionals out into
above new functions.
* Makefile.in (CGRAPH_H): Add is-a.h as used by cgraph.h.
Steven Bosscher [Wed, 31 Oct 2012 21:37:10 +0000 (21:37 +0000)]
re PR tree-optimization/55018 (CDDCE pass is too aggressive sometimes with infinite loops and with some functions)
gcc/
PR tree-optimization/55018
* basic-block.h (dfs_find_deadend): New prototype.
* cfganal.c (dfs_find_deadend): No longer static. Use bitmap
instead of sbitmap for visited.
(flow_dfs_compute_reverse_execute): Use dfs_find_deadend here, too.
* dominance.c (calc_dfs_tree): If saw_unconnected,
traverse from dfs_find_deadend of unconnected b
instead of b directly.
testsuite/
PR tree-optimization/55018
* gcc.dg/torture/pr55018.c: New test.
Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r193047
Tobias Burnus [Wed, 31 Oct 2012 15:46:59 +0000 (16:46 +0100)]
complex.c (csqrtq): NaN and INF fixes.
2012-10-31 Tobias Burnus <burnus@net-b.de>
Joseph Myers <joseph@codesourcery.com>
David S. Miller <davem@davemloft.net>
Ulrich Drepper <drepper@redhat.com>
Marek Polacek <polacek@redhat.com>:
Petr Baudis <pasky@suse.cz>
* math/complex.c (csqrtq): NaN and INF fixes.
* math/sqrtq.c (sqrt): NaN, INF and < 0 fixes.
* math/expm1q.c (expm1q): Changes from GLIBC. Use expq for
large parameters. Fix errno for boundary conditions.
* math/finiteq.c (finiteq): Add comment.
* math/fmaq.c (fmaq): Changes from GLIBC. Fix missing underflows
and bad results for some subnormal results. Fix sign of inexact
zero return. Fix sign of exact zero return.
Ensure additions are not scheduled after fetestexcept.
* math/jnq.c (jnq): Changes from GLIBC. Set up errno properly
for ynq. Fix jnq precision.
* math/nearbyintq.c (nearbyintq): Changes from GLIBC. Do not
manipulate bits before adding and subtracting TWO112[sx].
* math/rintq.c (rintq): Ditto.
* math/scalbnq.c (scalbnq): Changes from GLIBC. Fix integer
overflow.
Co-Authored-By: David S. Miller <davem@davemloft.net> Co-Authored-By: Joseph Myers <joseph@codesourcery.com> Co-Authored-By: Ulrich Drepper <drepper@redhat.com>
From-SVN: r193037
Dodji Seketeli [Wed, 31 Oct 2012 08:55:43 +0000 (08:55 +0000)]
PR c++/54955 - Fail to parse alignas expr at the beginning of a declaration
In this PR, g++ embarrassingly fails to parse the simple alignas
expression below:
alignas(double) int f;
even though the simple-declaration production in Clause 7 suggests
otherwise.
Fixed thus and tested on x86_64-unknown-linux-gnu against trunk.
gcc/cp
PR c++/54955
* parser.c (cp_nth_tokens_can_be_std_attribute_p): Recognize the
'Alignas' keyword as the beginning of a c++11 attribute specifier.
Update the comment of the function.
(cp_next_tokens_can_be_gnu_attribute_p): Update the comment of the
function.
gcc/testsuite/
PR c++/54955
* g++.dg/cpp0x/gen-attrs-48-2.C: New test.
Jakub Jelinek [Wed, 31 Oct 2012 08:45:27 +0000 (09:45 +0100)]
re PR tree-optimization/19105 (fold does not spot anti-range test)
PR tree-optimization/19105
PR tree-optimization/21643
PR tree-optimization/46309
* tree-ssa-reassoc.c (init_range_entry): Add STMT argument
and use it if EXP is NULL.
(update_range_test): Handle OPCODE equal to ERROR_MARK
and oe->op NULL.
(optimize_range_tests): Likewise.
(final_range_test_p, suitable_cond_bb, no_side_effect_bb, get_ops,
maybe_optimize_range_tests): New functions.
(reassociate_bb): Call maybe_optimize_range_tests if last
stmt of bb is GIMPLE_COND that hasn't been visited yet.
* gcc.dg/pr19105.c: New test.
* gcc.dg/pr21643.c: New test.
* gcc.dg/pr46309-2.c: New test.
* gcc.c-torture/execute/pr46309.c: New test.
expmed.c (store_bit_field_using_insv): New function, split out from...
gcc/
* expmed.c (store_bit_field_using_insv): New function,
split out from...
(store_bit_field_1): ...here.
(extract_bit_field_using_extv): New function, split out from...
(extract_bit_field_1): ...here.
expmed.c (store_bit_field_1): Use OP_MODE to check whether an insv pattern is available.
gcc/
* expmed.c (store_bit_field_1): Use OP_MODE to check whether an
insv pattern is available. Remove redundant checks for OP_MODE
being MAX_MACHINE_MODE.
(extract_bit_field_1): Remove redundant checks for EXT_MODE being
MAX_MACHINE_MODE.
Alexandre Oliva [Tue, 30 Oct 2012 23:47:35 +0000 (23:47 +0000)]
re PR debug/54551 (DF resets some DEBUG_INSNs unnecessarily)
PR debug/54551
PR debug/54693
* valtrack.c (dead_debug_promote_uses): Assert-check that
global used bit was clear and initialize entry
unconditionally.
Steve Ellcey [Tue, 30 Oct 2012 21:21:41 +0000 (21:21 +0000)]
mti-linux.h (SYSROOT_SUFFIX_SPEC): Change order and add mabi=64.
2012-10-30 Steve Ellcey <sellcey@mips.com>
* config/mips/mti-linux.h (SYSROOT_SUFFIX_SPEC): Change order
and add mabi=64.
(DRIVER_SELF_SPECS): Make -n32 the default on mips64* archs.
* config/mips/t-mti-linux (MULTILIB_OPTIONS): Change order.
(MULTILIB_DIRNAMES): Ditto.
(MULTILIB_EXCEPTIONS): New.
Jan Hubicka [Tue, 30 Oct 2012 16:50:05 +0000 (16:50 +0000)]
tree-ssa-loop-niter.c (number_of_iterations_exit): New parameter EVERY_ITERATION with implicit value of true.
* tree-ssa-loop-niter.c (number_of_iterations_exit): New parameter
EVERY_ITERATION with implicit value of true.
(record_estimate): Check dominance relationship of the basic block
we are estimating on instead of relying on UPPER to be false.
(struct ilb_data): Drop RELIABLE.
(idx_infer_loop_bounds): Update.
(infer_loop_bounds_from_ref): Drop parameter RELIABLE.
(infer_loop_bounds_from_array): Drop parameter RELIABLE.
(infer_loop_bounds_from_undefined): Update comments and handling
of RELIABLE.
(estimate_numbers_of_iterations_loop): Record all bounds.
Jan Hubicka [Tue, 30 Oct 2012 16:12:16 +0000 (17:12 +0100)]
tree-ssa-loop-niter.c (number_of_iterations_exit): New parameter EVERY_ITERATION with implicit value of true.
* tree-ssa-loop-niter.c (number_of_iterations_exit): New parameter
EVERY_ITERATION with implicit value of true.
(record_estimate): Check dominance relationship of the basic block
we are estimating on instead of relying on UPPER to be false.
(struct ilb_data): Drop RELIABLE.
(idx_infer_loop_bounds): Update.
(infer_loop_bounds_from_ref): Drop parameter RELIABLE.
(infer_loop_bounds_from_array): Drop parameter RELIABLE.
(infer_loop_bounds_from_undefined): Update comments and handling
of RELIABLE.
(estimate_numbers_of_iterations_loop): Record all bounds.
Oleg Endo [Tue, 30 Oct 2012 09:22:14 +0000 (09:22 +0000)]
re PR target/54963 (Wrong code generated for libgfortran/generated/eoshift3_8.c on SH)
PR target/54963
* config/sh/iterators.md (SIDI): New mode iterator.
* config/sh/sh.md (negdi2): Use parallel around operation and T_REG
clobber in expander.
(*negdi2): Mark output operand as early clobbered. Add T_REG clobber.
Split after reload. Simplify split code.
(abssi2, absdi2): Fold expanders into abs<mode>2.
(*abssi2, *absdi2): Fold into *abs<mode>2 insn_and_split. Split insns
before reload.
(*negabssi2, *negabsdi2): Fold into *negabs<mode>2. Add T_REG clobber.
Split insns before reload.
(negsi_cond): Reformat. Use emit_move_insn instead of
gen_movesi.
(negdi_cond): Reformat. Use emit_move_insn instead of a pair
of gen_movsi. Split insn before reload.
Jakub Jelinek [Tue, 30 Oct 2012 08:08:01 +0000 (09:08 +0100)]
re PR debug/54953 (New sra-1.c FAILs on powerpc)
PR debug/54953
* valtrack.h (DEBUG_TEMP_AFTER_WITH_REG_FORCE): New.
* valtrack.c (dead_debug_insert_temp): Use emit_debug_insn_after
even for where == DEBUG_TEMP_AFTER_WITH_REG_FORCE.
* dce.c (word_dce_process_block, dce_process_block): Pass
DEBUG_TEMP_AFTER_WITH_REG_FORCE if insn is needed and therefore
not going to be eliminated.
Lawrence Crowl [Tue, 30 Oct 2012 00:02:55 +0000 (00:02 +0000)]
This patch implements the unification of the *bitmap interfaces as discussed.
Essentially, we rename ebitmap and sbitmap functions to use the same names
as the bitmap functions. This rename works because we can now overload
on the bitmap type. Some macros now become inline functions to enable
that overloading.
The sbitmap non-bool returning bitwise operations have been merged with
the bool versions. Sometimes this merge involved modifying the non-bool
version to compute the bool value, and sometimes modifying bool version to
add additional work from the non-bool version. The redundant routines have
been removed.
The allocation functions have not been renamed, because we often do not
have an argument on which to overload. The cardinality functions have not
been renamed, because they have different parameters, and are thus not
interchangable. The iteration functions have not been renamed, because
they are functionally different.
Tested on x86_64, contrib/config-list.mk testing passed.
Jonathan Wakely [Mon, 29 Oct 2012 21:49:19 +0000 (21:49 +0000)]
re PR libstdc++/55123 ([C++11] Construction of shared_ptr<const T> from unique_ptr<const T> fails)
PR libstdc++/55123
* include/bits/shared_ptr_base.h (__shared_count::_S_create_from_up):
Do not instantiate allocator with element_type.
* testsuite/20_util/shared_ptr/cons/55123.cc: New.
re PR c/53066 (Wshadow should not warn for shadowing an extern function)
2012-10-29 Manuel López-Ibáñez <manu@gcc.gnu.org>
PR c/53066
c/
* c-decl.c (warn_if_shadowing): Do not warn if a variable
shadows a function, unless the variable is a function or a
pointer-to-function.
gcc/
* tree.h (FUNCTION_POINTER_TYPE_P): New.
testsuite/
* gcc.dg/Wshadow-4.c: New.
* gcc.dg/Wshadow-4.h: New.
Alexandre Oliva [Mon, 29 Oct 2012 19:27:31 +0000 (19:27 +0000)]
re PR debug/54551 (DF resets some DEBUG_INSNs unnecessarily)
PR debug/54551
PR debug/54693
* valtrack.c (dead_debug_global_find): Accept NULL dtemp.
(dead_debug_global_insert): Return new entry.
(dead_debug_global_replace_temp): Return early if REG is no
longer in place, or if dtemp was already substituted.
(dead_debug_promote_uses): Insert for all defs and replace all
debug uses at once.
(dead_debug_local_finish): Release used after promotion.
(dead_debug_insert_temp): Stop if dtemp is NULL.
Alexandre Oliva [Mon, 29 Oct 2012 19:26:16 +0000 (19:26 +0000)]
re PR debug/54693 (VTA guality issues with loops)
PR debug/54693
* config/i386/i386.c (add_parameter_dependencies): Stop
backward scan at the insn before the incoming head.
(ix86_dependencies_evaluation_hook): Skip debug insns. Stop
if first_arg is head.
Jan Hubicka [Mon, 29 Oct 2012 15:48:21 +0000 (16:48 +0100)]
ipa-inline.c (want_inline_function_called_once_p): Rename to ...
* ipa-inline.c (want_inline_function_called_once_p): Rename to ...
(want_inline_function_to_all_callers_p): check also functions with
multiple callers.
(ipa_inline): Handle inlining for size into multiple callers.
Arnaud Charlet [Mon, 29 Oct 2012 11:42:17 +0000 (12:42 +0100)]
[multiple changes]
2012-10-29 Thomas Quinot <quinot@adacore.com>
* xoscons.adb: Minor reformatting.
2012-10-29 Yannick Moy <moy@adacore.com>
* exp_alfa.adb (Expand_Alfa): Backtrack change that removed
qualification of names in formal verification mode. Instead,
the qualification should be modified.
* exp_dbug.adb (Qualify_Entity_Name): Modify qualification in formal
verification mode, so that only a suffix is added to distinguish
homonyms from the same scope.
Arnaud Charlet [Mon, 29 Oct 2012 11:41:01 +0000 (12:41 +0100)]
[multiple changes]
2012-10-29 Robert Dewar <dewar@adacore.com>
* gnat_rm.texi: Document that pragma Optimize_Alignment (Space) is
ignored with a warning for packed variable length records.
2012-10-29 Thomas Quinot <quinot@adacore.com>
* socket.c, g-socthi-dummy.adb, g-socthi-dummy.ads, g-socthi-vms.adb,
g-socthi-vms.ads, g-socthi-vxworks.adb, g-socthi-vxworks.ads,
s-oscons-tmplt.c, g-socthi-mingw.adb, g-socthi-mingw.ads, g-socthi.adb,
g-socthi.ads, xoscons.adb, g-socket.adb, g-sothco.ads: Introduce an
appropriate subtype for IOCTL requests, since these may be signed or
unsigned.
Arnaud Charlet [Mon, 29 Oct 2012 11:32:18 +0000 (12:32 +0100)]
[multiple changes]
2012-10-29 Robert Dewar <dewar@adacore.com>
* sem_prag.adb: Minor reformatting.
2012-10-29 Robert Dewar <dewar@adacore.com>
* gnat_rm.texi: Minor rewording.
2012-10-29 Javier Miranda <miranda@adacore.com>
* exp_disp.ads (Is_Expanded_Dispatching_Call): New subprogram.
* exp_disp.adb (Expand_Dispatching_Call): No action needed if the
call has been already expanded.
(Is_Expanded_Dispatching_Call): New subprogram.
* sem_disp.adb (Propagate_Tag): No action needed if the call
has been already expanded.
* exp_ch9.adb (Create_Index_And_Data): Remove local
variable Index_Typ and its uses. The type of the index is now
System.Tasking.Entry_Index. Update all related comments.
* rtsfind.ads: Add RE_Entry_Index in tables RE_Id and RE_Unit_Table.
* s-taskin.adb (Number_Of_Entries): The return type is now Entry_Index.
* s-taskin.ads: The index type of Task_Entry_Names_Array is now
Entry_Index.
(Number_Of_Entries): The return type is now Entry_Index.
* s-tpoben.adb (Number_Of_Entries): The return type is now Entry_Index.
* s-tpoben.ads: The index type of Protected_Entry_Names_Array
is now Entry_Index.
(Number_Of_Entries): The return type is now Entry_Index.
2012-10-29 Pascal Obry <obry@adacore.com>
* gnat_ugn.texi: Add note about SEH setup on x86-windows.
2012-10-29 Eric Botcazou <ebotcazou@adacore.com>
* s-bignum.adb (Allocate_Bignum): Use the exact layout of
Bignum_Data for the overlay.
Thomas Quinot [Mon, 29 Oct 2012 11:21:57 +0000 (11:21 +0000)]
gnat_rm.texi, [...] (Sem_Prag.Analyze_Pragma): Handle new pragma Attribute_Definition.
2012-10-29 Thomas Quinot <quinot@adacore.com>
* gnat_rm.texi, sem_prag.adb, sem_util.adb, sem_util.ads,
par-prag.adb, par-util.adb, snames.ads-tmpl (Sem_Prag.Analyze_Pragma):
Handle new pragma Attribute_Definition.
(Sem_Util.Bad_Attribute): New routine, moved here
from par-util, so that it can be used by the above.
(Par_Util.Signal_Bad_Attribute): Processing moved to
Sem_Util.Bad_Attribute.
Arnaud Charlet [Mon, 29 Oct 2012 11:09:46 +0000 (12:09 +0100)]
[multiple changes]
2012-10-29 Tristan Gingold <gingold@adacore.com>
* gnat_rm.texi: Document implementation advice for Pragma
Partition_Elaboration_Policy.
2012-10-29 Yannick Moy <moy@adacore.com>
* s-bignum.adb (Div_Rem): Reference that Algorithm_D is from
the second edition of TAOCP from Knuth, since the algo changed
in the third edition. Also correct the definition of 'd' which
could overflow.
* exp_ch3.adb (Build_Initialization_Call): Create static strings
which denote entry [family] names and associate them with the
object's Protection_Entries or ATCB.
(Build_Init_Statements):
Remove local variable Names. Do not generate the entry [family]
names inside the init proc because they are now static.
* exp_ch9.adb (Build_Entry_Names): Reimplemented. The strings
which denote entry [family] names are now generated statically
and associated with the concurrent object's Protection_Entries
or ATCB during initialization.
* exp_ch9.ads (Build_Entry_Names): Change subprogram profile
and associated comment on usage.
* rtsfind.ads: Add the following entries to tables RE_Id and
RE_Unit_Table:
Remove the following entries from tables RE_Id and RE_Unit_Table:
RO_PE_Set_Entry_Name RO_TS_Set_Entry_Name
* s-taskin.adb: Remove with clause for Ada.Unchecked_Deallocation.
(Free_Entry_Names_Array): Removed.
(Number_Of_Entries): New routine.
(Set_Entry_Names): New routine.
* s-taskin.ads: Rename type Entry_Names_Array to
Task_Entry_Names_Array. Rename type Entry_Names_Array_Access
to Task_Entry_Names_Access. Update the type of ACTB field
Entry_Names and add a comment on its protection status.
(Free_Entry_Names_Array): Removed.
(Number_Of_Entries): New routine.
(Set_Entry_Names): New routine.
* s-tassta.adb (Create_Task): Remove formal parameter
Build_Entry_Names. Do not allocate an array to hold the
string names of entries and families.
(Free_Entry_Names): Removed.
(Free_Task): Remove the call to Free_Entry_Names.
(Set_Entry_Name): Removed.
(Vulnerable_Free_Task): Remove the call to Free_Entry_Names.
* s-tassta.ads (Create_Task): Remove formal parameter
Build_Entry_Names along with associated comment.
(Set_Entry_Name): Removed.
* s-tpoben.adb: Remove with clause for Ada.Unchecked_Deallocation.
(Finalize): Remove the call to Free_Entry_Names.
(Free_Entry_Names): Removed.
(Initialize_Protection_Entries):
Remove formal parameter Build_Entry_Names. Do not allocate
an array to hold the string names of entries and families.
(Number_Of_Entries): New routine.
(Set_Entry_Name): Removed.
(Set_Entry_Names): New routine.
* s-tpoben.ads: Add types Protected_Entry_Names_Array and
Protected_Entry_Names_Access. Update the type of Protection_Enties
field Entry_Names.
(Initialize_Protection_Entries): Remove
formal parameter Build_Entry_Names along with associated comment.
(Number_Of_Entries): New routine.
(Set_Entry_Name): Removed.
(Set_Entry_Names): New routine.
projects / thirdparty / gcc.git / log
