Re: [PATCH] Fix vectorizable_reduction (PR tree-optimization/56350)
On Mon, 18 Feb 2013, Jakub Jelinek wrote:
Hi!
On the following testcase, a = a / 8; looks like reasonable reduction
statement, but we pattern recognize it as
patt_1 = a 0 ? 7 : 0;
patt_2 = a + patt_1;
a = patt2 3;
and in the first pattern stmt the reduction operand is in the condition of
COND_EXPR, not operands as vectorizable_reduction was asserting. Generally,
it couldn't be used anywhere in a pattern stmt, it could be just some
preparation statement.
Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok for
trunk?
Ok.
Thanks,
Richard.
2013-02-18 Jakub Jelinek ja...@redhat.com
PR tree-optimization/56350
* tree-vect-loop.c (vectorizable_reduction): If orig_stmt, return false
if haven't found reduction or nested cycle operand, rather than
asserting we must find it.
* gcc.dg/pr56350.c: New test.
--- gcc/tree-vect-loop.c.jj 2013-01-15 10:30:19.0 +0100
+++ gcc/tree-vect-loop.c 2013-02-18 11:32:30.306566358 +0100
@@ -4692,7 +4692,7 @@ vectorizable_reduction (gimple stmt, gim
The last use is the reduction variable. In case of nested cycle this
assumption is not true: we use reduc_index to record the index of the
reduction variable. */
- for (i = 0; i op_type-1; i++)
+ for (i = 0; i op_type - 1; i++)
{
/* The condition of COND_EXPR is checked in vectorizable_condition().
*/
if (i == 0 code == COND_EXPR)
@@ -4724,11 +4724,18 @@ vectorizable_reduction (gimple stmt, gim
if (!vectype_in)
vectype_in = tem;
gcc_assert (is_simple_use);
- gcc_assert (dt == vect_reduction_def
- || dt == vect_nested_cycle
- || ((dt == vect_internal_def || dt == vect_external_def
- || dt == vect_constant_def || dt == vect_induction_def)
-nested_cycle found_nested_cycle_def));
+ if (!(dt == vect_reduction_def
+ || dt == vect_nested_cycle
+ || ((dt == vect_internal_def || dt == vect_external_def
+ || dt == vect_constant_def || dt == vect_induction_def)
+ nested_cycle found_nested_cycle_def)))
+{
+ /* For pattern recognized stmts, orig_stmt might be a reduction,
+ but some helper statements for the pattern might not, or
+ might be COND_EXPRs with reduction uses in the condition. */
+ gcc_assert (orig_stmt);
+ return false;
+}
if (!found_nested_cycle_def)
reduc_def_stmt = def_stmt;
--- gcc/testsuite/gcc.dg/pr56350.c.jj 2013-02-18 11:30:12.117327599 +0100
+++ gcc/testsuite/gcc.dg/pr56350.c2013-02-18 11:29:59.0 +0100
@@ -0,0 +1,13 @@
+/* PR tree-optimization/56350 */
+/* { dg-do compile } */
+/* { dg-options -O -ftree-vectorize } */
+
+int a, b, c;
+
+void
+f (void)
+{
+ for (; c; c++)
+for (b = 0; b 2; b++)
+ a /= 8;
+}
Jakub
--
Richard Biener rguent...@suse.de
SUSE / SUSE Labs
SUSE LINUX Products GmbH - Nuernberg - AG Nuernberg - HRB 16746
GF: Jeff Hawn, Jennifer Guild, Felix Imend
Re: [PATCH] Fix create_expression_by_pieces (PR tree-optimization/56381)
On Mon, 18 Feb 2013, Jeff Law wrote: On 02/18/13 15:32, Jakub Jelinek wrote: Hi! This accesses one past the last initialized argument. Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk? 2013-02-18 Jakub Jelinek ja...@redhat.com PR tree-optimization/56381 * tree-ssa-pre.c (create_expression_by_pieces): Fix up last argument to fold_build3. * g++.dg/opt/pr56381.C: New test. OK. Arguably this could have gone in under the obvious rule. Btw, the 4.7 branch is affected, too. Richard.
Re: [patch] [libffi] do not install libffi library, headers and documentation
On Mon, Feb 18, 2013 at 6:02 PM, Matthias Klose d...@ubuntu.com wrote: Am 12.02.2013 13:45, schrieb Richard Biener: On Tue, Feb 12, 2013 at 1:44 PM, Richard Biener richard.guent...@gmail.com wrote: On Tue, Feb 12, 2013 at 1:30 PM, Matthias Klose d...@ubuntu.com wrote: The libffi library, headers and documentation are still installed, although libffi provides separate releases for a long time. So do not install these anymore as part of a GCC install. Tested with a build and an install with go and java enabled (both using libffi_convenience). Ok for the trunk? openSUSE is using the GCC provided libffi, so no, this is not ok (not at this stage anyway). Also proper not-installing libffi would work by disabling the maybe-install-target-libffi at the toplevel, not changing libffi makfiles (which are supposed to be imported from upstream, no?) Thus, add no_install= true; to the libffi target module updated patch attached, checked with a make install that no ffi headers and libraries are installed. If not ok for 4.8, ok for 4.9 when it opens? I'm fine with that variant but I'd like to see another ok. No preference as to whether to target 4.8 or 4.9. Richard. Matthias
Re: [PATCH, PR 55334] Disable IPA-CP on restrict pointers to arrays
On Mon, Feb 18, 2013 at 9:06 PM, Martin Jambor mjam...@suse.cz wrote:
Hi,
after much pondering about PR 55334 I came to conclusion that no nice
fix to the regression could be introduced in stage4. So for the sake
of the SPEC 200 benchmark I decided to cripple IPA-CP on restrict
pointers to arrays so that the restrict-ness of the memory references
which need to be vectorized is not lost.
I have tried various less aggressive tricks like only propagating the
one constant address that heuristics decided was profitable and not
all the others that came from the same calling contexts (and were
restrict) but then inlining came along and finished the deed with the
same consequences as IPA-CP had.
That only reinforced my feeling that the patch below is a spec hack
and that we ought to find a better way of preserving the restrict-ness
across IPA optimizations. Therefore I will keep the bug open and
revert this patch after 4.8 is branched and stage1 opens.
Meanwhile, this passes bootstrap and testsuite on x86_64-linux. OK
for trunk?
Thanks,
Martin
2013-02-07 Martin Jambor mjam...@suse.cz
PR tree-optimization/55334
* ipa-cp.c (initialize_node_lattices): Disable IPA-CP through and to
restricted pointers to arrays.
Index: src/gcc/ipa-cp.c
===
--- src.orig/gcc/ipa-cp.c
+++ src/gcc/ipa-cp.c
@@ -730,6 +730,22 @@ initialize_node_lattices (struct cgraph_
cgraph_node_name (node), node-uid,
disable ? BOTTOM : VARIABLE);
}
+ if (!disable)
+for (i = 0; i ipa_get_param_count (info) ; i++)
+ {
+ struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
+ tree t = TREE_TYPE (ipa_get_param(info, i));
+
+ if (TYPE_RESTRICT (t) POINTER_TYPE_P (t)
You want to first test for POINTER_TYPE_P and then for TYPE_RESTRICT.
+TREE_CODE (TREE_TYPE (t)) == ARRAY_TYPE)
With only testing for ARRAY_TYPE here you will still IPA-CP array
descriptors, right? Those are RECORD_TYPE, the parameter being
DECL_BY_REFERENCE. So I'd try instead
DECL_BY_REFERENCE (ipa_get_parm (info, i))
which leads me to the question - what is ipa_get_parm (info, i) returning?
Is it the lattice _value_? Or is it the PARM_DECL of the callee (which is
the only important piece of information!)?
Thanks,
Richard.
+ {
+ set_lattice_to_bottom (plats-itself);
+ if (dump_file (dump_flags TDF_DETAILS)
+!node-alias !node-thunk.thunk_p)
+ fprintf (dump_file, Going to ignore param %i of of %s/%i.\n,
+ i, cgraph_node_name (node), node-uid);
+ }
+ }
for (ie = node-indirect_calls; ie; ie = ie-next_callee)
if (ie-indirect_info-polymorphic)
Re: [PATCH, PR 55334] Disable IPA-CP on restrict pointers to arrays
Hi,
On Tue, Feb 19, 2013 at 10:21:32AM +0100, Richard Biener wrote:
On Mon, Feb 18, 2013 at 9:06 PM, Martin Jambor mjam...@suse.cz wrote:
Hi,
after much pondering about PR 55334 I came to conclusion that no nice
fix to the regression could be introduced in stage4. So for the sake
of the SPEC 200 benchmark I decided to cripple IPA-CP on restrict
pointers to arrays so that the restrict-ness of the memory references
which need to be vectorized is not lost.
I have tried various less aggressive tricks like only propagating the
one constant address that heuristics decided was profitable and not
all the others that came from the same calling contexts (and were
restrict) but then inlining came along and finished the deed with the
same consequences as IPA-CP had.
That only reinforced my feeling that the patch below is a spec hack
and that we ought to find a better way of preserving the restrict-ness
across IPA optimizations. Therefore I will keep the bug open and
revert this patch after 4.8 is branched and stage1 opens.
Meanwhile, this passes bootstrap and testsuite on x86_64-linux. OK
for trunk?
Thanks,
Martin
2013-02-07 Martin Jambor mjam...@suse.cz
PR tree-optimization/55334
* ipa-cp.c (initialize_node_lattices): Disable IPA-CP through and to
restricted pointers to arrays.
Index: src/gcc/ipa-cp.c
===
--- src.orig/gcc/ipa-cp.c
+++ src/gcc/ipa-cp.c
@@ -730,6 +730,22 @@ initialize_node_lattices (struct cgraph_
cgraph_node_name (node), node-uid,
disable ? BOTTOM : VARIABLE);
}
+ if (!disable)
+for (i = 0; i ipa_get_param_count (info) ; i++)
+ {
+ struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info, i);
+ tree t = TREE_TYPE (ipa_get_param(info, i));
+
+ if (TYPE_RESTRICT (t) POINTER_TYPE_P (t)
You want to first test for POINTER_TYPE_P and then for TYPE_RESTRICT.
I see, will do.
+TREE_CODE (TREE_TYPE (t)) == ARRAY_TYPE)
With only testing for ARRAY_TYPE here you will still IPA-CP array
descriptors, right? Those are RECORD_TYPE, the parameter being
DECL_BY_REFERENCE.
Possibly, why do you think that matters? In reality, can the
descriptors ever be global variables and thus propagated? (This is
all about the old scalar IPA-CP, it has nothing to do with the new
aggregate one. It's just that in 4.8 heuristics decides to clone
since an otherwise unrelated patch in r190015. In fact, aggregate
IPA-CP is not disabled by this patch).
So I'd try instead
DECL_BY_REFERENCE (ipa_get_parm (info, i))
This would also disable propagation of simple scalar constants passed
by reference the Fortran way (I have just checked by looking at of PR
53787) which I specifically did not want to do. Therefore, I'd rather
stick to my original proposal (modulo the order of pointer and
restrict tests order).
which leads me to the question - what is ipa_get_parm (info, i) returning?
Is it the lattice _value_? Or is it the PARM_DECL of the callee (which is
the only important piece of information!)?
The PARM_DECL.
Thanks,
Martin
Re: [PATCH, PR 55334] Disable IPA-CP on restrict pointers to arrays
On Tue, Feb 19, 2013 at 12:22 PM, Martin Jambor mjam...@suse.cz wrote:
Hi,
On Tue, Feb 19, 2013 at 10:21:32AM +0100, Richard Biener wrote:
On Mon, Feb 18, 2013 at 9:06 PM, Martin Jambor mjam...@suse.cz wrote:
Hi,
after much pondering about PR 55334 I came to conclusion that no nice
fix to the regression could be introduced in stage4. So for the sake
of the SPEC 200 benchmark I decided to cripple IPA-CP on restrict
pointers to arrays so that the restrict-ness of the memory references
which need to be vectorized is not lost.
I have tried various less aggressive tricks like only propagating the
one constant address that heuristics decided was profitable and not
all the others that came from the same calling contexts (and were
restrict) but then inlining came along and finished the deed with the
same consequences as IPA-CP had.
That only reinforced my feeling that the patch below is a spec hack
and that we ought to find a better way of preserving the restrict-ness
across IPA optimizations. Therefore I will keep the bug open and
revert this patch after 4.8 is branched and stage1 opens.
Meanwhile, this passes bootstrap and testsuite on x86_64-linux. OK
for trunk?
Thanks,
Martin
2013-02-07 Martin Jambor mjam...@suse.cz
PR tree-optimization/55334
* ipa-cp.c (initialize_node_lattices): Disable IPA-CP through and
to
restricted pointers to arrays.
Index: src/gcc/ipa-cp.c
===
--- src.orig/gcc/ipa-cp.c
+++ src/gcc/ipa-cp.c
@@ -730,6 +730,22 @@ initialize_node_lattices (struct cgraph_
cgraph_node_name (node), node-uid,
disable ? BOTTOM : VARIABLE);
}
+ if (!disable)
+for (i = 0; i ipa_get_param_count (info) ; i++)
+ {
+ struct ipcp_param_lattices *plats = ipa_get_parm_lattices (info,
i);
+ tree t = TREE_TYPE (ipa_get_param(info, i));
+
+ if (TYPE_RESTRICT (t) POINTER_TYPE_P (t)
You want to first test for POINTER_TYPE_P and then for TYPE_RESTRICT.
I see, will do.
+TREE_CODE (TREE_TYPE (t)) == ARRAY_TYPE)
With only testing for ARRAY_TYPE here you will still IPA-CP array
descriptors, right? Those are RECORD_TYPE, the parameter being
DECL_BY_REFERENCE.
Possibly, why do you think that matters? In reality, can the
descriptors ever be global variables and thus propagated? (This is
all about the old scalar IPA-CP, it has nothing to do with the new
aggregate one. It's just that in 4.8 heuristics decides to clone
since an otherwise unrelated patch in r190015. In fact, aggregate
IPA-CP is not disabled by this patch).
I'm not sure about global array descriptors but I seem to remember seeing
those.
So I'd try instead
DECL_BY_REFERENCE (ipa_get_parm (info, i))
This would also disable propagation of simple scalar constants passed
by reference the Fortran way (I have just checked by looking at of PR
53787) which I specifically did not want to do. Therefore, I'd rather
stick to my original proposal (modulo the order of pointer and
restrict tests order).
Ok, that makes sense.
which leads me to the question - what is ipa_get_parm (info, i) returning?
Is it the lattice _value_? Or is it the PARM_DECL of the callee (which is
the only important piece of information!)?
The PARM_DECL.
Fine.
The patch is ok then, with the restrict / pointer test re-ordered.
Thanks,
Richard.
Thanks,
Martin
[PATCH] Fix PR56384
This fixes PHI value-numbering which did not correctly distinguish
types in case of still VN_TOP PHI arguments.
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied.
Richard.
2013-02-19 Richard Biener rguent...@suse.de
PR tree-optimization/56384
* tree-ssa-sccvn.h (struct vn_phi_s): Add type member.
(vn_hash_type): Split out from ...
(vn_hash_constant_with_type): ... here.
* tree-ssa-sccvn.c (vn_phi_compute_hash): Use vn_hash_type.
(vn_phi_eq): Compare types from vn_phi_s structure.
(vn_phi_lookup): Populate vn_phi_s type.
(vn_phi_insert): Likewise.
* gcc.dg/torture/pr56384.c: New testcase.
Index: gcc/tree-ssa-sccvn.h
===
*** gcc/tree-ssa-sccvn.h(revision 196134)
--- gcc/tree-ssa-sccvn.h(working copy)
*** typedef struct vn_phi_s
*** 67,72
--- 67,73
hashval_t hashcode;
vectree phiargs;
basic_block block;
+ tree type;
tree result;
} *vn_phi_t;
typedef const struct vn_phi_s *const_vn_phi_t;
*** typedef struct vn_constant_s
*** 122,138
enum vn_kind { VN_NONE, VN_CONSTANT, VN_NARY, VN_REFERENCE, VN_PHI };
enum vn_kind vn_get_stmt_kind (gimple);
/* Hash the constant CONSTANT with distinguishing type incompatible
constants in the types_compatible_p sense. */
static inline hashval_t
vn_hash_constant_with_type (tree constant)
{
- tree type = TREE_TYPE (constant);
return (iterative_hash_expr (constant, 0)
! + INTEGRAL_TYPE_P (type)
! + (INTEGRAL_TYPE_P (type)
!? TYPE_PRECISION (type) + TYPE_UNSIGNED (type) : 0));
}
/* Compare the constants C1 and C2 with distinguishing type incompatible
--- 123,147
enum vn_kind { VN_NONE, VN_CONSTANT, VN_NARY, VN_REFERENCE, VN_PHI };
enum vn_kind vn_get_stmt_kind (gimple);
+ /* Hash the type TYPE using bits that distinguishes it in the
+types_compatible_p sense. */
+
+ static inline hashval_t
+ vn_hash_type (tree type)
+ {
+ return (INTEGRAL_TYPE_P (type)
+ + (INTEGRAL_TYPE_P (type)
+? TYPE_PRECISION (type) + TYPE_UNSIGNED (type) : 0));
+ }
+
/* Hash the constant CONSTANT with distinguishing type incompatible
constants in the types_compatible_p sense. */
static inline hashval_t
vn_hash_constant_with_type (tree constant)
{
return (iterative_hash_expr (constant, 0)
! + vn_hash_type (TREE_TYPE (constant)));
}
/* Compare the constants C1 and C2 with distinguishing type incompatible
Index: gcc/tree-ssa-sccvn.c
===
*** gcc/tree-ssa-sccvn.c(revision 196134)
--- gcc/tree-ssa-sccvn.c(working copy)
*** vn_phi_compute_hash (vn_phi_t vp1)
*** 2401,2410
/* If all PHI arguments are constants we need to distinguish
the PHI node via its type. */
! type = TREE_TYPE (vp1-phiargs[0]);
! result += (INTEGRAL_TYPE_P (type)
!+ (INTEGRAL_TYPE_P (type)
! ? TYPE_PRECISION (type) + TYPE_UNSIGNED (type) : 0));
FOR_EACH_VEC_ELT (vp1-phiargs, i, phi1op)
{
--- 2401,2408
/* If all PHI arguments are constants we need to distinguish
the PHI node via its type. */
! type = vp1-type;
! result += vn_hash_type (type);
FOR_EACH_VEC_ELT (vp1-phiargs, i, phi1op)
{
*** vn_phi_eq (const void *p1, const void *p
*** 2443,2450
/* If the PHI nodes do not have compatible types
they are not the same. */
! if (!types_compatible_p (TREE_TYPE (vp1-phiargs[0]),
! TREE_TYPE (vp2-phiargs[0])))
return false;
/* Any phi in the same block will have it's arguments in the
--- 2441,2447
/* If the PHI nodes do not have compatible types
they are not the same. */
! if (!types_compatible_p (vp1-type, vp2-type))
return false;
/* Any phi in the same block will have it's arguments in the
*** vn_phi_lookup (gimple phi)
*** 2484,2489
--- 2481,2487
def = TREE_CODE (def) == SSA_NAME ? SSA_VAL (def) : def;
shared_lookup_phiargs.safe_push (def);
}
+ vp1.type = TREE_TYPE (gimple_phi_result (phi));
vp1.phiargs = shared_lookup_phiargs;
vp1.block = gimple_bb (phi);
vp1.hashcode = vn_phi_compute_hash (vp1);
*** vn_phi_insert (gimple phi, tree result)
*** 2516,2521
--- 2514,2520
args.safe_push (def);
}
vp1-value_id = VN_INFO (result)-value_id;
+ vp1-type = TREE_TYPE (gimple_phi_result (phi));
vp1-phiargs = args;
vp1-block = gimple_bb (phi);
vp1-result = result;
[PATCH] Fix PR 55889
Hello,
As we discussed in the PR, the problem here is that the selective scheduler
does not expect that renaming a hard register to a pseudo would result in
extra dependencies. The dependencies come from implicit clobbers code of
sched-deps.c, so I've made a minimal patch that checks only for that case
using the same function from IRA and avoids it.
The patch fixes the test case on AIX as reported by David and also
bootstraps and tests fine on x86-64 and ia64, ok for trunk?
The second patch is restoring debug printing in the scheduler that was
accidentally broken, possibly by Steven's cleanups (now a pattern is
expected where an insn was previously). I will commit is as obvious
separately.
Andrey
2012-02-19 Andrey Belevantsev a...@ispras.ru
PR middle-end/55889
* sel-sched.c: Include ira.h.
(implicit_clobber_conflict_p): New function.
(moveup_expr): Use it.
* Makefile.in (sel-sched.o): Depend on ira.h.
And the second patch is:
2012-02-19 Andrey Belevantsev a...@ispras.ru
* sel-sched-dump.c (dump_insn_rtx_flags): Explicitly set
DUMP_INSN_RTX_UID.
(dump_insn_rtx_1): Pass PATTERN (insn) to str_pattern_slim.
diff --git a/gcc/Makefile.in b/gcc/Makefile.in
index f3bb168..557ab08 100644
--- a/gcc/Makefile.in
+++ b/gcc/Makefile.in
@@ -3338,7 +3338,7 @@ sel-sched.o : sel-sched.c $(CONFIG_H) $(SYSTEM_H) coretypes.h $(TM_H) \
$(FUNCTION_H) $(INSN_ATTR_H) $(RECOG_H) $(EXCEPT_H) $(PARAMS_H) \
$(TM_P_H) output.h $(TARGET_H) $(TREE_PASS_H) \
$(SCHED_INT_H) $(GGC_H) $(TREE_H) langhooks.h rtlhooks-def.h \
- $(SEL_SCHED_IR_H) $(SEL_SCHED_DUMP_H) sel-sched.h $(DBGCNT_H) $(EMIT_RTL_H)
+ $(SEL_SCHED_IR_H) $(SEL_SCHED_DUMP_H) sel-sched.h $(DBGCNT_H) $(EMIT_RTL_H) ira.h
sel-sched-dump.o : sel-sched-dump.c $(CONFIG_H) $(SYSTEM_H) coretypes.h $(TM_H) \
$(RTL_H) $(REGS_H) hard-reg-set.h $(FLAGS_H) insn-config.h \
$(FUNCTION_H) $(INSN_ATTR_H) $(DIAGNOSTIC_CORE_H) $(RECOG_H) $(EXCEPT_H) $(PARAMS_H) \
diff --git a/gcc/sel-sched.c b/gcc/sel-sched.c
index 6f60d70..2944489 100644
--- a/gcc/sel-sched.c
+++ b/gcc/sel-sched.c
@@ -41,6 +41,7 @@ along with GCC; see the file COPYING3. If not see
#include langhooks.h
#include rtlhooks-def.h
#include emit-rtl.h
+#include ira.h
#ifdef INSN_SCHEDULING
#include sel-sched-ir.h
@@ -2101,6 +2102,61 @@ moving_insn_creates_bookkeeping_block_p (insn_t insn,
return TRUE;
}
+/* Return true when the conflict with newly created implicit clobbers
+ between EXPR and THROUGH_INSN is found because of renaming. */
+static bool
+implicit_clobber_conflict_p (insn_t through_insn, expr_t expr)
+{
+ HARD_REG_SET temp;
+ rtx insn, reg, rhs, pat;
+ hard_reg_set_iterator hrsi;
+ unsigned regno;
+ bool valid;
+
+ /* Make a new pseudo register. */
+ reg = gen_reg_rtx (GET_MODE (EXPR_LHS (expr)));
+ max_regno = max_reg_num ();
+ maybe_extend_reg_info_p ();
+
+ /* Validate a change and bail out early. */
+ insn = EXPR_INSN_RTX (expr);
+ validate_change (insn, SET_DEST (PATTERN (insn)), reg, 1);
+ valid = verify_changes (0);
+ cancel_changes (0);
+ if (!valid)
+{
+ if (sched_verbose = 6)
+ sel_print (implicit clobbers failed validation, );
+ return true;
+}
+
+ /* Make a new insn with it. */
+ rhs = copy_rtx (VINSN_RHS (EXPR_VINSN (expr)));
+ pat = gen_rtx_SET (VOIDmode, reg, rhs);
+ start_sequence ();
+ insn = emit_insn (pat);
+ end_sequence ();
+
+ /* Calculate implicit clobbers. */
+ extract_insn (insn);
+ preprocess_constraints ();
+ ira_implicitly_set_insn_hard_regs (temp);
+ AND_COMPL_HARD_REG_SET (temp, ira_no_alloc_regs);
+
+ /* If any implicit clobber registers intersect with regular ones in
+ through_insn, we have a dependency and thus bail out. */
+ EXECUTE_IF_SET_IN_HARD_REG_SET (temp, 0, regno, hrsi)
+{
+ vinsn_t vi = INSN_VINSN (through_insn);
+ if (bitmap_bit_p (VINSN_REG_SETS (vi), regno)
+ || bitmap_bit_p (VINSN_REG_CLOBBERS (vi), regno)
+ || bitmap_bit_p (VINSN_REG_USES (vi), regno))
+ return true;
+}
+
+ return false;
+}
+
/* Modifies EXPR so it can be moved through the THROUGH_INSN,
performing necessary transformations. Record the type of transformation
made in PTRANS_TYPE, when it is not NULL. When INSIDE_INSN_GROUP,
@@ -2233,6 +2289,17 @@ moveup_expr (expr_t expr, insn_t through_insn, bool inside_insn_group,
if (!enable_schedule_as_rhs_p || !EXPR_SEPARABLE_P (expr))
return MOVEUP_EXPR_NULL;
+ /* When renaming a hard register to a pseudo before reload, extra
+ dependencies can occur from the implicit clobbers of the insn.
+ Filter out such cases here. */
+ if (!reload_completed REG_P (EXPR_LHS (expr))
+ HARD_REGISTER_P (EXPR_LHS (expr))
+ implicit_clobber_conflict_p (through_insn, expr))
+ {
+ if (sched_verbose = 6)
+ sel_print (implicit clobbers conflict detected, );
+ return MOVEUP_EXPR_NULL;
+ }
EXPR_TARGET_AVAILABLE (expr) =
[4.7, testsuite] Clear hwcap_2 with Sun ld
I'd like to backport [testsuite] Clear hwcap_2 with Sun ld http://gcc.gnu.org/ml/gcc-patches/2012-06/msg01246.html 2012-06-19 Rainer Orth r...@cebitec.uni-bielefeld.de * gcc.target/i386/clearcapv2.map: New file. * gcc.target/i386/i386.exp: Try it first before clearcap.map. to the 4.7 branch since recently all avx2 tests started to FAIL there on Solaris 10 and 11/x86 with Sun as. It applies cleanly as is, bootstrapped without regressions on i386-pc-solaris2.10 and x86_64-unknown-linux-gnu. Ok for 4.7 branch? Rainer -- - Rainer Orth, Center for Biotechnology, Bielefeld University
Re: [PATCH] Fix PR 55889
On Tue, 19 Feb 2013, Andrey Belevantsev wrote: Hello, As we discussed in the PR, the problem here is that the selective scheduler does not expect that renaming a hard register to a pseudo would result in extra dependencies. The dependencies come from implicit clobbers code of sched-deps.c, so I've made a minimal patch that checks only for that case using the same function from IRA and avoids it. The patch fixes the test case on AIX as reported by David and also bootstraps and tests fine on x86-64 and ia64, ok for trunk? OK (with 1 changed to true in arguments to validate_change as discussed offline), thanks. -- Alexander
[PATCH] Speedup IVOPTs
This speeds up IVOPTs by optimizing its hottest function when compiling
polyhedron linpk. The datastructure used for recording use, candidate
costs (a hashtable) should make O(1) queries on average - but it turns
out that for use, candidate queries that have no entry recorded in
it it is O(n) currently. That's because the collision handling does
not abort when coming across an empty hashtable entry (oops).
Overall compile-time effect is noise (linpk compiles in less than
a second for me ...), but callgrind tells me it results in a 3%
compile-time improvement.
There is still the degenerate cases when the hashtable is (almost)
full (doesn't happen for this testcase), but that's an unrelated issue.
Bootstrap and regtest running on x86_64-unknown-linux-gnu.
Jakub, does this look ok for trunk now or shall I wait for stage1?
Thanks,
Richard.
2013-02-19 Richard Biener rguent...@suse.de
* tree-ssa-loop-ivopts.c (alloc_use_cost_map): Use bitmap_count_bits
and ceil_log2.
(get_use_iv_cost): Terminate hashtable walk when coming across
an empty entry.
Index: gcc/tree-ssa-loop-ivopts.c
===
*** gcc/tree-ssa-loop-ivopts.c (revision 196135)
--- gcc/tree-ssa-loop-ivopts.c (working copy)
*** record_important_candidates (struct ivop
*** 2574,2599
static void
alloc_use_cost_map (struct ivopts_data *data)
{
! unsigned i, size, s, j;
for (i = 0; i n_iv_uses (data); i++)
{
struct iv_use *use = iv_use (data, i);
- bitmap_iterator bi;
if (data-consider_all_candidates)
size = n_iv_cands (data);
else
{
! s = 0;
! EXECUTE_IF_SET_IN_BITMAP (use-related_cands, 0, j, bi)
! {
! s++;
! }
/* Round up to the power of two, so that moduling by it is fast. */
! for (size = 1; size s; size = 1)
! continue;
}
use-n_map_members = size;
--- 2574,2593
static void
alloc_use_cost_map (struct ivopts_data *data)
{
! unsigned i, size, s;
for (i = 0; i n_iv_uses (data); i++)
{
struct iv_use *use = iv_use (data, i);
if (data-consider_all_candidates)
size = n_iv_cands (data);
else
{
! s = bitmap_count_bits (use-related_cands);
/* Round up to the power of two, so that moduling by it is fast. */
! size = 1 ceil_log2 (s + 1);
}
use-n_map_members = size;
*** get_use_iv_cost (struct ivopts_data *dat
*** 2731,2740
for (i = s; i use-n_map_members; i++)
if (use-cost_map[i].cand == cand)
return use-cost_map + i;
!
for (i = 0; i s; i++)
if (use-cost_map[i].cand == cand)
return use-cost_map + i;
return NULL;
}
--- 2725,2737
for (i = s; i use-n_map_members; i++)
if (use-cost_map[i].cand == cand)
return use-cost_map + i;
! else if (use-cost_map[i].cand == NULL)
! return NULL;
for (i = 0; i s; i++)
if (use-cost_map[i].cand == cand)
return use-cost_map + i;
+ else if (use-cost_map[i].cand == NULL)
+ return NULL;
return NULL;
}
Re: [PATCH] Speedup IVOPTs
On Tue, Feb 19, 2013 at 02:59:46PM +0100, Richard Biener wrote: This speeds up IVOPTs by optimizing its hottest function when compiling polyhedron linpk. The datastructure used for recording use, candidate costs (a hashtable) should make O(1) queries on average - but it turns out that for use, candidate queries that have no entry recorded in it it is O(n) currently. That's because the collision handling does not abort when coming across an empty hashtable entry (oops). Overall compile-time effect is noise (linpk compiles in less than a second for me ...), but callgrind tells me it results in a 3% compile-time improvement. There is still the degenerate cases when the hashtable is (almost) full (doesn't happen for this testcase), but that's an unrelated issue. Bootstrap and regtest running on x86_64-unknown-linux-gnu. Jakub, does this look ok for trunk now or shall I wait for stage1? This: /* Round up to the power of two, so that moduling by it is fast. */ ! for (size = 1; size s; size = 1) ! continue; } use-n_map_members = size; isn't equivalent to: /* Round up to the power of two, so that moduling by it is fast. */ ! size = 1 ceil_log2 (s + 1); } size = s ? (1 ceil_log2 (s)) : 1; would be I think. Is it intentional that for power of 2 you set now size to double that value? --- 2725,2737 for (i = s; i use-n_map_members; i++) if (use-cost_map[i].cand == cand) return use-cost_map + i; ! else if (use-cost_map[i].cand == NULL) ! return NULL; for (i = 0; i s; i++) if (use-cost_map[i].cand == cand) return use-cost_map + i; + else if (use-cost_map[i].cand == NULL) + return NULL; return NULL; } This looks good to me, even at this point. Jakub
[PATCH] Speedup lookup_constraint for 2-letter constraints
This speeds up lookup_constraint by allowing the generic folding
for n == 1 strncmp calls to trigger, handling two-letter constraints
by two char comparisons instead of one and a call to strncmp (which
ends up not being inlined because it's considered cold ...).
Bootstrap regtest pending on x86_64-unknown-linux-gnu.
Ok?
Thanks,
Richard.
2013-02-19 Richard Biener rguent...@suse.de
* genpreds.c (write_lookup_constraint): Do not compare first
letter of the constraint again.
Index: gcc/genpreds.c
===
--- gcc/genpreds.c (revision 196134)
+++ gcc/genpreds.c (working copy)
@@ -945,9 +945,10 @@ write_lookup_constraint (void)
{
do
{
- printf ( if (!strncmp (str, \%s\, %lu))\n
+ printf ( if (!strncmp (str + 1, \%s\, %lu))\n
return CONSTRAINT_%s;\n,
- c-name, (unsigned long int) c-namelen, c-c_name);
+ c-name + 1, (unsigned long int) c-namelen - 1,
+ c-c_name);
c = c-next_this_letter;
}
while (c);
Re: [PATCH] Speedup IVOPTs
On Tue, 19 Feb 2013, Jakub Jelinek wrote: On Tue, Feb 19, 2013 at 02:59:46PM +0100, Richard Biener wrote: This speeds up IVOPTs by optimizing its hottest function when compiling polyhedron linpk. The datastructure used for recording use, candidate costs (a hashtable) should make O(1) queries on average - but it turns out that for use, candidate queries that have no entry recorded in it it is O(n) currently. That's because the collision handling does not abort when coming across an empty hashtable entry (oops). Overall compile-time effect is noise (linpk compiles in less than a second for me ...), but callgrind tells me it results in a 3% compile-time improvement. There is still the degenerate cases when the hashtable is (almost) full (doesn't happen for this testcase), but that's an unrelated issue. Bootstrap and regtest running on x86_64-unknown-linux-gnu. Jakub, does this look ok for trunk now or shall I wait for stage1? This: /* Round up to the power of two, so that moduling by it is fast. */ ! for (size = 1; size s; size = 1) ! continue; } use-n_map_members = size; isn't equivalent to: /* Round up to the power of two, so that moduling by it is fast. */ ! size = 1 ceil_log2 (s + 1); } size = s ? (1 ceil_log2 (s)) : 1; would be I think. Is it intentional that for power of 2 you set now size to double that value? Well, I used the lame variant to avoid a zero argument to ceil_log2 ... I'll use yours instead. Ok? Thanks, Richard. --- 2725,2737 for (i = s; i use-n_map_members; i++) if (use-cost_map[i].cand == cand) return use-cost_map + i; ! else if (use-cost_map[i].cand == NULL) ! return NULL; for (i = 0; i s; i++) if (use-cost_map[i].cand == cand) return use-cost_map + i; + else if (use-cost_map[i].cand == NULL) + return NULL; return NULL; } This looks good to me, even at this point. Jakub -- Richard Biener rguent...@suse.de SUSE / SUSE Labs SUSE LINUX Products GmbH - Nuernberg - AG Nuernberg - HRB 16746 GF: Jeff Hawn, Jennifer Guild, Felix Imend
Re: [PATCH] Speedup lookup_constraint for 2-letter constraints
On Tue, Feb 19, 2013 at 04:07:49PM +0100, Richard Biener wrote: This speeds up lookup_constraint by allowing the generic folding for n == 1 strncmp calls to trigger, handling two-letter constraints by two char comparisons instead of one and a call to strncmp (which ends up not being inlined because it's considered cold ...). Bootstrap regtest pending on x86_64-unknown-linux-gnu. Ok? Yes, thanks. 2013-02-19 Richard Biener rguent...@suse.de * genpreds.c (write_lookup_constraint): Do not compare first letter of the constraint again. Jakub
Re: [PATCH] Speedup IVOPTs
On Tue, Feb 19, 2013 at 04:08:57PM +0100, Richard Biener wrote: Well, I used the lame variant to avoid a zero argument to ceil_log2 ... I'll use yours instead. Ok? Yep. Jakub
Speedup recognizing multi-letter constraints
Hi,
from IRC:
[15:45:21] richi ick - lookup_constraint for multi-letter constraints
is quite expensive ... strncmp is not expanded inline for some reason
Instead of fiddling with strncmp inlining, simply generate better code
from the start for two character constraints:
switch (str[0]) {
case 'Y':
switch (str[1])
{
case 'i':
return CONSTRAINT_Yi;
case 'm':
return CONSTRAINT_Ym;
...
Ciao,
Michael.
--
* genpreds (write_lookup_constraint): Special case two-character
constraints to also expand to a switch.
Index: genpreds.c
===
--- genpreds.c (revision 196053)
+++ genpreds.c (working copy)
@@ -941,6 +941,22 @@ write_lookup_constraint (void)
printf (case '%c':\n, i);
if (c-namelen == 1)
printf ( return CONSTRAINT_%s;\n, c-c_name);
+ else if (c-namelen == 2)
+ {
+ puts ( switch (str[1])\n
+ {);
+ do
+ {
+ printf (case '%c':\n
+ return CONSTRAINT_%s;\n,
+ c-name[1], c-c_name);
+ c = c-next_this_letter;
+ }
+ while (c);
+ puts (default: break;\n
+ }\n
+ break;);
+ }
else
{
do
Re: RFC: [PATCH,ARM] Fix 56110
On 18/02/13 21:47, Tilman Sauerbeck wrote: Hi, adding the instruction pattern below fixes my testcase for PR 56110; however I'm not sure if adding a new pattern is the correct way to go. I duplicated the andsi3_compare0_scratch pattern, and lifted the requirement that the 2nd operand be an arm_not_operand. I didn't copy over the clobber pattern because I don't know what it does ;) Comments? Can anyone put me in the right direction here? Or take my humble attempt and massage it into the proper fix? I can provide a dejagnu testcase if it helps. No ChangeLog entry because I know this diff won't go anywhere in its current incarnation. Thanks, Tilman diff --git a/gcc/config/arm/arm.md b/gcc/config/arm/arm.md index 64888f9..e47f8f7 100644 --- a/gcc/config/arm/arm.md +++ b/gcc/config/arm/arm.md @@ -2212,6 +2212,19 @@ (set_attr type simple_alu_imm,simple_alu_imm,*)] ) +(define_insn *andsi3_compare0_scratch2 + [(set (reg:CC_NOOV CC_REGNUM) + (compare:CC_NOOV +(and:SI (match_operand:SI 0 s_register_operand r,r,r) +(match_operand:SI 1 const_int_operand r,r,r)) +(const_int 0)))] + TARGET_32BIT + @ + tst%?\\t%0, %1 + [(set_attr conds set) + (set_attr type simple_alu_imm,simple_alu_imm,*)] +) + (define_insn *zeroextractsi_compare0_scratch [(set (reg:CC_NOOV CC_REGNUM) (compare:CC_NOOV (zero_extract:SI Sorry, this is not correct. You've got a constraint that requires a const_int_operand, but then specified only a register. This will force the compiler to reload the immediate operand into a register, just negating any saving you've made from getting rid of the compare instruction. Then you've added three identical variants all using two registers of the same class. However, the question you need to be asking is why the pattern immediately before the one you've added is not matching. The compiler knows how to add clobbers, so I'm surprised that you're finding a new pattern to be necessary. R.
Ping: [Patch,avr]: Add more devices take #2: 2/2
CCing Richard in case this needs RM approval (I don't think so because the patch just changes ternary parts). Ping #1 for: http://gcc.gnu.org/ml/gcc-patches/2013-02/msg00579.html Georg-Johann Lay schrieb: And here 2/2 with the device - arch mapping for gas. Ok for trunk? Johann * config/avr/avr.h (device_to_arch): Rename to device_to_ld. (avr_device_to_arch): Rename to avr_device_to_ld. (avr_device_to_as): New prototype. (EXTRA_SPEC_FUNCTIONS): Add device_to_as. (ASM_SPEC): Use device_to_as to get -mmcu= and -mno-skip-bug=. * config/avr/driver-avr.c (avr_device_to_as): New. (avr_device_to_arch): Rename to avr_device_to_ld.
Ping: [Patch,avr]: Add more devices take #2: 1/2
CCing Richard in case this needs RM approval (I don't think so because
the patch just changes ternary parts).
Ping #1 for:
http://gcc.gnu.org/ml/gcc-patches/2013-02/msg00577.html
http://gcc.gnu.org/ml/gcc-patches/2013-02/msg00576.html
Georg-Johann Lay schrieb:
Better attach the patch...
Georg-Johann Lay wrote:
Let me drop this, I was thinking too complicated.
src/gas/config/tc-avr.c
reads:
static struct mcu_type_s mcu_types[] =
{
{avr1, AVR_ISA_AVR1,bfd_mach_avr1},
/* TODO: insruction set for avr2 architecture should be AVR_ISA_AVR2,
but set to AVR_ISA_AVR25 for some following version
of GCC (from 4.3) for backward compatibility. */
{avr2, AVR_ISA_AVR25, bfd_mach_avr2},
{avr25, AVR_ISA_AVR25, bfd_mach_avr25},
/* TODO: insruction set for avr3 architecture should be AVR_ISA_AVR3,
but set to AVR_ISA_AVR3_ALL for some following version
of GCC (from 4.3) for backward compatibility. */
{avr3, AVR_ISA_AVR3_ALL, bfd_mach_avr3},
{avr31, AVR_ISA_AVR31, bfd_mach_avr31},
{avr35, AVR_ISA_AVR35, bfd_mach_avr35},
{avr4, AVR_ISA_AVR4,bfd_mach_avr4},
/* TODO: insruction set for avr5 architecture should be AVR_ISA_AVR5,
but set to AVR_ISA_AVR51 for some following version
of GCC (from 4.3) for backward compatibility. */
{avr5, AVR_ISA_AVR51, bfd_mach_avr5},
{avr51, AVR_ISA_AVR51, bfd_mach_avr51},
{avr6, AVR_ISA_AVR6,bfd_mach_avr6},
This means that cores liek avr31, avr35 etc. are supported by binutils and
there is no more the need for avr-gcc to do mappings like avr31 - avr3.
It appears that avr-gcc implemented these mappings to overcome missing binutils
support for specific cores, and in turn binutils arranged for that kludge...
This means that avr_arch_t need not to be extended to hold the arch mapping for
gas calls and the gas call can use the core name from avr_arch_t.
I broke the patch into two parts:
1/2: New devices
2/2: Map device to core to disentangle gcc/binutils interdependencies.
Okay for trunk and 4.7 backport?
Johann
* avr/avr-mcus.def (ata5272, ata5505, attiny1634, ata6285)
(ata6286, atmega8a, atmega48pa, ata5790, ata5790n, ata5795)
(atmega164pa, atmega165pa, atmega168pa, atmega16hva, atmega16hvb)
(atmega16hvbrevb, atmega16m1, atmega16u4, atmega26hvg, atmega32a)
(atmega32a, atmega3250pa, atmega3290pa, atmega32c1, atmega32m1)
(atmega32u4, atmega32u6, atmega64a, atmega6490a, atmega6490p)
(atmega64c1, atmega64m1, atmega64rfa2, atmega64rfr2, atmega32hvb)
(atmega32hvbrevb, atmega16hva2, atmega48hvf, at90pwm161)
(atmega128a, atmega1284, atmxt112sl, atmxt224, atmxt224e)
(atmxt336s, atxmega16a4u, atxmega16c4, atxmega32a4u, atxmega32c4)
(atxmega32e5, atxmega64a3u, atxmega64a4u, atxmega64b1, atxmega64b3)
(atxmega64c3, atxmega64d4, atxmega128a3u, atxmega128b1)
(atxmega128b3, atxmega128c3, atxmega128d4, atmxt540s, atmxt540sreva)
(atxmega192a3u, atxmega192c3, atxmega256a3u, atxmega256c3)
(atxmega384c3, atxmega384d3, atxmega128a4u): New AVR_MCU.
(avrxmega6): Increase max flash segments from 5 to 6.
* config/avr/t-multilib: Regenerate.
* config/avr/avr-tables.opt: Regenerate.
* doc/avr-mmcu.texi: Regenerate.
Re: [PATCH] Temporarily revert Steven's PCH changes for 4.8 (PR pch/54117)
On Thu, Feb 14, 2013 at 7:09 PM, Joel Brobecker brobec...@adacore.com wrote: AFAICT, for gcc+gas it should already work with binutils that include the AdaCore patch for DWARF support. But this has apparently not been tested with AIX ld, and there are AdaCore local patches pending. http://sourceware.org/ml/binutils/2011-04/msg00250.html http://www.sourceware.org/ml/gdb/2013-01/msg00030.html What is the status of merging Adacore's patches for DWARF support on AIX? Sorry, David. I have been meaning to work on that ever since we talked about it... I spent some time today isolating the patches. I am currently testing the patches we wrote for support on AIX 5.3. I then see a few more patches to extend support to AIX 7.1 - section alignment constraints, and stuff related to TLS. Just for anyone who is curious, this is what I am testing. It is missing the pieces for the other tools such as objdump, for instance. But I think those can be submitted separately. Joel, Thanks! That's great progress on the Binutils side. What is the status of patches for GCC to generate DWARF for AIX XCOFF and do the AIX assembler and linker recognize, consume and process the directives produced by GCC correctly? Thanks, David
Re: [C++ Patch] PR 56373
... in case, we may want to do something like the below too.
Thanks,
Paolo.
///
Index: cp/cp-tree.h
===
--- cp/cp-tree.h(revision 196136)
+++ cp/cp-tree.h(working copy)
@@ -5834,6 +5834,7 @@ extern bool cast_valid_in_integral_constant_expres
extern bool cxx_type_hash_eq (const_tree, const_tree);
extern void cxx_print_statistics (void);
+extern bool maybe_warn_zero_as_null_pointer_constant (tree, location_t);
/* in ptree.c */
extern void cxx_print_xnode(FILE *, tree, int);
Index: cp/cvt.c
===
--- cp/cvt.c(revision 196136)
+++ cp/cvt.c(working copy)
@@ -203,11 +203,8 @@ cp_convert_to_pointer (tree type, tree expr, tsubs
if (null_ptr_cst_p (expr))
{
- if ((complain tf_warning)
- c_inhibit_evaluation_warnings == 0
- !NULLPTR_TYPE_P (TREE_TYPE (expr)))
- warning_at (loc, OPT_Wzero_as_null_pointer_constant,
- zero as null pointer constant);
+ if (complain tf_warning)
+ maybe_warn_zero_as_null_pointer_constant (expr, loc);
if (TYPE_PTRMEMFUNC_P (type))
return build_ptrmemfunc (TYPE_PTRMEMFUNC_FN_TYPE (type), expr, 0,
@@ -783,7 +780,11 @@ ocp_convert (tree type, tree expr, int convtype, i
return ignore_overflows (converted, e);
}
if (NULLPTR_TYPE_P (type) e null_ptr_cst_p (e))
-return nullptr_node;
+{
+ if (complain tf_warning)
+ maybe_warn_zero_as_null_pointer_constant (e, loc);
+ return nullptr_node;
+}
if (POINTER_TYPE_P (type) || TYPE_PTRMEM_P (type))
return fold_if_not_in_template (cp_convert_to_pointer (type, e, complain));
if (code == VECTOR_TYPE)
Index: cp/decl.c
===
--- cp/decl.c (revision 196136)
+++ cp/decl.c (working copy)
@@ -10861,15 +10861,10 @@ check_default_argument (tree decl, tree arg)
--cp_unevaluated_operand;
if (warn_zero_as_null_pointer_constant
- c_inhibit_evaluation_warnings == 0
TYPE_PTR_OR_PTRMEM_P (decl_type)
null_ptr_cst_p (arg)
- !NULLPTR_TYPE_P (TREE_TYPE (arg)))
-{
- warning (OPT_Wzero_as_null_pointer_constant,
- zero as null pointer constant);
- return nullptr_node;
-}
+ maybe_warn_zero_as_null_pointer_constant (arg, input_location))
+return nullptr_node;
/* [dcl.fct.default]
Index: cp/tree.c
===
--- cp/tree.c (revision 196136)
+++ cp/tree.c (working copy)
@@ -3939,6 +3939,21 @@ cp_tree_operand_length (const_tree t)
return TREE_OPERAND_LENGTH (t);
}
}
+
+/* Implement -Wzero_as_null_pointer_constant. Return true if the
+ conditions for the warning hold, false otherwise. */
+bool
+maybe_warn_zero_as_null_pointer_constant (tree expr, location_t loc)
+{
+ if (c_inhibit_evaluation_warnings == 0
+ !NULLPTR_TYPE_P (TREE_TYPE (expr)))
+{
+ warning_at (loc, OPT_Wzero_as_null_pointer_constant,
+ zero as null pointer constant);
+ return true;
+}
+ return false;
+}
#if defined ENABLE_TREE_CHECKING (GCC_VERSION = 2007)
/* Complain that some language-specific thing hanging off a tree
Index: cp/typeck.c
===
--- cp/typeck.c (revision 196136)
+++ cp/typeck.c (working copy)
@@ -4293,13 +4293,10 @@ cp_build_binary_op (location_t location,
delta0,
integer_one_node,
complain);
-
- if ((complain tf_warning)
- c_inhibit_evaluation_warnings == 0
- !NULLPTR_TYPE_P (TREE_TYPE (op1)))
- warning (OPT_Wzero_as_null_pointer_constant,
-zero as null pointer constant);
+ if (complain tf_warning)
+ maybe_warn_zero_as_null_pointer_constant (op1, input_location);
+
e2 = cp_build_binary_op (location,
EQ_EXPR, e2, integer_zero_node,
complain);
Index: testsuite/g++.dg/cpp0x/Wzero-as-null-pointer-constant-2.C
===
--- testsuite/g++.dg/cpp0x/Wzero-as-null-pointer-constant-2.C (revision 0)
+++ testsuite/g++.dg/cpp0x/Wzero-as-null-pointer-constant-2.C (working copy)
@@ -0,0 +1,14 @@
+// PR c++/56373
+// { dg-options -std=c++11 -Wzero-as-null-pointer-constant }
+
+struct shared_ptr
+{
+ shared_ptr(decltype(nullptr));
+};
+
+void f()
+{
+ shared_ptr a = 0; // { dg-warning zero as null pointer }
+ shared_ptr b(0); // { dg-warning zero as null pointer }
+ shared_ptr c{0}; // { dg-warning zero as null pointer }
+}
Re: [PATCH] Temporarily revert Steven's PCH changes for 4.8 (PR pch/54117)
Thanks! That's great progress on the Binutils side. You are very welcome. What is the status of patches for GCC to generate DWARF for AIX XCOFF and do the AIX assembler and linker recognize, consume and process the directives produced by GCC correctly? I haven't looked at this part in details, but our recollection is that it's a matter of configuring GCC with --with-gnu-as and --with-gnu-ld, together with a small patch to make DWARF the default. AFAIK, the rest should be working as expected. -- Joel
[Patch, fortran] [0/5] PR54730 ICE: confused by type-like fonctions
Hello, this is a fix for cases like:
program main
implicit none
intrinsic :: real
print *,(/ real(a = 1) /)
end
where `real(a = 1)' is initially parsed as a typespec, creating
a symbol for 'a' along the way. The match fails, and then it is parsed
as a constructor element and accepted that way. However, accepting the
statement implies accepting all the symbols created so far including 'a',
which is wrong and leads to the ICE.
To handle correctly this, we have to remove 'a' before proceeding with
the second parse attempt. However, we can't use gfc_undo_symbols, as
it would also remove 'b' in the following case.
b = (/ real(a = 1) /)
The fix proposed here implements a partial undo framework. It packs the
changed_syms and tentative_tbp variables into a single 'gfc_change_set'
struct, and makes it possible to have more than one of those structs,
organised as a stack. That change makes the current linked list
implementation using in-symbol 'tlink' pointer impractical as it prevents
the same symbol from being in more than one changeset. I don't really know
whether that is a true limitation, but have decided to lift it anyway by
registering the symbols in a vector instead. This makes backporting a bit
more difficult unfortunately; I will submit the (yet nonexisting) backport
patches separately.
The work is divided as follows:
1/5: Pack the changed_syms and tentative_tbp variables in a 'gfc_change_set'
struct and move to the vec API.
2/5: New function restore_old_symbol, extracted from gfc_undo_symbols.
3/5: Fix restore_old_symbol
4/5: Add support for more than one 'gfc_change_set' variable.
5/5: Fix gfc_match_array_constructor using the just introduced functions.
The patches are attached to the follow-up mails; the full diff is also provided
here.
Bootstrap-asan'ed and regression tested on x86_64-unknown-linux-gnu.
OK for trunk?
diff --git a/Make-lang.in b/Make-lang.in
index 3584dd8..8c9e7ea 100644
--- a/Make-lang.in
+++ b/Make-lang.in
@@ -327,7 +327,7 @@ $(F95_PARSER_OBJS): fortran/gfortran.h fortran/libgfortran.h \
fortran/intrinsic.h fortran/match.h fortran/constructor.h \
fortran/parse.h fortran/arith.h fortran/target-memory.h \
$(CONFIG_H) $(SYSTEM_H) $(TM_H) $(TM_P_H) coretypes.h \
- dumpfile.h $(TREE_H) dumpfile.h $(GGC_H) \
+ dumpfile.h $(TREE_H) dumpfile.h $(GGC_H) $(VEC_H) \
$(FLAGS_H) $(DIAGNOSTIC_H) errors.h $(FUNCTION_H) \
fortran/iso-c-binding.def fortran/iso-fortran-env.def
fortran/openmp.o: pointer-set.h $(TARGET_H) toplev.h
diff --git a/array.c b/array.c
index 6787c05..b4a028b 100644
--- a/array.c
+++ b/array.c
@@ -1046,6 +1046,7 @@ match
gfc_match_array_constructor (gfc_expr **result)
{
gfc_constructor_base head, new_cons;
+ gfc_change_set changed_syms;
gfc_expr *expr;
gfc_typespec ts;
locus where;
@@ -1074,6 +1075,7 @@ gfc_match_array_constructor (gfc_expr **result)
/* Try to match an optional type-spec :: */
gfc_clear_ts (ts);
+ gfc_new_checkpoint (changed_syms);
if (gfc_match_decl_type_spec (ts, 0) == MATCH_YES)
{
seen_ts = (gfc_match ( ::) == MATCH_YES);
@@ -1082,19 +1084,28 @@ gfc_match_array_constructor (gfc_expr **result)
{
if (gfc_notify_std (GFC_STD_F2003, Array constructor
including type specification at %C) == FAILURE)
- goto cleanup;
+ {
+ gfc_restore_last_checkpoint ();
+ goto cleanup;
+ }
if (ts.deferred)
{
gfc_error (Type-spec at %L cannot contain a deferred
type parameter, where);
+ gfc_restore_last_checkpoint ();
goto cleanup;
}
}
}
- if (! seen_ts)
-gfc_current_locus = where;
+ if (seen_ts)
+gfc_drop_last_checkpoint ();
+ else
+{
+ gfc_restore_last_checkpoint ();
+ gfc_current_locus = where;
+}
if (gfc_match (end_delim) == MATCH_YES)
{
diff --git a/gfortran.h b/gfortran.h
index 3b4b473..7a18c6c 100644
--- a/gfortran.h
+++ b/gfortran.h
@@ -39,6 +39,7 @@ along with GCC; see the file COPYING3. If not see
#include intl.h
#include input.h
#include splay-tree.h
+#include vec.h
/* Major control parameters. */
@@ -1275,6 +1276,15 @@ typedef struct gfc_symbol
}
gfc_symbol;
+
+struct gfc_change_set
+{
+ vecgfc_symbol * syms;
+ vecgfc_typebound_proc * tbps;
+ gfc_change_set *previous;
+};
+
+
/* This structure is used to keep track of symbols in common blocks. */
typedef struct gfc_common_head
{
@@ -2632,6 +2642,9 @@ int gfc_get_sym_tree (const char *, gfc_namespace *, gfc_symtree **, bool);
int gfc_get_ha_symbol (const char *, gfc_symbol **);
int gfc_get_ha_sym_tree (const char *, gfc_symtree **);
+void gfc_new_checkpoint (gfc_change_set );
+void gfc_drop_last_checkpoint (void);
+void gfc_restore_last_checkpoint (void);
void gfc_undo_symbols (void);
void gfc_commit_symbols (void);
void gfc_commit_symbol (gfc_symbol *);
diff --git a/symbol.c b/symbol.c
index acfebc5..f040431 100644
--- a/symbol.c
+++ b/symbol.c
@@ -97,21 +97,10 @@ gfc_namespace
[Patch, fortran] [2/5] PR54730 ICE: confused by type-like fonctions: Extract restore_old_symbol.
The ChangeLog says it all.
2013-02-15 Mikael Morin mik...@gcc.gnu.org
PR fortran/54730
* symbol.c (gfc_undo_symbols): Move code...
(restore_old_symbol): ... here as a new function.
diff --git a/symbol.c b/symbol.c
index 2c122d0..4f1752d 100644
--- a/symbol.c
+++ b/symbol.c
@@ -2878,6 +2878,64 @@ find_common_symtree (gfc_symtree *st, gfc_common_head *head)
}
+/* Restore previous state of symbol. Just copy simple stuff. */
+
+static void
+restore_old_symbol (gfc_symbol *p)
+{
+ gfc_symbol *old;
+
+ p-mark = 0;
+ old = p-old_symbol;
+
+ p-ts.type = old-ts.type;
+ p-ts.kind = old-ts.kind;
+
+ p-attr = old-attr;
+
+ if (p-value != old-value)
+{
+ gfc_free_expr (old-value);
+ p-value = NULL;
+}
+
+ if (p-as != old-as)
+{
+ if (p-as)
+ gfc_free_array_spec (p-as);
+ p-as = old-as;
+}
+
+ p-generic = old-generic;
+ p-component_access = old-component_access;
+
+ if (p-namelist != NULL old-namelist == NULL)
+{
+ gfc_free_namelist (p-namelist);
+ p-namelist = NULL;
+}
+ else
+{
+ if (p-namelist_tail != old-namelist_tail)
+ {
+ gfc_free_namelist (old-namelist_tail-next);
+ old-namelist_tail-next = NULL;
+ }
+}
+
+ p-namelist_tail = old-namelist_tail;
+
+ if (p-formal != old-formal)
+{
+ gfc_free_formal_arglist (p-formal);
+ p-formal = old-formal;
+}
+
+ free (p-old_symbol);
+ p-old_symbol = NULL;
+}
+
+
/* Undoes all the changes made to symbols in the current statement.
This subroutine is made simpler due to the fact that attributes are
never removed once added. */
@@ -2885,7 +2943,7 @@ find_common_symtree (gfc_symtree *st, gfc_common_head *head)
void
gfc_undo_symbols (void)
{
- gfc_symbol *p, *old;
+ gfc_symbol *p;
unsigned i;
FOR_EACH_VEC_ELT (changes-syms, i, p)
@@ -2944,58 +3002,9 @@ gfc_undo_symbols (void)
gfc_delete_symtree (p-ns-sym_root, p-name);
gfc_release_symbol (p);
- continue;
- }
-
- /* Restore previous state of symbol. Just copy simple stuff. */
- p-mark = 0;
- old = p-old_symbol;
-
- p-ts.type = old-ts.type;
- p-ts.kind = old-ts.kind;
-
- p-attr = old-attr;
-
- if (p-value != old-value)
- {
- gfc_free_expr (old-value);
- p-value = NULL;
- }
-
- if (p-as != old-as)
- {
- if (p-as)
- gfc_free_array_spec (p-as);
- p-as = old-as;
- }
-
- p-generic = old-generic;
- p-component_access = old-component_access;
-
- if (p-namelist != NULL old-namelist == NULL)
- {
- gfc_free_namelist (p-namelist);
- p-namelist = NULL;
}
else
- {
- if (p-namelist_tail != old-namelist_tail)
- {
- gfc_free_namelist (old-namelist_tail-next);
- old-namelist_tail-next = NULL;
- }
- }
-
- p-namelist_tail = old-namelist_tail;
-
- if (p-formal != old-formal)
- {
- gfc_free_formal_arglist (p-formal);
- p-formal = old-formal;
- }
-
- free (p-old_symbol);
- p-old_symbol = NULL;
+ restore_old_symbol (p);
}
changes-syms.truncate (0);
[Patch, fortran] [1/5] PR54730 ICE: confused by type-like fonctions: Move to the vec API.
This packs symbol.c's 'changed_syms' and 'tentative_tbp' variables into a new
'gfc_change_set' structure using the vec API.
2013-02-15 Mikael Morin mik...@gcc.gnu.org
PR fortran/54730
* Make-lang.in (F95_PARSER_OBJS): Add dependency to vec.h
* gfortran.h: Include vec.h
(gfc_change_set): New struct.
* symbol.c (tentative_tbp): Remove struct.
(changed_syms, tentative_tbp_list): Remove variables.
(change_set_var, changes): New variables.
(save_symbol_data, gfc_get_sym_tree, gfc_undo_symbols,
gfc_commit_symbols, gfc_commit_symbol,
gfc_enforce_clean_symbol_state, gfc_get_typebound_proc):
Use changes instead of changed_syms and tentative_tbp_list.
diff --git a/Make-lang.in b/Make-lang.in
index 3584dd8..8c9e7ea 100644
--- a/Make-lang.in
+++ b/Make-lang.in
@@ -327,7 +327,7 @@ $(F95_PARSER_OBJS): fortran/gfortran.h fortran/libgfortran.h \
fortran/intrinsic.h fortran/match.h fortran/constructor.h \
fortran/parse.h fortran/arith.h fortran/target-memory.h \
$(CONFIG_H) $(SYSTEM_H) $(TM_H) $(TM_P_H) coretypes.h \
- dumpfile.h $(TREE_H) dumpfile.h $(GGC_H) \
+ dumpfile.h $(TREE_H) dumpfile.h $(GGC_H) $(VEC_H) \
$(FLAGS_H) $(DIAGNOSTIC_H) errors.h $(FUNCTION_H) \
fortran/iso-c-binding.def fortran/iso-fortran-env.def
fortran/openmp.o: pointer-set.h $(TARGET_H) toplev.h
diff --git a/gfortran.h b/gfortran.h
index 3b4b473..31b0d42 100644
--- a/gfortran.h
+++ b/gfortran.h
@@ -39,6 +39,7 @@ along with GCC; see the file COPYING3. If not see
#include intl.h
#include input.h
#include splay-tree.h
+#include vec.h
/* Major control parameters. */
@@ -1275,6 +1276,14 @@ typedef struct gfc_symbol
}
gfc_symbol;
+
+struct gfc_change_set
+{
+ vecgfc_symbol * syms;
+ vecgfc_typebound_proc * tbps;
+};
+
+
/* This structure is used to keep track of symbols in common blocks. */
typedef struct gfc_common_head
{
diff --git a/symbol.c b/symbol.c
index acfebc5..2c122d0 100644
--- a/symbol.c
+++ b/symbol.c
@@ -97,21 +97,10 @@ gfc_namespace *gfc_global_ns_list;
gfc_gsymbol *gfc_gsym_root = NULL;
-static gfc_symbol *changed_syms = NULL;
-
gfc_dt_list *gfc_derived_types;
-
-/* List of tentative typebound-procedures. */
-
-typedef struct tentative_tbp
-{
- gfc_typebound_proc *proc;
- struct tentative_tbp *next;
-}
-tentative_tbp;
-
-static tentative_tbp *tentative_tbp_list = NULL;
+static gfc_change_set change_set_var = { vNULL, vNULL };
+static gfc_change_set *changes = change_set_var;
/*** IMPLICIT NONE and IMPLICIT statement handlers ***/
@@ -2720,8 +2709,7 @@ save_symbol_data (gfc_symbol *sym)
sym-old_symbol = XCNEW (gfc_symbol);
*(sym-old_symbol) = *sym;
- sym-tlink = changed_syms;
- changed_syms = sym;
+ changes-syms.safe_push (sym);
}
@@ -2757,10 +2745,9 @@ gfc_get_sym_tree (const char *name, gfc_namespace *ns, gfc_symtree **result,
/* Add to the list of tentative symbols. */
p-old_symbol = NULL;
- p-tlink = changed_syms;
p-mark = 1;
p-gfc_new = 1;
- changed_syms = p;
+ changes-syms.safe_push (p);
st = gfc_new_symtree (ns-sym_root, name);
st-n.sym = p;
@@ -2898,13 +2885,11 @@ find_common_symtree (gfc_symtree *st, gfc_common_head *head)
void
gfc_undo_symbols (void)
{
- gfc_symbol *p, *q, *old;
- tentative_tbp *tbp, *tbq;
+ gfc_symbol *p, *old;
+ unsigned i;
- for (p = changed_syms; p; p = q)
+ FOR_EACH_VEC_ELT (changes-syms, i, p)
{
- q = p-tlink;
-
if (p-gfc_new)
{
/* Symbol was new. */
@@ -3011,18 +2996,10 @@ gfc_undo_symbols (void)
free (p-old_symbol);
p-old_symbol = NULL;
- p-tlink = NULL;
}
- changed_syms = NULL;
-
- for (tbp = tentative_tbp_list; tbp; tbp = tbq)
-{
- tbq = tbp-next;
- /* Procedure is already marked `error' by default. */
- free (tbp);
-}
- tentative_tbp_list = NULL;
+ changes-syms.truncate (0);
+ changes-tbps.truncate (0);
}
@@ -3059,26 +3036,21 @@ free_old_symbol (gfc_symbol *sym)
void
gfc_commit_symbols (void)
{
- gfc_symbol *p, *q;
- tentative_tbp *tbp, *tbq;
+ gfc_symbol *p;
+ gfc_typebound_proc *tbp;
+ unsigned i;
- for (p = changed_syms; p; p = q)
+ FOR_EACH_VEC_ELT (changes-syms, i, p)
{
- q = p-tlink;
- p-tlink = NULL;
p-mark = 0;
p-gfc_new = 0;
free_old_symbol (p);
}
- changed_syms = NULL;
+ changes-syms.truncate (0);
- for (tbp = tentative_tbp_list; tbp; tbp = tbq)
-{
- tbq = tbp-next;
- tbp-proc-error = 0;
- free (tbp);
-}
- tentative_tbp_list = NULL;
+ FOR_EACH_VEC_ELT (changes-tbps, i, tbp)
+tbp-error = 0;
+ changes-tbps.truncate (0);
}
@@ -3089,20 +3061,15 @@ void
gfc_commit_symbol (gfc_symbol *sym)
{
gfc_symbol *p;
+ unsigned i;
- if (changed_syms == sym)
-changed_syms = sym-tlink;
- else
-{
- for (p = changed_syms; p; p = p-tlink)
-if (p-tlink == sym)
[Patch, fortran] [3/5] PR54730 ICE: confused by type-like fonctions: Fix restore_old_symbol
This is a fix for what looks like a thinko.
The other patches don't depend on it, so it can well wait for 4.9 if
that is preferred.
2013-02-15 Mikael Morin mik...@gcc.gnu.org
PR fortran/54730
* symbol.c (restore_old_symbol): Fix thinko.
diff --git a/symbol.c b/symbol.c
index 4f1752d..e4dbb41 100644
--- a/symbol.c
+++ b/symbol.c
@@ -2895,7 +2895,8 @@ restore_old_symbol (gfc_symbol *p)
if (p-value != old-value)
{
- gfc_free_expr (old-value);
+ gcc_checking_assert (old-value == NULL);
+ gfc_free_expr (p-value);
p-value = NULL;
}
[Patch, fortran] [4/5] PR54730 ICE: confused by type-like fonctions: Support multiple change sets.
A new field 'previous' is added to the 'gfc_change_set' structure so that it can
be used as a stack.
New procedures are added to use the new partial undoing feature, namely:
gfc_new_checkpoint, gfc_drop_last_checkpoint and gfc_restore_last_checkpoint.
They will be used in the next patch.
2013-02-15 Mikael Morin mik...@gcc.gnu.org
PR fortran/54730
* gfortran.h (struct gfc_change_set): New field 'previous'.
(gfc_new_checkpoint, gfc_drop_last_checkpoint,
gfc_restore_last_checkpoint):
New prototypes.
* symbol.c (change_set_var): Update initialization.
(single_undo_checkpoint_p, gfc_new_checkpoint, free_change_set_data,
pop_change_set, gfc_drop_last_checkpiont,
enforce_single_undo_checkpoint):
New functions.
(save_symbol_data): Handle multiple change sets. Make sure old_symbol
field's
previous value is not overwritten. Clear gfc_new field.
(restore_old_symbol): Restore previous old_symbol field.
(gfc_restore_last_checkpoint): New function, using body renamed from
gfc_undo_symbols. Restore the previous change set as current one.
(gfc_undo_symbols): New body.
(gfc_commit_symbols, gfc_commit_symbol, gfc_enforce_clean_symbol_state):
Call enforce_single_undo_checkpoint.
(gfc_symbol_done_2): Free change set data.
diff --git a/gfortran.h b/gfortran.h
index 31b0d42..7a18c6c 100644
--- a/gfortran.h
+++ b/gfortran.h
@@ -1281,6 +1281,7 @@ struct gfc_change_set
{
vecgfc_symbol * syms;
vecgfc_typebound_proc * tbps;
+ gfc_change_set *previous;
};
@@ -2641,6 +2642,9 @@ int gfc_get_sym_tree (const char *, gfc_namespace *, gfc_symtree **, bool);
int gfc_get_ha_symbol (const char *, gfc_symbol **);
int gfc_get_ha_sym_tree (const char *, gfc_symtree **);
+void gfc_new_checkpoint (gfc_change_set );
+void gfc_drop_last_checkpoint (void);
+void gfc_restore_last_checkpoint (void);
void gfc_undo_symbols (void);
void gfc_commit_symbols (void);
void gfc_commit_symbol (gfc_symbol *);
diff --git a/symbol.c b/symbol.c
index e4dbb41..f040431 100644
--- a/symbol.c
+++ b/symbol.c
@@ -99,7 +99,7 @@ gfc_gsymbol *gfc_gsym_root = NULL;
gfc_dt_list *gfc_derived_types;
-static gfc_change_set change_set_var = { vNULL, vNULL };
+static gfc_change_set change_set_var = { vNULL, vNULL, NULL };
static gfc_change_set *changes = change_set_var;
@@ -2697,17 +2697,49 @@ gfc_find_symbol (const char *name, gfc_namespace *ns, int parent_flag,
}
+/* Tells whether there is only one set of changes in the stack. */
+
+static bool
+single_undo_checkpoint_p (void)
+{
+ if (changes == change_set_var)
+{
+ gcc_assert (changes-previous == NULL);
+ return true;
+}
+ else
+{
+ gcc_assert (changes-previous != NULL);
+ return false;
+}
+}
+
/* Save symbol with the information necessary to back it out. */
static void
save_symbol_data (gfc_symbol *sym)
{
+ gfc_symbol *s;
+ unsigned i;
- if (sym-gfc_new || sym-old_symbol != NULL)
+ if (!single_undo_checkpoint_p ())
+{
+ /* If there is more than one change set, look for the symbol in the
+ current one. If it is found there, we can reuse it. */
+ FOR_EACH_VEC_ELT (changes-syms, i, s)
+ if (s == sym)
+ {
+ gcc_assert (sym-gfc_new || sym-old_symbol != NULL);
+ return;
+ }
+}
+ else if (sym-gfc_new || sym-old_symbol != NULL)
return;
- sym-old_symbol = XCNEW (gfc_symbol);
- *(sym-old_symbol) = *sym;
+ s = XCNEW (gfc_symbol);
+ *s = *sym;
+ sym-old_symbol = s;
+ sym-gfc_new = 0;
changes-syms.safe_push (sym);
}
@@ -2878,6 +2910,22 @@ find_common_symtree (gfc_symtree *st, gfc_common_head *head)
}
+/* Clear the given storage, and make it the current change set for registering
+ changed symbols. Its contents are freed after a call to
+ gfc_restore_last_checkpoint or gfc_drop_last_checkpoint, but it is up to the
+ caller to free the storage itself. It is usually a local variable, so there
+ is nothing to do anyway. */
+
+void
+gfc_new_checkpoint (gfc_change_set chg_syms)
+{
+ chg_syms.syms = vNULL;
+ chg_syms.tbps = vNULL;
+ chg_syms.previous = changes;
+ changes = chg_syms;
+}
+
+
/* Restore previous state of symbol. Just copy simple stuff. */
static void
@@ -2932,17 +2980,88 @@ restore_old_symbol (gfc_symbol *p)
p-formal = old-formal;
}
- free (p-old_symbol);
- p-old_symbol = NULL;
+ p-old_symbol = old-old_symbol;
+ free (old);
+}
+
+
+/* Frees the internal data of a gfc_change_set structure. Doesn't free the
+ structure itself. */
+
+static void
+free_change_set_data (gfc_change_set cs)
+{
+ cs.syms.release ();
+ cs.tbps.release ();
+}
+
+
+/* Given a change set pointer, free its target's contents and update it with
+ the address of the previous change set. Note that only the contents are
+ freed, not the target itself (the contents' container). It is not a problem
+ as the latter will
[Patch, fortran] [5/5] PR54730 ICE: confused by type-like fonctions: Fix gfc_match_array_constructor.
This is the fix for the PR.
2013-02-15 Mikael Morin mik...@gcc.gnu.org
PR fortran/54730
* array.c (gfc_match_array_constructor): Set a checkpoint before
matching
a typespec. Drop it on success, restore it otherwise.
diff --git a/array.c b/array.c
index 6787c05..b4a028b 100644
--- a/array.c
+++ b/array.c
@@ -1046,6 +1046,7 @@ match
gfc_match_array_constructor (gfc_expr **result)
{
gfc_constructor_base head, new_cons;
+ gfc_change_set changed_syms;
gfc_expr *expr;
gfc_typespec ts;
locus where;
@@ -1074,6 +1075,7 @@ gfc_match_array_constructor (gfc_expr **result)
/* Try to match an optional type-spec :: */
gfc_clear_ts (ts);
+ gfc_new_checkpoint (changed_syms);
if (gfc_match_decl_type_spec (ts, 0) == MATCH_YES)
{
seen_ts = (gfc_match ( ::) == MATCH_YES);
@@ -1082,19 +1084,28 @@ gfc_match_array_constructor (gfc_expr **result)
{
if (gfc_notify_std (GFC_STD_F2003, Array constructor
including type specification at %C) == FAILURE)
- goto cleanup;
+ {
+ gfc_restore_last_checkpoint ();
+ goto cleanup;
+ }
if (ts.deferred)
{
gfc_error (Type-spec at %L cannot contain a deferred
type parameter, where);
+ gfc_restore_last_checkpoint ();
goto cleanup;
}
}
}
- if (! seen_ts)
-gfc_current_locus = where;
+ if (seen_ts)
+gfc_drop_last_checkpoint ();
+ else
+{
+ gfc_restore_last_checkpoint ();
+ gfc_current_locus = where;
+}
if (gfc_match (end_delim) == MATCH_YES)
{
2013-02-15 Mikael Morin mik...@gcc.gnu.org
PR fortran/54730
* gfortran.dg/array_constructor_42.f90: New test.
! { dg-do compile }
!
! PR fortran/54730
! A symbol 'a' was created while attempting to parse a typespec in the array
! constructor. That (invalid) symbol was kept until translation stage
! where it was leading to an ICE.
!
! Original testcase from Paul Kapinos kapi...@rz.rwth-aachen.de
!
subroutine s
implicit none
intrinsic :: real
real :: vec(1:2)
vec = (/ real(a = 1), 1. /)
end subroutine s
program main
implicit none
intrinsic :: real
print *,(/ real(a = 1) /)
end
4.7 backports
Hi!
I've backported a bunch of patches from trunk to 4.7, after
bootstrapping/regtesting them on x86_64-linux and i686-linux.
Jakub
2013-02-19 Jakub Jelinek ja...@redhat.com
Backported from mainline
2012-11-27 Jakub Jelinek ja...@redhat.com
PR tree-optimization/55110
* tree-vect-loop.c (vectorizable_reduction): Don't assert
that STMT_VINFO_RELATED_STMT of orig_stmt is stmt.
* gcc.dg/pr55110.c: New test.
--- gcc/tree-vect-loop.c(revision 193844)
+++ gcc/tree-vect-loop.c(revision 193845)
@@ -4624,7 +4624,6 @@ vectorizable_reduction (gimple stmt, gim
if (orig_stmt)
{
orig_stmt_info = vinfo_for_stmt (orig_stmt);
- gcc_assert (STMT_VINFO_RELATED_STMT (orig_stmt_info) == stmt);
gcc_assert (STMT_VINFO_IN_PATTERN_P (orig_stmt_info));
gcc_assert (!STMT_VINFO_IN_PATTERN_P (stmt_info));
}
--- gcc/testsuite/gcc.dg/pr55110.c (revision 0)
+++ gcc/testsuite/gcc.dg/pr55110.c (revision 193845)
@@ -0,0 +1,13 @@
+/* PR tree-optimization/55110 */
+/* { dg-do compile } */
+/* { dg-options -O1 -ftree-vectorize } */
+
+int
+foo (int x)
+{
+ int a, b;
+ for (b = 0; b 8; b++)
+for (a = 0; a 2; a++)
+ x /= 3;
+ return x;
+}
2013-02-19 Jakub Jelinek ja...@redhat.com
Backported from mainline
2013-02-06 Jakub Jelinek ja...@redhat.com
PR middle-end/56217
* omp-low.c (use_pointer_for_field): Return false if
lower_send_shared_vars doesn't generate any copy-out code.
* g++.dg/gomp/pr56217.C: New test.
* testsuite/libgomp.c++/pr56217.C: New test.
--- gcc/omp-low.c (revision 195795)
+++ gcc/omp-low.c (revision 195796)
@@ -757,12 +757,20 @@ use_pointer_for_field (tree decl, omp_co
if (TREE_ADDRESSABLE (decl))
return true;
+ /* lower_send_shared_vars only uses copy-in, but not copy-out
+for these. */
+ if (TREE_READONLY (decl)
+ || ((TREE_CODE (decl) == RESULT_DECL
+ || TREE_CODE (decl) == PARM_DECL)
+ DECL_BY_REFERENCE (decl)))
+ return false;
+
/* Disallow copy-in/out in nested parallel if
decl is shared in outer parallel, otherwise
each thread could store the shared variable
in its own copy-in location, making the
variable no longer really shared. */
- if (!TREE_READONLY (decl) shared_ctx-is_nested)
+ if (shared_ctx-is_nested)
{
omp_context *up;
@@ -785,11 +793,10 @@ use_pointer_for_field (tree decl, omp_co
}
}
- /* For tasks avoid using copy-in/out, unless they are readonly
-(in which case just copy-in is used). As tasks can be
+ /* For tasks avoid using copy-in/out. As tasks can be
deferred or executed in different thread, when GOMP_task
returns, the task hasn't necessarily terminated. */
- if (!TREE_READONLY (decl) is_task_ctx (shared_ctx))
+ if (is_task_ctx (shared_ctx))
{
tree outer;
maybe_mark_addressable_and_ret:
--- gcc/testsuite/g++.dg/gomp/pr56217.C (revision 0)
+++ gcc/testsuite/g++.dg/gomp/pr56217.C (revision 195796)
@@ -0,0 +1,14 @@
+// PR middle-end/56217
+// { dg-do compile }
+// { dg-options -fopenmp }
+
+struct S { int *p; S (); S (S ); };
+
+S
+foo ()
+{
+ S s;
+ #pragma omp task shared (s)
+s.p = 0;
+ return s;
+}
--- libgomp/testsuite/libgomp.c++/pr56217.C (revision 0)
+++ libgomp/testsuite/libgomp.c++/pr56217.C (revision 195796)
@@ -0,0 +1,36 @@
+// PR middle-end/56217
+// { dg-do run }
+// { dg-options -std=c++0x }
+
+extern C void abort ();
+
+template typename T
+struct ptr {
+ T *p;
+ ptr () : p () {}
+ ptr (ptr ) = delete;
+ ptr (ptr o) : p(o) {}
+ operator T * () { return p; }
+};
+
+int a[6] = { 100, 101, 102, 103, 104, 105 };
+
+static ptrint
+f ()
+{
+ ptrint pt;
+ #pragma omp task shared (pt)
+pt.p = a + 2;
+ #pragma omp taskwait
+ return pt;
+}
+
+int
+main ()
+{
+ ptrint pt;
+ #pragma omp parallel
+ #pragma omp single
+ if (f () != a + 2 || *f () != 102)
+abort ();
+}
2013-02-19 Jakub Jelinek ja...@redhat.com
Backported from mainline
2013-02-07 Jakub Jelinek ja...@redhat.com
PR c++/56237
* decl.c (push_local_name): Look at DECL_DISCRIMINATOR (t)
only if DECL_DISCRIMINATOR_SET_P (t) rather than just
DECL_LANG_SPECIFIC (t).
* g++.dg/abi/mangle61.C: New test.
--- gcc/cp/decl.c (revision 195857)
+++ gcc/cp/decl.c (revision 195858)
@@ -920,7 +920,7 @@ push_local_name (tree decl)
if (!DECL_LANG_SPECIFIC (decl))
retrofit_lang_decl (decl);
DECL_LANG_SPECIFIC (decl)-u.base.u2sel = 1;
- if (DECL_LANG_SPECIFIC (t))
+ if (DECL_DISCRIMINATOR_SET_P (t))
DECL_DISCRIMINATOR (decl) = DECL_DISCRIMINATOR (t) + 1;
else
DECL_DISCRIMINATOR (decl) = 1;
---
Re: [4.7, testsuite] Clear hwcap_2 with Sun ld
On 02/19/2013 05:25 AM, Rainer Orth wrote: I'd like to backport [testsuite] Clear hwcap_2 with Sun ld http://gcc.gnu.org/ml/gcc-patches/2012-06/msg01246.html 2012-06-19 Rainer Orth r...@cebitec.uni-bielefeld.de * gcc.target/i386/clearcapv2.map: New file. * gcc.target/i386/i386.exp: Try it first before clearcap.map. to the 4.7 branch since recently all avx2 tests started to FAIL there on Solaris 10 and 11/x86 with Sun as. It applies cleanly as is, bootstrapped without regressions on i386-pc-solaris2.10 and x86_64-unknown-linux-gnu. Ok for 4.7 branch? Rainer OK. Janis
Re: FW: [PATCH] [MIPS] microMIPS gcc support
Thanks, this is looking much better now.
Moore, Catherine catherine_mo...@mentor.com writes:
I'd also like to support -mno-jals for backward compatibility. Are you
okay with that? If so, I'll submit as a separate patch.
Hmm, to be honest, I'd rather you kept it local to the Mentor toolchain.
The way the patch is written, we (rightly) use JALS even with
-minterlink-compressed in cases where we can prove that the call
is to another microMIPS function, so calling -mno-jals an alias of
-minterlink-compressed would be a bit confusing.
-mno-jals sounds like it ought to disable JALS in all situations,
which would need special code to handle. I think doing that would be
extra baggage with little benefit.
Index: gcc.target/mips/umips-constraints-1.c
===
--- gcc.target/mips/umips-constraints-1.c (revision 0)
+++ gcc.target/mips/umips-constraints-1.c (revision 0)
@@ -0,0 +1,14 @@
+/* { dg-options (-mmicromips) } */
+/* { dg-skip-if code quality test { *-*-* } { -O0 } { } } */
+
+MICROMIPS void
+foo (int *x)
+{
+ asm volatile (insn1\t%a0 :: ZD (x[0]));
+ asm volatile (insn2\t%a0 :: ZD (x[511]));
+ asm volatile (insn3\t%a0 :: ZD (x[512]));
+}
+
+/* { dg-final { scan-assembler \tinsn1\t0\\( } } */
+/* { dg-final { scan-assembler \tinsn2\t2044\\( } } */
+/* { dg-final { scan-assembler \tinsn3\t2048\\( } } */
But the insn3 is wrong, isn't it? I suggested scan-assembler-not for
that one because I thought it had to be a 12-bit signed offset on
microMIPS. The patch has:
(define_address_constraint ZD
When compiling microMIPS code, this constraint matches an address operand
that is formed from a base register and a 12-bit offset. These operands
can be used for microMIPS instructions such as @code{prefetch}. When
not compiling for microMIPS code, @code{ZD} is equivalent to @code{p}.
(ior (and (match_test TARGET_MICROMIPS)
(match_test umips_12bit_offset_address_p (op, mode)))
(match_test mips_address_insns (op, mode, false
but it looks like it ought to be:
(if_then_else (match_test TARGET_MICROMIPS)
(match_test umips_12bit_offset_address_p (op, mode))
(match_test mips_address_insns (op, mode, false)))
Same for ZC.
@@ -1246,6 +1247,10 @@ proc mips-dg-options { args } {
append extra_tool_flags -DMIPS16=__attribute__((mips16))
}
+if { [mips_have_test_option_p options -mmicromips] } {
+ append extra_tool_flags -DMICROMIPS=__attribute__((micromips))
+}
+
# Use our version of gcc-dg-test for this test.
if { ![string equal [info procs mips-gcc-dg-test] ] } {
rename gcc-dg-test mips-old-gcc-dg-test
@@ -1275,6 +1280,6 @@ proc mips-gcc-dg-test { prog do_what extra_tool_fl
dg-init
mips-dg-init
gcc-dg-runtest [lsort [glob -nocomplain $srcdir/$subdir/*.c]] \
--DNOMIPS16=__attribute__((nomips16))
+-DNOMIPS16=__attribute__((nomips16))
-DNOMICROMIPS=__attribute__((nomicromips))
-DNOCOMPRESSION=__attribute__((nocompression))
-DMICROMIPS=__attribute__((micromips))
mips-dg-finish
dg-finish
Please drop the -DMICROMIPS in the second hunk. The first hunk is the
right way to add it. The idea is that we want a compilation failure if
the test uses MICROMIPS but forgets to add:
/* { dg-options (-mmicromips) } */
The -DNOMICROMIPS and -DNOCOMPRESSION in the second hunk are fine though.
+/* { dg-options (-mmicromips) } */
+/* { dg-skip-if code quality test { *-*-* } { -O0 -O1 } { } } */
Sorry, only just realised you were skipping -O1. Please add a comment saying
why it fails at -O1 or (preferably) add whichever other option is needed for
the test to pass at -O1. I assume it's -fpeephole2 in this case.
Same for all tests with this skip.
Index: gcc.target/mips/umips-lwp-swp-2.c
===
--- gcc.target/mips/umips-lwp-swp-2.c (revision 0)
+++ gcc.target/mips/umips-lwp-swp-2.c (revision 0)
@@ -0,0 +1,19 @@
+/* { dg-options -mgp32 (-mmicromips) } */
+/* { dg-skip-if code quality test { *-*-* } { -O0 -O1 -Os } { } }
*/
+int a[2];
+
+MICROMIPS f (b)
+{
+ unsigned int i;
+ int *p;
+ for (p = a[b], i = b; --i ~0; )
+*--p = i * 3 + (int)a;
+
+}
+
+MICROMIPS main ()
+{
+ a[0] = a[1] = 0;
+ f (2);
+}
+/* { dg-final { scan-assembler \tswp\t\\\$3,0\\(\\\$3\\) } }*/
Is this a test of the swap_p case? Thanks if so, but could you add a
comment saying where the SWP gets generated, and why the test needs
to be skipped at -O1 and -Os?
Index: gcc.target/mips/umips-lwp-swp-volatile.c
===
--- gcc.target/mips/umips-lwp-swp-volatile.c (revision 0)
+++ gcc.target/mips/umips-lwp-swp-volatile.c (revision 0)
@@ -0,0 +1,41 @@
+/* { dg-do compile } */
+
+/* This test ensures that we do not generate microMIPS SWP or LWP
+ instructions
Re: [PATCH] Temporarily revert Steven's PCH changes for 4.8 (PR pch/54117)
On Tue, Feb 19, 2013 at 11:11 AM, Joel Brobecker brobec...@adacore.com wrote: I haven't looked at this part in details, but our recollection is that it's a matter of configuring GCC with --with-gnu-as and --with-gnu-ld, together with a small patch to make DWARF the default. AFAIK, the rest should be working as expected. A start is the patch that I sent you or Tristan a long time ago, but simply enabling DWARF is not enough because XCOFF only supports some of the sections and uses slightly different pseudo-ops. * AIX 7.1 supports DWARF3 debugging, but XCOFF remains the default. */ #define DWARF2_DEBUGGING_INFO 1 #define PREFERRED_DEBUGGING_TYPE XCOFF_DEBUG #define DEBUG_INFO_SECTION \t.dwsect 0x1,Ldebug_info..0 #define DEBUG_ABBREV_SECTION \t.dwsect 0x6,Ldebug_abbrev..0 #define DEBUG_ARANGES_SECTION \t.dwsect 0x5 #define DEBUG_LINE_SECTION \t.dwsect 0x2,Ldebug_line..0 #define DEBUG_PUBNAMES_SECTION \t.dwsect 0x3 #define DEBUG_PUBTYPES_SECTION \t.dwsect 0x4 #define DEBUG_STR_SECTION \t.dwsect 0x7 #define DEBUG_RANGES_SECTION \t.dwsect 0x8,Ldebug_ranges..0 Thanks, David
Re: [PATCH] Temporarily revert Steven's PCH changes for 4.8 (PR pch/54117)
Il 13/02/2013 22:12, Steven Bosscher ha scritto: On Wed, Feb 13, 2013 at 4:54 PM, Jakub Jelinek wrote: Hi! As agreed on in the PR, here is the revertion of 3 commits, so that PCH works again for -gstabs and other debug info formats for 4.8 release. For 4.9 we should either remove support for anything non-DWARF, or hope somebody steps up and fixes dbxout.c etc. not to emit debug info right away, but queue it till end of compilation. Hello, I'd still like to propose deprecating all stabs support for GCC 4.8 and remove it for GCC 4.9 (i.e. dbxout, sdbout, xcoffout). It's not a very useful debug info format, after all. The only argument against DWARF, that it is so much larger than stabs, is something I haven't been able to reproduce (see http://sourceware.org/ml/binutils/2013-01/msg00221.html and http://sourceware.org/ml/binutils/2013-01/msg00239.html). DWARF compression via dwz, and debug-fission will improve things even further. An important question is of course: Who still *needs* stabs? If there are important platforms that need stabs support to continue to work, then deprecating is probably not a realistic option yet... stabs are needed to use VTune on Windows. Paolo The list of platforms that are IMHO blockers rights now, with possible solutions for GCC 4.9, are: * hppa2.0w-hp-hpux11.11 - no solution immediately available * powerpc-ibm-aix* - only support AIX7 and later - deprecate AIX6 and older in GCC 4.8 * ix86-*-openbsd2.*, ix86-*openbsd3.[0123] - OpenBSD 3.3 was released in May 2003 - deprecate for GCC 4.8 * m68k*-*-openbsd - port has seen no active maintenance, and has no test results posted, ever - deprecate for GCC 4.8 * pdp11-*-* - toy port - default to DEBUG_NONE * ix86-*-interix* - no solution immediately available, and no-one listed in MAINTAINERS to ask for help, so maybe Doug can say something about this one? Platforms that support DWARF2+ but currently have another default preferred debug info type, would be changed to default to DWARF2+. So older AIX and 32-bits HPUX are the only real problem cases. Ciao! Steven All GCC 4.8 primary platforms support DWARF2+, and AFAICT they all have it as their PREFERRED_DEBUGGING_TYPE (including freebsd): * arm-linux-gnueabi * i386-unknown-freebsd * i686-pc-linux-gnu * mipsisa64-elf * powerpc64-unknown-linux-gnu * sparc-sun-solaris2.10 * x86_64-unknown-linux-gnu Most of the above also include other debug info types by default, but DWARF2+ is the default. For the secondary platforms list things are not so simple. Some of them have a non-DWARF PREFERRED_DEBUGGING_TYPE by default, and some only conditionally (old configurations): * hppa2.0w-hp-hpux11.11 - DBX_DEBUGGING_INFO For 32-bits only, but that's the most common subtarget. According to Dave, it should be possible to add DRAWF2+ support: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54117#c14 Perhaps Dave can explain what would have to be done to move this platform to DWARF2...? * powerpc-ibm-aix7.1.0.0 - XCOFF_DEBUGGING_INFO I think this is for historical reasons. AIX7 has DWARF support, but it looks like that's not yet implemented in GCC. Older AIX do not support DWARF, AFAICT (David E.?). * i686-apple-darwin - DBX_DEBUGGING_INFO MacOSX10.4, 32-bits (darwin8). All 64bits and darwin9+ prefer DWARF2. * i686-pc-cygwin, i686-mingw32 - DBX_DEBUGGING_INFO With binutils 2.16 and later, DWARF2_DEBUGGING_INFO is the default: http://sourceware.org/ml/binutils/2004-04/msg00327.html http://gcc.gnu.org/ml/gcc-patches/2004-04/msg01885.html http://cygwin.com/ml/cygwin/2006-06/msg00865.html * s390x-linux-gnu - DWARF2_DEBUGGING_INFO Other platforms that support DWARF2+ and have it as the preferred type: * *-*-elf* via: config/elfos.h:#define PREFERRED_DEBUGGING_TYPE DWARF2_DEBUG This is practically all ELF platforms. A few target headers redefine PREFERRED_DEBUGGING_TYPE, usually to DWARF2_DEBUG again... Exceptions are microblaze*-*-rtems*, microblaze*-*-elf* * tic6x-*-*, ix86-*-nto-qnx* via: config/tm-dwarf2.h:#define PREFERRED_DEBUGGING_TYPE DWARF2_DEBUG * *-*-vxworks* via: config/vx-common.h:#define PREFERRED_DEBUGGING_TYPE DWARF2_DEBUG * alpha*-dec-*vms* via: config/alpha/vms.h:#define PREFERRED_DEBUGGING_TYPE VMS_AND_DWARF2_DEBUG * hppa*64*-*-hpux11* via: config/pa/pa64-hpux.h:#define PREFERRED_DEBUGGING_TYPE DWARF2_DEBUG Other platforms that support DWARF2+ but prefer another debug info type by default: * rx-*-elf* if using TARGET_AS100_SYNTAX, via: config/rx/rx.h:#define PREFERRED_DEBUGGING_TYPE (TARGET_AS100_SYNTAX \ config/rx/rx.h:? DBX_DEBUG : DWARF2_DEBUG) * avr-*-* includes elfos.h but changes preffered debug info type via: config/avr/elf.h:#define PREFERRED_DEBUGGING_TYPE DBX_DEBUG * powerpc-*-lynxos*, ix86-*-lynxos* include elfos.h
linaro/gcc-4_8-branch created and documented
All, I have just created a distribution branch: 'linaro/gcc-4_8-branch'. I have committed the attached patch to the wwwdocs CVS repository to document this branch (and future Linaro branches). The branch will track the equivalent FSF release branch (once created) and also accept backports of patches accepted for trunk which are of interest to the ARM and AArch64 backends. Anyone from Linaro with write access to the GCC Subversion repository can commit and approve patches for this branch. The branch has been created early as there are some Stage 1 pending patches which I want to backport early to this branch. Thanks, Matt -- Matthew Gretton-Dann Toolchain Working Group, Linaro ? htdocs/.svn.html.swp ? htdocs/svn.html~ Index: htdocs/svn.html === RCS file: /cvs/gcc/wwwdocs/htdocs/svn.html,v retrieving revision 1.179 diff -u -p -r1.179 svn.html --- htdocs/svn.html 12 Feb 2013 15:53:34 - 1.179 +++ htdocs/svn.html 19 Feb 2013 21:20:11 - @@ -621,6 +621,11 @@ be prefixed with the initials of the dis The branch is maintained by Michael Meissner, a href=mailto:meiss...@linux.vnet.ibm.com;meiss...@linux.vnet.ibm.com/a./dd + dtlinaro/gcc-emx_y/em-branch/dt + ddLinaro compilers based on GCC emx.y/em releases. These branches + only accept backports of patches which have been accepted to trunk. This + family of branches is maintained by personnel from Linaro./dd + dtredhat/gcc-3_2-branch/dt ddRed Hat GNU/Linux compilers based on GCC 3.2.x./dd
[committed] Fix libquadmath strtoflt128 at -O0 (PR libquadmath/56379)
Hi!
I've committed to trunk what has been accepted earlier today for glibc
strtod_l.c, a fix for -O0 compilation of the file.
Bootstrapped/regtested on x86_64-linux and i686-linux.
2013-02-19 Jakub Jelinek ja...@redhat.com
PR libquadmath/56379
* strtod/strtod_l.c (mpn_lshift_1): Rewritten as function-like
macro.
--- libquadmath/strtod/strtod_l.c
+++ libquadmath/strtod/strtod_l.c
@@ -444,28 +444,30 @@ str_to_mpn (const STRING_TYPE *str, int digcnt, mp_limb_t
*n, mp_size_t *nsize,
/* Shift {PTR, SIZE} COUNT bits to the left, and fill the vacated bits
with the COUNT most significant bits of LIMB.
- Tege doesn't like this function so I have to write it here myself. :)
+ Implemented as a macro, so that __builtin_constant_p works even at -O0.
+
+ Tege doesn't like this macro so I have to write it here myself. :)
--drepper */
-static inline void
-__attribute ((always_inline))
-mpn_lshift_1 (mp_limb_t *ptr, mp_size_t size, unsigned int count,
- mp_limb_t limb)
-{
- if (__builtin_constant_p (count) count == BITS_PER_MP_LIMB)
-{
- /* Optimize the case of shifting by exactly a word:
-just copy words, with no actual bit-shifting. */
- mp_size_t i;
- for (i = size - 1; i 0; --i)
- ptr[i] = ptr[i - 1];
- ptr[0] = limb;
-}
- else
-{
- (void) mpn_lshift (ptr, ptr, size, count);
- ptr[0] |= limb (BITS_PER_MP_LIMB - count);
-}
-}
+#define mpn_lshift_1(ptr, size, count, limb) \
+ do \
+{ \
+ mp_limb_t *__ptr = (ptr);
\
+ if (__builtin_constant_p (count) count == BITS_PER_MP_LIMB) \
+ { \
+ mp_size_t i; \
+ for (i = (size) - 1; i 0; --i) \
+ __ptr[i] = __ptr[i - 1];\
+ __ptr[0] = (limb);\
+ } \
+ else \
+ { \
+ /* We assume count 0 count BITS_PER_MP_LIMB here. */ \
+ unsigned int __count = (count); \
+ (void) mpn_lshift (__ptr, __ptr, size, __count); \
+ __ptr[0] |= (limb) (BITS_PER_MP_LIMB - __count); \
+ } \
+} \
+ while (0)
#define INTERNAL(x) INTERNAL1(x)
Jakub
Re: RFC: [PATCH,ARM] Fix 56110
Richard Earnshaw [2013-02-19 15:12]:
Hi,
thanks for your reply.
[...]
However, the question you need to be asking is why the pattern immediately
before the one you've added is not matching. The compiler knows how to add
clobbers, so I'm surprised that you're finding a new pattern to be
necessary.
From the andsi3_compare0_scratch pattern definition:
[(set (reg:CC_NOOV CC_REGNUM)
(compare:CC_NOOV
(and:SI (match_operand:SI 0 s_register_operand r,r,r)
(match_operand:SI 1 arm_not_operand I,K,r))
(const_int 0)))
So the insn pattern only matches if the 2nd operand is an
arm_not_operand. In my test program, the 2nd operand is 0x8080
(const_int 32896 [0x8080]) which cannot be used as an immediate operand
(const_ok_for_arm doesn't match 0x8080 nor ~0x8080).
I don't see _why_ we would want the pattern to only be applied
to registers or immediates though.
Actually the diff below makes it so that gcc recognizes the pattern even if
0x8080 is used as the 2nd operand. It looks like using
reg_or_int_operand should be the correct thing to do, since we do have
the I/K constraints in the pattern...
However it breaks the case where the 2nd operand is a const_int that
*can* be used as an immediate (eg 0x80), and ends up generating the
AND/CMP combination.
For the latter case, RTL looks like this before the combiner pass:
(insn 7 4 8 2 (set (reg:SI 114 [ D.4127 ])
(and:SI (reg/v:SI 113 [ m ])
(const_int 128 [0x80]))) /home/user/g.c:3 76 {*arm_andsi3_insn}
(expr_list:REG_DEAD (reg/v:SI 113 [ m ])
(nil)))
(insn 8 7 9 2 (set (reg:CC 100 cc)
(compare:CC (reg:SI 114 [ D.4127 ])
(const_int 0 [0]))) /home/user/g.c:3 217 {*arm_cmpsi_insn}
(expr_list:REG_DEAD (reg:SI 114 [ D.4127 ])
(nil)))
In the combiner pass:
Successfully matched this instruction:
(set (reg:SI 114 [ D.4127 ])
(and:SI (reg:SI 1 r1 [ m ])
(const_int 128 [0x80])))
[...]
Failed to match this instruction:
(set (reg:CC_NOOV 100 cc)
(compare:CC_NOOV (zero_extract:SI (reg:SI 1 r1 [ m ])
(const_int 1 [0x1])
(const_int 7 [0x7]))
(const_int 0 [0])))
I don't get why relaxing the restrictions for the
andsi3_compare0_scratch pattern results in a mismatch for the
zeroextractsi_compare0_scratch one.
Any ideas?
Thanks,
Tilman
diff --git a/gcc/config/arm/arm.md b/gcc/config/arm/arm.md
index 64888f9..3724a8d 100644
--- a/gcc/config/arm/arm.md
+++ b/gcc/config/arm/arm.md
@@ -2200,7 +2200,7 @@
[(set (reg:CC_NOOV CC_REGNUM)
(compare:CC_NOOV
(and:SI (match_operand:SI 0 s_register_operand r,r,r)
-(match_operand:SI 1 arm_not_operand I,K,r))
+(match_operand:SI 1 reg_or_int_operand I,K,r))
(const_int 0)))
(clobber (match_scratch:SI 2 =X,r,X))]
TARGET_32BIT
Re: FW: [PATCH] [MIPS] microMIPS gcc support
On Tue, 19 Feb 2013, Richard Sandiford wrote:
Index: gcc.target/mips/umips-lwp-swp-2.c
===
--- gcc.target/mips/umips-lwp-swp-2.c (revision 0)
+++ gcc.target/mips/umips-lwp-swp-2.c (revision 0)
@@ -0,0 +1,19 @@
+/* { dg-options -mgp32 (-mmicromips) } */
+/* { dg-skip-if code quality test { *-*-* } { -O0 -O1 -Os } { }
} */
+int a[2];
+
+MICROMIPS f (b)
+{
+ unsigned int i;
+ int *p;
+ for (p = a[b], i = b; --i ~0; )
+*--p = i * 3 + (int)a;
+
+}
+
+MICROMIPS main ()
+{
+ a[0] = a[1] = 0;
+ f (2);
+}
+/* { dg-final { scan-assembler \tswp\t\\\$3,0\\(\\\$3\\) } }*/
Is this a test of the swap_p case? Thanks if so, but could you add a
comment saying where the SWP gets generated, and why the test needs
to be skipped at -O1 and -Os?
TBH, it is -Os where size really matters, so IMHO it is the -Os
optimisation level where LWP/SWP should be used where possible in the
first place, even if it hurts performance for some reason (and where e.g.
-O2 might decide to go for individual accesses instead).
Not to be meant as a show-stopper for this initial implementation, but if
the optimisation does not work for LWP/SWP at -Os for some reason, then I
think there's something going seriously wrong somewhere (instruction
lengths set wrong for example?) which looks to me worth investigating and
acting upon accordingly as the next step.
I saw cases with instruction lengths set too high with our trunk just
about yesterday and MIPS16 code, e.g.:
.file 1 foo.c
.section .mdebug.abi32
.previous
.gnu_attribute 4, 1
.abicalls
.option pic0
.text
.align 2
.globl foo
.setmips16
.entfoo
.type foo, @function
foo:
.frame $sp,32,$31 # vars= 0, regs= 1/0, args= 16, gp= 8
.mask 0x8000,-4
.fmask 0x,0
save32,$31 # 11 *mips16e_save_restore [length = 4]
.setnoreorder
.setnomacro
jal bar # 22 call_split/2[length = 8]
li $4,0 # 5*movsi_mips16/4 [length = 4]
.setmacro
.setreorder
restore 32,$31 # 18 *mips16e_save_restore [length = 4]
j $31 # 20 simple_return_internal [length = 2]
.endfoo
.size foo, .-foo
-- it looks all wrong to me except for the final J. Of course I realise
where it originally comes from, but I think it would be good if the
initial pessimistic estimates were actually adjusted as more is known
about actual insns produced.
I wonder if we may have similar issues with microMIPS instruction size
calculation triggering here.
Maciej
[PATCH] Fix ccp (PR tree-optimization/56396)
Hi!
On the following patch gcc ICEs because malloc memory is corrupted.
The problem is that const_val array is allocated at the start of the pass,
but during the execution of ccp some new SSA_NAMEs are created
(update_call_from_tree if I remember well).
The following patch fixes that by making const_val a vector instead of
array, and growing it when needed.
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
2013-02-19 Jakub Jelinek ja...@redhat.com
PR tree-optimization/56396
* tree-ssa-ccp.c (const_val): Change into vector.
(get_value): Use !const_val.exists () instead of const_val == NULL,
call safe_grow_cleared method if SSA_NAME_VERSION (var) is above
last vector element.
(set_value_varying): Call safe_grow_cleared method if
SSA_NAME_VERSION (var) is above last vector element.
(set_lattice_value): Likewise.
(ccp_initialize): Call const_val.create (num_ssa_names) instead
of XCNEWVEC.
(do_dbg_cnt): Call safe_grow_cleared method if num_ssa_names
is above vector length.
(ccp_finalize): Call const_val.release ().
* gcc.dg/pr56396.c: New test.
--- gcc/tree-ssa-ccp.c.jj 2013-01-11 09:02:48.0 +0100
+++ gcc/tree-ssa-ccp.c 2013-02-19 20:04:21.194681227 +0100
@@ -155,13 +155,13 @@ struct prop_value_d {
typedef struct prop_value_d prop_value_t;
-/* Array of propagated constant values. After propagation,
+/* Vector of propagated constant values. After propagation,
CONST_VAL[I].VALUE holds the constant value for SSA_NAME(I). If
the constant is held in an SSA name representing a memory store
(i.e., a VDEF), CONST_VAL[I].MEM_REF will contain the actual
memory reference used to store (i.e., the LHS of the assignment
doing the store). */
-static prop_value_t *const_val;
+static vecprop_value_t const_val;
static void canonicalize_float_value (prop_value_t *);
static bool ccp_fold_stmt (gimple_stmt_iterator *);
@@ -295,9 +295,12 @@ get_value (tree var)
{
prop_value_t *val;
- if (const_val == NULL)
+ if (!const_val.exists ())
return NULL;
+ if (SSA_NAME_VERSION (var) = const_val.length ())
+const_val.safe_grow_cleared (SSA_NAME_VERSION (var) + 32);
+
val = const_val[SSA_NAME_VERSION (var)];
if (val-lattice_val == UNINITIALIZED)
*val = get_default_value (var);
@@ -333,8 +336,12 @@ get_constant_value (tree var)
static inline void
set_value_varying (tree var)
{
- prop_value_t *val = const_val[SSA_NAME_VERSION (var)];
+ prop_value_t *val;
+ if (SSA_NAME_VERSION (var) = const_val.length ())
+const_val.safe_grow_cleared (SSA_NAME_VERSION (var) + 32);
+
+ val = const_val[SSA_NAME_VERSION (var)];
val-lattice_val = VARYING;
val-value = NULL_TREE;
val-mask = double_int_minus_one;
@@ -428,7 +435,12 @@ static bool
set_lattice_value (tree var, prop_value_t new_val)
{
/* We can deal with old UNINITIALIZED values just fine here. */
- prop_value_t *old_val = const_val[SSA_NAME_VERSION (var)];
+ prop_value_t *old_val;
+
+ if (SSA_NAME_VERSION (var) = const_val.length ())
+const_val.safe_grow_cleared (SSA_NAME_VERSION (var) + 32);
+
+ old_val = const_val[SSA_NAME_VERSION (var)];
canonicalize_float_value (new_val);
@@ -713,7 +725,7 @@ ccp_initialize (void)
{
basic_block bb;
- const_val = XCNEWVEC (prop_value_t, num_ssa_names);
+ const_val.create (num_ssa_names);
/* Initialize simulation flags for PHI nodes and statements. */
FOR_EACH_BB (bb)
@@ -774,6 +786,9 @@ static void
do_dbg_cnt (void)
{
unsigned i;
+
+ if (num_ssa_names = const_val.length ())
+const_val.safe_grow_cleared (num_ssa_names);
for (i = 0; i num_ssa_names; i++)
{
if (!dbg_cnt (ccp))
@@ -829,9 +844,8 @@ ccp_finalize (void)
something_changed = substitute_and_fold (get_constant_value,
ccp_fold_stmt, true);
- free (const_val);
- const_val = NULL;
- return something_changed;;
+ const_val.release ();
+ return something_changed;
}
--- gcc/testsuite/gcc.dg/pr56396.c.jj 2013-02-19 20:02:48.663057957 +0100
+++ gcc/testsuite/gcc.dg/pr56396.c 2013-02-19 19:59:00.0 +0100
@@ -0,0 +1,22 @@
+/* PR tree-optimization/56396 */
+/* { dg-do compile } */
+/* { dg-options -O2 -fpic -g } */
+
+struct S { char *s; int z; };
+struct T { int t; } *c, u;
+void bar (int, const char *);
+
+inline void *
+foo (void *x, char *y, int z)
+{
+ struct S s;
+ char b[256];
+ s.s = b;
+ s.z = __builtin___sprintf_chk (s.s, 1, __builtin_object_size (s.s, 2),
Require);
+ if (s.z 0)
+bar (u.t | c-t, rls);
+ if (foo (x, s.s, s.z))
+{
+}
+ return (void *) 0;
+}
Jakub
[Patch, libfortran] PR 30162 pipe I/O regression with 4.7/4.8
Hi, attached is an attempt to fix writing formatted sequential I/O to a pipe (The PR was reopened in comment #22, which refers to formatted I/O so the PR title is incorrect). I think the underlying reason was that the introduction of the ssize() member function led to a change in semantics (size of a non-seekable fd is now 0 rather than -1), and thus in io/open.c test_endfile() we incorrectly concluded that the file was not positioned at the end, and thus we try to truncate after writing, leading to the failure. At the same time, the patch reverts the previous fix; Unformatted sequential requires seeking due to updating the record markers, I think that issue can be closed as WONTFIX. Regtested on x86_64-unknown-linux-gnu, Ok for trunk/4.7? 2013-02-19 Janne Blomqvist j...@gcc.gnu.org PR libfortran/30162 * io/open.c (test_endfile): Call stell only if size != 0. * io/unix.c (raw_tell): Revert r194679. (raw_size): Return size field only for regular files, otherwise 0. -- Janne Blomqvist pipe2.diff Description: Binary data
Re: Patch for 4.7: Avoid subreg'ing VFP D registers in big-endian mode
On Fri, Feb 15, 2013 at 10:40 PM, Seth LaForge set...@google.com wrote: gcc 4.7.2 generates incorrect code for big-endian ARM VFP processors when storing a local double to a packed memory location, as described in bug 56351. http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56351 A fix has been submitted to the trunk by Julian Brown: http://gcc.gnu.org/git/?p=gcc.git;a=patch;h=4f0e4afcb2a9d345ed7e9c14233fb97d3a3e8d3c This fix is not included in the latest 4.7 snapshot. Can we please include it in the 4.7 branch so that this serious codegen bug is fixed in 4.7.3? Unfortunately, it's a not a trivial backport, as it appears that support for ARM FPA floating point has been removed from gcc since 4.7. Below is my best attempt at a backport - it works for me, on a big-endian VFP processor. I'm not familiar with this code, though - Julian, can you take a look and confirm whether I'm doing this right? 2013-02-15 Seth LaForge set...@google.com * config/arm/arm.h (CANNOT_CHANGE_MODE_CLASS): Avoid subreg'ing VFP D registers in big-endian mode. Backported from a change by Julian Brown jul...@codesourcery.com This is not the correct form of a changelog entry . Please see other examples of Backports in the Changelog file for the 4.7 branch. Instead please use something like this. DATE yourname youremailid TABBackport TAB2012-10-22 Julian Brown jul...@codesourcery.com * config/arm/arm.h (CANNOT_CHANGE_MODE_CLASS): Avoid subreg'ing VFP D registers in big-endian mode. Ok to backport provided no regressions when running the testsuite in big endian mode. diff -urp gcc-4.7.2/gcc/config/arm/arm.h gcc-4.7.2.new/gcc/config/arm/arm.h --- gcc-4.7.2/gcc/config/arm/arm.h 2012-01-09 20:14:09.0 -0800 +++ gcc-4.7.2.new/gcc/config/arm/arm.h 2013-02-15 14:03:09.113531299 -0800 @@ -1123,11 +1123,18 @@ enum reg_class /* FPA registers can't do subreg as all values are reformatted to internal precision. In VFPv1, VFP registers could only be accessed in the mode they were set, so subregs would be invalid there too. However, we don't - support VFPv1 at the moment, and the restriction was lifted in VFPv2. */ + support VFPv1 at the moment, and the restriction was lifted in VFPv2. + In big-endian mode, modes greater than word size (i.e. DFmode) are stored in + VFP registers in little-endian order. We can't describe that accurately to + GCC, so avoid taking subregs of such values. */ #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \ - (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \ - ? reg_classes_intersect_p (FPA_REGS, (CLASS)) \ - : 0) + (TARGET_VFP \ + ? TARGET_BIG_END \ + (GET_MODE_SIZE (FROM) UNITS_PER_WORD \ + || GET_MODE_SIZE (TO) UNITS_PER_WORD) \ + reg_classes_intersect_p (VFP_REGS, (CLASS)) \ + : GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \ + reg_classes_intersect_p (FPA_REGS, (CLASS))) Watch out for alignment of the \'s though it might be your mailer munging things rather than anything else. Can you please make sure that the \'s are aligned to the same column please ? This is why I prefer to get git to generate out patches using git-format-patch and attach them as text files to all my mails otherwise it seems to mess things up like this. Thanks, Ramana
Re: [Patch, microblaze]: Handle 0x80000000 as 32bit signed dividend
On 02/10/2013 10:38 PM, David Holsgrove wrote: Handle 0x8000 as 32bit signed dividend Changelog 2013-02-11 Edgar E. Iglesias edgar.igles...@gmail.com * config/microblaze/modsi3.S (modsi3): Fix case with 0x8000 as dividend. Committed revision 196156. -- Michael Eagerea...@eagercon.com 1960 Park Blvd., Palo Alto, CA 94306 650-325-8077
Re: [Patch, microblaze]: Avoid PC relative branches between sections
On 02/10/2013 10:38 PM, David Holsgrove wrote: Avoid PC relative branches between sections by further checking that the symbol we are calling is declared as a function before using PC relative calls. Changelog 2013-02-11 Edgar E. Iglesias edgar.igles...@gmail.com * config/microblaze/microblaze.md (call_value_intern): Check symbol is function before branching Committed revision 196157. -- Michael Eagerea...@eagercon.com 1960 Park Blvd., Palo Alto, CA 94306 650-325-8077
Re: [Patch, microblaze]: Add support for the CLZ insn
On 02/10/2013 10:39 PM, David Holsgrove wrote: Add support for the CLZ insn Will be used if pattern-compare is enabled and the targeted core is newer than v8.10.a. Changelog 2013-02-11 Edgar E. Iglesias edgar.igles...@gmail.com * config/microblaze/microblaze.c: microblaze_has_clz = 0 Add version check for v8.10.a to enable microblaze_has_clz * config/microblaze/microblaze.h: Add TARGET_HAS_CLZ as combined version and TARGET_PATTERN_COMPARE check * config/microblaze/microblaze.md: New clzsi2 instruction Committed revision 196158. -- Michael Eagerea...@eagercon.com 1960 Park Blvd., Palo Alto, CA 94306 650-325-8077
[v3] correct doxygen parse
Fixups for doxygen. Since 2012-11-20, most of C++11 has been ignored by doxygen, resulting in some missing modules documentation even in HTML output. Jakub, this fixes the libstdc++.tag validation errors. I've regenerated the gcc.gnu.org docs for 4.8.0. They look equivalent or better than 4.7.2 to me now. tested x86/linux -benjamin2013-02-19 Benjamin Kosnik b...@redhat.com * doc/doxygen/user.cfg.in: Set __cplusplus to 201103L. Change to _GLIBCXX_INCLUDE_AS_CXX11. DIRECTORY_GRAPH, MARKDOWN_SUPPORT, AUTOLINK_SUPPORT to NO. Update to doxygen 1.8.3.1. * include/bits/stl_pair.h: Add to utilities group. * include/std/tuple: Same. * include/std/typeindex: Same. * include/bits/stringfwd.h: Fix markup. * include/std/limits: Same. * include/std/type_traits: Same. * include/tr1/memory: Same. * include/tr1/regex: Same. * scripts/run_doxygen: Comment. * testsuite/20_util/uses_allocator/cons_neg.cc: Fixup line numbers. diff --git a/libstdc++-v3/doc/doxygen/user.cfg.in b/libstdc++-v3/doc/doxygen/user.cfg.in index 5af636b..4a3afc4 100644 --- a/libstdc++-v3/doc/doxygen/user.cfg.in +++ b/libstdc++-v3/doc/doxygen/user.cfg.in @@ -1,4 +1,4 @@ -# Doxyfile 1.8.2 +# Doxyfile 1.8.3.1 # This file describes the settings to be used by the documentation system # doxygen (www.doxygen.org) for a project. @@ -252,14 +252,15 @@ EXTENSION_MAPPING = # can mix doxygen, HTML, and XML commands with Markdown formatting. # Disable only in case of backward compatibilities issues. -MARKDOWN_SUPPORT = YES +MARKDOWN_SUPPORT = NO -# When enabled doxygen tries to link words that correspond to documented classes, -# or namespaces to their corresponding documentation. Such a link can be -# prevented in individual cases by by putting a % sign in front of the word or -# globally by setting AUTOLINK_SUPPORT to NO. +# When enabled doxygen tries to link words that correspond to +# documented classes, or namespaces to their corresponding +# documentation. Such a link can be prevented in individual cases by +# by putting a % sign in front of the word or globally by setting +# AUTOLINK_SUPPORT to NO. -AUTOLINK_SUPPORT = YES +AUTOLINK_SUPPORT = NO # If you use STL classes (i.e. std::string, std::vector, etc.) but do not want # to include (a tag file for) the STL sources as input, then you should @@ -281,7 +282,12 @@ CPP_CLI_SUPPORT= NO SIP_SUPPORT= NO -# For Microsoft's IDL there are propget and propput attributes to indicate getter and setter methods for a property. Setting this option to YES (the default) will make doxygen replace the get and set methods by a property in the documentation. This will only work if the methods are indeed getting or setting a simple type. If this is not the case, or you want to show the methods anyway, you should set this option to NO. +# For Microsoft's IDL there are propget and propput attributes to indicate +# getter and setter methods for a property. Setting this option to YES (the +# default) will make doxygen replace the get and set methods by a property in +# the documentation. This will only work if the methods are indeed getting or +# setting a simple type. If this is not the case, or you want to show the +# methods anyway, you should set this option to NO. IDL_PROPERTY_SUPPORT = NO @@ -305,8 +311,7 @@ SUBGROUPING= YES # @ingroup) instead of on a separate page (for HTML and Man pages) or # section (for LaTeX and RTF). -#INLINE_GROUPED_CLASSES = NO -INLINE_GROUPED_CLASSES = YES +INLINE_GROUPED_CLASSES = NO # When the INLINE_SIMPLE_STRUCTS tag is set to YES, structs, classes, and # unions with only public data fields will be shown inline in the documentation @@ -314,7 +319,7 @@ INLINE_GROUPED_CLASSES = YES # documentation), provided this scope is documented. If set to NO (the default), # structs, classes, and unions are shown on a separate page (for HTML and Man # pages) or section (for LaTeX and RTF). - + INLINE_SIMPLE_STRUCTS = NO # When TYPEDEF_HIDES_STRUCT is enabled, a typedef of a struct, union, or enum @@ -542,7 +547,8 @@ GENERATE_BUGLIST = YES GENERATE_DEPRECATEDLIST= YES # The ENABLED_SECTIONS tag can be used to enable conditional -# documentation sections, marked by \if sectionname ... \endif. +# documentation sections, marked by \if section-label ... \endif +# and \cond section-label ... \endcond blocks. ENABLED_SECTIONS = @enabled_sections@ @@ -600,7 +606,8 @@ LAYOUT_FILE= # requires the bibtex tool to be installed. See also # http://en.wikipedia.org/wiki/BibTeX for more info. For LaTeX the style # of the bibliography can be controlled using LATEX_BIB_STYLE. To use this -# feature you need bibtex and perl available in the search path. +# feature you need bibtex and perl available in the search path. Do not use +# file names with spaces, bibtex cannot handle them. CITE_BIB_FILES = @@ -952,6 +959,13 @@ FILTER_SOURCE_FILES= NO
Re: Version specific onlinedocs page
On Mon, 18 Feb 2013, Tobias Burnus wrote: If we means that you are volunteering ;-), absolutely, go for it! ;-) How about the following patch? If it is okay, I will add the remaining index.html for 4.6 and 4.7 and update gcc-4.6/index.html – and then commit it. Looks good to me. Perhaps you can add a note for the release manager how to create index.html (copy from previous release and adjust)? Gerald
Re: [C++ Patch] PR 56373
OK. Jason
Re: RFC: [PATCH,ARM] Fix 56110
Tilman Sauerbeck [2013-02-19 23:26]: However it breaks the case where the 2nd operand is a const_int that *can* be used as an immediate (eg 0x80), and ends up generating the AND/CMP combination. ... and that would be because I changed the operand patterns in zeroextractsi_compare0_scratch and didn't restore the originals when trying the new diff :/ So the diff I posted last night might do the trick. If there's no obvious reason why it would be wrong, I'll start a bootstrap and regression test tonight. Thanks, Tilman -- A: Because it messes up the order in which people normally read text. Q: Why is top-posting such a bad thing? A: Top-posting. Q: What is the most annoying thing on usenet and in e-mail?
