Patchwork PING: Re: [patch] implement simd loops in trunk (OMP_SIMD)

login
register
mail settings
Submitter Aldy Hernandez
Date July 29, 2013, 5:13 a.m.
Message ID <51F5F9EB.2070205@redhat.com>
Download mbox | patch
Permalink /patch/262673/
State New
Headers show

Comments

Aldy Hernandez - July 29, 2013, 5:13 a.m.
On 07/09/13 18:02, Aldy Hernandez wrote:

PING PING folks.  Mr. Biener and/or Mr. Henderson.  Could either of you 
look at this, especially the vectorizer bits.  This is needed for the 
CILK_SIMD patchset that Jason is currently looking at.

Thanks.
Aldy

> Hi folks.
>
> I have distilled the relevant machinery to implement and handle simd
> loops from the gomp-4_0-branch.  This will be used by both gomp4
> (#pragma omp simd) and Cilk Plus (#pragma simd).
>
> All of it is Jakub's code.  There is nothing new here, as I believe
> Jakub and the Richards have talked about all this at length.  Included
> in the patch are the vectorizer changes that have been previously
> discussed, all from the gomp-4_0-branch.
>
> It would be ideal to get an ok from either Mr. Biener or Mr. Henderson
> on the vectorizer changes.
>
> The patch I am formally submitting is the attached patch named OMP_SIMD.
>   I am including the second (CILK_SIMD), which I will submit
> independently, as reference to how it will be used (if for some reason
> it isn't clear from the gomp4 branch usage of OMP_SIMD).
>
> Tested on x86-64 Linux with and without the Cilk Plus bits, not to
> mention that this has been extensively tested on the gomp4 branch.
>
> OK for trunk?
* Makefile.in (omp-low.o): Depend on $(TARGET_H).
	* cfgloop.h (struct loop): Add safelen, force_vect, simduid.
	* function.h (struct function): Add has_force_vect_loops and
	has_simduid_loops.
	* gimple-pretty-print.c (dump_gimple_omp_for): Handle
	GF_OMP_FOR_KIND*.
	* gimple.c (gimple_build_omp_critical): Add KIND argument and
	handle it.
	* gimple.def: Update CLAUSES comments.
	* gimple.h (enum gf_mask): Add GF_OMP_FOR_KIND_{FOR,SIMD}.
	(gimple_build_omp_for): Add argument to prototype.
	(gimple_omp_for_kind): New.
	(gimple_omp_for_set_kind): New.
	* gimplify.c (enum gimplify_omp_var_data): Add GOVD_LINEAR to
	GOVD_DATA_SHARE_CLASS.
	(enum omp_region_type): Add ORT_SIMD.
	(gimple_add_tmp_var): Handle ORT_SIMD.
	(gimplify_var_or_parm_decl): Same.
	(is_gimple_stmt): Same.
	(omp_firstprivatize_variable): Same.
	(omp_add_variable): Only use splay_tree_insert if lookup failed.
	(omp_notice_variable): Handle ORT_SIMD.
	(omp_is_private): Add SIMD argument and handle it as well as
	ORT_SIMD.
	(omp_check_private): Handle ORT_SIMD.
	(gimplify_scan_omp_clauses): Handle OMP_CLAUSE_LINEAR and
	OMP_CLAUSE_SAFELEN.
	(gimplify_adjust_omp_clauses_1): Handle GOVD_LINEAR.
	Handle OMP_CLAUSE_LASTPRIVATE.
	(gimplify_adjust_omp_clauses): Handle OMP_CLAUSE_LINEAR and
	OMP_CLAUSE_SAFELEN.
	(gimplify_omp_for): Handle OMP_SIMD and OMP_CLAUSE_LINEAR.
	(gimplify_expr): Handle OMP_SIMD.
	* internal-fn.c (expand_GOMP_SIMD_LANE): New.
	(expand_GOMP_SIMD_VF): New.
	(expand_GOMP_SIMD_LAST_LANE): New.
	* internal-fn.def (GOMP_SIMD_LANE): New.
	(GOMP_SIMD_VF): New.
	(GOMP_SIMD_LAST_LANE): New.
	* omp-low.c: Include target.h.
	(extract_omp_for_data): Handle OMP_SIMD, OMP_CLAUSE_LINEAR,
	OMP_CLAUSE_SAFELEN.
	(check_omp_nesting_restrictions): Same.
	(omp_max_vf): New.
	(lower_rec_simd_input_clauses): New.
	(lower_rec_input_clauses): Handle OMP_SIMD, GF_OMP_FOR_KIND_SIMD,
	OMP_CLAUSE_LINEAR.
	(lower_lastprivate_clauses): Handle OMP_CLAUSE_LINEAR,
	GF_OMP_FOR_KIND_SIMD, OMP_SIMD.
	(expand_omp_build_assign): New.
	(expand_omp_for_init_counts): New.
	(expand_omp_for_init_vars): New.
	(extract_omp_for_update_vars): New.
	(expand_omp_for_generic): Use expand_omp_for_{init,update}_vars
	and rewrite accordingly.
	(expand_omp_simd): New.
	(expand_omp_for): Use expand_omp_simd.
	(lower_omp_for_lastprivate): Unshare vinit when appropriate.
	(lower_omp_for): Do not lower the body.
	* tree-data-ref (get_references_in_stmt): Allow IFN_GOMP_SIMD_LANE
	in their own loops.
	* tree-flow.h (find_omp_clause): Remove prototype.
	* tree-if-conv.c (main_tree_if_conversion): Run if doing if
	conversion, forcing vectorization of the loop, or if
	flag_tree_vectorize.
	(gate_tree_if_conversion): Similarly.
	* tree-inline.c (remap_gimple_stmt): Pass for kind argument to
	gimple_build_omp_for.
	(copy_cfg_body): set has_force_vect_loops and has_simduid_loops.
	* tree-parloops (create_parallel_loop): Pass kind argument to
	gimple_build_omp_for.
	* tree-pretty-print.c (dump_omp_clause): Add cases for
	OMP_CLAUSE_UNIFORM, OMP_CLAUSE_LINEAR, OMP_CLAUSE_SAFELEN,
	OMP_CLAUSE__SIMDUID_.
	(dump_generic_node): Handle OMP_SIMD.
	* tree-ssa-ccp.c (likely_value): Handle IFN_GOMP_SIMD*.
	* tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Do not
	unroll OMP_SIMD loops here.
	* tree-ssa-loop.c (gate_tree_vectorize): Run if
	has_force_vect_loops.
	* tree-vect-data-refs.c (vect_analyze_data_ref_dependence): Handle
	loop->safelen
	(vect_analyze_data_refs): Handle simd loops.
	* tree-vect-loop.c (vectorizable_live_operation): Handle
	IFN_GOMP_SIMD*.
	* tree-vect-stmts.c (vectorizable_call): Handle
	IFN_GOMP_SIMD_LANE.
	(vectorizable_store): Handle STMT_VINFO_SIMD_LANE_ACCESS_P.
	(vectorizable_load): Same.
	* tree-vectorizer.c: Include hash-table.h and
	tree-ssa-propagate.h.
	(struct simduid_to_vf): New.
	(simduid_to_vf::hash): New.
	(simduid_to-vf::equal): New.
	(struct decl_to_simduid): New.
	(decl_to_simduid::hash): New.
	(decl_to_simduid::equal): New.
	(adjust_simduid_builtins): New.
	(struct note_simd_array_uses_struct): New.
	(note_simd_array_uses_cb): New.
	(note_simd_array_uses): New.
	(vectorize_loops): Handle simd hints and adjust simd builtins
	accordingly.
	* tree-vectorizer.h (struct _stmt_vec_info): Add
	simd_lane_access_p field.
	(STMT_VINFO_SIMD_LANE_ACCESS_P): New macro.
	* tree.c (omp_clause_num_ops): Add entries for OMP_CLAUSE_LINEAR,
	OMP_CLAUSE_SAFELEN, OMP_CLAUSE__SIMDUID_, OMP_CLAUSE_UNIFORM.
	(omp_clause_code_name): Same.
	(walk_tree_1): Handle OMP_CLAUSE_UNIFORM, OMP_CLAUSE_SAFELEN,
	OMP_CLAUSE__SIMDUID_, OMP_CLAUSE_LINEAR.
	* tree.def (OMP_SIMD): New entry.
	* tree.h (enum omp_clause_code): Add entries for
	OMP_CLAUSE_LINEAR, OMP_CLAUSE_UNIFORM, OMP_CLAUSE_SAFELEN,
	OMP_CLAUSE__SIMDUID_.
	(OMP_CLAUSE_DECL): Adjust range for new clauses.
	(OMP_CLAUSE_LINEAR_NO_COPYIN): New.
	(OMP_CLAUSE_LINEAR_NO_COPYOUT): New.
	(OMP_CLAUSE_LINEAR_STEP): New.
	(OMP_CLAUSE_SAFELEN_EXPR): New.
	(OMP_CLAUSE__SIMDUID__DECL): New.
	(find_omp_clause): New prototype.
cp/
	* cp-tree.h (CP_OMP_CLAUSE_INFO): Adjust range for new clauses.
Richard Henderson - Aug. 1, 2013, 7:38 p.m.
> +      if (simd
> +         /*
> +         || (fd->sched_kind == OMP_CLAUSE_SCHEDULE_STATIC
> +         && !fd->have_ordered)*/)

Debugging leftovers or what?

> +  /* Enforce simdlen 1 in simd loops with data sharing clauses referencing
> +     variable sized vars.  That is unnecessarily hard to support and very
> +     unlikely to result in vectorized code anyway.  */
> +  if (is_simd)
> +    for (c = clauses; c ; c = OMP_CLAUSE_CHAIN (c))
> +      switch (OMP_CLAUSE_CODE (c))
> +       {
> +       case OMP_CLAUSE_REDUCTION:
> +         if (OMP_CLAUSE_REDUCTION_PLACEHOLDER (c))
> +           max_vf = 1;
> +         /* FALLTHRU */
> +       case OMP_CLAUSE_PRIVATE:
> +       case OMP_CLAUSE_FIRSTPRIVATE:
> +       case OMP_CLAUSE_LASTPRIVATE:
> +       case OMP_CLAUSE_LINEAR:
> +         if (is_variable_sized (OMP_CLAUSE_DECL (c)))
> +           max_vf = 1;
> +         break;
> +       default:
> +         continue;
> +       }

Comment is wrong, code is right, adjusting max_vf not simdlen.

> +  /* If max_vf is non-NULL, then we can use only vectorization factor
> +     up to the max_vf we chose.  So stick it into safelen clause.  */
> +  if (max_vf)
> +    {
> +      tree c = find_omp_clause (gimple_omp_for_clauses (ctx->stmt),
> +                               OMP_CLAUSE_SAFELEN);

First, non-zero, not non-null -- max_vf is not a pointer.

Second, I don't understand why we're adjusting safelen.  The VF we choose
for optimizing the loop (even 1), does not change the fact that there are
no dependencies between loop iterations larger than that.

Did you want to be adding a _max_vf_ attribute to record stuff that we
learned while examining the omp loop?  E.g. the max_vf=1 reduction above?

Given the only adjustment we make to max_vf is to disable vectorization,
did we in fact want to discard the simd attribute instead, making this a
"regular" openmp loop?

> +       if (__builtin_expect (N32 cond3 N31, 0)) goto ZERO_ITER_BB;
> +       if (cond3 is <)
> +         adj = STEP3 - 1;
> +       else
> +         adj = STEP3 + 1;
> +       count3 = (adj + N32 - N31) / STEP3;
> +       if (__builtin_expect (N22 cond2 N21, 0)) goto ZERO_ITER_BB;
> +       if (cond2 is <)
> +         adj = STEP2 - 1;
> +       else
> +         adj = STEP2 + 1;
> +       count2 = (adj + N22 - N21) / STEP2;
> +       if (__builtin_expect (N12 cond1 N11, 0)) goto ZERO_ITER_BB;

CEIL_DIV_EXPR, instead of TRUNC_DIV_EXPR and adjusting by hand?  Unless we
can't generate the same code in the end because generically we don't know that
the values involved must be negative for GT?

I wonder if we do better mooshing all of the BBs together, creating one larger
BB with all the computation and the unexpected jump at the end.  I.e.

  bool zero3, zero2, zero1, zero;

  zero3 = N32 c3 N31;
  count3 = (N32 - N31) /[cl] STEP3;
  zero2 = N22 c2 N21;
  count2 = (N22 - N21) /[cl] STEP2;
  zero1 = N12 c1 N11;
  count1 = (N12 - N11) /[cl] STEP1;
  zero = zero3 || zero2 || zero1;
  count = count1 * count2 * count3;
  if (__builtin_expect(zero, false)) goto zero_iter_bb;

After all, we expect the zero=false, and thus we expect to have to evaluate all
of the comparison expressions, so short-circuiting oughtn't be a win.  Since
the condition isn't protecting a denominator, we're not concerned about
divide-by-zero, so we can fully evaluate count even if a numerator turned out
to be wrong.

It seems like putting this all together would create much better scheduling
opportunities, and less pressure on the chip's branch predictor.

> @@ -291,6 +300,15 @@ vect_analyze_data_ref_dependence (struct data_dependence_re

You've two adjustments in this function.  First hunk for "dont_know", which is
fine; second hunk for "data is bad", which I suppose is really a form of
dont_know and thus also fine.

But it seems like there should be a third hunk in the "know" section, in which
we perform some MAX(dist, loop->safelen) computation inside that loop.

Alternately, why are we patching this function at all, rather than whatever bit
of code creates the data_dependence_relation data?

> +  hash_table <simduid_to_vf> simduid_to_vf_htab;
> +  hash_table <decl_to_simduid> decl_to_simduid_htab;

Why two hashes?  Seems like you can map from decl to vf directly.  At what
point do you have a simduid integer, but not the decl from whence it came?



r~
Aldy Hernandez - Aug. 8, 2013, 12:06 p.m.
>
>> +  hash_table <simduid_to_vf> simduid_to_vf_htab;
>> +  hash_table <decl_to_simduid> decl_to_simduid_htab;
>
> Why two hashes?  Seems like you can map from decl to vf directly.  At what
> point do you have a simduid integer, but not the decl from whence it came?

decl_to_simduid seems to be a misnomer.  What it really is, is a 
"omp_simd_array_to_simduid" map.  OMP simd array, being the temporary 
array used to hold simd lane private variable copies and such.

The hash is not really mapping the simd DECL to the simduid, since 
that's just a matter of DECL_UID(simduid), but the OMP simd array to the 
index used to reference it (simduid), like thus:

		_7 = GOMP_SIMD_LANE (simduid.0)
		...
		...
		D.1737[_7] = stuff;

decl_to_simduid is a mapping from the simduid.0 to the D.1737[] (the OMP 
simd array).

Agreed, or am I missing something?
Richard Henderson - Aug. 8, 2013, 4:42 p.m.
On 08/08/2013 02:06 AM, Aldy Hernandez wrote:
> The hash is not really mapping the simd DECL to the simduid, since that's just
> a matter of DECL_UID(simduid), but the OMP simd array to the index used to
> reference it (simduid), like thus:
> 
>         _7 = GOMP_SIMD_LANE (simduid.0)
>         ...
>         ...
>         D.1737[_7] = stuff;
> 
> decl_to_simduid is a mapping from the simduid.0 to the D.1737[] (the OMP simd
> array).
> 
> Agreed, or am I missing something?

Agreed.  Yeah, that's a bit confusing.


r~
Aldy Hernandez - Aug. 8, 2013, 7:15 p.m.
On 08/08/13 18:42, Richard Henderson wrote:
> On 08/08/2013 02:06 AM, Aldy Hernandez wrote:
>> The hash is not really mapping the simd DECL to the simduid, since that's just
>> a matter of DECL_UID(simduid), but the OMP simd array to the index used to
>> reference it (simduid), like thus:
>>
>>          _7 = GOMP_SIMD_LANE (simduid.0)
>>          ...
>>          ...
>>          D.1737[_7] = stuff;
>>
>> decl_to_simduid is a mapping from the simduid.0 to the D.1737[] (the OMP simd
>> array).
>>
>> Agreed, or am I missing something?
>
> Agreed.  Yeah, that's a bit confusing.

Ok, I will change the hash name and add some comments.
Jakub Jelinek - Aug. 19, 2013, 8:55 a.m.
On Thu, Aug 08, 2013 at 09:15:54PM +0200, Aldy Hernandez wrote:
> On 08/08/13 18:42, Richard Henderson wrote:
> >On 08/08/2013 02:06 AM, Aldy Hernandez wrote:
> >>The hash is not really mapping the simd DECL to the simduid, since that's just
> >>a matter of DECL_UID(simduid), but the OMP simd array to the index used to
> >>reference it (simduid), like thus:
> >>
> >>         _7 = GOMP_SIMD_LANE (simduid.0)
> >>         ...
> >>         ...
> >>         D.1737[_7] = stuff;
> >>
> >>decl_to_simduid is a mapping from the simduid.0 to the D.1737[] (the OMP simd
> >>array).
> >>
> >>Agreed, or am I missing something?
> >
> >Agreed.  Yeah, that's a bit confusing.
> 
> Ok, I will change the hash name and add some comments.

Do you have such an incremental patch?  I'd like to apply it to
gomp-4_0-branch so that we are in sync.

	Jakub
Jakub Jelinek - Aug. 19, 2013, 9:35 a.m.
On Thu, Aug 01, 2013 at 09:38:31AM -1000, Richard Henderson wrote:
> > +      if (simd
> > +         /*
> > +         || (fd->sched_kind == OMP_CLAUSE_SCHEDULE_STATIC
> > +         && !fd->have_ordered)*/)
> 
> Debugging leftovers or what?

gomp-4_0-branch contains also the || there, but for the trunk merge
I think Aldy can leave that comment out, I'll deal with it when merging the
trunk changes back to gomp-4_0-branch.

> > +  /* Enforce simdlen 1 in simd loops with data sharing clauses referencing
> > +     variable sized vars.  That is unnecessarily hard to support and very
> > +     unlikely to result in vectorized code anyway.  */
> > +  if (is_simd)
> > +    for (c = clauses; c ; c = OMP_CLAUSE_CHAIN (c))
> > +      switch (OMP_CLAUSE_CODE (c))
> > +       {
> > +       case OMP_CLAUSE_REDUCTION:
> > +         if (OMP_CLAUSE_REDUCTION_PLACEHOLDER (c))
> > +           max_vf = 1;
> > +         /* FALLTHRU */
> > +       case OMP_CLAUSE_PRIVATE:
> > +       case OMP_CLAUSE_FIRSTPRIVATE:
> > +       case OMP_CLAUSE_LASTPRIVATE:
> > +       case OMP_CLAUSE_LINEAR:
> > +         if (is_variable_sized (OMP_CLAUSE_DECL (c)))
> > +           max_vf = 1;
> > +         break;
> > +       default:
> > +         continue;
> > +       }
> 
> Comment is wrong, code is right, adjusting max_vf not simdlen.

max_vf = 1 causes
          c = build_omp_clause (UNKNOWN_LOCATION, OMP_CLAUSE_SAFELEN);
          OMP_CLAUSE_SAFELEN_EXPR (c) = build_int_cst (integer_type_node,
                                                       max_vf);
so I think the comment should just be changed s/simdlen/safelen/, always get
those 2 wrong...  simdlen is for elemental functions, safelen for simd
loops.
> 
> > +  /* If max_vf is non-NULL, then we can use only vectorization factor
> > +     up to the max_vf we chose.  So stick it into safelen clause.  */
> > +  if (max_vf)
> > +    {
> > +      tree c = find_omp_clause (gimple_omp_for_clauses (ctx->stmt),
> > +                               OMP_CLAUSE_SAFELEN);
> 
> First, non-zero, not non-null -- max_vf is not a pointer.

Ack.

> Second, I don't understand why we're adjusting safelen.  The VF we choose
> for optimizing the loop (even 1), does not change the fact that there are
> no dependencies between loop iterations larger than that.

No, it adds depenencies.  We choose some size of the per-simd lane arrays,
and we must not allow larger vectorization factor than that, as every simd
lane should have different object.  The size is normally the maximum
possible value for a target, so it should be big enough, but we really need
it to be recorded, so that we don't generate invalid code.  max_vf = 1
is for the cases where we punt on handling it properly, those loops for now
can't be vectorized just because they had safelen clause (normal
vectorization rules will probably also punt on those).  In the future if
e.g. privatized per-SIMD lane VLAs are common we can improve that, but right
now I somehow doubt that would be the case.

> Did you want to be adding a _max_vf_ attribute to record stuff that we
> learned while examining the omp loop?  E.g. the max_vf=1 reduction above?
> 
> Given the only adjustment we make to max_vf is to disable vectorization,
> did we in fact want to discard the simd attribute instead, making this a
> "regular" openmp loop?

See above, it is not only about disabling it, we record the sizes of the
arrays there.  And, for SAFELEN clause with 1, we actually don't set
loop->force_vect and set loop->safelen to 0 (i.e. normal loop).

> > +       if (__builtin_expect (N32 cond3 N31, 0)) goto ZERO_ITER_BB;
> > +       if (cond3 is <)
> > +         adj = STEP3 - 1;
> > +       else
> > +         adj = STEP3 + 1;
> > +       count3 = (adj + N32 - N31) / STEP3;
> > +       if (__builtin_expect (N22 cond2 N21, 0)) goto ZERO_ITER_BB;
> > +       if (cond2 is <)
> > +         adj = STEP2 - 1;
> > +       else
> > +         adj = STEP2 + 1;
> > +       count2 = (adj + N22 - N21) / STEP2;
> > +       if (__builtin_expect (N12 cond1 N11, 0)) goto ZERO_ITER_BB;
> 
> CEIL_DIV_EXPR, instead of TRUNC_DIV_EXPR and adjusting by hand?  Unless we
> can't generate the same code in the end because generically we don't know that
> the values involved must be negative for GT?

This is just moving code around (and merging multiple copies of it), the old
code was also using TRUNC_DIV_EXPR.  I think we can try CEIL_DIV_EXPR, but
I'd strongly prefer to only do it incrementally and separately from these
changes.

> I wonder if we do better mooshing all of the BBs together, creating one larger
> BB with all the computation and the unexpected jump at the end.  I.e.
> 
>   bool zero3, zero2, zero1, zero;
> 
>   zero3 = N32 c3 N31;
>   count3 = (N32 - N31) /[cl] STEP3;
>   zero2 = N22 c2 N21;
>   count2 = (N22 - N21) /[cl] STEP2;
>   zero1 = N12 c1 N11;
>   count1 = (N12 - N11) /[cl] STEP1;
>   zero = zero3 || zero2 || zero1;
>   count = count1 * count2 * count3;
>   if (__builtin_expect(zero, false)) goto zero_iter_bb;
> 
> After all, we expect the zero=false, and thus we expect to have to evaluate all
> of the comparison expressions, so short-circuiting oughtn't be a win.  Since
> the condition isn't protecting a denominator, we're not concerned about
> divide-by-zero, so we can fully evaluate count even if a numerator turned out
> to be wrong.
> 
> It seems like putting this all together would create much better scheduling
> opportunities, and less pressure on the chip's branch predictor.

Ditto this.  Though, from my experience, on i?86/x86_64 the above results in
tons of setne/sete insns and it would surprise me if the generated code was
faster.

> > @@ -291,6 +300,15 @@ vect_analyze_data_ref_dependence (struct data_dependence_re
> 
> You've two adjustments in this function.  First hunk for "dont_know", which is
> fine; second hunk for "data is bad", which I suppose is really a form of
> dont_know and thus also fine.
> 
> But it seems like there should be a third hunk in the "know" section, in which
> we perform some MAX(dist, loop->safelen) computation inside that loop.

The user guarantee IMHO is useful only if you don't know or would give up
otherwise, if you can compute the exact distance, it is better to just use
what you've computed.

> Alternately, why are we patching this function at all, rather than whatever bit
> of code creates the data_dependence_relation data?

This is the spot where we first see *max_vf and can tweak it, in
tree-data-ref.c vectorization factor can't be adjusted.

> > +  hash_table <simduid_to_vf> simduid_to_vf_htab;
> > +  hash_table <decl_to_simduid> decl_to_simduid_htab;
> 
> Why two hashes?  Seems like you can map from decl to vf directly.  At what
> point do you have a simduid integer, but not the decl from whence it came?

simduid_to_vf is for mapping simduid to vectorization factor of that loop.
decl_to_simduid maps the array VAR_DECLs with "omp declare simd" attribute
to the simduid of the loop they are used in; you can have several such
arrays in a loop, or none.

	Jakub

Patch

diff --git a/gcc/Makefile.in b/gcc/Makefile.in
index ffd85e7..b264f1b 100644
--- a/gcc/Makefile.in
+++ b/gcc/Makefile.in
@@ -2557,7 +2557,7 @@  omp-low.o : omp-low.c $(CONFIG_H) $(SYSTEM_H) coretypes.h $(TM_H) $(TREE_H) \
    $(RTL_H) $(GIMPLE_H) $(TREE_INLINE_H) langhooks.h $(DIAGNOSTIC_CORE_H) \
    $(TREE_FLOW_H) $(FLAGS_H) $(EXPR_H) $(DIAGNOSTIC_CORE_H) \
    $(TREE_PASS_H) $(GGC_H) $(EXCEPT_H) $(SPLAY_TREE_H) $(OPTABS_H) \
-   $(CFGLOOP_H) tree-iterator.h gt-omp-low.h
+   $(CFGLOOP_H) tree-iterator.h $(TARGET_H) gt-omp-low.h
 tree-browser.o : tree-browser.c tree-browser.def $(CONFIG_H) $(SYSTEM_H) \
    coretypes.h $(HASH_TABLE_H) $(TREE_H) $(TREE_PRETTY_PRINT_H)
 omega.o : omega.c $(OMEGA_H) $(CONFIG_H) $(SYSTEM_H) coretypes.h $(DUMPFILE_H) \
diff --git a/gcc/cfgloop.h b/gcc/cfgloop.h
index 0f24799..cd2f527 100644
--- a/gcc/cfgloop.h
+++ b/gcc/cfgloop.h
@@ -168,6 +168,20 @@  struct GTY ((chain_next ("%h.next"))) loop {
      describes what is the state of the estimation.  */
   enum loop_estimation estimate_state;
 
+  /* If > 0, an integer, where the user asserted that for any
+     I in [ 0, nb_iterations ) and for any J in
+     [ I, min ( I + safelen, nb_iterations ) ), the Ith and Jth iterations
+     of the loop can be safely evaluated concurrently.  */
+  int safelen;
+
+  /* True if we should try harder to vectorize this loop.  */
+  bool force_vect;
+
+  /* For SIMD loops, this is a unique identifier of the loop, referenced
+     by IFN_GOMP_SIMD_VF, IFN_GOMP_SIMD_LANE and IFN_GOMP_SIMD_LAST_LANE
+     builtins.  */
+  tree simduid;
+
   /* Upper bound on number of iterations of a loop.  */
   struct nb_iter_bound *bounds;
 
diff --git a/gcc/cp/cp-tree.h b/gcc/cp/cp-tree.h
index 3e8043a..01d906e 100644
--- a/gcc/cp/cp-tree.h
+++ b/gcc/cp/cp-tree.h
@@ -4024,7 +4024,7 @@  more_aggr_init_expr_args_p (const aggr_init_expr_arg_iterator *iter)
    See semantics.c for details.  */
 #define CP_OMP_CLAUSE_INFO(NODE) \
   TREE_TYPE (OMP_CLAUSE_RANGE_CHECK (NODE, OMP_CLAUSE_PRIVATE, \
-				     OMP_CLAUSE_COPYPRIVATE))
+				     OMP_CLAUSE_LINEAR))
 
 /* Nonzero if this transaction expression's body contains statements.  */
 #define TRANSACTION_EXPR_IS_STMT(NODE) \
diff --git a/gcc/function.h b/gcc/function.h
index c651f50..d1f4ffc 100644
--- a/gcc/function.h
+++ b/gcc/function.h
@@ -650,6 +650,14 @@  struct GTY(()) function {
      adjusts one of its arguments and forwards to another
      function.  */
   unsigned int is_thunk : 1;
+
+  /* Nonzero if the current function contains any loops with
+     loop->force_vect set.  */
+  unsigned int has_force_vect_loops : 1;
+
+  /* Nonzero if the current function contains any loops with
+     nonzero value in loop->simduid.  */
+  unsigned int has_simduid_loops : 1;
 };
 
 /* Add the decl D to the local_decls list of FUN.  */
diff --git a/gcc/gimple-pretty-print.c b/gcc/gimple-pretty-print.c
index ddb086c..a698251 100644
--- a/gcc/gimple-pretty-print.c
+++ b/gcc/gimple-pretty-print.c
@@ -1101,8 +1101,20 @@  dump_gimple_omp_for (pretty_printer *buffer, gimple gs, int spc, int flags)
 
   if (flags & TDF_RAW)
     {
-      dump_gimple_fmt (buffer, spc, flags, "%G <%+BODY <%S>%nCLAUSES <", gs,
-                       gimple_omp_body (gs));
+      const char *kind;
+      switch (gimple_omp_for_kind (gs))
+	{
+	case GF_OMP_FOR_KIND_FOR:
+	  kind = "";
+	  break;
+	case GF_OMP_FOR_KIND_SIMD:
+	  kind = " simd";
+	  break;
+	default:
+	  gcc_unreachable ();
+	}
+      dump_gimple_fmt (buffer, spc, flags, "%G%s <%+BODY <%S>%nCLAUSES <", gs,
+		       kind, gimple_omp_body (gs));
       dump_omp_clauses (buffer, gimple_omp_for_clauses (gs), spc, flags);
       dump_gimple_fmt (buffer, spc, flags, " >,");
       for (i = 0; i < gimple_omp_for_collapse (gs); i++)
@@ -1118,7 +1130,17 @@  dump_gimple_omp_for (pretty_printer *buffer, gimple gs, int spc, int flags)
     }
   else
     {
-      pp_string (buffer, "#pragma omp for");
+      switch (gimple_omp_for_kind (gs))
+	{
+	case GF_OMP_FOR_KIND_FOR:
+	  pp_string (buffer, "#pragma omp for");
+	  break;
+	case GF_OMP_FOR_KIND_SIMD:
+	  pp_string (buffer, "#pragma omp simd");
+	  break;
+	default:
+	  gcc_unreachable ();
+	}
       dump_omp_clauses (buffer, gimple_omp_for_clauses (gs), spc, flags);
       for (i = 0; i < gimple_omp_for_collapse (gs); i++)
 	{
diff --git a/gcc/gimple.c b/gcc/gimple.c
index f507419..8f3b938 100644
--- a/gcc/gimple.c
+++ b/gcc/gimple.c
@@ -902,19 +902,21 @@  gimple_build_omp_critical (gimple_seq body, tree name)
 /* Build a GIMPLE_OMP_FOR statement.
 
    BODY is sequence of statements inside the for loop.
+   KIND is the `for' variant.
    CLAUSES, are any of the OMP loop construct's clauses: private, firstprivate,
    lastprivate, reductions, ordered, schedule, and nowait.
    COLLAPSE is the collapse count.
    PRE_BODY is the sequence of statements that are loop invariant.  */
 
 gimple
-gimple_build_omp_for (gimple_seq body, tree clauses, size_t collapse,
+gimple_build_omp_for (gimple_seq body, int kind, tree clauses, size_t collapse,
 		      gimple_seq pre_body)
 {
   gimple p = gimple_alloc (GIMPLE_OMP_FOR, 0);
   if (body)
     gimple_omp_set_body (p, body);
   gimple_omp_for_set_clauses (p, clauses);
+  gimple_omp_for_set_kind (p, kind);
   p->gimple_omp_for.collapse = collapse;
   p->gimple_omp_for.iter
       = ggc_alloc_cleared_vec_gimple_omp_for_iter (collapse);
diff --git a/gcc/gimple.def b/gcc/gimple.def
index acad572..f3652f4 100644
--- a/gcc/gimple.def
+++ b/gcc/gimple.def
@@ -287,7 +287,7 @@  DEFGSCODE(GIMPLE_OMP_ORDERED, "gimple_omp_ordered", GSS_OMP)
 
    BODY is a the sequence of statements to be executed by all threads.
 
-   CLAUSES is a TREE_LIST node with all the clauses.
+   CLAUSES is an OMP_CLAUSE chain with all the clauses.
 
    CHILD_FN is set when outlining the body of the parallel region.
    All the statements in BODY are moved into this newly created
@@ -306,7 +306,7 @@  DEFGSCODE(GIMPLE_OMP_PARALLEL, "gimple_omp_parallel", GSS_OMP_PARALLEL)
 
    BODY is a the sequence of statements to be executed by all threads.
 
-   CLAUSES is a TREE_LIST node with all the clauses.
+   CLAUSES is an OMP_CLAUSE chain with all the clauses.
 
    CHILD_FN is set when outlining the body of the explicit task region.
    All the statements in BODY are moved into this newly created
@@ -334,7 +334,7 @@  DEFGSCODE(GIMPLE_OMP_SECTION, "gimple_omp_section", GSS_OMP)
 /* OMP_SECTIONS <BODY, CLAUSES, CONTROL> represents #pragma omp sections.
 
    BODY is the sequence of statements in the sections body.
-   CLAUSES is a TREE_LIST node holding the list of associated clauses.
+   CLAUSES is an OMP_CLAUSE chain holding the list of associated clauses.
    CONTROL is a VAR_DECL used for deciding which of the sections
    to execute.  */
 DEFGSCODE(GIMPLE_OMP_SECTIONS, "gimple_omp_sections", GSS_OMP_SECTIONS)
@@ -346,7 +346,7 @@  DEFGSCODE(GIMPLE_OMP_SECTIONS_SWITCH, "gimple_omp_sections_switch", GSS_BASE)
 
 /* GIMPLE_OMP_SINGLE <BODY, CLAUSES> represents #pragma omp single
    BODY is the sequence of statements inside the single section.
-   CLAUSES is a TREE_LIST node holding the associated clauses.  */
+   CLAUSES is an OMP_CLAUSE chain holding the associated clauses.  */
 DEFGSCODE(GIMPLE_OMP_SINGLE, "gimple_omp_single", GSS_OMP_SINGLE)
 
 /* GIMPLE_PREDICT <PREDICT, OUTCOME> specifies a hint for branch prediction.
diff --git a/gcc/gimple.h b/gcc/gimple.h
index 8ae07c9..37c37a6 100644
--- a/gcc/gimple.h
+++ b/gcc/gimple.h
@@ -110,6 +110,9 @@  enum gf_mask {
     GF_CALL_ALLOCA_FOR_VAR	= 1 << 5,
     GF_CALL_INTERNAL		= 1 << 6,
     GF_OMP_PARALLEL_COMBINED	= 1 << 0,
+    GF_OMP_FOR_KIND_MASK	= 3 << 0,
+    GF_OMP_FOR_KIND_FOR		= 0 << 0,
+    GF_OMP_FOR_KIND_SIMD	= 1 << 0,
 
     /* True on an GIMPLE_OMP_RETURN statement if the return does not require
        a thread synchronization via some sort of barrier.  The exact barrier
@@ -799,7 +802,7 @@  gimple gimple_build_switch_nlabels (unsigned, tree, tree);
 gimple gimple_build_switch (tree, tree, vec<tree> );
 gimple gimple_build_omp_parallel (gimple_seq, tree, tree, tree);
 gimple gimple_build_omp_task (gimple_seq, tree, tree, tree, tree, tree, tree);
-gimple gimple_build_omp_for (gimple_seq, tree, size_t, gimple_seq);
+gimple gimple_build_omp_for (gimple_seq, int, tree, size_t, gimple_seq);
 gimple gimple_build_omp_critical (gimple_seq, tree);
 gimple gimple_build_omp_section (gimple_seq);
 gimple gimple_build_omp_continue (tree, tree);
@@ -3948,6 +3951,27 @@  gimple_omp_critical_set_name (gimple gs, tree name)
 }
 
 
+/* Return the kind of OMP for statemement.  */
+
+static inline int
+gimple_omp_for_kind (const_gimple g)
+{
+  GIMPLE_CHECK (g, GIMPLE_OMP_FOR);
+  return (gimple_omp_subcode (g) & GF_OMP_FOR_KIND_MASK);
+}
+
+
+/* Set the OMP for kind.  */
+
+static inline void
+gimple_omp_for_set_kind (gimple g, int kind)
+{
+  GIMPLE_CHECK (g, GIMPLE_OMP_FOR);
+  g->gsbase.subcode = (g->gsbase.subcode & ~GF_OMP_FOR_KIND_MASK)
+		      | (kind & GF_OMP_FOR_KIND_MASK);
+}
+
+
 /* Return the clauses associated with OMP_FOR GS.  */
 
 static inline tree
diff --git a/gcc/gimplify.c b/gcc/gimplify.c
index e2ae893..7b16f87 100644
--- a/gcc/gimplify.c
+++ b/gcc/gimplify.c
@@ -59,14 +59,17 @@  enum gimplify_omp_var_data
   GOVD_LOCAL = 128,
   GOVD_DEBUG_PRIVATE = 256,
   GOVD_PRIVATE_OUTER_REF = 512,
+  GOVD_LINEAR = 2048,
   GOVD_DATA_SHARE_CLASS = (GOVD_SHARED | GOVD_PRIVATE | GOVD_FIRSTPRIVATE
-			   | GOVD_LASTPRIVATE | GOVD_REDUCTION | GOVD_LOCAL)
+			   | GOVD_LASTPRIVATE | GOVD_REDUCTION | GOVD_LINEAR
+			   | GOVD_LOCAL)
 };
 
 
 enum omp_region_type
 {
   ORT_WORKSHARE = 0,
+  ORT_SIMD = 1,
   ORT_PARALLEL = 2,
   ORT_COMBINED_PARALLEL = 3,
   ORT_TASK = 4,
@@ -711,7 +714,9 @@  gimple_add_tmp_var (tree tmp)
       if (gimplify_omp_ctxp)
 	{
 	  struct gimplify_omp_ctx *ctx = gimplify_omp_ctxp;
-	  while (ctx && ctx->region_type == ORT_WORKSHARE)
+	  while (ctx
+		 && (ctx->region_type == ORT_WORKSHARE
+		     || ctx->region_type == ORT_SIMD))
 	    ctx = ctx->outer_context;
 	  if (ctx)
 	    omp_add_variable (ctx, tmp, GOVD_LOCAL | GOVD_SEEN);
@@ -2062,7 +2067,9 @@  gimplify_var_or_parm_decl (tree *expr_p)
 	  && decl_function_context (decl) != current_function_decl)
 	{
 	  struct gimplify_omp_ctx *ctx = gimplify_omp_ctxp;
-	  while (ctx && ctx->region_type == ORT_WORKSHARE)
+	  while (ctx
+		 && (ctx->region_type == ORT_WORKSHARE
+		     || ctx->region_type == ORT_SIMD))
 	    ctx = ctx->outer_context;
 	  if (!ctx && !pointer_set_insert (nonlocal_vlas, decl))
 	    {
@@ -4703,6 +4710,7 @@  is_gimple_stmt (tree t)
     case STATEMENT_LIST:
     case OMP_PARALLEL:
     case OMP_FOR:
+    case OMP_SIMD:
     case OMP_SECTIONS:
     case OMP_SECTION:
     case OMP_SINGLE:
@@ -5715,7 +5723,8 @@  omp_firstprivatize_variable (struct gimplify_omp_ctx *ctx, tree decl)
 	  else
 	    return;
 	}
-      else if (ctx->region_type != ORT_WORKSHARE)
+      else if (ctx->region_type != ORT_WORKSHARE
+	       && ctx->region_type != ORT_SIMD)
 	omp_add_variable (ctx, decl, GOVD_FIRSTPRIVATE);
 
       ctx = ctx->outer_context;
@@ -5807,7 +5816,8 @@  omp_add_variable (struct gimplify_omp_ctx *ctx, tree decl, unsigned int flags)
 	 FIRSTPRIVATE and LASTPRIVATE.  */
       nflags = n->value | flags;
       gcc_assert ((nflags & GOVD_DATA_SHARE_CLASS)
-		  == (GOVD_FIRSTPRIVATE | GOVD_LASTPRIVATE));
+		  == (GOVD_FIRSTPRIVATE | GOVD_LASTPRIVATE)
+		  || (flags & GOVD_DATA_SHARE_CLASS) == 0);
       n->value = nflags;
       return;
     }
@@ -5871,7 +5881,10 @@  omp_add_variable (struct gimplify_omp_ctx *ctx, tree decl, unsigned int flags)
 	}
     }
 
-  splay_tree_insert (ctx->variables, (splay_tree_key)decl, flags);
+  if (n != NULL)
+    n->value |= flags;
+  else
+    splay_tree_insert (ctx->variables, (splay_tree_key)decl, flags);
 }
 
 /* Notice a threadprivate variable DECL used in OpenMP context CTX.
@@ -5937,7 +5950,8 @@  omp_notice_variable (struct gimplify_omp_ctx *ctx, tree decl, bool in_code)
       enum omp_clause_default_kind default_kind, kind;
       struct gimplify_omp_ctx *octx;
 
-      if (ctx->region_type == ORT_WORKSHARE)
+      if (ctx->region_type == ORT_WORKSHARE
+	  || ctx->region_type == ORT_SIMD)
 	goto do_outer;
 
       /* ??? Some compiler-generated variables (like SAVE_EXPRs) could be
@@ -6050,7 +6064,7 @@  omp_notice_variable (struct gimplify_omp_ctx *ctx, tree decl, bool in_code)
    to the contrary in the innermost scope, generate an error.  */
 
 static bool
-omp_is_private (struct gimplify_omp_ctx *ctx, tree decl)
+omp_is_private (struct gimplify_omp_ctx *ctx, tree decl, bool simd)
 {
   splay_tree_node n;
 
@@ -6061,8 +6075,12 @@  omp_is_private (struct gimplify_omp_ctx *ctx, tree decl)
 	{
 	  if (ctx == gimplify_omp_ctxp)
 	    {
-	      error ("iteration variable %qE should be private",
-		     DECL_NAME (decl));
+	      if (simd)
+		error ("iteration variable %qE is predetermined linear",
+		       DECL_NAME (decl));
+	      else
+		error ("iteration variable %qE should be private",
+		       DECL_NAME (decl));
 	      n->value = GOVD_PRIVATE;
 	      return true;
 	    }
@@ -6080,16 +6098,26 @@  omp_is_private (struct gimplify_omp_ctx *ctx, tree decl)
 	  else if ((n->value & GOVD_REDUCTION) != 0)
 	    error ("iteration variable %qE should not be reduction",
 		   DECL_NAME (decl));
+	  else if (simd && (n->value & GOVD_LASTPRIVATE) != 0)
+	    error ("iteration variable %qE should not be lastprivate",
+		   DECL_NAME (decl));
+	  else if (simd && (n->value & GOVD_PRIVATE) != 0)
+	    error ("iteration variable %qE should not be private",
+		   DECL_NAME (decl));
+	  else if (simd && (n->value & GOVD_LINEAR) != 0)
+	    error ("iteration variable %qE is predetermined linear",
+		   DECL_NAME (decl));
 	}
       return (ctx == gimplify_omp_ctxp
 	      || (ctx->region_type == ORT_COMBINED_PARALLEL
 		  && gimplify_omp_ctxp->outer_context == ctx));
     }
 
-  if (ctx->region_type != ORT_WORKSHARE)
+  if (ctx->region_type != ORT_WORKSHARE
+      && ctx->region_type != ORT_SIMD)
     return false;
   else if (ctx->outer_context)
-    return omp_is_private (ctx->outer_context, decl);
+    return omp_is_private (ctx->outer_context, decl, simd);
   return false;
 }
 
@@ -6114,7 +6142,8 @@  omp_check_private (struct gimplify_omp_ctx *ctx, tree decl)
       if (n != NULL)
 	return (n->value & GOVD_SHARED) == 0;
     }
-  while (ctx->region_type == ORT_WORKSHARE);
+  while (ctx->region_type == ORT_WORKSHARE
+	 || ctx->region_type == ORT_SIMD);
   return false;
 }
 
@@ -6167,6 +6196,15 @@  gimplify_scan_omp_clauses (tree *list_p, gimple_seq *pre_p,
 	  flags = GOVD_REDUCTION | GOVD_SEEN | GOVD_EXPLICIT;
 	  check_non_private = "reduction";
 	  goto do_add;
+       case OMP_CLAUSE_LINEAR:
+	 if (gimplify_expr (&OMP_CLAUSE_LINEAR_STEP (c), pre_p, NULL,
+			    is_gimple_val, fb_rvalue) == GS_ERROR)
+	   {
+	     remove = true;
+	     break;
+	   }
+	 flags = GOVD_LINEAR | GOVD_EXPLICIT;
+	 goto do_add;
 
 	do_add:
 	  decl = OMP_CLAUSE_DECL (c);
@@ -6265,6 +6303,7 @@  gimplify_scan_omp_clauses (tree *list_p, gimple_seq *pre_p,
 	case OMP_CLAUSE_UNTIED:
 	case OMP_CLAUSE_COLLAPSE:
 	case OMP_CLAUSE_MERGEABLE:
+	case OMP_CLAUSE_SAFELEN:
 	  break;
 
 	case OMP_CLAUSE_DEFAULT:
@@ -6322,7 +6361,8 @@  gimplify_adjust_omp_clauses_1 (splay_tree_node n, void *data)
 	      splay_tree_node on
 		= splay_tree_lookup (ctx->variables, (splay_tree_key) decl);
 	      if (on && (on->value & (GOVD_FIRSTPRIVATE | GOVD_LASTPRIVATE
-				      | GOVD_PRIVATE | GOVD_REDUCTION)) != 0)
+				      | GOVD_PRIVATE | GOVD_REDUCTION
+				      | GOVD_LINEAR)) != 0)
 		break;
 	      ctx = ctx->outer_context;
 	    }
@@ -6335,6 +6375,8 @@  gimplify_adjust_omp_clauses_1 (splay_tree_node n, void *data)
     code = OMP_CLAUSE_PRIVATE;
   else if (flags & GOVD_FIRSTPRIVATE)
     code = OMP_CLAUSE_FIRSTPRIVATE;
+  else if (flags & GOVD_LASTPRIVATE)
+    code = OMP_CLAUSE_LASTPRIVATE;
   else
     gcc_unreachable ();
 
@@ -6367,6 +6409,7 @@  gimplify_adjust_omp_clauses (tree *list_p)
 	case OMP_CLAUSE_PRIVATE:
 	case OMP_CLAUSE_SHARED:
 	case OMP_CLAUSE_FIRSTPRIVATE:
+	case OMP_CLAUSE_LINEAR:
 	  decl = OMP_CLAUSE_DECL (c);
 	  n = splay_tree_lookup (ctx->variables, (splay_tree_key) decl);
 	  remove = !(n->value & GOVD_SEEN);
@@ -6382,6 +6425,31 @@  gimplify_adjust_omp_clauses (tree *list_p)
 		  OMP_CLAUSE_SET_CODE (c, OMP_CLAUSE_PRIVATE);
 		  OMP_CLAUSE_PRIVATE_DEBUG (c) = 1;
 		}
+	      if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_LINEAR
+		  && ctx->outer_context
+		  && !(OMP_CLAUSE_LINEAR_NO_COPYIN (c)
+		       && OMP_CLAUSE_LINEAR_NO_COPYOUT (c))
+		  && !is_global_var (decl))
+		{
+		  if (ctx->outer_context->region_type == ORT_COMBINED_PARALLEL)
+		    {
+		      n = splay_tree_lookup (ctx->outer_context->variables,
+					     (splay_tree_key) decl);
+		      if (n == NULL
+			  || (n->value & GOVD_DATA_SHARE_CLASS) == 0)
+			{
+			  int flags = OMP_CLAUSE_LINEAR_NO_COPYIN (c)
+				      ? GOVD_LASTPRIVATE : GOVD_SHARED;
+			  if (n == NULL)
+			    omp_add_variable (ctx->outer_context, decl,
+					      flags | GOVD_SEEN);
+			  else
+			    n->value |= flags | GOVD_SEEN;
+			}
+		    }
+		  else
+		    omp_notice_variable (ctx->outer_context, decl, true);
+		}
 	    }
 	  break;
 
@@ -6407,6 +6475,7 @@  gimplify_adjust_omp_clauses (tree *list_p)
 	case OMP_CLAUSE_COLLAPSE:
 	case OMP_CLAUSE_FINAL:
 	case OMP_CLAUSE_MERGEABLE:
+	case OMP_CLAUSE_SAFELEN:
 	  break;
 
 	default:
@@ -6510,14 +6579,40 @@  gimplify_omp_for (tree *expr_p, gimple_seq *pre_p)
   gimple gfor;
   gimple_seq for_body, for_pre_body;
   int i;
+  bool simd;
+  bitmap has_decl_expr = NULL;
 
   for_stmt = *expr_p;
 
+  simd = TREE_CODE (for_stmt) == OMP_SIMD; 
   gimplify_scan_omp_clauses (&OMP_FOR_CLAUSES (for_stmt), pre_p,
-			     ORT_WORKSHARE);
+			     simd ? ORT_SIMD : ORT_WORKSHARE);
 
   /* Handle OMP_FOR_INIT.  */
   for_pre_body = NULL;
+  if (simd && OMP_FOR_PRE_BODY (for_stmt))
+    {
+      has_decl_expr = BITMAP_ALLOC (NULL);
+      if (TREE_CODE (OMP_FOR_PRE_BODY (for_stmt)) == DECL_EXPR
+	  && TREE_CODE (DECL_EXPR_DECL (OMP_FOR_PRE_BODY (for_stmt)))
+	  == VAR_DECL)
+	{
+	  t = OMP_FOR_PRE_BODY (for_stmt);
+	  bitmap_set_bit (has_decl_expr, DECL_UID (DECL_EXPR_DECL (t)));
+	}
+      else if (TREE_CODE (OMP_FOR_PRE_BODY (for_stmt)) == STATEMENT_LIST)
+	{
+	  tree_stmt_iterator si;
+	  for (si = tsi_start (OMP_FOR_PRE_BODY (for_stmt)); !tsi_end_p (si);
+	       tsi_next (&si))
+	    {
+	      t = tsi_stmt (si);
+	      if (TREE_CODE (t) == DECL_EXPR
+		  && TREE_CODE (DECL_EXPR_DECL (t)) == VAR_DECL)
+		bitmap_set_bit (has_decl_expr, DECL_UID (DECL_EXPR_DECL (t)));
+	    }
+	}
+    }
   gimplify_and_add (OMP_FOR_PRE_BODY (for_stmt), &for_pre_body);
   OMP_FOR_PRE_BODY (for_stmt) = NULL_TREE;
 
@@ -6536,7 +6631,44 @@  gimplify_omp_for (tree *expr_p, gimple_seq *pre_p)
 		  || POINTER_TYPE_P (TREE_TYPE (decl)));
 
       /* Make sure the iteration variable is private.  */
-      if (omp_is_private (gimplify_omp_ctxp, decl))
+      tree c = NULL_TREE;
+      if (simd)
+	{
+	  splay_tree_node n = splay_tree_lookup (gimplify_omp_ctxp->variables,
+						 (splay_tree_key)decl);
+	  omp_is_private (gimplify_omp_ctxp, decl, simd);
+	  if (n != NULL && (n->value & GOVD_DATA_SHARE_CLASS) != 0)
+	    omp_notice_variable (gimplify_omp_ctxp, decl, true);
+	  else if (TREE_VEC_LENGTH (OMP_FOR_INIT (for_stmt)) == 1)
+	    {
+	      c = build_omp_clause (input_location, OMP_CLAUSE_LINEAR);
+	      OMP_CLAUSE_LINEAR_NO_COPYIN (c) = 1;
+	      if (has_decl_expr
+		  && bitmap_bit_p (has_decl_expr, DECL_UID (decl)))
+		OMP_CLAUSE_LINEAR_NO_COPYOUT (c) = 1;
+	      OMP_CLAUSE_DECL (c) = decl;
+	      OMP_CLAUSE_CHAIN (c) = OMP_FOR_CLAUSES (for_stmt);
+	      OMP_FOR_CLAUSES (for_stmt) = c;
+	      omp_add_variable (gimplify_omp_ctxp, decl,
+				GOVD_LINEAR | GOVD_EXPLICIT | GOVD_SEEN);
+	    }
+	  else
+	    {
+	      bool lastprivate
+		= (!has_decl_expr
+		   || !bitmap_bit_p (has_decl_expr, DECL_UID (decl)));
+	      c = build_omp_clause (input_location,
+				    lastprivate ? OMP_CLAUSE_LASTPRIVATE
+						: OMP_CLAUSE_PRIVATE);
+	      OMP_CLAUSE_DECL (c) = decl;
+	      OMP_CLAUSE_CHAIN (c) = OMP_FOR_CLAUSES (for_stmt);
+	      omp_add_variable (gimplify_omp_ctxp, decl,
+				(lastprivate ? GOVD_LASTPRIVATE : GOVD_PRIVATE)
+				| GOVD_SEEN);
+	      c = NULL_TREE;
+	    }
+	}
+      else if (omp_is_private (gimplify_omp_ctxp, decl, simd))
 	omp_notice_variable (gimplify_omp_ctxp, decl, true);
       else
 	omp_add_variable (gimplify_omp_ctxp, decl, GOVD_PRIVATE | GOVD_SEEN);
@@ -6578,6 +6710,8 @@  gimplify_omp_for (tree *expr_p, gimple_seq *pre_p)
 	case PREINCREMENT_EXPR:
 	case POSTINCREMENT_EXPR:
 	  t = build_int_cst (TREE_TYPE (decl), 1);
+	  if (c)
+	    OMP_CLAUSE_LINEAR_STEP (c) = t;
 	  t = build2 (PLUS_EXPR, TREE_TYPE (decl), var, t);
 	  t = build2 (MODIFY_EXPR, TREE_TYPE (var), var, t);
 	  TREE_VEC_ELT (OMP_FOR_INCR (for_stmt), i) = t;
@@ -6586,6 +6720,8 @@  gimplify_omp_for (tree *expr_p, gimple_seq *pre_p)
 	case PREDECREMENT_EXPR:
 	case POSTDECREMENT_EXPR:
 	  t = build_int_cst (TREE_TYPE (decl), -1);
+	  if (c)
+	    OMP_CLAUSE_LINEAR_STEP (c) = t;
 	  t = build2 (PLUS_EXPR, TREE_TYPE (decl), var, t);
 	  t = build2 (MODIFY_EXPR, TREE_TYPE (var), var, t);
 	  TREE_VEC_ELT (OMP_FOR_INCR (for_stmt), i) = t;
@@ -6619,6 +6755,20 @@  gimplify_omp_for (tree *expr_p, gimple_seq *pre_p)
 	  tret = gimplify_expr (&TREE_OPERAND (t, 1), &for_pre_body, NULL,
 				is_gimple_val, fb_rvalue);
 	  ret = MIN (ret, tret);
+	  if (c)
+	    {
+	      OMP_CLAUSE_LINEAR_STEP (c) = TREE_OPERAND (t, 1);
+	      if (TREE_CODE (t) == MINUS_EXPR)
+		{
+		  t = TREE_OPERAND (t, 1);
+		  OMP_CLAUSE_LINEAR_STEP (c)
+		    = fold_build1 (NEGATE_EXPR, TREE_TYPE (t), t);
+		  tret = gimplify_expr (&OMP_CLAUSE_LINEAR_STEP (c),
+					&for_pre_body, NULL,
+					is_gimple_val, fb_rvalue);
+		  ret = MIN (ret, tret);
+		}
+	    }
 	  break;
 
 	default:
@@ -6649,11 +6799,21 @@  gimplify_omp_for (tree *expr_p, gimple_seq *pre_p)
 	}
     }
 
+  BITMAP_FREE (has_decl_expr);
+
   gimplify_and_add (OMP_FOR_BODY (for_stmt), &for_body);
 
   gimplify_adjust_omp_clauses (&OMP_FOR_CLAUSES (for_stmt));
 
-  gfor = gimple_build_omp_for (for_body, OMP_FOR_CLAUSES (for_stmt),
+  int kind;
+  switch (TREE_CODE (for_stmt))
+    {
+    case OMP_FOR: kind = GF_OMP_FOR_KIND_FOR; break;
+    case OMP_SIMD: kind = GF_OMP_FOR_KIND_SIMD; break;
+    default:
+      gcc_unreachable ();
+    }
+  gfor = gimple_build_omp_for (for_body, kind, OMP_FOR_CLAUSES (for_stmt),
 			       TREE_VEC_LENGTH (OMP_FOR_INIT (for_stmt)),
 			       for_pre_body);
 
@@ -6670,7 +6830,10 @@  gimplify_omp_for (tree *expr_p, gimple_seq *pre_p)
     }
 
   gimplify_seq_add_stmt (pre_p, gfor);
-  return ret == GS_ALL_DONE ? GS_ALL_DONE : GS_ERROR;
+  if (ret != GS_ALL_DONE)
+    return GS_ERROR;
+  *expr_p = NULL_TREE;
+  return GS_ALL_DONE;
 }
 
 /* Gimplify the gross structure of other OpenMP worksharing constructs.
@@ -7588,6 +7751,7 @@  gimplify_expr (tree *expr_p, gimple_seq *pre_p, gimple_seq *post_p,
 	  break;
 
 	case OMP_FOR:
+	case OMP_SIMD:
 	  ret = gimplify_omp_for (expr_p, pre_p);
 	  break;
 
diff --git a/gcc/internal-fn.c b/gcc/internal-fn.c
index b841abd..983efeb 100644
--- a/gcc/internal-fn.c
+++ b/gcc/internal-fn.c
@@ -109,6 +109,30 @@  expand_STORE_LANES (gimple stmt)
   expand_insn (get_multi_vector_move (type, vec_store_lanes_optab), 2, ops);
 }
 
+/* This should get expanded in adjust_simduid_builtins.  */
+
+static void
+expand_GOMP_SIMD_LANE (gimple stmt ATTRIBUTE_UNUSED)
+{
+  gcc_unreachable ();
+}
+
+/* This should get expanded in adjust_simduid_builtins.  */
+
+static void
+expand_GOMP_SIMD_VF (gimple stmt ATTRIBUTE_UNUSED)
+{
+  gcc_unreachable ();
+}
+
+/* This should get expanded in adjust_simduid_builtins.  */
+
+static void
+expand_GOMP_SIMD_LAST_LANE (gimple stmt ATTRIBUTE_UNUSED)
+{
+  gcc_unreachable ();
+}
+
 /* Routines to expand each internal function, indexed by function number.
    Each routine has the prototype:
 
diff --git a/gcc/internal-fn.def b/gcc/internal-fn.def
index 8900d90..5427664 100644
--- a/gcc/internal-fn.def
+++ b/gcc/internal-fn.def
@@ -40,3 +40,6 @@  along with GCC; see the file COPYING3.  If not see
 
 DEF_INTERNAL_FN (LOAD_LANES, ECF_CONST | ECF_LEAF)
 DEF_INTERNAL_FN (STORE_LANES, ECF_CONST | ECF_LEAF)
+DEF_INTERNAL_FN (GOMP_SIMD_LANE, ECF_NOVOPS | ECF_LEAF | ECF_NOTHROW)
+DEF_INTERNAL_FN (GOMP_SIMD_VF, ECF_CONST | ECF_LEAF | ECF_NOTHROW)
+DEF_INTERNAL_FN (GOMP_SIMD_LAST_LANE, ECF_CONST | ECF_LEAF | ECF_NOTHROW)
diff --git a/gcc/omp-low.c b/gcc/omp-low.c
index afddf37..faa01ca 100644
--- a/gcc/omp-low.c
+++ b/gcc/omp-low.c
@@ -42,6 +42,7 @@  along with GCC; see the file COPYING3.  If not see
 #include "splay-tree.h"
 #include "optabs.h"
 #include "cfgloop.h"
+#include "target.h"
 
 
 /* Lowering of OpenMP parallel and workshare constructs proceeds in two
@@ -222,6 +223,7 @@  extract_omp_for_data (gimple for_stmt, struct omp_for_data *fd,
   int i;
   struct omp_for_data_loop dummy_loop;
   location_t loc = gimple_location (for_stmt);
+  bool simd = gimple_omp_for_kind (for_stmt) == GF_OMP_FOR_KIND_SIMD;
 
   fd->for_stmt = for_stmt;
   fd->pre = NULL;
@@ -349,7 +351,21 @@  extract_omp_for_data (gimple for_stmt, struct omp_for_data *fd,
 	  gcc_unreachable ();
 	}
 
-      if (iter_type != long_long_unsigned_type_node)
+      if (simd
+	  /*
+	  || (fd->sched_kind == OMP_CLAUSE_SCHEDULE_STATIC
+	  && !fd->have_ordered)*/)
+	{
+	  if (fd->collapse == 1)
+	    iter_type = TREE_TYPE (loop->v);
+	  else if (i == 0
+		   || TYPE_PRECISION (iter_type)
+		      < TYPE_PRECISION (TREE_TYPE (loop->v)))
+	    iter_type
+	      = build_nonstandard_integer_type
+	      (TYPE_PRECISION (TREE_TYPE (loop->v)), 1);
+	}
+      else if (iter_type != long_long_unsigned_type_node)
 	{
 	  if (POINTER_TYPE_P (TREE_TYPE (loop->v)))
 	    iter_type = long_long_unsigned_type_node;
@@ -445,7 +461,8 @@  extract_omp_for_data (gimple for_stmt, struct omp_for_data *fd,
 	}
     }
 
-  if (count)
+  if (count
+      && !simd)
     {
       if (!tree_int_cst_lt (count, TYPE_MAX_VALUE (long_integer_type_node)))
 	iter_type = long_long_unsigned_type_node;
@@ -918,6 +935,19 @@  build_outer_var_ref (tree var, omp_context *ctx)
       bool by_ref = use_pointer_for_field (var, NULL);
       x = build_receiver_ref (var, by_ref, ctx);
     }
+  else if (gimple_code (ctx->stmt) == GIMPLE_OMP_FOR
+	   && gimple_omp_for_kind (ctx->stmt) == GF_OMP_FOR_KIND_SIMD)
+    {
+      /* #pragma omp simd isn't a worksharing construct, and can reference even
+	 private vars in its linear etc. clauses.  */
+      x = NULL_TREE;
+      if (ctx->outer && is_taskreg_ctx (ctx))
+	x = lookup_decl (var, ctx->outer);
+      else if (ctx->outer)
+	x = maybe_lookup_decl (var, ctx->outer);
+      if (x == NULL_TREE)
+	x = var;
+    }
   else if (ctx->outer)
     x = lookup_decl (var, ctx->outer);
   else if (is_reference (var))
@@ -1423,6 +1453,7 @@  scan_sharing_clauses (tree clauses, omp_context *ctx)
 
 	case OMP_CLAUSE_FIRSTPRIVATE:
 	case OMP_CLAUSE_REDUCTION:
+	case OMP_CLAUSE_LINEAR:
 	  decl = OMP_CLAUSE_DECL (c);
 	do_private:
 	  if (is_variable_sized (decl))
@@ -1474,6 +1505,7 @@  scan_sharing_clauses (tree clauses, omp_context *ctx)
 	case OMP_CLAUSE_COLLAPSE:
 	case OMP_CLAUSE_UNTIED:
 	case OMP_CLAUSE_MERGEABLE:
+	case OMP_CLAUSE_SAFELEN:
 	  break;
 
 	default:
@@ -1497,6 +1529,7 @@  scan_sharing_clauses (tree clauses, omp_context *ctx)
 	case OMP_CLAUSE_PRIVATE:
 	case OMP_CLAUSE_FIRSTPRIVATE:
 	case OMP_CLAUSE_REDUCTION:
+	case OMP_CLAUSE_LINEAR:
 	  decl = OMP_CLAUSE_DECL (c);
 	  if (is_variable_sized (decl))
 	    install_var_local (decl, ctx);
@@ -1526,6 +1559,7 @@  scan_sharing_clauses (tree clauses, omp_context *ctx)
 	case OMP_CLAUSE_UNTIED:
 	case OMP_CLAUSE_FINAL:
 	case OMP_CLAUSE_MERGEABLE:
+	case OMP_CLAUSE_SAFELEN:
 	  break;
 
 	default:
@@ -1631,7 +1665,6 @@  create_omp_child_function (omp_context *ctx, bool task_copy)
   pop_cfun ();
 }
 
-
 /* Scan an OpenMP parallel directive.  */
 
 static void
@@ -1831,9 +1864,22 @@  scan_omp_single (gimple stmt, omp_context *outer_ctx)
 static bool
 check_omp_nesting_restrictions (gimple stmt, omp_context *ctx)
 {
+  if (ctx != NULL)
+    {
+      if (gimple_code (ctx->stmt) == GIMPLE_OMP_FOR
+	  && gimple_omp_for_kind (ctx->stmt) == GF_OMP_FOR_KIND_SIMD)
+	{
+	  error_at (gimple_location (stmt),
+		    "OpenMP constructs may not be nested inside simd region");
+	  return false;
+	}
+    }
   switch (gimple_code (stmt))
     {
     case GIMPLE_OMP_FOR:
+      if (gimple_omp_for_kind (stmt) == GF_OMP_FOR_KIND_SIMD)
+	return true;
+      /* FALLTHRU */
     case GIMPLE_OMP_SECTIONS:
     case GIMPLE_OMP_SINGLE:
     case GIMPLE_CALL:
@@ -2254,6 +2300,73 @@  omp_reduction_init (tree clause, tree type)
     }
 }
 
+/* Return maximum possible vectorization factor for the target.  */
+
+static int
+omp_max_vf (void)
+{
+  if (!optimize
+      || optimize_debug
+      || (!flag_tree_vectorize
+	  && global_options_set.x_flag_tree_vectorize))
+    return 1;
+
+  int vs = targetm.vectorize.autovectorize_vector_sizes ();
+  if (vs)
+    {
+      vs = 1 << floor_log2 (vs);
+      return vs;
+    }
+  enum machine_mode vqimode = targetm.vectorize.preferred_simd_mode (QImode);
+  if (GET_MODE_CLASS (vqimode) == MODE_VECTOR_INT)
+    return GET_MODE_NUNITS (vqimode);
+  return 1;
+}
+
+/* Helper function of lower_rec_input_clauses, used for #pragma omp simd
+   privatization.  */
+
+static bool
+lower_rec_simd_input_clauses (tree new_var, omp_context *ctx, int &max_vf,
+			      tree &idx, tree &lane, tree &ivar, tree &lvar)
+{
+  if (max_vf == 0)
+    {
+      max_vf = omp_max_vf ();
+      if (max_vf > 1)
+	{
+	  tree c = find_omp_clause (gimple_omp_for_clauses (ctx->stmt),
+				    OMP_CLAUSE_SAFELEN);
+	  if (c
+	      && compare_tree_int (OMP_CLAUSE_SAFELEN_EXPR (c), max_vf) == -1)
+	    max_vf = tree_low_cst (OMP_CLAUSE_SAFELEN_EXPR (c), 0);
+	}
+      if (max_vf > 1)
+	{
+	  idx = create_tmp_var (unsigned_type_node, NULL);
+	  lane = create_tmp_var (unsigned_type_node, NULL);
+	}
+    }
+  if (max_vf == 1)
+    return false;
+
+  tree atype = build_array_type_nelts (TREE_TYPE (new_var), max_vf);
+  tree avar = create_tmp_var_raw (atype, NULL);
+  if (TREE_ADDRESSABLE (new_var))
+    TREE_ADDRESSABLE (avar) = 1;
+  DECL_ATTRIBUTES (avar)
+    = tree_cons (get_identifier ("omp simd array"), NULL,
+		 DECL_ATTRIBUTES (avar));
+  gimple_add_tmp_var (avar);
+  ivar = build4 (ARRAY_REF, TREE_TYPE (new_var), avar, idx,
+		 NULL_TREE, NULL_TREE);
+  lvar = build4 (ARRAY_REF, TREE_TYPE (new_var), avar, lane,
+		 NULL_TREE, NULL_TREE);
+  SET_DECL_VALUE_EXPR (new_var, lvar);
+  DECL_HAS_VALUE_EXPR_P (new_var) = 1;
+  return true;
+}
+
 /* Generate code to implement the input clauses, FIRSTPRIVATE and COPYIN,
    from the receiver (aka child) side and initializers for REFERENCE_TYPE
    private variables.  Initialization statements go in ILIST, while calls
@@ -2267,9 +2380,37 @@  lower_rec_input_clauses (tree clauses, gimple_seq *ilist, gimple_seq *dlist,
   bool copyin_by_ref = false;
   bool lastprivate_firstprivate = false;
   int pass;
+  bool is_simd = (gimple_code (ctx->stmt) == GIMPLE_OMP_FOR
+		  && gimple_omp_for_kind (ctx->stmt) == GF_OMP_FOR_KIND_SIMD);
+  int max_vf = 0;
+  tree lane = NULL_TREE, idx = NULL_TREE;
+  tree ivar = NULL_TREE, lvar = NULL_TREE;
+  gimple_seq llist[2] = { NULL, NULL };
 
   copyin_seq = NULL;
 
+  /* Enforce simdlen 1 in simd loops with data sharing clauses referencing
+     variable sized vars.  That is unnecessarily hard to support and very
+     unlikely to result in vectorized code anyway.  */
+  if (is_simd)
+    for (c = clauses; c ; c = OMP_CLAUSE_CHAIN (c))
+      switch (OMP_CLAUSE_CODE (c))
+	{
+	case OMP_CLAUSE_REDUCTION:
+	  if (OMP_CLAUSE_REDUCTION_PLACEHOLDER (c))
+	    max_vf = 1;
+	  /* FALLTHRU */
+	case OMP_CLAUSE_PRIVATE:
+	case OMP_CLAUSE_FIRSTPRIVATE:
+	case OMP_CLAUSE_LASTPRIVATE:
+	case OMP_CLAUSE_LINEAR:
+	  if (is_variable_sized (OMP_CLAUSE_DECL (c)))
+	    max_vf = 1;
+	  break;
+	default:
+	  continue;
+	}
+
   /* Do all the fixed sized types in the first pass, and the variable sized
      types in the second pass.  This makes sure that the scalar arguments to
      the variable sized types are processed before we use them in the
@@ -2299,6 +2440,8 @@  lower_rec_input_clauses (tree clauses, gimple_seq *ilist, gimple_seq *dlist,
 	    case OMP_CLAUSE_COPYIN:
 	    case OMP_CLAUSE_REDUCTION:
 	      break;
+	    case OMP_CLAUSE_LINEAR:
+	      break;
 	    case OMP_CLAUSE_LASTPRIVATE:
 	      if (OMP_CLAUSE_LASTPRIVATE_FIRSTPRIVATE (c))
 		{
@@ -2443,7 +2586,36 @@  lower_rec_input_clauses (tree clauses, gimple_seq *ilist, gimple_seq *dlist,
 		}
 	      else
 		x = NULL;
+	    do_private:
 	      x = lang_hooks.decls.omp_clause_default_ctor (c, new_var, x);
+	      if (is_simd)
+		{
+		  tree y = lang_hooks.decls.omp_clause_dtor (c, new_var);
+		  if ((TREE_ADDRESSABLE (new_var) || x || y
+		       || OMP_CLAUSE_CODE (c) == OMP_CLAUSE_LASTPRIVATE)
+		      && lower_rec_simd_input_clauses (new_var, ctx, max_vf,
+						       idx, lane, ivar, lvar))
+		    {
+		      if (x)
+			x = lang_hooks.decls.omp_clause_default_ctor
+						(c, unshare_expr (ivar), x);
+		      if (x)
+			gimplify_and_add (x, &llist[0]);
+		      if (y)
+			{
+			  y = lang_hooks.decls.omp_clause_dtor (c, ivar);
+			  if (y)
+			    {
+			      gimple_seq tseq = NULL;
+
+			      dtor = y;
+			      gimplify_stmt (&dtor, &tseq);
+			      gimple_seq_add_seq (&llist[1], tseq);
+			    }
+			}
+		      break;
+		    }
+		}
 	      if (x)
 		gimplify_and_add (x, ilist);
 	      /* FALLTHRU */
@@ -2460,6 +2632,15 @@  lower_rec_input_clauses (tree clauses, gimple_seq *ilist, gimple_seq *dlist,
 		}
 	      break;
 
+	    case OMP_CLAUSE_LINEAR:
+	      if (!OMP_CLAUSE_LINEAR_NO_COPYIN (c))
+		goto do_firstprivate;
+	      if (OMP_CLAUSE_LINEAR_NO_COPYOUT (c))
+		x = NULL;
+	      else
+		x = build_outer_var_ref (var, ctx);
+	      goto do_private;
+
 	    case OMP_CLAUSE_FIRSTPRIVATE:
 	      if (is_task_ctx (ctx))
 		{
@@ -2475,11 +2656,56 @@  lower_rec_input_clauses (tree clauses, gimple_seq *ilist, gimple_seq *dlist,
 		      goto do_dtor;
 		    }
 		}
+	    do_firstprivate:
 	      x = build_outer_var_ref (var, ctx);
+	      if (is_simd)
+		{
+		  if ((OMP_CLAUSE_CODE (c) != OMP_CLAUSE_LINEAR
+		       || TREE_ADDRESSABLE (new_var))
+		      && lower_rec_simd_input_clauses (new_var, ctx, max_vf,
+						       idx, lane, ivar, lvar))
+		    {
+		      if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_LINEAR)
+			{
+			  tree iv = create_tmp_var (TREE_TYPE (new_var), NULL);
+			  x = lang_hooks.decls.omp_clause_copy_ctor (c, iv, x);
+			  gimplify_and_add (x, ilist);
+			  gimple_stmt_iterator gsi
+			    = gsi_start_1 (gimple_omp_body_ptr (ctx->stmt));
+			  gimple g
+			    = gimple_build_assign (unshare_expr (lvar), iv);
+			  gsi_insert_before_without_update (&gsi, g,
+							    GSI_SAME_STMT);
+			  tree stept = POINTER_TYPE_P (TREE_TYPE (x))
+				       ? sizetype : TREE_TYPE (x);
+			  tree t = fold_convert (stept,
+						 OMP_CLAUSE_LINEAR_STEP (c));
+			  enum tree_code code = PLUS_EXPR;
+			  if (POINTER_TYPE_P (TREE_TYPE (new_var)))
+			    code = POINTER_PLUS_EXPR;
+			  g = gimple_build_assign_with_ops (code, iv, iv, t);
+			  gsi_insert_before_without_update (&gsi, g,
+							    GSI_SAME_STMT);
+			  break;
+			}
+		      x = lang_hooks.decls.omp_clause_copy_ctor
+						(c, unshare_expr (ivar), x);
+		      gimplify_and_add (x, &llist[0]);
+		      x = lang_hooks.decls.omp_clause_dtor (c, ivar);
+		      if (x)
+			{
+			  gimple_seq tseq = NULL;
+
+			  dtor = x;
+			  gimplify_stmt (&dtor, &tseq);
+			  gimple_seq_add_seq (&llist[1], tseq);
+			}
+		      break;
+		    }
+		}
 	      x = lang_hooks.decls.omp_clause_copy_ctor (c, new_var, x);
 	      gimplify_and_add (x, ilist);
 	      goto do_dtor;
-	      break;
 
 	    case OMP_CLAUSE_COPYIN:
 	      by_ref = use_pointer_for_field (var, NULL);
@@ -2495,6 +2721,8 @@  lower_rec_input_clauses (tree clauses, gimple_seq *ilist, gimple_seq *dlist,
 		  tree placeholder = OMP_CLAUSE_REDUCTION_PLACEHOLDER (c);
 		  x = build_outer_var_ref (var, ctx);
 
+		  /* FIXME: Not handled yet.  */
+		  gcc_assert (!is_simd);
 		  if (is_reference (var))
 		    x = build_fold_addr_expr_loc (clause_loc, x);
 		  SET_DECL_VALUE_EXPR (placeholder, x);
@@ -2509,7 +2737,31 @@  lower_rec_input_clauses (tree clauses, gimple_seq *ilist, gimple_seq *dlist,
 		{
 		  x = omp_reduction_init (c, TREE_TYPE (new_var));
 		  gcc_assert (TREE_CODE (TREE_TYPE (new_var)) != ARRAY_TYPE);
-		  gimplify_assign (new_var, x, ilist);
+		  if (is_simd
+		      && lower_rec_simd_input_clauses (new_var, ctx, max_vf,
+						       idx, lane, ivar, lvar))
+		    {
+		      enum tree_code code = OMP_CLAUSE_REDUCTION_CODE (c);
+		      tree ref = build_outer_var_ref (var, ctx);
+
+		      gimplify_assign (unshare_expr (ivar), x, &llist[0]);
+
+		      /* reduction(-:var) sums up the partial results, so it
+			 acts identically to reduction(+:var).  */
+		      if (code == MINUS_EXPR)
+			code = PLUS_EXPR;
+
+		      x = build2 (code, TREE_TYPE (ref), ref, ivar);
+		      ref = build_outer_var_ref (var, ctx);
+		      gimplify_assign (ref, x, &llist[1]);
+		    }
+		  else
+		    {
+		      gimplify_assign (new_var, x, ilist);
+		      if (is_simd)
+			gimplify_assign (build_outer_var_ref (var, ctx),
+					 new_var, dlist);
+		    }
 		}
 	      break;
 
@@ -2519,6 +2771,49 @@  lower_rec_input_clauses (tree clauses, gimple_seq *ilist, gimple_seq *dlist,
 	}
     }
 
+  if (lane)
+    {
+      tree uid = create_tmp_var (ptr_type_node, "simduid");
+      gimple g
+	= gimple_build_call_internal (IFN_GOMP_SIMD_LANE, 1, uid);
+      gimple_call_set_lhs (g, lane);
+      gimple_stmt_iterator gsi = gsi_start_1 (gimple_omp_body_ptr (ctx->stmt));
+      gsi_insert_before_without_update (&gsi, g, GSI_SAME_STMT);
+      c = build_omp_clause (UNKNOWN_LOCATION, OMP_CLAUSE__SIMDUID_);
+      OMP_CLAUSE__SIMDUID__DECL (c) = uid;
+      OMP_CLAUSE_CHAIN (c) = gimple_omp_for_clauses (ctx->stmt);
+      gimple_omp_for_set_clauses (ctx->stmt, c);
+      g = gimple_build_assign_with_ops (INTEGER_CST, lane,
+					build_int_cst (unsigned_type_node, 0),
+					NULL_TREE);
+      gimple_seq_add_stmt (ilist, g);
+      for (int i = 0; i < 2; i++)
+	if (llist[i])
+	  {
+	    tree vf = create_tmp_var (unsigned_type_node, NULL);
+	    g = gimple_build_call_internal (IFN_GOMP_SIMD_VF, 1, uid);
+	    gimple_call_set_lhs (g, vf);
+	    gimple_seq *seq = i == 0 ? ilist : dlist;
+	    gimple_seq_add_stmt (seq, g);
+	    tree t = build_int_cst (unsigned_type_node, 0);
+	    g = gimple_build_assign_with_ops (INTEGER_CST, idx, t, NULL_TREE);
+	    gimple_seq_add_stmt (seq, g);
+	    tree body = create_artificial_label (UNKNOWN_LOCATION);
+	    tree header = create_artificial_label (UNKNOWN_LOCATION);
+	    tree end = create_artificial_label (UNKNOWN_LOCATION);
+	    gimple_seq_add_stmt (seq, gimple_build_goto (header));
+	    gimple_seq_add_stmt (seq, gimple_build_label (body));
+	    gimple_seq_add_seq (seq, llist[i]);
+	    t = build_int_cst (unsigned_type_node, 1);
+	    g = gimple_build_assign_with_ops (PLUS_EXPR, idx, idx, t);
+	    gimple_seq_add_stmt (seq, g);
+	    gimple_seq_add_stmt (seq, gimple_build_label (header));
+	    g = gimple_build_cond (LT_EXPR, idx, vf, body, end);
+	    gimple_seq_add_stmt (seq, g);
+	    gimple_seq_add_stmt (seq, gimple_build_label (end));
+	  }
+    }
+
   /* The copyin sequence is not to be executed by the main thread, since
      that would result in self-copies.  Perhaps not visible to scalars,
      but it certainly is to C++ operator=.  */
@@ -2538,7 +2833,31 @@  lower_rec_input_clauses (tree clauses, gimple_seq *ilist, gimple_seq *dlist,
      lastprivate clauses we need to ensure the lastprivate copying
      happens after firstprivate copying in all threads.  */
   if (copyin_by_ref || lastprivate_firstprivate)
-    gimplify_and_add (build_omp_barrier (), ilist);
+    {
+      /* Don't add any barrier for #pragma omp simd or
+	 #pragma omp distribute.  */
+      if (gimple_code (ctx->stmt) != GIMPLE_OMP_FOR
+	  || gimple_omp_for_kind (ctx->stmt) == GF_OMP_FOR_KIND_FOR)
+	gimplify_and_add (build_omp_barrier (), ilist);
+    }
+
+  /* If max_vf is non-NULL, then we can use only vectorization factor
+     up to the max_vf we chose.  So stick it into safelen clause.  */
+  if (max_vf)
+    {
+      tree c = find_omp_clause (gimple_omp_for_clauses (ctx->stmt),
+				OMP_CLAUSE_SAFELEN);
+      if (c == NULL_TREE
+	  || compare_tree_int (OMP_CLAUSE_SAFELEN_EXPR (c),
+			       max_vf) == 1)
+	{
+	  c = build_omp_clause (UNKNOWN_LOCATION, OMP_CLAUSE_SAFELEN);
+	  OMP_CLAUSE_SAFELEN_EXPR (c) = build_int_cst (integer_type_node,
+						       max_vf);
+	  OMP_CLAUSE_CHAIN (c) = gimple_omp_for_clauses (ctx->stmt);
+	  gimple_omp_for_set_clauses (ctx->stmt, c);
+	}
+    }
 }
 
 
@@ -2550,11 +2869,16 @@  static void
 lower_lastprivate_clauses (tree clauses, tree predicate, gimple_seq *stmt_list,
 			    omp_context *ctx)
 {
-  tree x, c, label = NULL;
+  tree x, c, label = NULL, orig_clauses = clauses;
   bool par_clauses = false;
+  tree simduid = NULL, lastlane = NULL;
 
-  /* Early exit if there are no lastprivate clauses.  */
-  clauses = find_omp_clause (clauses, OMP_CLAUSE_LASTPRIVATE);
+  /* Early exit if there are no lastprivate or linear clauses.  */
+  for (; clauses ; clauses = OMP_CLAUSE_CHAIN (clauses))
+    if (OMP_CLAUSE_CODE (clauses) == OMP_CLAUSE_LASTPRIVATE
+	|| (OMP_CLAUSE_CODE (clauses) == OMP_CLAUSE_LINEAR
+	    && !OMP_CLAUSE_LINEAR_NO_COPYOUT (clauses)))
+      break;
   if (clauses == NULL)
     {
       /* If this was a workshare clause, see if it had been combined
@@ -2591,23 +2915,59 @@  lower_lastprivate_clauses (tree clauses, tree predicate, gimple_seq *stmt_list,
       gimple_seq_add_stmt (stmt_list, gimple_build_label (label_true));
     }
 
+  if (gimple_code (ctx->stmt) == GIMPLE_OMP_FOR
+      && gimple_omp_for_kind (ctx->stmt) == GF_OMP_FOR_KIND_SIMD)
+    {
+      simduid = find_omp_clause (orig_clauses, OMP_CLAUSE__SIMDUID_);
+      if (simduid)
+	simduid = OMP_CLAUSE__SIMDUID__DECL (simduid);
+    }
+
   for (c = clauses; c ;)
     {
       tree var, new_var;
       location_t clause_loc = OMP_CLAUSE_LOCATION (c);
 
-      if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_LASTPRIVATE)
+      if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_LASTPRIVATE
+	  || (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_LINEAR
+	      && !OMP_CLAUSE_LINEAR_NO_COPYOUT (c)))
 	{
 	  var = OMP_CLAUSE_DECL (c);
 	  new_var = lookup_decl (var, ctx);
 
-	  if (OMP_CLAUSE_LASTPRIVATE_GIMPLE_SEQ (c))
+	  if (simduid && DECL_HAS_VALUE_EXPR_P (new_var))
+	    {
+	      tree val = DECL_VALUE_EXPR (new_var);
+	      if (TREE_CODE (val) == ARRAY_REF
+		  && VAR_P (TREE_OPERAND (val, 0))
+		  && lookup_attribute ("omp simd array",
+				       DECL_ATTRIBUTES (TREE_OPERAND (val,
+								      0))))
+		{
+		  if (lastlane == NULL)
+		    {
+		      lastlane = create_tmp_var (unsigned_type_node, NULL);
+		      gimple g
+			= gimple_build_call_internal (IFN_GOMP_SIMD_LAST_LANE,
+						      2, simduid,
+						      TREE_OPERAND (val, 1));
+		      gimple_call_set_lhs (g, lastlane);
+		      gimple_seq_add_stmt (stmt_list, g);
+		    }
+		  new_var = build4 (ARRAY_REF, TREE_TYPE (val),
+				    TREE_OPERAND (val, 0), lastlane,
+				    NULL_TREE, NULL_TREE);
+		}
+	    }
+
+	  if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_LASTPRIVATE
+	      && OMP_CLAUSE_LASTPRIVATE_GIMPLE_SEQ (c))
 	    {
 	      lower_omp (&OMP_CLAUSE_LASTPRIVATE_GIMPLE_SEQ (c), ctx);
 	      gimple_seq_add_seq (stmt_list,
 				  OMP_CLAUSE_LASTPRIVATE_GIMPLE_SEQ (c));
+	      OMP_CLAUSE_LASTPRIVATE_GIMPLE_SEQ (c) = NULL;
 	    }
-	  OMP_CLAUSE_LASTPRIVATE_GIMPLE_SEQ (c) = NULL;
 
 	  x = build_outer_var_ref (var, ctx);
 	  if (is_reference (var))
@@ -2649,6 +3009,11 @@  lower_reduction_clauses (tree clauses, gimple_seq *stmt_seqp, omp_context *ctx)
   tree x, c;
   int count = 0;
 
+  /* SIMD reductions are handled in lower_rec_input_clauses.  */
+  if (gimple_code (ctx->stmt) == GIMPLE_OMP_FOR
+      && gimple_omp_for_kind (ctx->stmt) == GF_OMP_FOR_KIND_SIMD)
+    return;
+
   /* First see if there is exactly one reduction clause.  Use OMP_ATOMIC
      update in that case, otherwise use a lock.  */
   for (c = clauses; c && count < 2; c = OMP_CLAUSE_CHAIN (c))
@@ -3411,6 +3776,24 @@  expand_omp_regimplify_p (tree *tp, int *walk_subtrees, void *)
   return NULL_TREE;
 }
 
+/* Prepend TO = FROM assignment before *GSI_P.  */
+
+static void
+expand_omp_build_assign (gimple_stmt_iterator *gsi_p, tree to, tree from)
+{
+  bool simple_p = DECL_P (to) && TREE_ADDRESSABLE (to);
+  from = force_gimple_operand_gsi (gsi_p, from, simple_p, NULL_TREE,
+				   true, GSI_SAME_STMT);
+  gimple stmt = gimple_build_assign (to, from);
+  gsi_insert_before (gsi_p, stmt, GSI_SAME_STMT);
+  if (walk_tree (&from, expand_omp_regimplify_p, NULL, NULL)
+      || walk_tree (&to, expand_omp_regimplify_p, NULL, NULL))
+    {
+      gimple_stmt_iterator gsi = gsi_for_stmt (stmt);
+      gimple_regimplify_operands (stmt, &gsi);
+    }
+}
+
 /* Expand the OpenMP parallel or task directive starting at REGION.  */
 
 static void
@@ -3654,6 +4037,280 @@  expand_omp_taskreg (struct omp_region *region)
 }
 
 
+/* Helper function for expand_omp_{for_*,simd}.  If this is the outermost
+   of the combined collapse > 1 loop constructs, generate code like:
+	if (__builtin_expect (N32 cond3 N31, 0)) goto ZERO_ITER_BB;
+	if (cond3 is <)
+	  adj = STEP3 - 1;
+	else
+	  adj = STEP3 + 1;
+	count3 = (adj + N32 - N31) / STEP3;
+	if (__builtin_expect (N22 cond2 N21, 0)) goto ZERO_ITER_BB;
+	if (cond2 is <)
+	  adj = STEP2 - 1;
+	else
+	  adj = STEP2 + 1;
+	count2 = (adj + N22 - N21) / STEP2;
+	if (__builtin_expect (N12 cond1 N11, 0)) goto ZERO_ITER_BB;
+	if (cond1 is <)
+	  adj = STEP1 - 1;
+	else
+	  adj = STEP1 + 1;
+	count1 = (adj + N12 - N11) / STEP1;
+	count = count1 * count2 * count3;
+   Furthermore, if ZERO_ITER_BB is NULL, create a BB which does:
+	count = 0;
+   and set ZERO_ITER_BB to that bb.  */
+
+static void
+expand_omp_for_init_counts (struct omp_for_data *fd, gimple_stmt_iterator *gsi,
+			    basic_block &entry_bb, tree *counts,
+			    basic_block &zero_iter_bb, int &first_zero_iter,
+			    basic_block &l2_dom_bb)
+{
+  tree t, type = TREE_TYPE (fd->loop.v);
+  gimple stmt;
+  edge e, ne;
+  int i;
+
+  /* Collapsed loops need work for expansion into SSA form.  */
+  gcc_assert (!gimple_in_ssa_p (cfun));
+
+  for (i = 0; i < fd->collapse; i++)
+    {
+      tree itype = TREE_TYPE (fd->loops[i].v);
+
+      if (SSA_VAR_P (fd->loop.n2)
+	  && ((t = fold_binary (fd->loops[i].cond_code, boolean_type_node,
+				fold_convert (itype, fd->loops[i].n1),
+				fold_convert (itype, fd->loops[i].n2)))
+	      == NULL_TREE || !integer_onep (t)))
+	{
+	  tree n1, n2;
+	  n1 = fold_convert (itype, unshare_expr (fd->loops[i].n1));
+	  n1 = force_gimple_operand_gsi (gsi, n1, true, NULL_TREE,
+					 true, GSI_SAME_STMT);
+	  n2 = fold_convert (itype, unshare_expr (fd->loops[i].n2));
+	  n2 = force_gimple_operand_gsi (gsi, n2, true, NULL_TREE,
+					 true, GSI_SAME_STMT);
+	  stmt = gimple_build_cond (fd->loops[i].cond_code, n1, n2,
+				    NULL_TREE, NULL_TREE);
+	  gsi_insert_before (gsi, stmt, GSI_SAME_STMT);
+	  if (walk_tree (gimple_cond_lhs_ptr (stmt),
+			 expand_omp_regimplify_p, NULL, NULL)
+	      || walk_tree (gimple_cond_rhs_ptr (stmt),
+			    expand_omp_regimplify_p, NULL, NULL))
+	    {
+	      *gsi = gsi_for_stmt (stmt);
+	      gimple_regimplify_operands (stmt, gsi);
+	    }
+	  e = split_block (entry_bb, stmt);
+	  if (zero_iter_bb == NULL)
+	    {
+	      first_zero_iter = i;
+	      zero_iter_bb = create_empty_bb (entry_bb);
+	      if (current_loops)
+		add_bb_to_loop (zero_iter_bb, entry_bb->loop_father);
+	      *gsi = gsi_after_labels (zero_iter_bb);
+	      stmt = gimple_build_assign (fd->loop.n2,
+					  build_zero_cst (type));
+	      gsi_insert_before (gsi, stmt, GSI_SAME_STMT);
+	      set_immediate_dominator (CDI_DOMINATORS, zero_iter_bb,
+				       entry_bb);
+	    }
+	  ne = make_edge (entry_bb, zero_iter_bb, EDGE_FALSE_VALUE);
+	  ne->probability = REG_BR_PROB_BASE / 2000 - 1;
+	  e->flags = EDGE_TRUE_VALUE;
+	  e->probability = REG_BR_PROB_BASE - ne->probability;
+	  if (l2_dom_bb == NULL)
+	    l2_dom_bb = entry_bb;
+	  entry_bb = e->dest;
+	  *gsi = gsi_last_bb (entry_bb);
+	}
+
+      if (POINTER_TYPE_P (itype))
+	itype = signed_type_for (itype);
+      t = build_int_cst (itype, (fd->loops[i].cond_code == LT_EXPR
+				 ? -1 : 1));
+      t = fold_build2 (PLUS_EXPR, itype,
+		       fold_convert (itype, fd->loops[i].step), t);
+      t = fold_build2 (PLUS_EXPR, itype, t,
+		       fold_convert (itype, fd->loops[i].n2));
+      t = fold_build2 (MINUS_EXPR, itype, t,
+		       fold_convert (itype, fd->loops[i].n1));
+      if (TYPE_UNSIGNED (itype) && fd->loops[i].cond_code == GT_EXPR)
+	t = fold_build2 (TRUNC_DIV_EXPR, itype,
+			 fold_build1 (NEGATE_EXPR, itype, t),
+			 fold_build1 (NEGATE_EXPR, itype,
+				      fold_convert (itype,
+						    fd->loops[i].step)));
+      else
+	t = fold_build2 (TRUNC_DIV_EXPR, itype, t,
+			 fold_convert (itype, fd->loops[i].step));
+      t = fold_convert (type, t);
+      if (TREE_CODE (t) == INTEGER_CST)
+	counts[i] = t;
+      else
+	{
+	  counts[i] = create_tmp_reg (type, ".count");
+	  expand_omp_build_assign (gsi, counts[i], t);
+	}
+      if (SSA_VAR_P (fd->loop.n2))
+	{
+	  if (i == 0)
+	    t = counts[0];
+	  else
+	    t = fold_build2 (MULT_EXPR, type, fd->loop.n2, counts[i]);
+	  expand_omp_build_assign (gsi, fd->loop.n2, t);
+	}
+    }
+}
+
+
+/* Helper function for expand_omp_{for_*,simd}.  Generate code like:
+	T = V;
+	V3 = N31 + (T % count3) * STEP3;
+	T = T / count3;
+	V2 = N21 + (T % count2) * STEP2;
+	T = T / count2;
+	V1 = N11 + T * STEP1;
+   if this loop doesn't have an inner loop construct combined with it.  */
+
+static void
+expand_omp_for_init_vars (struct omp_for_data *fd, gimple_stmt_iterator *gsi,
+			  tree *counts, tree startvar)
+{
+  int i;
+  tree type = TREE_TYPE (fd->loop.v);
+  tree tem = create_tmp_reg (type, ".tem");
+  gimple stmt = gimple_build_assign (tem, startvar);
+  gsi_insert_after (gsi, stmt, GSI_CONTINUE_LINKING);
+
+  for (i = fd->collapse - 1; i >= 0; i--)
+    {
+      tree vtype = TREE_TYPE (fd->loops[i].v), itype, t;
+      itype = vtype;
+      if (POINTER_TYPE_P (vtype))
+	itype = signed_type_for (vtype);
+      if (i != 0)
+	t = fold_build2 (TRUNC_MOD_EXPR, type, tem, counts[i]);
+      else
+	t = tem;
+      t = fold_convert (itype, t);
+      t = fold_build2 (MULT_EXPR, itype, t,
+		       fold_convert (itype, fd->loops[i].step));
+      if (POINTER_TYPE_P (vtype))
+	t = fold_build_pointer_plus (fd->loops[i].n1, t);
+      else
+	t = fold_build2 (PLUS_EXPR, itype, fd->loops[i].n1, t);
+      t = force_gimple_operand_gsi (gsi, t,
+				    DECL_P (fd->loops[i].v)
+				    && TREE_ADDRESSABLE (fd->loops[i].v),
+				    NULL_TREE, false,
+				    GSI_CONTINUE_LINKING);
+      stmt = gimple_build_assign (fd->loops[i].v, t);
+      gsi_insert_after (gsi, stmt, GSI_CONTINUE_LINKING);
+      if (i != 0)
+	{
+	  t = fold_build2 (TRUNC_DIV_EXPR, type, tem, counts[i]);
+	  t = force_gimple_operand_gsi (gsi, t, false, NULL_TREE,
+					false, GSI_CONTINUE_LINKING);
+	  stmt = gimple_build_assign (tem, t);
+	  gsi_insert_after (gsi, stmt, GSI_CONTINUE_LINKING);
+	}
+    }
+}
+
+
+/* Helper function for expand_omp_for_*.  Generate code like:
+    L10:
+	V3 += STEP3;
+	if (V3 cond3 N32) goto BODY_BB; else goto L11;
+    L11:
+	V3 = N31;
+	V2 += STEP2;
+	if (V2 cond2 N22) goto BODY_BB; else goto L12;
+    L12:
+	V2 = N21;
+	V1 += STEP1;
+	goto BODY_BB;  */
+
+static basic_block
+extract_omp_for_update_vars (struct omp_for_data *fd, basic_block cont_bb,
+			     basic_block body_bb)
+{
+  basic_block last_bb, bb, collapse_bb = NULL;
+  int i;
+  gimple_stmt_iterator gsi;
+  edge e;
+  tree t;
+  gimple stmt;
+
+  last_bb = cont_bb;
+  for (i = fd->collapse - 1; i >= 0; i--)
+    {
+      tree vtype = TREE_TYPE (fd->loops[i].v);
+
+      bb = create_empty_bb (last_bb);
+      if (current_loops)
+	add_bb_to_loop (bb, last_bb->loop_father);
+      gsi = gsi_start_bb (bb);
+
+      if (i < fd->collapse - 1)
+	{
+	  e = make_edge (last_bb, bb, EDGE_FALSE_VALUE);
+	  e->probability = REG_BR_PROB_BASE / 8;
+
+	  t = fd->loops[i + 1].n1;
+	  t = force_gimple_operand_gsi (&gsi, t,
+					DECL_P (fd->loops[i + 1].v)
+					&& TREE_ADDRESSABLE (fd->loops[i
+								       + 1].v),
+					NULL_TREE, false,
+					GSI_CONTINUE_LINKING);
+	  stmt = gimple_build_assign (fd->loops[i + 1].v, t);
+	  gsi_insert_after (&gsi, stmt, GSI_CONTINUE_LINKING);
+	}
+      else
+	collapse_bb = bb;
+
+      set_immediate_dominator (CDI_DOMINATORS, bb, last_bb);
+
+      if (POINTER_TYPE_P (vtype))
+	t = fold_build_pointer_plus (fd->loops[i].v, fd->loops[i].step);
+      else
+	t = fold_build2 (PLUS_EXPR, vtype, fd->loops[i].v, fd->loops[i].step);
+      t = force_gimple_operand_gsi (&gsi, t,
+				    DECL_P (fd->loops[i].v)
+				    && TREE_ADDRESSABLE (fd->loops[i].v),
+				    NULL_TREE, false, GSI_CONTINUE_LINKING);
+      stmt = gimple_build_assign (fd->loops[i].v, t);
+      gsi_insert_after (&gsi, stmt, GSI_CONTINUE_LINKING);
+
+      if (i > 0)
+	{
+	  t = fd->loops[i].n2;
+	  t = force_gimple_operand_gsi (&gsi, t, true, NULL_TREE,
+					false, GSI_CONTINUE_LINKING);
+	  tree v = fd->loops[i].v;
+	  if (DECL_P (v) && TREE_ADDRESSABLE (v))
+	    v = force_gimple_operand_gsi (&gsi, v, true, NULL_TREE,
+					  false, GSI_CONTINUE_LINKING);
+	  t = fold_build2 (fd->loops[i].cond_code, boolean_type_node, v, t);
+	  stmt = gimple_build_cond_empty (t);
+	  gsi_insert_after (&gsi, stmt, GSI_CONTINUE_LINKING);
+	  e = make_edge (bb, body_bb, EDGE_TRUE_VALUE);
+	  e->probability = REG_BR_PROB_BASE * 7 / 8;
+	}
+      else
+	make_edge (bb, body_bb, EDGE_FALLTHRU);
+      last_bb = bb;
+    }
+
+  return collapse_bb;
+}
+
+
 /* A subroutine of expand_omp_for.  Generate code for a parallel
    loop with any schedule.  Given parameters:
 
@@ -3816,105 +4473,14 @@  expand_omp_for_generic (struct omp_region *region,
   gcc_assert (gimple_code (gsi_stmt (gsi)) == GIMPLE_OMP_FOR);
   if (fd->collapse > 1)
     {
-      basic_block zero_iter_bb = NULL;
       int first_zero_iter = -1;
+      basic_block zero_iter_bb = NULL, l2_dom_bb = NULL;
 
-      /* collapsed loops need work for expansion in SSA form.  */
-      gcc_assert (!gimple_in_ssa_p (cfun));
-      counts = (tree *) alloca (fd->collapse * sizeof (tree));
-      for (i = 0; i < fd->collapse; i++)
-	{
-	  tree itype = TREE_TYPE (fd->loops[i].v);
+      counts = XALLOCAVEC (tree, fd->collapse);
+      expand_omp_for_init_counts (fd, &gsi, entry_bb, counts,
+				  zero_iter_bb, first_zero_iter,
+				  l2_dom_bb);
 
-	  if (SSA_VAR_P (fd->loop.n2)
-	      && ((t = fold_binary (fd->loops[i].cond_code, boolean_type_node,
-				    fold_convert (itype, fd->loops[i].n1),
-				    fold_convert (itype, fd->loops[i].n2)))
-		  == NULL_TREE || !integer_onep (t)))
-	    {
-	      tree n1, n2;
-	      n1 = fold_convert (itype, unshare_expr (fd->loops[i].n1));
-	      n1 = force_gimple_operand_gsi (&gsi, n1, true, NULL_TREE,
-					     true, GSI_SAME_STMT);
-	      n2 = fold_convert (itype, unshare_expr (fd->loops[i].n2));
-	      n2 = force_gimple_operand_gsi (&gsi, n2, true, NULL_TREE,
-					     true, GSI_SAME_STMT);
-	      stmt = gimple_build_cond (fd->loops[i].cond_code, n1, n2,
-					NULL_TREE, NULL_TREE);
-	      gsi_insert_before (&gsi, stmt, GSI_SAME_STMT);
-	      if (walk_tree (gimple_cond_lhs_ptr (stmt),
-			     expand_omp_regimplify_p, NULL, NULL)
-		  || walk_tree (gimple_cond_rhs_ptr (stmt),
-				expand_omp_regimplify_p, NULL, NULL))
-		{
-		  gsi = gsi_for_stmt (stmt);
-		  gimple_regimplify_operands (stmt, &gsi);
-		}
-	      e = split_block (entry_bb, stmt);
-	      if (zero_iter_bb == NULL)
-		{
-		  first_zero_iter = i;
-		  zero_iter_bb = create_empty_bb (entry_bb);
-		  if (current_loops)
-		    add_bb_to_loop (zero_iter_bb, entry_bb->loop_father);
-		  gsi = gsi_after_labels (zero_iter_bb);
-		  stmt = gimple_build_assign (fd->loop.n2,
-					      build_zero_cst (type));
-		  gsi_insert_before (&gsi, stmt, GSI_SAME_STMT);
-		  set_immediate_dominator (CDI_DOMINATORS, zero_iter_bb,
-					   entry_bb);
-		}
-	      ne = make_edge (entry_bb, zero_iter_bb, EDGE_FALSE_VALUE);
-	      ne->probability = REG_BR_PROB_BASE / 2000 - 1;
-	      e->flags = EDGE_TRUE_VALUE;
-	      e->probability = REG_BR_PROB_BASE - ne->probability;
-	      entry_bb = e->dest;
-	      gsi = gsi_last_bb (entry_bb);
-	    }
-	  if (POINTER_TYPE_P (itype))
-	    itype = signed_type_for (itype);
-	  t = build_int_cst (itype, (fd->loops[i].cond_code == LT_EXPR
-				     ? -1 : 1));
-	  t = fold_build2 (PLUS_EXPR, itype,
-			   fold_convert (itype, fd->loops[i].step), t);
-	  t = fold_build2 (PLUS_EXPR, itype, t,
-			   fold_convert (itype, fd->loops[i].n2));
-	  t = fold_build2 (MINUS_EXPR, itype, t,
-			   fold_convert (itype, fd->loops[i].n1));
-	  if (TYPE_UNSIGNED (itype) && fd->loops[i].cond_code == GT_EXPR)
-	    t = fold_build2 (TRUNC_DIV_EXPR, itype,
-			     fold_build1 (NEGATE_EXPR, itype, t),
-			     fold_build1 (NEGATE_EXPR, itype,
-					  fold_convert (itype,
-							fd->loops[i].step)));
-	  else
-	    t = fold_build2 (TRUNC_DIV_EXPR, itype, t,
-			     fold_convert (itype, fd->loops[i].step));
-	  t = fold_convert (type, t);
-	  if (TREE_CODE (t) == INTEGER_CST)
-	    counts[i] = t;
-	  else
-	    {
-	      counts[i] = create_tmp_reg (type, ".count");
-	      t = force_gimple_operand_gsi (&gsi, t, false, NULL_TREE,
-					    true, GSI_SAME_STMT);
-	      stmt = gimple_build_assign (counts[i], t);
-	      gsi_insert_before (&gsi, stmt, GSI_SAME_STMT);
-	    }
-	  if (SSA_VAR_P (fd->loop.n2))
-	    {
-	      if (i == 0)
-		t = counts[0];
-	      else
-		{
-		  t = fold_build2 (MULT_EXPR, type, fd->loop.n2, counts[i]);
-		  t = force_gimple_operand_gsi (&gsi, t, false, NULL_TREE,
-						true, GSI_SAME_STMT);
-		}
-	      stmt = gimple_build_assign (fd->loop.n2, t);
-	      gsi_insert_before (&gsi, stmt, GSI_SAME_STMT);
-	    }
-	}
       if (zero_iter_bb)
 	{
 	  /* Some counts[i] vars might be uninitialized if
@@ -3949,18 +4515,21 @@  expand_omp_for_generic (struct omp_region *region,
       t4 = build_fold_addr_expr (iend0);
       t3 = build_fold_addr_expr (istart0);
       t2 = fold_convert (fd->iter_type, fd->loop.step);
-      if (POINTER_TYPE_P (type)
-	  && TYPE_PRECISION (type) != TYPE_PRECISION (fd->iter_type))
+      t1 = fd->loop.n2;
+      t0 = fd->loop.n1;
+      if (POINTER_TYPE_P (TREE_TYPE (t0))
+	  && TYPE_PRECISION (TREE_TYPE (t0))
+	     != TYPE_PRECISION (fd->iter_type))
 	{
 	  /* Avoid casting pointers to integer of a different size.  */
 	  tree itype = signed_type_for (type);
-	  t1 = fold_convert (fd->iter_type, fold_convert (itype, fd->loop.n2));
-	  t0 = fold_convert (fd->iter_type, fold_convert (itype, fd->loop.n1));
+	  t1 = fold_convert (fd->iter_type, fold_convert (itype, t1));
+	  t0 = fold_convert (fd->iter_type, fold_convert (itype, t0));
 	}
       else
 	{
-	  t1 = fold_convert (fd->iter_type, fd->loop.n2);
-	  t0 = fold_convert (fd->iter_type, fd->loop.n1);
+	  t1 = fold_convert (fd->iter_type, t1);
+	  t0 = fold_convert (fd->iter_type, t0);
 	}
       if (bias)
 	{
@@ -4015,64 +4584,38 @@  expand_omp_for_generic (struct omp_region *region,
   gsi_remove (&gsi, true);
 
   /* Iteration setup for sequential loop goes in L0_BB.  */
+  tree startvar = fd->loop.v;
+  tree endvar = NULL_TREE;
+
   gsi = gsi_start_bb (l0_bb);
   t = istart0;
   if (bias)
     t = fold_build2 (MINUS_EXPR, fd->iter_type, t, bias);
-  if (POINTER_TYPE_P (type))
-    t = fold_convert (signed_type_for (type), t);
-  t = fold_convert (type, t);
+  if (POINTER_TYPE_P (TREE_TYPE (startvar)))
+    t = fold_convert (signed_type_for (TREE_TYPE (startvar)), t);
+  t = fold_convert (TREE_TYPE (startvar), t);
   t = force_gimple_operand_gsi (&gsi, t,
-				DECL_P (fd->loop.v)
-				&& TREE_ADDRESSABLE (fd->loop.v),
+				DECL_P (startvar)
+				&& TREE_ADDRESSABLE (startvar),
 				NULL_TREE, false, GSI_CONTINUE_LINKING);
-  stmt = gimple_build_assign (fd->loop.v, t);
+  stmt = gimple_build_assign (startvar, t);
   gsi_insert_after (&gsi, stmt, GSI_CONTINUE_LINKING);
 
   t = iend0;
   if (bias)
     t = fold_build2 (MINUS_EXPR, fd->iter_type, t, bias);
-  if (POINTER_TYPE_P (type))
-    t = fold_convert (signed_type_for (type), t);
-  t = fold_convert (type, t);
+  if (POINTER_TYPE_P (TREE_TYPE (startvar)))
+    t = fold_convert (signed_type_for (TREE_TYPE (startvar)), t);
+  t = fold_convert (TREE_TYPE (startvar), t);
   iend = force_gimple_operand_gsi (&gsi, t, true, NULL_TREE,
 				   false, GSI_CONTINUE_LINKING);
-  if (fd->collapse > 1)
+  if (endvar)
     {
-      tree tem = create_tmp_reg (type, ".tem");
-      stmt = gimple_build_assign (tem, fd->loop.v);
+      stmt = gimple_build_assign (endvar, iend);
       gsi_insert_after (&gsi, stmt, GSI_CONTINUE_LINKING);
-      for (i = fd->collapse - 1; i >= 0; i--)
-	{
-	  tree vtype = TREE_TYPE (fd->loops[i].v), itype;
-	  itype = vtype;
-	  if (POINTER_TYPE_P (vtype))
-	    itype = signed_type_for (vtype);
-	  t = fold_build2 (TRUNC_MOD_EXPR, type, tem, counts[i]);
-	  t = fold_convert (itype, t);
-	  t = fold_build2 (MULT_EXPR, itype, t,
-			   fold_convert (itype, fd->loops[i].step));
-	  if (POINTER_TYPE_P (vtype))
-	    t = fold_build_pointer_plus (fd->loops[i].n1, t);
-	  else
-	    t = fold_build2 (PLUS_EXPR, itype, fd->loops[i].n1, t);
-	  t = force_gimple_operand_gsi (&gsi, t,
-					DECL_P (fd->loops[i].v)
-					&& TREE_ADDRESSABLE (fd->loops[i].v),
-					NULL_TREE, false,
-					GSI_CONTINUE_LINKING);
-	  stmt = gimple_build_assign (fd->loops[i].v, t);
-	  gsi_insert_after (&gsi, stmt, GSI_CONTINUE_LINKING);
-	  if (i != 0)
-	    {
-	      t = fold_build2 (TRUNC_DIV_EXPR, type, tem, counts[i]);
-	      t = force_gimple_operand_gsi (&gsi, t, false, NULL_TREE,
-					    false, GSI_CONTINUE_LINKING);
-	      stmt = gimple_build_assign (tem, t);
-	      gsi_insert_after (&gsi, stmt, GSI_CONTINUE_LINKING);
-	    }
-	}
     }
+  if (fd->collapse > 1)
+    expand_omp_for_init_vars (fd, &gsi, counts, startvar);
 
   if (!broken_loop)
     {
@@ -4084,93 +4627,32 @@  expand_omp_for_generic (struct omp_region *region,
       vmain = gimple_omp_continue_control_use (stmt);
       vback = gimple_omp_continue_control_def (stmt);
 
-      if (POINTER_TYPE_P (type))
-	t = fold_build_pointer_plus (vmain, fd->loop.step);
-      else
-	t = fold_build2 (PLUS_EXPR, type, vmain, fd->loop.step);
-      t = force_gimple_operand_gsi (&gsi, t,
-				    DECL_P (vback) && TREE_ADDRESSABLE (vback),
-				    NULL_TREE, true, GSI_SAME_STMT);
-      stmt = gimple_build_assign (vback, t);
-      gsi_insert_before (&gsi, stmt, GSI_SAME_STMT);
-
-      t = build2 (fd->loop.cond_code, boolean_type_node,
-		  DECL_P (vback) && TREE_ADDRESSABLE (vback) ? t : vback,
-		  iend);
-      stmt = gimple_build_cond_empty (t);
-      gsi_insert_before (&gsi, stmt, GSI_SAME_STMT);
+      /* OMP4 placeholder: if (!gimple_omp_for_combined_p (fd->for_stmt)).  */
+      if (1)
+	{
+	  if (POINTER_TYPE_P (type))
+	    t = fold_build_pointer_plus (vmain, fd->loop.step);
+	  else
+	    t = fold_build2 (PLUS_EXPR, type, vmain, fd->loop.step);
+	  t = force_gimple_operand_gsi (&gsi, t,
+					DECL_P (vback)
+					&& TREE_ADDRESSABLE (vback),
+					NULL_TREE, true, GSI_SAME_STMT);
+	  stmt = gimple_build_assign (vback, t);
+	  gsi_insert_before (&gsi, stmt, GSI_SAME_STMT);
+
+	  t = build2 (fd->loop.cond_code, boolean_type_node,
+		      DECL_P (vback) && TREE_ADDRESSABLE (vback) ? t : vback,
+		      iend);
+	  stmt = gimple_build_cond_empty (t);
+	  gsi_insert_before (&gsi, stmt, GSI_SAME_STMT);
+	}
 
       /* Remove GIMPLE_OMP_CONTINUE.  */
       gsi_remove (&gsi, true);
 
       if (fd->collapse > 1)
-	{
-	  basic_block last_bb, bb;
-
-	  last_bb = cont_bb;
-	  for (i = fd->collapse - 1; i >= 0; i--)
-	    {
-	      tree vtype = TREE_TYPE (fd->loops[i].v);
-
-	      bb = create_empty_bb (last_bb);
-	      if (current_loops)
-		add_bb_to_loop (bb, last_bb->loop_father);
-	      gsi = gsi_start_bb (bb);
-
-	      if (i < fd->collapse - 1)
-		{
-		  e = make_edge (last_bb, bb, EDGE_FALSE_VALUE);
-		  e->probability = REG_BR_PROB_BASE / 8;
-
-		  t = fd->loops[i + 1].n1;
-		  t = force_gimple_operand_gsi (&gsi, t,
-						DECL_P (fd->loops[i + 1].v)
-						&& TREE_ADDRESSABLE
-							(fd->loops[i + 1].v),
-						NULL_TREE, false,
-						GSI_CONTINUE_LINKING);
-		  stmt = gimple_build_assign (fd->loops[i + 1].v, t);
-		  gsi_insert_after (&gsi, stmt, GSI_CONTINUE_LINKING);
-		}
-	      else
-		collapse_bb = bb;
-
-	      set_immediate_dominator (CDI_DOMINATORS, bb, last_bb);
-
-	      if (POINTER_TYPE_P (vtype))
-		t = fold_build_pointer_plus (fd->loops[i].v, fd->loops[i].step);
-	      else
-		t = fold_build2 (PLUS_EXPR, vtype, fd->loops[i].v,
-				 fd->loops[i].step);
-	      t = force_gimple_operand_gsi (&gsi, t,
-					    DECL_P (fd->loops[i].v)
-					    && TREE_ADDRESSABLE (fd->loops[i].v),
-					    NULL_TREE, false,
-					    GSI_CONTINUE_LINKING);
-	      stmt = gimple_build_assign (fd->loops[i].v, t);
-	      gsi_insert_after (&gsi, stmt, GSI_CONTINUE_LINKING);
-
-	      if (i > 0)
-		{
-		  t = fd->loops[i].n2;
-		  t = force_gimple_operand_gsi (&gsi, t, true, NULL_TREE,
-						false, GSI_CONTINUE_LINKING);
-		  tree v = fd->loops[i].v;
-		  if (DECL_P (v) && TREE_ADDRESSABLE (v))
-		    v = force_gimple_operand_gsi (&gsi, v, true, NULL_TREE,
-						  false, GSI_CONTINUE_LINKING);
-		  t = fold_build2 (fd->loops[i].cond_code, boolean_type_node,
-				   v, t);
-		  stmt = gimple_build_cond_empty (t);
-		  gsi_insert_after (&gsi, stmt, GSI_CONTINUE_LINKING);
-		  e = make_edge (bb, l1_bb, EDGE_TRUE_VALUE);
-		  e->probability = REG_BR_PROB_BASE * 7 / 8;
-		}
-	      else
-		make_edge (bb, l1_bb, EDGE_FALLTHRU);
-	      last_bb = bb;
-	    }
-	}
+	collapse_bb = extract_omp_for_update_vars (fd, cont_bb, l1_bb);
 
       /* Emit code to get the next parallel iteration in L2_BB.  */
       gsi = gsi_start_bb (l2_bb);
@@ -4220,19 +4702,27 @@  expand_omp_for_generic (struct omp_region *region,
       make_edge (cont_bb, l2_bb, EDGE_FALSE_VALUE);
       if (current_loops)
 	add_bb_to_loop (l2_bb, cont_bb->loop_father);
-      if (fd->collapse > 1)
+      e = find_edge (cont_bb, l1_bb);
+      /* OMP4 placeholder for gimple_omp_for_combined_p (fd->for_stmt).  */
+      if (0)
+	;
+      else if (fd->collapse > 1)
 	{
-	  e = find_edge (cont_bb, l1_bb);
 	  remove_edge (e);
 	  e = make_edge (cont_bb, collapse_bb, EDGE_TRUE_VALUE);
 	}
       else
+	e->flags = EDGE_TRUE_VALUE;
+      if (e)
 	{
-	  e = find_edge (cont_bb, l1_bb);
-	  e->flags = EDGE_TRUE_VALUE;
+	  e->probability = REG_BR_PROB_BASE * 7 / 8;
+	  find_edge (cont_bb, l2_bb)->probability = REG_BR_PROB_BASE / 8;
+	}
+      else
+	{
+	  e = find_edge (cont_bb, l2_bb);
+	  e->flags = EDGE_FALLTHRU;
 	}
-      e->probability = REG_BR_PROB_BASE * 7 / 8;
-      find_edge (cont_bb, l2_bb)->probability = REG_BR_PROB_BASE / 8;
       make_edge (l2_bb, l0_bb, EDGE_TRUE_VALUE);
 
       set_immediate_dominator (CDI_DOMINATORS, l2_bb,
@@ -4249,10 +4739,14 @@  expand_omp_for_generic (struct omp_region *region,
       outer_loop->latch = l2_bb;
       add_loop (outer_loop, l0_bb->loop_father);
 
-      struct loop *loop = alloc_loop ();
-      loop->header = l1_bb;
-      /* The loop may have multiple latches.  */
-      add_loop (loop, outer_loop);
+      /* OMP4 placeholder: if (!gimple_omp_for_combined_p (fd->for_stmt)).  */
+      if (1)
+	{
+	  struct loop *loop = alloc_loop ();
+	  loop->header = l1_bb;
+	  /* The loop may have multiple latches.  */
+	  add_loop (loop, outer_loop);
+	}
     }
 }
 
@@ -4883,6 +5377,295 @@  expand_omp_for_static_chunk (struct omp_region *region, struct omp_for_data *fd)
   add_loop (loop, trip_loop);
 }
 
+/* A subroutine of expand_omp_for.  Generate code for a simd non-worksharing
+   loop.  Given parameters:
+
+	for (V = N1; V cond N2; V += STEP) BODY;
+
+   where COND is "<" or ">", we generate pseudocode
+
+	V = N1;
+	goto L1;
+    L0:
+	BODY;
+	V += STEP;
+    L1:
+	if (V cond N2) goto L0; else goto L2;
+    L2:
+
+    For collapsed loops, given parameters:
+      collapse(3)
+      for (V1 = N11; V1 cond1 N12; V1 += STEP1)
+	for (V2 = N21; V2 cond2 N22; V2 += STEP2)
+	  for (V3 = N31; V3 cond3 N32; V3 += STEP3)
+	    BODY;
+
+    we generate pseudocode
+
+	if (cond3 is <)
+	  adj = STEP3 - 1;
+	else
+	  adj = STEP3 + 1;
+	count3 = (adj + N32 - N31) / STEP3;
+	if (cond2 is <)
+	  adj = STEP2 - 1;
+	else
+	  adj = STEP2 + 1;
+	count2 = (adj + N22 - N21) / STEP2;
+	if (cond1 is <)
+	  adj = STEP1 - 1;
+	else
+	  adj = STEP1 + 1;
+	count1 = (adj + N12 - N11) / STEP1;
+	count = count1 * count2 * count3;
+	V = 0;
+	V1 = N11;
+	V2 = N21;
+	V3 = N31;
+	goto L1;
+    L0:
+	BODY;
+	V += 1;
+	V3 += STEP3;
+	V2 += (V3 cond3 N32) ? 0 : STEP2;
+	V3 = (V3 cond3 N32) ? V3 : N31;
+	V1 += (V2 cond2 N22) ? 0 : STEP1;
+	V2 = (V2 cond2 N22) ? V2 : N21;
+    L1:
+	if (V < count) goto L0; else goto L2;
+    L2:
+
+      */
+
+static void
+expand_omp_simd (struct omp_region *region, struct omp_for_data *fd)
+{
+  tree type, t;
+  basic_block entry_bb, cont_bb, exit_bb, l0_bb, l1_bb, l2_bb, l2_dom_bb;
+  gimple_stmt_iterator gsi;
+  gimple stmt;
+  bool broken_loop = region->cont == NULL;
+  edge e, ne;
+  tree *counts = NULL;
+  int i;
+  tree safelen = find_omp_clause (gimple_omp_for_clauses (fd->for_stmt),
+				  OMP_CLAUSE_SAFELEN);
+  tree simduid = find_omp_clause (gimple_omp_for_clauses (fd->for_stmt),
+				  OMP_CLAUSE__SIMDUID_);
+  tree n2;
+
+  type = TREE_TYPE (fd->loop.v);
+  entry_bb = region->entry;
+  cont_bb = region->cont;
+  gcc_assert (EDGE_COUNT (entry_bb->succs) == 2);
+  gcc_assert (broken_loop
+	      || BRANCH_EDGE (entry_bb)->dest == FALLTHRU_EDGE (cont_bb)->dest);
+  l0_bb = FALLTHRU_EDGE (entry_bb)->dest;
+  if (!broken_loop)
+    {
+      gcc_assert (BRANCH_EDGE (cont_bb)->dest == l0_bb);
+      gcc_assert (EDGE_COUNT (cont_bb->succs) == 2);
+      l1_bb = split_block (cont_bb, last_stmt (cont_bb))->dest;
+      l2_bb = BRANCH_EDGE (entry_bb)->dest;
+    }
+  else
+    {
+      BRANCH_EDGE (entry_bb)->flags &= ~EDGE_ABNORMAL;
+      l1_bb = split_edge (BRANCH_EDGE (entry_bb));
+      l2_bb = single_succ (l1_bb);
+    }
+  exit_bb = region->exit;
+  l2_dom_bb = NULL;
+
+  gsi = gsi_last_bb (entry_bb);
+
+  gcc_assert (gimple_code (gsi_stmt (gsi)) == GIMPLE_OMP_FOR);
+  /* Not needed in SSA form right now.  */
+  gcc_assert (!gimple_in_ssa_p (cfun));
+  if (fd->collapse > 1)
+    {
+      int first_zero_iter = -1;
+      basic_block zero_iter_bb = l2_bb;
+
+      counts = XALLOCAVEC (tree, fd->collapse);
+      expand_omp_for_init_counts (fd, &gsi, entry_bb, counts,
+				  zero_iter_bb, first_zero_iter,
+				  l2_dom_bb);
+    }
+  if (l2_dom_bb == NULL)
+    l2_dom_bb = l1_bb;
+
+  n2 = fd->loop.n2;
+  if (0)
+    /* Place holder for gimple_omp_for_combined_into_p() in
+       the upcoming gomp-4_0-branch merge.  */;
+  else
+    {
+      expand_omp_build_assign (&gsi, fd->loop.v,
+			       fold_convert (type, fd->loop.n1));
+      if (fd->collapse > 1)
+	for (i = 0; i < fd->collapse; i++)
+	  {
+	    tree itype = TREE_TYPE (fd->loops[i].v);
+	    if (POINTER_TYPE_P (itype))
+	      itype = signed_type_for (itype);
+	    t = fold_convert (TREE_TYPE (fd->loops[i].v), fd->loops[i].n1);
+	    expand_omp_build_assign (&gsi, fd->loops[i].v, t);
+	  }
+      }
+
+  /* Remove the GIMPLE_OMP_FOR statement.  */
+  gsi_remove (&gsi, true);
+
+  if (!broken_loop)
+    {
+      /* Code to control the increment goes in the CONT_BB.  */
+      gsi = gsi_last_bb (cont_bb);
+      stmt = gsi_stmt (gsi);
+      gcc_assert (gimple_code (stmt) == GIMPLE_OMP_CONTINUE);
+
+      if (POINTER_TYPE_P (type))
+	t = fold_build_pointer_plus (fd->loop.v, fd->loop.step);
+      else
+	t = fold_build2 (PLUS_EXPR, type, fd->loop.v, fd->loop.step);
+      expand_omp_build_assign (&gsi, fd->loop.v, t);
+
+      if (fd->collapse > 1)
+	{
+	  i = fd->collapse - 1;
+	  if (POINTER_TYPE_P (TREE_TYPE (fd->loops[i].v)))
+	    {
+	      t = fold_convert (sizetype, fd->loops[i].step);
+	      t = fold_build_pointer_plus (fd->loops[i].v, t);
+	    }
+	  else
+	    {
+	      t = fold_convert (TREE_TYPE (fd->loops[i].v),
+				fd->loops[i].step);
+	      t = fold_build2 (PLUS_EXPR, TREE_TYPE (fd->loops[i].v),
+			       fd->loops[i].v, t);
+	    }
+	  expand_omp_build_assign (&gsi, fd->loops[i].v, t);
+
+	  for (i = fd->collapse - 1; i > 0; i--)
+	    {
+	      tree itype = TREE_TYPE (fd->loops[i].v);
+	      tree itype2 = TREE_TYPE (fd->loops[i - 1].v);
+	      if (POINTER_TYPE_P (itype2))
+		itype2 = signed_type_for (itype2);
+	      t = build3 (COND_EXPR, itype2,
+			  build2 (fd->loops[i].cond_code, boolean_type_node,
+				  fd->loops[i].v,
+				  fold_convert (itype, fd->loops[i].n2)),
+			  build_int_cst (itype2, 0),
+			  fold_convert (itype2, fd->loops[i - 1].step));
+	      if (POINTER_TYPE_P (TREE_TYPE (fd->loops[i - 1].v)))
+		t = fold_build_pointer_plus (fd->loops[i - 1].v, t);
+	      else
+		t = fold_build2 (PLUS_EXPR, itype2, fd->loops[i - 1].v, t);
+	      expand_omp_build_assign (&gsi, fd->loops[i - 1].v, t);
+
+	      t = build3 (COND_EXPR, itype,
+			  build2 (fd->loops[i].cond_code, boolean_type_node,
+				  fd->loops[i].v,
+				  fold_convert (itype, fd->loops[i].n2)),
+			  fd->loops[i].v,
+			  fold_convert (itype, fd->loops[i].n1));
+	      expand_omp_build_assign (&gsi, fd->loops[i].v, t);
+	    }
+	}
+
+      /* Remove GIMPLE_OMP_CONTINUE.  */
+      gsi_remove (&gsi, true);
+    }
+
+  /* Emit the condition in L1_BB.  */
+  gsi = gsi_start_bb (l1_bb);
+
+  t = fold_convert (type, n2);
+  t = force_gimple_operand_gsi (&gsi, t, true, NULL_TREE,
+				false, GSI_CONTINUE_LINKING);
+  t = build2 (fd->loop.cond_code, boolean_type_node, fd->loop.v, t);
+  stmt = gimple_build_cond_empty (t);
+  gsi_insert_after (&gsi, stmt, GSI_CONTINUE_LINKING);
+  if (walk_tree (gimple_cond_lhs_ptr (stmt), expand_omp_regimplify_p,
+		 NULL, NULL)
+      || walk_tree (gimple_cond_rhs_ptr (stmt), expand_omp_regimplify_p,
+		    NULL, NULL))
+    {
+      gsi = gsi_for_stmt (stmt);
+      gimple_regimplify_operands (stmt, &gsi);
+    }
+
+  /* Remove GIMPLE_OMP_RETURN.  */
+  gsi = gsi_last_bb (exit_bb);
+  gsi_remove (&gsi, true);
+
+  /* Connect the new blocks.  */
+  remove_edge (FALLTHRU_EDGE (entry_bb));
+
+  if (!broken_loop)
+    {
+      remove_edge (BRANCH_EDGE (entry_bb));
+      make_edge (entry_bb, l1_bb, EDGE_FALLTHRU);
+
+      e = BRANCH_EDGE (l1_bb);
+      ne = FALLTHRU_EDGE (l1_bb);
+      e->flags = EDGE_TRUE_VALUE;
+    }
+  else
+    {
+      single_succ_edge (entry_bb)->flags = EDGE_FALLTHRU;
+
+      ne = single_succ_edge (l1_bb);
+      e = make_edge (l1_bb, l0_bb, EDGE_TRUE_VALUE);
+
+    }
+  ne->flags = EDGE_FALSE_VALUE;
+  e->probability = REG_BR_PROB_BASE * 7 / 8;
+  ne->probability = REG_BR_PROB_BASE / 8;
+
+  set_immediate_dominator (CDI_DOMINATORS, l1_bb, entry_bb);
+  set_immediate_dominator (CDI_DOMINATORS, l2_bb, l2_dom_bb);
+  set_immediate_dominator (CDI_DOMINATORS, l0_bb, l1_bb);
+
+  if (!broken_loop)
+    {
+      struct loop *loop = alloc_loop ();
+      loop->header = l1_bb;
+      loop->latch = e->dest;
+      add_loop (loop, l1_bb->loop_father);
+      if (safelen == NULL_TREE)
+	loop->safelen = INT_MAX;
+      else
+	{
+	  safelen = OMP_CLAUSE_SAFELEN_EXPR (safelen);
+	  if (!host_integerp (safelen, 1)
+	      || (unsigned HOST_WIDE_INT) tree_low_cst (safelen, 1)
+		 > INT_MAX)
+	    loop->safelen = INT_MAX;
+	  else
+	    loop->safelen = tree_low_cst (safelen, 1);
+	  if (loop->safelen == 1)
+	    loop->safelen = 0;
+	}
+      if (simduid)
+	{
+	  loop->simduid = OMP_CLAUSE__SIMDUID__DECL (simduid);
+	  cfun->has_simduid_loops = true;
+	}
+      /* If not -fno-tree-vectorize, hint that we want to vectorize
+	 the loop.  */
+      if ((flag_tree_vectorize
+	   || !global_options_set.x_flag_tree_vectorize)
+	  && loop->safelen > 1)
+	{
+	  loop->force_vect = true;
+	  cfun->has_force_vect_loops = true;
+	}
+    }
+}
+
 
 /* Expand the OpenMP loop defined by REGION.  */
 
@@ -4914,7 +5697,9 @@  expand_omp_for (struct omp_region *region)
        original loops from being detected.  Fix that up.  */
     loops_state_set (LOOPS_NEED_FIXUP);
 
-  if (fd.sched_kind == OMP_CLAUSE_SCHEDULE_STATIC
+  if (gimple_omp_for_kind (fd.for_stmt) == GF_OMP_FOR_KIND_SIMD)
+    expand_omp_simd (region, &fd);
+  else if (fd.sched_kind == OMP_CLAUSE_SCHEDULE_STATIC
       && !fd.have_ordered
       && fd.collapse == 1
       && region->cont != NULL)
@@ -4928,6 +5713,8 @@  expand_omp_for (struct omp_region *region)
     {
       int fn_index, start_ix, next_ix;
 
+      gcc_assert (gimple_omp_for_kind (fd.for_stmt)
+		  == GF_OMP_FOR_KIND_FOR);
       if (fd.chunk_size == NULL
 	  && fd.sched_kind == OMP_CLAUSE_SCHEDULE_STATIC)
 	fd.chunk_size = integer_zero_node;
@@ -6516,6 +7303,8 @@  lower_omp_for_lastprivate (struct omp_for_data *fd, gimple_seq *body_p,
 	  && host_integerp (fd->loop.n2, 0)
 	  && ! integer_zerop (fd->loop.n2))
 	vinit = build_int_cst (TREE_TYPE (fd->loop.v), 0);
+      else
+	vinit = unshare_expr (vinit);
 
       /* Initialize the iterator variable, so that threads that don't execute
 	 any iterations don't execute the lastprivate clauses by accident.  */
@@ -6539,7 +7328,6 @@  lower_omp_for (gimple_stmt_iterator *gsi_p, omp_context *ctx)
   push_gimplify_context (&gctx);
 
   lower_omp (gimple_omp_for_pre_body_ptr (stmt), ctx);
-  lower_omp (gimple_omp_body_ptr (stmt), ctx);
 
   block = make_node (BLOCK);
   new_stmt = gimple_build_bind (NULL, NULL, block);
@@ -6564,6 +7352,8 @@  lower_omp_for (gimple_stmt_iterator *gsi_p, omp_context *ctx)
   lower_rec_input_clauses (gimple_omp_for_clauses (stmt), &body, &dlist, ctx);
   gimple_seq_add_seq (&body, gimple_omp_for_pre_body (stmt));
 
+  lower_omp (gimple_omp_body_ptr (stmt), ctx);
+
   /* Lower the header expressions.  At this point, we can assume that
      the header is of the form:
 
diff --git a/gcc/tree-data-ref.c b/gcc/tree-data-ref.c
index 10431c0..5bd7719 100644
--- a/gcc/tree-data-ref.c
+++ b/gcc/tree-data-ref.c
@@ -4331,10 +4331,25 @@  get_references_in_stmt (gimple stmt, vec<data_ref_loc, va_stack> *references)
   /* ASM_EXPR and CALL_EXPR may embed arbitrary side effects.
      As we cannot model data-references to not spelled out
      accesses give up if they may occur.  */
-  if ((stmt_code == GIMPLE_CALL
-       && !(gimple_call_flags (stmt) & ECF_CONST))
-      || (stmt_code == GIMPLE_ASM
-	  && (gimple_asm_volatile_p (stmt) || gimple_vuse (stmt))))
+  if (stmt_code == GIMPLE_CALL
+      && !(gimple_call_flags (stmt) & ECF_CONST))
+    {
+      /* Allow IFN_GOMP_SIMD_LANE in their own loops.  */
+      if (gimple_call_internal_p (stmt)
+	  && gimple_call_internal_fn (stmt) == IFN_GOMP_SIMD_LANE)
+	{
+	  struct loop *loop = gimple_bb (stmt)->loop_father;
+	  tree uid = gimple_call_arg (stmt, 0);
+	  gcc_assert (TREE_CODE (uid) == SSA_NAME);
+	  if (loop == NULL
+	      || loop->simduid != SSA_NAME_VAR (uid))
+	    clobbers_memory = true;
+	}
+      else
+	clobbers_memory = true;
+    }
+  else if (stmt_code == GIMPLE_ASM
+	   && (gimple_asm_volatile_p (stmt) || gimple_vuse (stmt)))
     clobbers_memory = true;
 
   if (!gimple_vuse (stmt))
diff --git a/gcc/tree-flow.h b/gcc/tree-flow.h
index caa8d74..fe9ecee 100644
--- a/gcc/tree-flow.h
+++ b/gcc/tree-flow.h
@@ -344,7 +344,6 @@  extern struct omp_region *new_omp_region (basic_block, enum gimple_code,
 					  struct omp_region *);
 extern void free_omp_regions (void);
 void omp_expand_local (basic_block);
-extern tree find_omp_clause (tree, enum omp_clause_code);
 tree copy_var_decl (tree, tree, tree);
 
 /*---------------------------------------------------------------------------
diff --git a/gcc/tree-if-conv.c b/gcc/tree-if-conv.c
index 0ebb8c3..eb3a3fa 100644
--- a/gcc/tree-if-conv.c
+++ b/gcc/tree-if-conv.c
@@ -1822,6 +1822,10 @@  main_tree_if_conversion (void)
     return 0;
 
   FOR_EACH_LOOP (li, loop, 0)
+    if (flag_tree_loop_if_convert == 1
+	|| flag_tree_loop_if_convert_stores == 1
+	|| flag_tree_vectorize
+	|| loop->force_vect)
     changed |= tree_if_conversion (loop);
 
   if (changed)
@@ -1848,7 +1852,8 @@  main_tree_if_conversion (void)
 static bool
 gate_tree_if_conversion (void)
 {
-  return ((flag_tree_vectorize && flag_tree_loop_if_convert != 0)
+  return (((flag_tree_vectorize || cfun->has_force_vect_loops)
+	   && flag_tree_loop_if_convert != 0)
 	  || flag_tree_loop_if_convert == 1
 	  || flag_tree_loop_if_convert_stores == 1);
 }
diff --git a/gcc/tree-inline.c b/gcc/tree-inline.c
index f524771..4ec1d66 100644
--- a/gcc/tree-inline.c
+++ b/gcc/tree-inline.c
@@ -1298,7 +1298,8 @@  remap_gimple_stmt (gimple stmt, copy_body_data *id)
 	case GIMPLE_OMP_FOR:
 	  s1 = remap_gimple_seq (gimple_omp_body (stmt), id);
 	  s2 = remap_gimple_seq (gimple_omp_for_pre_body (stmt), id);
-	  copy = gimple_build_omp_for (s1, gimple_omp_for_clauses (stmt),
+	  copy = gimple_build_omp_for (s1, gimple_omp_for_kind (stmt),
+				       gimple_omp_for_clauses (stmt),
 				       gimple_omp_for_collapse (stmt), s2);
 	  {
 	    size_t i;
@@ -2331,6 +2332,8 @@  copy_cfg_body (copy_body_data * id, gcov_type count, int frequency_scale,
 		  get_loop (src_cfun, 0));
       /* Defer to cfgcleanup to update loop-father fields of basic-blocks.  */
       loops_state_set (LOOPS_NEED_FIXUP);
+      cfun->has_force_vect_loops |= src_cfun->has_force_vect_loops;
+      cfun->has_simduid_loops |= src_cfun->has_simduid_loops;
     }
 
   /* If the loop tree in the source function needed fixup, mark the
diff --git a/gcc/tree-parloops.c b/gcc/tree-parloops.c
index cea6f03..1e6bb07 100644
--- a/gcc/tree-parloops.c
+++ b/gcc/tree-parloops.c
@@ -1686,7 +1686,7 @@  create_parallel_loop (struct loop *loop, tree loop_fn, tree data,
   t = build_omp_clause (loc, OMP_CLAUSE_SCHEDULE);
   OMP_CLAUSE_SCHEDULE_KIND (t) = OMP_CLAUSE_SCHEDULE_STATIC;
 
-  for_stmt = gimple_build_omp_for (NULL, t, 1, NULL);
+  for_stmt = gimple_build_omp_for (NULL, GF_OMP_FOR_KIND_FOR, t, 1, NULL);
   gimple_set_location (for_stmt, loc);
   gimple_omp_for_set_index (for_stmt, 0, initvar);
   gimple_omp_for_set_initial (for_stmt, 0, cvar_init);
diff --git a/gcc/tree-pretty-print.c b/gcc/tree-pretty-print.c
index 7745f73..b2d32fa8 100644
--- a/gcc/tree-pretty-print.c
+++ b/gcc/tree-pretty-print.c
@@ -314,11 +314,14 @@  dump_omp_clause (pretty_printer *buffer, tree clause, int spc, int flags)
     case OMP_CLAUSE_COPYPRIVATE:
       name = "copyprivate";
       goto print_remap;
+    case OMP_CLAUSE_UNIFORM:
+      name = "uniform";
+      goto print_remap;
   print_remap:
       pp_string (buffer, name);
       pp_character (buffer, '(');
       dump_generic_node (buffer, OMP_CLAUSE_DECL (clause),
-	  spc, flags, false);
+			 spc, flags, false);
       pp_character (buffer, ')');
       break;
 
@@ -431,6 +434,30 @@  dump_omp_clause (pretty_printer *buffer, tree clause, int spc, int flags)
       pp_string (buffer, "mergeable");
       break;
 
+    case OMP_CLAUSE_LINEAR:
+      pp_string (buffer, "linear(");
+      dump_generic_node (buffer, OMP_CLAUSE_DECL (clause),
+			 spc, flags, false);
+      pp_character (buffer, ':');
+      dump_generic_node (buffer, OMP_CLAUSE_LINEAR_STEP (clause),
+			 spc, flags, false);
+      pp_character (buffer, ')');
+      break;
+
+    case OMP_CLAUSE_SAFELEN:
+      pp_string (buffer, "safelen(");
+      dump_generic_node (buffer, OMP_CLAUSE_SAFELEN_EXPR (clause),
+			 spc, flags, false);
+      pp_character (buffer, ')');
+      break;
+
+    case OMP_CLAUSE__SIMDUID_:
+      pp_string (buffer, "_simduid_(");
+      dump_generic_node (buffer, OMP_CLAUSE__SIMDUID__DECL (clause),
+			 spc, flags, false);
+      pp_character (buffer, ')');
+      break;
+
     default:
       /* Should never happen.  */
       dump_generic_node (buffer, clause, spc, flags, false);
@@ -2178,6 +2205,13 @@  dump_generic_node (pretty_printer *buffer, tree node, int spc, int flags,
 
     case OMP_FOR:
       pp_string (buffer, "#pragma omp for");
+      goto dump_omp_loop;
+
+    case OMP_SIMD:
+      pp_string (buffer, "#pragma omp simd");
+      goto dump_omp_loop;
+
+    dump_omp_loop:
       dump_omp_clauses (buffer, OMP_FOR_CLAUSES (node), spc, flags);
 
       if (!(flags & TDF_SLIM))
diff --git a/gcc/tree-ssa-ccp.c b/gcc/tree-ssa-ccp.c
index 1bc4c2f..7f66bda 100644
--- a/gcc/tree-ssa-ccp.c
+++ b/gcc/tree-ssa-ccp.c
@@ -626,6 +626,22 @@  likely_value (gimple stmt)
   if (has_constant_operand)
     all_undefined_operands = false;
 
+  if (has_undefined_operand
+      && code == GIMPLE_CALL
+      && gimple_call_internal_p (stmt))
+    switch (gimple_call_internal_fn (stmt))
+      {
+	/* These 3 builtins use the first argument just as a magic
+	   way how to find out a decl uid.  */
+      case IFN_GOMP_SIMD_LANE:
+      case IFN_GOMP_SIMD_VF:
+      case IFN_GOMP_SIMD_LAST_LANE:
+	has_undefined_operand = false;
+	break;
+      default:
+	break;
+      }
+
   /* If the operation combines operands like COMPLEX_EXPR make sure to
      not mark the result UNDEFINED if only one part of the result is
      undefined.  */
diff --git a/gcc/tree-ssa-loop-ivcanon.c b/gcc/tree-ssa-loop-ivcanon.c
index 91cf8c1..abe4557 100644
--- a/gcc/tree-ssa-loop-ivcanon.c
+++ b/gcc/tree-ssa-loop-ivcanon.c
@@ -1125,6 +1125,11 @@  tree_unroll_loops_completely_1 (bool may_increase_size, bool unroll_outer,
   if (changed)
     return true;
 
+  /* Don't unroll #pragma omp simd loops until the vectorizer
+     attempts to vectorize those.  */
+  if (loop->force_vect)
+    return false;
+
   /* Try to unroll this loop.  */
   loop_father = loop_outer (loop);
   if (!loop_father)
diff --git a/gcc/tree-ssa-loop.c b/gcc/tree-ssa-loop.c
index 99e27a1..2160318 100644
--- a/gcc/tree-ssa-loop.c
+++ b/gcc/tree-ssa-loop.c
@@ -225,7 +225,7 @@  tree_vectorize (void)
 static bool
 gate_tree_vectorize (void)
 {
-  return flag_tree_vectorize;
+  return flag_tree_vectorize || cfun->has_force_vect_loops;
 }
 
 struct gimple_opt_pass pass_vectorize =
diff --git a/gcc/tree-vect-data-refs.c b/gcc/tree-vect-data-refs.c
index 47ecad3..f6e2131 100644
--- a/gcc/tree-vect-data-refs.c
+++ b/gcc/tree-vect-data-refs.c
@@ -255,6 +255,15 @@  vect_analyze_data_ref_dependence (struct data_dependence_relation *ddr,
   /* Unknown data dependence.  */
   if (DDR_ARE_DEPENDENT (ddr) == chrec_dont_know)
     {
+      /* If user asserted safelen consecutive iterations can be
+	 executed concurrently, assume independence.  */
+      if (loop->safelen >= 2)
+	{
+	  if (loop->safelen < *max_vf)
+	    *max_vf = loop->safelen;
+	  return false;
+	}
+
       if (STMT_VINFO_GATHER_P (stmtinfo_a)
 	  || STMT_VINFO_GATHER_P (stmtinfo_b))
 	{
@@ -291,6 +300,15 @@  vect_analyze_data_ref_dependence (struct data_dependence_relation *ddr,
   /* Known data dependence.  */
   if (DDR_NUM_DIST_VECTS (ddr) == 0)
     {
+      /* If user asserted safelen consecutive iterations can be
+	 executed concurrently, assume independence.  */
+      if (loop->safelen >= 2)
+	{
+	  if (loop->safelen < *max_vf)
+	    *max_vf = loop->safelen;
+	  return false;
+	}
+
       if (STMT_VINFO_GATHER_P (stmtinfo_a)
 	  || STMT_VINFO_GATHER_P (stmtinfo_b))
 	{
@@ -2859,6 +2877,7 @@  vect_analyze_data_refs (loop_vec_info loop_vinfo,
       stmt_vec_info stmt_info;
       tree base, offset, init;
       bool gather = false;
+      bool simd_lane_access = false;
       int vf;
 
 again:
@@ -2890,12 +2909,17 @@  again:
       if (!DR_BASE_ADDRESS (dr) || !DR_OFFSET (dr) || !DR_INIT (dr)
 	  || !DR_STEP (dr))
         {
-	  /* If target supports vector gather loads, see if they can't
-	     be used.  */
-	  if (loop_vinfo
-	      && DR_IS_READ (dr)
+	  bool maybe_gather
+	    = DR_IS_READ (dr)
 	      && !TREE_THIS_VOLATILE (DR_REF (dr))
-	      && targetm.vectorize.builtin_gather != NULL
+	      && targetm.vectorize.builtin_gather != NULL;
+	  bool maybe_simd_lane_access
+	    = loop_vinfo && loop->simduid;
+
+	  /* If target supports vector gather loads, or if this might be
+	     a SIMD lane access, see if they can't be used.  */
+	  if (loop_vinfo
+	      && (maybe_gather || maybe_simd_lane_access)
 	      && !nested_in_vect_loop_p (loop, stmt))
 	    {
 	      struct data_reference *newdr
@@ -2908,14 +2932,59 @@  again:
 		  && DR_STEP (newdr)
 		  && integer_zerop (DR_STEP (newdr)))
 		{
-		  dr = newdr;
-		  gather = true;
+		  if (maybe_simd_lane_access)
+		    {
+		      tree off = DR_OFFSET (newdr);
+		      STRIP_NOPS (off);
+		      if (TREE_CODE (DR_INIT (newdr)) == INTEGER_CST
+			  && TREE_CODE (off) == MULT_EXPR
+			  && host_integerp (TREE_OPERAND (off, 1), 1))
+			{
+			  tree step = TREE_OPERAND (off, 1);
+			  off = TREE_OPERAND (off, 0);
+			  STRIP_NOPS (off);
+			  if (CONVERT_EXPR_P (off)
+			      && TYPE_PRECISION (TREE_TYPE (TREE_OPERAND (off,
+									  0)))
+				 < TYPE_PRECISION (TREE_TYPE (off)))
+			    off = TREE_OPERAND (off, 0);
+			  if (TREE_CODE (off) == SSA_NAME)
+			    {
+			      gimple def = SSA_NAME_DEF_STMT (off);
+			      tree reft = TREE_TYPE (DR_REF (newdr));
+			      if (gimple_call_internal_p (def)
+				  && gimple_call_internal_fn (def)
+				  == IFN_GOMP_SIMD_LANE)
+				{
+				  tree arg = gimple_call_arg (def, 0);
+				  gcc_assert (TREE_CODE (arg) == SSA_NAME);
+				  arg = SSA_NAME_VAR (arg);
+				  if (arg == loop->simduid
+				      /* For now.  */
+				      && tree_int_cst_equal
+					   (TYPE_SIZE_UNIT (reft),
+					    step))
+				    {
+				      DR_OFFSET (newdr) = ssize_int (0);
+				      DR_STEP (newdr) = step;
+				      dr = newdr;
+				      simd_lane_access = true;
+				    }
+				}
+			    }
+			}
+		    }
+		  if (!simd_lane_access && maybe_gather)
+		    {
+		      dr = newdr;
+		      gather = true;
+		    }
 		}
-	      else
+	      if (!gather && !simd_lane_access)
 		free_data_ref (newdr);
 	    }
 
-	  if (!gather)
+	  if (!gather && !simd_lane_access)
 	    {
 	      if (dump_enabled_p ())
 		{
@@ -2942,7 +3011,7 @@  again:
           if (bb_vinfo)
 	    break;
 
-	  if (gather)
+	  if (gather || simd_lane_access)
 	    free_data_ref (dr);
 	  return false;
         }
@@ -2975,7 +3044,7 @@  again:
           if (bb_vinfo)
 	    break;
 
-	  if (gather)
+	  if (gather || simd_lane_access)
 	    free_data_ref (dr);
           return false;
         }
@@ -2994,7 +3063,7 @@  again:
           if (bb_vinfo)
 	    break;
 
-	  if (gather)
+	  if (gather || simd_lane_access)
 	    free_data_ref (dr);
           return false;
 	}
@@ -3015,7 +3084,7 @@  again:
 	  if (bb_vinfo)
 	    break;
 
-	  if (gather)
+	  if (gather || simd_lane_access)
 	    free_data_ref (dr);
 	  return false;
 	}
@@ -3150,12 +3219,17 @@  again:
           if (bb_vinfo)
 	    break;
 
-	  if (gather)
+	  if (gather || simd_lane_access)
 	    free_data_ref (dr);
           return false;
         }
 
       STMT_VINFO_DATA_REF (stmt_info) = dr;
+      if (simd_lane_access)
+	{
+	  STMT_VINFO_SIMD_LANE_ACCESS_P (stmt_info) = true;
+	  datarefs[i] = dr;
+	}
 
       /* Set vectype for STMT.  */
       scalar_type = TREE_TYPE (DR_REF (dr));
@@ -3176,7 +3250,7 @@  again:
           if (bb_vinfo)
 	    break;
 
-	  if (gather)
+	  if (gather || simd_lane_access)
 	    {
 	      STMT_VINFO_DATA_REF (stmt_info) = NULL;
 	      free_data_ref (dr);
diff --git a/gcc/tree-vect-loop.c b/gcc/tree-vect-loop.c
index c9b1021..3346dbf 100644
--- a/gcc/tree-vect-loop.c
+++ b/gcc/tree-vect-loop.c
@@ -5361,7 +5361,7 @@  vectorizable_induction (gimple phi, gimple_stmt_iterator *gsi ATTRIBUTE_UNUSED,
 bool
 vectorizable_live_operation (gimple stmt,
 			     gimple_stmt_iterator *gsi ATTRIBUTE_UNUSED,
-			     gimple *vec_stmt ATTRIBUTE_UNUSED)
+			     gimple *vec_stmt)
 {
   stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
   loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
@@ -5381,7 +5381,41 @@  vectorizable_live_operation (gimple stmt,
     return false;
 
   if (!is_gimple_assign (stmt))
-    return false;
+    {
+      if (gimple_call_internal_p (stmt)
+	  && gimple_call_internal_fn (stmt) == IFN_GOMP_SIMD_LANE
+	  && gimple_call_lhs (stmt)
+	  && loop->simduid
+	  && TREE_CODE (gimple_call_arg (stmt, 0)) == SSA_NAME
+	  && loop->simduid
+	     == SSA_NAME_VAR (gimple_call_arg (stmt, 0)))
+	{
+	  edge e = single_exit (loop);
+	  basic_block merge_bb = e->dest;
+	  imm_use_iterator imm_iter;
+	  use_operand_p use_p;
+	  tree lhs = gimple_call_lhs (stmt);
+
+	  FOR_EACH_IMM_USE_FAST (use_p, imm_iter, lhs)
+	    {
+	      gimple use_stmt = USE_STMT (use_p);
+	      if (gimple_code (use_stmt) == GIMPLE_PHI
+		  || gimple_bb (use_stmt) == merge_bb)
+		{
+		  if (vec_stmt)
+		    {
+		      tree vfm1
+			= build_int_cst (unsigned_type_node,
+					 loop_vinfo->vectorization_factor - 1);
+		      SET_PHI_ARG_DEF (use_stmt, e->dest_idx, vfm1);
+		    }
+		  return true;
+		}
+	    }
+	}
+
+      return false;
+    }
 
   if (TREE_CODE (gimple_assign_lhs (stmt)) != SSA_NAME)
     return false;
diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
index 0580f7d..3768dcd 100644
--- a/gcc/tree-vect-stmts.c
+++ b/gcc/tree-vect-stmts.c
@@ -1755,6 +1755,14 @@  vectorizable_call (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
   if (nargs == 0 || nargs > 3)
     return false;
 
+  /* Ignore the argument of IFN_GOMP_SIMD_LANE, it is magic.  */
+  if (gimple_call_internal_p (stmt)
+      && gimple_call_internal_fn (stmt) == IFN_GOMP_SIMD_LANE)
+    {
+      nargs = 0;
+      rhs_type = unsigned_type_node;
+    }
+
   for (i = 0; i < nargs; i++)
     {
       tree opvectype;
@@ -1830,11 +1838,26 @@  vectorizable_call (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
   fndecl = vectorizable_function (stmt, vectype_out, vectype_in);
   if (fndecl == NULL_TREE)
     {
-      if (dump_enabled_p ())
-	dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
-                         "function is not vectorizable.");
-
-      return false;
+      if (gimple_call_internal_p (stmt)
+	  && gimple_call_internal_fn (stmt) == IFN_GOMP_SIMD_LANE
+	  && !slp_node
+	  && loop_vinfo
+	  && LOOP_VINFO_LOOP (loop_vinfo)->simduid
+	  && TREE_CODE (gimple_call_arg (stmt, 0)) == SSA_NAME
+	  && LOOP_VINFO_LOOP (loop_vinfo)->simduid
+	     == SSA_NAME_VAR (gimple_call_arg (stmt, 0)))
+	{
+	  /* We can handle IFN_GOMP_SIMD_LANE by returning a
+	     { 0, 1, 2, ... vf - 1 } vector.  */
+	  gcc_assert (nargs == 0);
+	}
+      else
+	{
+	  if (dump_enabled_p ())
+	    dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
+			     "function is not vectorizable.");
+	  return false;
+	}
     }
 
   gcc_assert (!gimple_vuse (stmt));
@@ -1932,9 +1955,30 @@  vectorizable_call (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
 	      vargs.quick_push (vec_oprnd0);
 	    }
 
-	  new_stmt = gimple_build_call_vec (fndecl, vargs);
-	  new_temp = make_ssa_name (vec_dest, new_stmt);
-	  gimple_call_set_lhs (new_stmt, new_temp);
+	  if (gimple_call_internal_p (stmt)
+	      && gimple_call_internal_fn (stmt) == IFN_GOMP_SIMD_LANE)
+	    {
+	      tree *v = XALLOCAVEC (tree, nunits_out);
+	      int k;
+	      for (k = 0; k < nunits_out; ++k)
+		v[k] = build_int_cst (unsigned_type_node, j * nunits_out + k);
+	      tree cst = build_vector (vectype_out, v);
+	      tree new_var
+		= vect_get_new_vect_var (vectype_out, vect_simple_var, "cst_");
+	      gimple init_stmt = gimple_build_assign (new_var, cst);
+	      new_temp = make_ssa_name (new_var, init_stmt);
+	      gimple_assign_set_lhs (init_stmt, new_temp);
+	      vect_init_vector_1 (stmt, init_stmt, NULL);
+	      new_temp = make_ssa_name (vec_dest, NULL);
+	      new_stmt = gimple_build_assign (new_temp,
+					      gimple_assign_lhs (init_stmt));
+	    }
+	  else
+	    {
+	      new_stmt = gimple_build_call_vec (fndecl, vargs);
+	      new_temp = make_ssa_name (vec_dest, new_stmt);
+	      gimple_call_set_lhs (new_stmt, new_temp);
+	    }
 	  vect_finish_stmt_generation (stmt, new_stmt, gsi);
 
 	  if (j == 0)
@@ -3796,6 +3840,7 @@  vectorizable_store (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
   enum vect_def_type dt;
   stmt_vec_info prev_stmt_info = NULL;
   tree dataref_ptr = NULL_TREE;
+  tree dataref_offset = NULL_TREE;
   gimple ptr_incr = NULL;
   int nunits = TYPE_VECTOR_SUBPARTS (vectype);
   int ncopies;
@@ -4085,9 +4130,26 @@  vectorizable_store (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
 	  /* We should have catched mismatched types earlier.  */
 	  gcc_assert (useless_type_conversion_p (vectype,
 						 TREE_TYPE (vec_oprnd)));
-	  dataref_ptr = vect_create_data_ref_ptr (first_stmt, aggr_type, NULL,
-						  NULL_TREE, &dummy, gsi,
-						  &ptr_incr, false, &inv_p);
+	  bool simd_lane_access_p
+	    = STMT_VINFO_SIMD_LANE_ACCESS_P (stmt_info);
+	  if (simd_lane_access_p
+	      && TREE_CODE (DR_BASE_ADDRESS (first_dr)) == ADDR_EXPR
+	      && VAR_P (TREE_OPERAND (DR_BASE_ADDRESS (first_dr), 0))
+	      && integer_zerop (DR_OFFSET (first_dr))
+	      && integer_zerop (DR_INIT (first_dr))
+	      && alias_sets_conflict_p (get_alias_set (aggr_type),
+					get_alias_set (DR_REF (first_dr))))
+	    {
+	      dataref_ptr = unshare_expr (DR_BASE_ADDRESS (first_dr));
+	      dataref_offset = build_int_cst (reference_alias_ptr_type
+					      (DR_REF (first_dr)), 0);
+	    }
+	  else
+	    dataref_ptr
+	      = vect_create_data_ref_ptr (first_stmt, aggr_type,
+					  simd_lane_access_p ? loop : NULL,
+					  NULL_TREE, &dummy, gsi, &ptr_incr,
+					  simd_lane_access_p, &inv_p);
 	  gcc_assert (bb_vinfo || !inv_p);
 	}
       else
@@ -4108,8 +4170,13 @@  vectorizable_store (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
 	      dr_chain[i] = vec_oprnd;
 	      oprnds[i] = vec_oprnd;
 	    }
-	  dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt,
-					 TYPE_SIZE_UNIT (aggr_type));
+	  if (dataref_offset)
+	    dataref_offset
+	      = int_const_binop (PLUS_EXPR, dataref_offset,
+				 TYPE_SIZE_UNIT (aggr_type));
+	  else
+	    dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt,
+					   TYPE_SIZE_UNIT (aggr_type));
 	}
 
       if (store_lanes_p)
@@ -4161,8 +4228,10 @@  vectorizable_store (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
 		vec_oprnd = result_chain[i];
 
 	      data_ref = build2 (MEM_REF, TREE_TYPE (vec_oprnd), dataref_ptr,
-				 build_int_cst (reference_alias_ptr_type
-						(DR_REF (first_dr)), 0));
+				 dataref_offset
+				 ? dataref_offset
+				 : build_int_cst (reference_alias_ptr_type
+						  (DR_REF (first_dr)), 0));
 	      align = TYPE_ALIGN_UNIT (vectype);
 	      if (aligned_access_p (first_dr))
 		misalign = 0;
@@ -4181,8 +4250,9 @@  vectorizable_store (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
 					  TYPE_ALIGN (elem_type));
 		  misalign = DR_MISALIGNMENT (first_dr);
 		}
-	      set_ptr_info_alignment (get_ptr_info (dataref_ptr), align,
-				      misalign);
+	      if (dataref_offset == NULL_TREE)
+		set_ptr_info_alignment (get_ptr_info (dataref_ptr), align,
+					misalign);
 
 	      /* Arguments are ready.  Create the new vector stmt.  */
 	      new_stmt = gimple_build_assign (data_ref, vec_oprnd);
@@ -4314,6 +4384,7 @@  vectorizable_load (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
   tree dummy;
   enum dr_alignment_support alignment_support_scheme;
   tree dataref_ptr = NULL_TREE;
+  tree dataref_offset = NULL_TREE;
   gimple ptr_incr = NULL;
   int nunits = TYPE_VECTOR_SUBPARTS (vectype);
   int ncopies;
@@ -4947,9 +5018,32 @@  vectorizable_load (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
     {
       /* 1. Create the vector or array pointer update chain.  */
       if (j == 0)
-        dataref_ptr = vect_create_data_ref_ptr (first_stmt, aggr_type, at_loop,
-						offset, &dummy, gsi,
-						&ptr_incr, false, &inv_p);
+	{
+	  bool simd_lane_access_p
+	    = STMT_VINFO_SIMD_LANE_ACCESS_P (stmt_info);
+	  if (simd_lane_access_p
+	      && TREE_CODE (DR_BASE_ADDRESS (first_dr)) == ADDR_EXPR
+	      && VAR_P (TREE_OPERAND (DR_BASE_ADDRESS (first_dr), 0))
+	      && integer_zerop (DR_OFFSET (first_dr))
+	      && integer_zerop (DR_INIT (first_dr))
+	      && alias_sets_conflict_p (get_alias_set (aggr_type),
+					get_alias_set (DR_REF (first_dr)))
+	      && (alignment_support_scheme == dr_aligned
+		  || alignment_support_scheme == dr_unaligned_supported))
+	    {
+	      dataref_ptr = unshare_expr (DR_BASE_ADDRESS (first_dr));
+	      dataref_offset = build_int_cst (reference_alias_ptr_type
+					      (DR_REF (first_dr)), 0);
+	    }
+	  else
+	    dataref_ptr
+	      = vect_create_data_ref_ptr (first_stmt, aggr_type, at_loop,
+					  offset, &dummy, gsi, &ptr_incr,
+					  simd_lane_access_p, &inv_p);
+	}
+      else if (dataref_offset)
+	dataref_offset = int_const_binop (PLUS_EXPR, dataref_offset,
+					  TYPE_SIZE_UNIT (aggr_type));
       else
         dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt,
 				       TYPE_SIZE_UNIT (aggr_type));
@@ -4999,8 +5093,10 @@  vectorizable_load (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
 
 		    data_ref
 		      = build2 (MEM_REF, vectype, dataref_ptr,
-				build_int_cst (reference_alias_ptr_type
-					       (DR_REF (first_dr)), 0));
+				dataref_offset
+				? dataref_offset
+				: build_int_cst (reference_alias_ptr_type
+						 (DR_REF (first_dr)), 0));
 		    align = TYPE_ALIGN_UNIT (vectype);
 		    if (alignment_support_scheme == dr_aligned)
 		      {
@@ -5022,8 +5118,9 @@  vectorizable_load (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
 						TYPE_ALIGN (elem_type));
 			misalign = DR_MISALIGNMENT (first_dr);
 		      }
-		    set_ptr_info_alignment (get_ptr_info (dataref_ptr),
-					    align, misalign);
+		    if (dataref_offset == NULL_TREE)
+		      set_ptr_info_alignment (get_ptr_info (dataref_ptr),
+					      align, misalign);
 		    break;
 		  }
 		case dr_explicit_realign:
diff --git a/gcc/tree-vectorizer.c b/gcc/tree-vectorizer.c
index 1ef31ee..6de914f 100644
--- a/gcc/tree-vectorizer.c
+++ b/gcc/tree-vectorizer.c
@@ -66,13 +66,209 @@  along with GCC; see the file COPYING3.  If not see
 #include "cfgloop.h"
 #include "tree-vectorizer.h"
 #include "tree-pass.h"
+#include "hash-table.h"
+#include "tree-ssa-propagate.h"
 
 /* Loop or bb location.  */
 LOC vect_location;
 
 /* Vector mapping GIMPLE stmt to stmt_vec_info. */
 vec<vec_void_p> stmt_vec_info_vec;
+
+/* For mapping simduid to vectorization factor.  */
+
+struct simduid_to_vf : typed_free_remove<simduid_to_vf>
+{
+  unsigned int simduid;
+  int vf;
+
+  /* hash_table support.  */
+  typedef simduid_to_vf value_type;
+  typedef simduid_to_vf compare_type;
+  static inline hashval_t hash (const value_type *);
+  static inline int equal (const value_type *, const compare_type *);
+};
+
+inline hashval_t
+simduid_to_vf::hash (const value_type *p)
+{
+  return p->simduid;
+}
+
+inline int
+simduid_to_vf::equal (const value_type *p1, const value_type *p2)
+{
+  return p1->simduid == p2->simduid;
+}
+
+/* For mapping decl to simduid.  */
+
+struct decl_to_simduid : typed_free_remove<decl_to_simduid>
+{
+  tree decl;
+  unsigned int simduid;
+
+  /* hash_table support.  */
+  typedef decl_to_simduid value_type;
+  typedef decl_to_simduid compare_type;
+  static inline hashval_t hash (const value_type *);
+  static inline int equal (const value_type *, const compare_type *);
+};
+
+inline hashval_t
+decl_to_simduid::hash (const value_type *p)
+{
+  return DECL_UID (p->decl);
+}
+
+inline int
+decl_to_simduid::equal (const value_type *p1, const value_type *p2)
+{
+  return p1->decl == p2->decl;
+}
+
+/* Fold IFN_GOMP_SIMD_LANE, IFN_GOMP_SIMD_VF and IFN_GOMP_SIMD_LAST_LANE
+   into their corresponding constants.  */
+
+static void
+adjust_simduid_builtins (hash_table <simduid_to_vf> &htab)
+{
+  basic_block bb;
+
+  FOR_EACH_BB (bb)
+    {
+      gimple_stmt_iterator i;
+
+      for (i = gsi_start_bb (bb); !gsi_end_p (i); gsi_next (&i))
+	{
+	  unsigned int vf = 1;
+	  enum internal_fn ifn;
+	  gimple stmt = gsi_stmt (i);
+	  tree t;
+	  if (!is_gimple_call (stmt)
+	      || !gimple_call_internal_p (stmt))
+	    continue;
+	  ifn = gimple_call_internal_fn (stmt);
+	  switch (ifn)
+	    {
+	    case IFN_GOMP_SIMD_LANE:
+	    case IFN_GOMP_SIMD_VF:
+	    case IFN_GOMP_SIMD_LAST_LANE:
+	      break;
+	    default:
+	      continue;
+	    }
+	  tree arg = gimple_call_arg (stmt, 0);
+	  gcc_assert (arg != NULL_TREE);
+	  gcc_assert (TREE_CODE (arg) == SSA_NAME);
+	  simduid_to_vf *p = NULL, data;
+	  data.simduid = DECL_UID (SSA_NAME_VAR (arg));
+	  if (htab.is_created ())
+	    p = htab.find (&data);
+	  if (p)
+	    vf = p->vf;
+	  switch (ifn)
+	    {
+	    case IFN_GOMP_SIMD_VF:
+	      t = build_int_cst (unsigned_type_node, vf);
+	      break;
+	    case IFN_GOMP_SIMD_LANE:
+	      t = build_int_cst (unsigned_type_node, 0);
+	      break;
+	    case IFN_GOMP_SIMD_LAST_LANE:
+	      t = gimple_call_arg (stmt, 1);
+	      break;
+	    default:
+	      gcc_unreachable ();
+	    }
+	  update_call_from_tree (&i, t);
+	}
+    }
+}
 
+/* Helper structure for note_simd_array_uses.  */
+
+struct note_simd_array_uses_struct
+{
+  hash_table <decl_to_simduid> *htab;
+  unsigned int simduid;
+};
+
+/* Callback for note_simd_array_uses, called through walk_gimple_op.  */
+
+static tree
+note_simd_array_uses_cb (tree *tp, int *walk_subtrees, void *data)
+{
+  struct walk_stmt_info *wi = (struct walk_stmt_info *) data;
+  struct note_simd_array_uses_struct *ns
+    = (struct note_simd_array_uses_struct *) wi->info;
+
+  if (TYPE_P (*tp))
+    *walk_subtrees = 0;
+  else if (VAR_P (*tp)
+	   && lookup_attribute ("omp simd array", DECL_ATTRIBUTES (*tp))
+	   && DECL_CONTEXT (*tp) == current_function_decl)
+    {
+      decl_to_simduid data;
+      if (!ns->htab->is_created ())
+	ns->htab->create (15);
+      data.decl = *tp;
+      data.simduid = ns->simduid;
+      decl_to_simduid **slot = ns->htab->find_slot (&data, INSERT);
+      if (*slot == NULL)
+	{
+	  decl_to_simduid *p = XNEW (decl_to_simduid);
+	  *p = data;
+	  *slot = p;
+	}
+      else if ((*slot)->simduid != ns->simduid)
+	(*slot)->simduid = -1U;
+      *walk_subtrees = 0;
+    }
+  return NULL_TREE;
+}
+
+/* Find "omp simd array" temporaries and map them to corresponding
+   simduid.  */
+
+static void
+note_simd_array_uses (hash_table <decl_to_simduid> *htab)
+{
+  basic_block bb;
+  gimple_stmt_iterator gsi;
+  struct walk_stmt_info wi;
+  struct note_simd_array_uses_struct ns;
+
+  memset (&wi, 0, sizeof (wi));
+  wi.info = &ns;
+  ns.htab = htab;
+
+  FOR_EACH_BB (bb)
+    for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
+      {
+	gimple stmt = gsi_stmt (gsi);
+	if (!is_gimple_call (stmt) || !gimple_call_internal_p (stmt))
+	  continue;
+	switch (gimple_call_internal_fn (stmt))
+	  {
+	  case IFN_GOMP_SIMD_LANE:
+	  case IFN_GOMP_SIMD_VF:
+	  case IFN_GOMP_SIMD_LAST_LANE:
+	    break;
+	  default:
+	    continue;
+	  }
+	tree lhs = gimple_call_lhs (stmt);
+	if (lhs == NULL_TREE)
+	  continue;
+	imm_use_iterator use_iter;
+	gimple use_stmt;
+	ns.simduid = DECL_UID (SSA_NAME_VAR (gimple_call_arg (stmt, 0)));
+	FOR_EACH_IMM_USE_STMT (use_stmt, use_iter, lhs)
+	  if (!is_gimple_debug (use_stmt))
+	    walk_gimple_op (use_stmt, note_simd_array_uses_cb, &wi);
+      }
+}
 
 /* Function vectorize_loops.
 
@@ -86,12 +282,21 @@  vectorize_loops (void)
   unsigned int vect_loops_num;
   loop_iterator li;
   struct loop *loop;
+  hash_table <simduid_to_vf> simduid_to_vf_htab;
+  hash_table <decl_to_simduid> decl_to_simduid_htab;
 
   vect_loops_num = number_of_loops (cfun);
 
   /* Bail out if there are no loops.  */
   if (vect_loops_num <= 1)
-    return 0;
+    {
+      if (cfun->has_simduid_loops)
+	adjust_simduid_builtins (simduid_to_vf_htab);
+      return 0;
+    }
+
+  if (cfun->has_simduid_loops)
+    note_simd_array_uses (&decl_to_simduid_htab);
 
   init_stmt_vec_info_vec ();
 
@@ -101,7 +306,8 @@  vectorize_loops (void)
      than all previously defined loops.  This fact allows us to run
      only over initial loops skipping newly generated ones.  */
   FOR_EACH_LOOP (li, loop, 0)
-    if (optimize_loop_nest_for_speed_p (loop))
+    if ((flag_tree_vectorize && optimize_loop_nest_for_speed_p (loop))
+	|| loop->force_vect)
       {
 	loop_vec_info loop_vinfo;
 	vect_location = find_loop_location (loop);
@@ -122,6 +328,20 @@  vectorize_loops (void)
                            "Vectorized loop\n");
 	vect_transform_loop (loop_vinfo);
 	num_vectorized_loops++;
+	/* Now that the loop has been vectorized, allow it to be unrolled
+	   etc.  */
+	loop->force_vect = false;
+
+	if (loop->simduid)
+	  {
+	    simduid_to_vf *simduid_to_vf_data = XNEW (simduid_to_vf);
+	    if (!simduid_to_vf_htab.is_created ())
+	      simduid_to_vf_htab.create (15);
+	    simduid_to_vf_data->simduid = DECL_UID (loop->simduid);
+	    simduid_to_vf_data->vf = loop_vinfo->vectorization_factor;
+	    *simduid_to_vf_htab.find_slot (simduid_to_vf_data, INSERT)
+	      = simduid_to_vf_data;
+	  }
       }
 
   vect_location = UNKNOWN_LOC;
@@ -149,6 +369,40 @@  vectorize_loops (void)
 
   free_stmt_vec_info_vec ();
 
+  /* Fold IFN_GOMP_SIMD_{VF,LANE,LAST_LANE} builtins.  */
+  if (cfun->has_simduid_loops)
+    adjust_simduid_builtins (simduid_to_vf_htab);
+
+  /* Shrink any "omp array simd" temporary arrays to the
+     actual vectorization factors.  */
+  if (decl_to_simduid_htab.is_created ())
+    {
+      for (hash_table <decl_to_simduid>::iterator iter
+	   = decl_to_simduid_htab.begin ();
+	   iter != decl_to_simduid_htab.end (); ++iter)
+	if ((*iter).simduid != -1U)
+	  {
+	    tree decl = (*iter).decl;
+	    int vf = 1;
+	    if (simduid_to_vf_htab.is_created ())
+	      {
+		simduid_to_vf *p = NULL, data;
+		data.simduid = (*iter).simduid;
+		p = simduid_to_vf_htab.find (&data);
+		if (p)
+		  vf = p->vf;
+	      }
+	    tree atype
+	      = build_array_type_nelts (TREE_TYPE (TREE_TYPE (decl)), vf);
+	    TREE_TYPE (decl) = atype;
+	    relayout_decl (decl);
+	  }
+
+      decl_to_simduid_htab.dispose ();
+    }
+  if (simduid_to_vf_htab.is_created ())
+    simduid_to_vf_htab.dispose ();
+
   if (num_vectorized_loops > 0)
     {
       /* If we vectorized any loop only virtual SSA form needs to be updated.
diff --git a/gcc/tree-vectorizer.h b/gcc/tree-vectorizer.h
index 7c5dfe8..3570dc9 100644
--- a/gcc/tree-vectorizer.h
+++ b/gcc/tree-vectorizer.h
@@ -576,6 +576,9 @@  typedef struct _stmt_vec_info {
   /* For loads only, true if this is a gather load.  */
   bool gather_p;
   bool stride_load_p;
+
+  /* For both loads and stores.  */
+  bool simd_lane_access_p;
 } *stmt_vec_info;
 
 /* Access Functions.  */
@@ -591,6 +594,7 @@  typedef struct _stmt_vec_info {
 #define STMT_VINFO_DATA_REF(S)             (S)->data_ref_info
 #define STMT_VINFO_GATHER_P(S)		   (S)->gather_p
 #define STMT_VINFO_STRIDE_LOAD_P(S)	   (S)->stride_load_p
+#define STMT_VINFO_SIMD_LANE_ACCESS_P(S)   (S)->simd_lane_access_p
 
 #define STMT_VINFO_DR_BASE_ADDRESS(S)      (S)->dr_base_address
 #define STMT_VINFO_DR_INIT(S)              (S)->dr_init
diff --git a/gcc/tree.c b/gcc/tree.c
index ab11735..c1a0e93 100644
--- a/gcc/tree.c
+++ b/gcc/tree.c
@@ -236,6 +236,8 @@  unsigned const char omp_clause_num_ops[] =
   4, /* OMP_CLAUSE_REDUCTION  */
   1, /* OMP_CLAUSE_COPYIN  */
   1, /* OMP_CLAUSE_COPYPRIVATE  */
+  2, /* OMP_CLAUSE_LINEAR  */
+  1, /* OMP_CLAUSE_UNIFORM  */
   1, /* OMP_CLAUSE_IF  */
   1, /* OMP_CLAUSE_NUM_THREADS  */
   1, /* OMP_CLAUSE_SCHEDULE  */
@@ -245,7 +247,9 @@  unsigned const char omp_clause_num_ops[] =
   3, /* OMP_CLAUSE_COLLAPSE  */
   0, /* OMP_CLAUSE_UNTIED   */
   1, /* OMP_CLAUSE_FINAL  */
-  0  /* OMP_CLAUSE_MERGEABLE  */
+  0, /* OMP_CLAUSE_MERGEABLE  */
+  1, /* OMP_CLAUSE_SAFELEN  */
+  1, /* OMP_CLAUSE__SIMDUID_  */
 };
 
 const char * const omp_clause_code_name[] =
@@ -258,6 +262,8 @@  const char * const omp_clause_code_name[] =
   "reduction",
   "copyin",
   "copyprivate",
+  "linear",
+  "uniform",
   "if",
   "num_threads",
   "schedule",
@@ -267,7 +273,9 @@  const char * const omp_clause_code_name[] =
   "collapse",
   "untied",
   "final",
-  "mergeable"
+  "mergeable",
+  "safelen",
+  "_simduid_"
 };
 
 
@@ -11033,6 +11041,9 @@  walk_tree_1 (tree *tp, walk_tree_fn func, void *data,
 	case OMP_CLAUSE_IF:
 	case OMP_CLAUSE_NUM_THREADS:
 	case OMP_CLAUSE_SCHEDULE:
+	case OMP_CLAUSE_UNIFORM:
+	case OMP_CLAUSE_SAFELEN:
+	case OMP_CLAUSE__SIMDUID_:
 	  WALK_SUBTREE (OMP_CLAUSE_OPERAND (*tp, 0));
 	  /* FALLTHRU */
 
@@ -11056,6 +11067,11 @@  walk_tree_1 (tree *tp, walk_tree_fn func, void *data,
 	    WALK_SUBTREE_TAIL (OMP_CLAUSE_CHAIN (*tp));
 	  }
 
+	case OMP_CLAUSE_LINEAR:
+	  WALK_SUBTREE (OMP_CLAUSE_DECL (*tp));
+	  WALK_SUBTREE (OMP_CLAUSE_OPERAND (*tp, 1));
+	  WALK_SUBTREE_TAIL (OMP_CLAUSE_CHAIN (*tp));
+
 	case OMP_CLAUSE_REDUCTION:
 	  {
 	    int i;
diff --git a/gcc/tree.def b/gcc/tree.def
index da30074..f825aad 100644
--- a/gcc/tree.def
+++ b/gcc/tree.def
@@ -1030,6 +1030,10 @@  DEFTREECODE (OMP_TASK, "omp_task", tcc_statement, 2)
    unspecified by the standard.  */
 DEFTREECODE (OMP_FOR, "omp_for", tcc_statement, 6)
 
+/* OpenMP - #pragma omp simd [clause1 ... clauseN]
+   Operands like for OMP_FOR.  */
+DEFTREECODE (OMP_SIMD, "omp_simd", tcc_statement, 6)
+
 /* OpenMP - #pragma omp sections [clause1 ... clauseN]
    Operand 0: OMP_SECTIONS_BODY: Sections body.
    Operand 1: OMP_SECTIONS_CLAUSES: List of clauses.  */
diff --git a/gcc/tree.h b/gcc/tree.h
index b444517..b902e39 100644
--- a/gcc/tree.h
+++ b/gcc/tree.h
@@ -365,6 +365,12 @@  enum omp_clause_code
   /* OpenMP clause: copyprivate (variable_list).  */
   OMP_CLAUSE_COPYPRIVATE,
 
+  /* OpenMP clause: linear (variable-list[:linear-step]).  */
+  OMP_CLAUSE_LINEAR,
+
+  /* OpenMP clause: uniform (argument-list).  */
+  OMP_CLAUSE_UNIFORM,
+
   /* OpenMP clause: if (scalar-expression).  */
   OMP_CLAUSE_IF,
 
@@ -393,7 +399,13 @@  enum omp_clause_code
   OMP_CLAUSE_FINAL,
 
   /* OpenMP clause: mergeable.  */
-  OMP_CLAUSE_MERGEABLE
+  OMP_CLAUSE_MERGEABLE,
+
+  /* OpenMP clause: safelen (constant-integer-expression).  */
+  OMP_CLAUSE_SAFELEN,
+
+  /* Internally used only clause, holding SIMD uid.  */
+  OMP_CLAUSE__SIMDUID_
 };
 
 /* The definition of tree nodes fills the next several pages.  */
@@ -560,6 +572,9 @@  struct GTY(()) tree_base {
        OMP_CLAUSE_PRIVATE_DEBUG in
            OMP_CLAUSE_PRIVATE
 
+       OMP_CLAUSE_LINEAR_NO_COPYIN in
+	   OMP_CLAUSE_LINEAR
+
        TRANSACTION_EXPR_RELAXED in
 	   TRANSACTION_EXPR
 
@@ -580,6 +595,9 @@  struct GTY(()) tree_base {
        OMP_CLAUSE_PRIVATE_OUTER_REF in
 	   OMP_CLAUSE_PRIVATE
 
+       OMP_CLAUSE_LINEAR_NO_COPYOUT in
+	   OMP_CLAUSE_LINEAR
+
        TYPE_REF_IS_RVALUE in
 	   REFERENCE_TYPE
 
@@ -1800,7 +1818,7 @@  extern void protected_set_expr_location (tree, location_t);
 #define OMP_CLAUSE_DECL(NODE)      					\
   OMP_CLAUSE_OPERAND (OMP_CLAUSE_RANGE_CHECK (OMP_CLAUSE_CHECK (NODE),	\
 					      OMP_CLAUSE_PRIVATE,	\
-	                                      OMP_CLAUSE_COPYPRIVATE), 0)
+	                                      OMP_CLAUSE_UNIFORM), 0)
 #define OMP_CLAUSE_HAS_LOCATION(NODE) \
   (LOCATION_LOCUS ((OMP_CLAUSE_CHECK (NODE))->omp_clause.locus)		\
   != UNKNOWN_LOCATION)
@@ -1867,6 +1885,25 @@  extern void protected_set_expr_location (tree, location_t);
 #define OMP_CLAUSE_REDUCTION_PLACEHOLDER(NODE) \
   OMP_CLAUSE_OPERAND (OMP_CLAUSE_SUBCODE_CHECK (NODE, OMP_CLAUSE_REDUCTION), 3)
 
+/* True if a LINEAR clause doesn't need copy in.  True for iterator vars which
+   are always initialized inside of the loop construct, false otherwise.  */
+#define OMP_CLAUSE_LINEAR_NO_COPYIN(NODE) \
+  (OMP_CLAUSE_SUBCODE_CHECK (NODE, OMP_CLAUSE_LINEAR)->base.public_flag)
+
+/* True if a LINEAR clause doesn't need copy out.  True for iterator vars which
+   are declared inside of the simd construct.  */
+#define OMP_CLAUSE_LINEAR_NO_COPYOUT(NODE) \
+  TREE_PRIVATE (OMP_CLAUSE_SUBCODE_CHECK (NODE, OMP_CLAUSE_LINEAR))
+
+#define OMP_CLAUSE_LINEAR_STEP(NODE) \
+  OMP_CLAUSE_OPERAND (OMP_CLAUSE_SUBCODE_CHECK (NODE, OMP_CLAUSE_LINEAR), 1)
+
+#define OMP_CLAUSE_SAFELEN_EXPR(NODE) \
+  OMP_CLAUSE_OPERAND (OMP_CLAUSE_SUBCODE_CHECK (NODE, OMP_CLAUSE_SAFELEN), 0)
+
+#define OMP_CLAUSE__SIMDUID__DECL(NODE) \
+  OMP_CLAUSE_OPERAND (OMP_CLAUSE_SUBCODE_CHECK (NODE, OMP_CLAUSE__SIMDUID_), 0)
+
 enum omp_clause_schedule_kind
 {
   OMP_CLAUSE_SCHEDULE_STATIC,
@@ -4783,6 +4820,7 @@  extern tree build_translation_unit_decl (tree);
 extern tree build_block (tree, tree, tree, tree);
 extern tree build_empty_stmt (location_t);
 extern tree build_omp_clause (location_t, enum omp_clause_code);
+extern tree find_omp_clause (tree, enum omp_clause_code);
 
 extern tree build_vl_exp_stat (enum tree_code, int MEM_STAT_DECL);
 #define build_vl_exp(c,n) build_vl_exp_stat (c,n MEM_STAT_INFO)