8193130: Bad graph when unrolled loop bounds conflicts with range checks
Reviewed-by: kvn, thartmann
--- a/src/hotspot/share/opto/compile.cpp Thu Mar 22 16:39:02 2018 -0700
+++ b/src/hotspot/share/opto/compile.cpp Thu Mar 22 20:21:19 2018 -0700
@@ -397,13 +397,20 @@
remove_range_check_cast(cast);
}
}
- // Remove useless expensive node
+ // Remove useless expensive nodes
for (int i = C->expensive_count()-1; i >= 0; i--) {
Node* n = C->expensive_node(i);
if (!useful.member(n)) {
remove_expensive_node(n);
}
}
+ // Remove useless Opaque4 nodes
+ for (int i = opaque4_count() - 1; i >= 0; i--) {
+ Node* opaq = opaque4_node(i);
+ if (!useful.member(opaq)) {
+ remove_opaque4_node(opaq);
+ }
+ }
// clean up the late inline lists
remove_useless_late_inlines(&_string_late_inlines, useful);
remove_useless_late_inlines(&_boxing_late_inlines, useful);
@@ -1179,6 +1186,7 @@
_predicate_opaqs = new(comp_arena()) GrowableArray<Node*>(comp_arena(), 8, 0, NULL);
_expensive_nodes = new(comp_arena()) GrowableArray<Node*>(comp_arena(), 8, 0, NULL);
_range_check_casts = new(comp_arena()) GrowableArray<Node*>(comp_arena(), 8, 0, NULL);
+ _opaque4_nodes = new(comp_arena()) GrowableArray<Node*>(comp_arena(), 8, 0, NULL);
register_library_intrinsics();
}
@@ -1957,6 +1965,22 @@
assert(range_check_cast_count() == 0, "should be empty");
}
+void Compile::add_opaque4_node(Node* n) {
+ assert(n->Opcode() == Op_Opaque4, "Opaque4 only");
+ assert(!_opaque4_nodes->contains(n), "duplicate entry in Opaque4 list");
+ _opaque4_nodes->append(n);
+}
+
+// Remove all Opaque4 nodes.
+void Compile::remove_opaque4_nodes(PhaseIterGVN &igvn) {
+ for (int i = opaque4_count(); i > 0; i--) {
+ Node* opaq = opaque4_node(i-1);
+ assert(opaq->Opcode() == Op_Opaque4, "Opaque4 only");
+ igvn.replace_node(opaq, opaq->in(2));
+ }
+ assert(opaque4_count() == 0, "should be empty");
+}
+
// StringOpts and late inlining of string methods
void Compile::inline_string_calls(bool parse_time) {
{
@@ -2332,6 +2356,11 @@
}
}
+ if (opaque4_count() > 0) {
+ C->remove_opaque4_nodes(igvn);
+ igvn.optimize();
+ }
+
DEBUG_ONLY( _modified_nodes = NULL; )
} // (End scope of igvn; run destructor if necessary for asserts.)
@@ -3332,6 +3361,20 @@
}
break;
}
+ case Op_CmpUL: {
+ if (!Matcher::has_match_rule(Op_CmpUL)) {
+ // We don't support unsigned long comparisons. Set 'max_idx_expr'
+ // to max_julong if < 0 to make the signed comparison fail.
+ ConINode* sign_pos = new ConINode(TypeInt::make(BitsPerLong - 1));
+ Node* sign_bit_mask = new RShiftLNode(n->in(1), sign_pos);
+ Node* orl = new OrLNode(n->in(1), sign_bit_mask);
+ ConLNode* remove_sign_mask = new ConLNode(TypeLong::make(max_jlong));
+ Node* andl = new AndLNode(orl, remove_sign_mask);
+ Node* cmp = new CmpLNode(andl, n->in(2));
+ n->subsume_by(cmp, this);
+ }
+ break;
+ }
default:
assert( !n->is_Call(), "" );
assert( !n->is_Mem(), "" );
--- a/src/hotspot/share/opto/compile.hpp Thu Mar 22 16:39:02 2018 -0700
+++ b/src/hotspot/share/opto/compile.hpp Thu Mar 22 20:21:19 2018 -0700
@@ -416,6 +416,7 @@
GrowableArray<Node*>* _predicate_opaqs; // List of Opaque1 nodes for the loop predicates.
GrowableArray<Node*>* _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
GrowableArray<Node*>* _range_check_casts; // List of CastII nodes with a range check dependency
+ GrowableArray<Node*>* _opaque4_nodes; // List of Opaque4 nodes that have a default value
ConnectionGraph* _congraph;
#ifndef PRODUCT
IdealGraphPrinter* _printer;
@@ -810,6 +811,16 @@
// Remove all range check dependent CastIINodes.
void remove_range_check_casts(PhaseIterGVN &igvn);
+ void add_opaque4_node(Node* n);
+ void remove_opaque4_node(Node* n) {
+ if (_opaque4_nodes->contains(n)) {
+ _opaque4_nodes->remove(n);
+ }
+ }
+ Node* opaque4_node(int idx) const { return _opaque4_nodes->at(idx); }
+ int opaque4_count() const { return _opaque4_nodes->length(); }
+ void remove_opaque4_nodes(PhaseIterGVN &igvn);
+
// remove the opaque nodes that protect the predicates so that the unused checks and
// uncommon traps will be eliminated from the graph.
void cleanup_loop_predicates(PhaseIterGVN &igvn);
--- a/src/hotspot/share/opto/loopPredicate.cpp Thu Mar 22 16:39:02 2018 -0700
+++ b/src/hotspot/share/opto/loopPredicate.cpp Thu Mar 22 20:21:19 2018 -0700
@@ -632,7 +632,7 @@
int scale, Node* offset,
Node* init, Node* limit, jint stride,
Node* range, bool upper, bool &overflow) {
- jint con_limit = limit->is_Con() ? limit->get_int() : 0;
+ jint con_limit = (limit != NULL && limit->is_Con()) ? limit->get_int() : 0;
jint con_init = init->is_Con() ? init->get_int() : 0;
jint con_offset = offset->is_Con() ? offset->get_int() : 0;
@@ -751,26 +751,7 @@
// Integer expressions may overflow, do long comparison
range = new ConvI2LNode(range);
register_new_node(range, ctrl);
- if (!Matcher::has_match_rule(Op_CmpUL)) {
- // We don't support unsigned long comparisons. Set 'max_idx_expr'
- // to max_julong if < 0 to make the signed comparison fail.
- ConINode* sign_pos = _igvn.intcon(BitsPerLong - 1);
- set_ctrl(sign_pos, C->root());
- Node* sign_bit_mask = new RShiftLNode(max_idx_expr, sign_pos);
- register_new_node(sign_bit_mask, ctrl);
- // OR with sign bit to set all bits to 1 if negative (otherwise no change)
- max_idx_expr = new OrLNode(max_idx_expr, sign_bit_mask);
- register_new_node(max_idx_expr, ctrl);
- // AND with 0x7ff... to unset the sign bit
- ConLNode* remove_sign_mask = _igvn.longcon(max_jlong);
- set_ctrl(remove_sign_mask, C->root());
- max_idx_expr = new AndLNode(max_idx_expr, remove_sign_mask);
- register_new_node(max_idx_expr, ctrl);
-
- cmp = new CmpLNode(max_idx_expr, range);
- } else {
- cmp = new CmpULNode(max_idx_expr, range);
- }
+ cmp = new CmpULNode(max_idx_expr, range);
} else {
cmp = new CmpUNode(max_idx_expr, range);
}
@@ -785,6 +766,29 @@
return bol;
}
+// After pre/main/post loops are created, we'll put a copy of some
+// range checks between the pre and main loop to validate the initial
+// value of the induction variable for the main loop. Make a copy of
+// the predicates here with an opaque node as a place holder for the
+// initial value.
+ProjNode* PhaseIdealLoop::insert_skeleton_predicate(IfNode* iff, IdealLoopTree *loop,
+ ProjNode* proj, ProjNode *predicate_proj,
+ ProjNode* upper_bound_proj,
+ int scale, Node* offset,
+ Node* init, Node* limit, jint stride,
+ Node* rng, bool &overflow) {
+ assert(proj->_con && predicate_proj->_con, "not a range check?");
+ Node* opaque_init = new Opaque1Node(C, init);
+ register_new_node(opaque_init, upper_bound_proj);
+ BoolNode* bol = rc_predicate(loop, upper_bound_proj, scale, offset, opaque_init, limit, stride, rng, (stride > 0) != (scale > 0), overflow);
+ Node* opaque_bol = new Opaque4Node(C, bol, _igvn.intcon(1)); // This will go away once loop opts are over
+ register_new_node(opaque_bol, upper_bound_proj);
+ ProjNode* new_proj = create_new_if_for_predicate(predicate_proj, NULL, Deoptimization::Reason_predicate, overflow ? Op_If : iff->Opcode());
+ _igvn.replace_input_of(new_proj->in(0), 1, opaque_bol);
+ assert(opaque_init->outcnt() > 0, "should be used");
+ return new_proj;
+}
+
//------------------------------ loop_predication_impl--------------------------
// Insert loop predicates for null checks and range checks
bool PhaseIdealLoop::loop_predication_impl(IdealLoopTree *loop) {
@@ -980,6 +984,10 @@
// any dependent nodes onto the upper bound test.
new_predicate_proj = upper_bound_proj;
+ if (iff->is_RangeCheck()) {
+ new_predicate_proj = insert_skeleton_predicate(iff, loop, proj, predicate_proj, upper_bound_proj, scale, offset, init, limit, stride, rng, overflow);
+ }
+
#ifndef PRODUCT
if (TraceLoopOpts && !TraceLoopPredicate) {
tty->print("Predicate RC ");
--- a/src/hotspot/share/opto/loopTransform.cpp Thu Mar 22 16:39:02 2018 -0700
+++ b/src/hotspot/share/opto/loopTransform.cpp Thu Mar 22 20:21:19 2018 -0700
@@ -982,7 +982,7 @@
return n;
}
-bool PhaseIdealLoop::cast_incr_before_loop(Node* incr, Node* ctrl, Node* loop) {
+Node* PhaseIdealLoop::cast_incr_before_loop(Node* incr, Node* ctrl, Node* loop) {
Node* castii = new CastIINode(incr, TypeInt::INT, true);
castii->set_req(0, ctrl);
register_new_node(castii, ctrl);
@@ -990,10 +990,138 @@
Node* n = incr->fast_out(i);
if (n->is_Phi() && n->in(0) == loop) {
int nrep = n->replace_edge(incr, castii);
- return true;
+ return castii;
}
}
- return false;
+ return NULL;
+}
+
+// Make a copy of the skeleton range check predicates before the main
+// loop and set the initial value of loop as input. After unrolling,
+// the range of values for the induction variable in the main loop can
+// fall outside the allowed range of values by the array access (main
+// loop is never executed). When that happens, range check
+// CastII/ConvI2L nodes cause some data paths to die. For consistency,
+// the control paths must die too but the range checks were removed by
+// predication. The range checks that we add here guarantee that they
+// do.
+void PhaseIdealLoop::duplicate_predicates(CountedLoopNode* pre_head, Node* min_taken, Node* castii,
+ IdealLoopTree* outer_loop, LoopNode* outer_main_head,
+ uint dd_main_head) {
+ if (UseLoopPredicate) {
+ Node* entry = pre_head->in(LoopNode::EntryControl);
+ Node* predicate = NULL;
+ predicate = find_predicate_insertion_point(entry, Deoptimization::Reason_loop_limit_check);
+ if (predicate != NULL) {
+ entry = entry->in(0)->in(0);
+ }
+ predicate = find_predicate_insertion_point(entry, Deoptimization::Reason_predicate);
+ if (predicate != NULL) {
+ IfNode* iff = entry->in(0)->as_If();
+ ProjNode* uncommon_proj = iff->proj_out(1 - entry->as_Proj()->_con);
+ Node* rgn = uncommon_proj->unique_ctrl_out();
+ assert(rgn->is_Region() || rgn->is_Call(), "must be a region or call uct");
+ assert(iff->in(1)->in(1)->Opcode() == Op_Opaque1, "unexpected predicate shape");
+ entry = entry->in(0)->in(0);
+ Node* prev_proj = min_taken;
+ while (entry != NULL && entry->is_Proj() && entry->in(0)->is_If()) {
+ uncommon_proj = entry->in(0)->as_If()->proj_out(1 - entry->as_Proj()->_con);
+ if (uncommon_proj->unique_ctrl_out() != rgn)
+ break;
+ iff = entry->in(0)->as_If();
+ if (iff->in(1)->Opcode() == Op_Opaque4) {
+ Node_Stack to_clone(2);
+ to_clone.push(iff->in(1), 1);
+ uint current = C->unique();
+ Node* result = NULL;
+ // Look for the opaque node to replace with the init value
+ // and clone everything in between. We keep the Opaque4 node
+ // so the duplicated predicates are eliminated once loop
+ // opts are over: they are here only to keep the IR graph
+ // consistent.
+ do {
+ Node* n = to_clone.node();
+ uint i = to_clone.index();
+ Node* m = n->in(i);
+ int op = m->Opcode();
+ if (m->is_Bool() ||
+ m->is_Cmp() ||
+ op == Op_AndL ||
+ op == Op_OrL ||
+ op == Op_RShiftL ||
+ op == Op_LShiftL ||
+ op == Op_AddL ||
+ op == Op_AddI ||
+ op == Op_MulL ||
+ op == Op_MulI ||
+ op == Op_SubL ||
+ op == Op_SubI ||
+ op == Op_ConvI2L) {
+ to_clone.push(m, 1);
+ continue;
+ }
+ if (op == Op_Opaque1) {
+ if (n->_idx < current) {
+ n = n->clone();
+ }
+ n->set_req(i, castii);
+ register_new_node(n, min_taken);
+ to_clone.set_node(n);
+ }
+ for (;;) {
+ Node* cur = to_clone.node();
+ uint j = to_clone.index();
+ if (j+1 < cur->req()) {
+ to_clone.set_index(j+1);
+ break;
+ }
+ to_clone.pop();
+ if (to_clone.size() == 0) {
+ result = cur;
+ break;
+ }
+ Node* next = to_clone.node();
+ j = to_clone.index();
+ if (cur->_idx >= current) {
+ if (next->_idx < current) {
+ next = next->clone();
+ register_new_node(next, min_taken);
+ to_clone.set_node(next);
+ }
+ assert(next->in(j) != cur, "input should have been cloned");
+ next->set_req(j, cur);
+ }
+ }
+ } while (result == NULL);
+ assert(result->_idx >= current, "new node expected");
+
+ Node* proj = entry->clone();
+ Node* other_proj = uncommon_proj->clone();
+ Node* new_iff = iff->clone();
+ new_iff->set_req(1, result);
+ proj->set_req(0, new_iff);
+ other_proj->set_req(0, new_iff);
+ Node *frame = new ParmNode(C->start(), TypeFunc::FramePtr);
+ register_new_node(frame, C->start());
+ // It's impossible for the predicate to fail at runtime. Use
+ // an Halt node.
+ Node* halt = new HaltNode(other_proj, frame);
+ C->root()->add_req(halt);
+ new_iff->set_req(0, prev_proj);
+
+ register_control(new_iff, outer_loop->_parent, prev_proj);
+ register_control(proj, outer_loop->_parent, new_iff);
+ register_control(other_proj, _ltree_root, new_iff);
+ register_control(halt, _ltree_root, other_proj);
+
+ prev_proj = proj;
+ }
+ entry = entry->in(0)->in(0);
+ }
+ _igvn.replace_input_of(outer_main_head, LoopNode::EntryControl, prev_proj);
+ set_idom(outer_main_head, prev_proj, dd_main_head);
+ }
+ }
}
//------------------------------insert_pre_post_loops--------------------------
@@ -1137,8 +1265,9 @@
// dependencies.
// CastII for the main loop:
- bool inserted = cast_incr_before_loop( pre_incr, min_taken, main_head );
- assert(inserted, "no castII inserted");
+ Node* castii = cast_incr_before_loop( pre_incr, min_taken, main_head );
+ assert(castii != NULL, "no castII inserted");
+ duplicate_predicates(pre_head, min_taken, castii, outer_loop, outer_main_head, dd_main_head);
// Step B4: Shorten the pre-loop to run only 1 iteration (for now).
// RCE and alignment may change this later.
@@ -1403,8 +1532,8 @@
}
// CastII for the new post loop:
- bool inserted = cast_incr_before_loop(zer_opaq->in(1), zer_taken, post_head);
- assert(inserted, "no castII inserted");
+ Node* castii = cast_incr_before_loop(zer_opaq->in(1), zer_taken, post_head);
+ assert(castii != NULL, "no castII inserted");
return new_main_exit;
}
@@ -1467,7 +1596,7 @@
if (!is_canonical_loop_entry(loop_head)) {
return;
}
- opaq = ctrl->in(0)->in(1)->in(1)->in(2);
+ opaq = loop_head->skip_predicates()->in(0)->in(1)->in(1)->in(2);
// Zero-trip test uses an 'opaque' node which is not shared.
assert(opaq->outcnt() == 1 && opaq->in(1) == limit, "");
}
@@ -2034,6 +2163,34 @@
return false;
}
+// Same as PhaseIdealLoop::duplicate_predicates() but for range checks
+// eliminated by iteration splitting.
+Node* PhaseIdealLoop::add_range_check_predicate(IdealLoopTree* loop, CountedLoopNode* cl,
+ Node* predicate_proj, int scale_con, Node* offset,
+ Node* limit, jint stride_con) {
+ bool overflow = false;
+ BoolNode* bol = rc_predicate(loop, predicate_proj, scale_con, offset, cl->init_trip(), NULL, stride_con, limit, (stride_con > 0) != (scale_con > 0), overflow);
+ Node* opaque_bol = new Opaque4Node(C, bol, _igvn.intcon(1));
+ register_new_node(opaque_bol, predicate_proj);
+ IfNode* new_iff = NULL;
+ if (overflow) {
+ new_iff = new IfNode(predicate_proj, bol, PROB_MAX, COUNT_UNKNOWN);
+ } else {
+ new_iff = new RangeCheckNode(predicate_proj, bol, PROB_MAX, COUNT_UNKNOWN);
+ }
+ register_control(new_iff, loop->_parent, predicate_proj);
+ Node* iffalse = new IfFalseNode(new_iff);
+ register_control(iffalse, _ltree_root, new_iff);
+ ProjNode* iftrue = new IfTrueNode(new_iff);
+ register_control(iftrue, loop->_parent, new_iff);
+ Node *frame = new ParmNode(C->start(), TypeFunc::FramePtr);
+ register_new_node(frame, C->start());
+ Node* halt = new HaltNode(iffalse, frame);
+ register_control(halt, _ltree_root, iffalse);
+ C->root()->add_req(halt);
+ return iftrue;
+}
+
//------------------------------do_range_check---------------------------------
// Eliminate range-checks and other trip-counter vs loop-invariant tests.
int PhaseIdealLoop::do_range_check( IdealLoopTree *loop, Node_List &old_new ) {
@@ -2069,7 +2226,7 @@
}
// Need to find the main-loop zero-trip guard
- Node *ctrl = cl->skip_strip_mined()->in(LoopNode::EntryControl);
+ Node *ctrl = cl->skip_predicates();
Node *iffm = ctrl->in(0);
Node *opqzm = iffm->in(1)->in(1)->in(2);
assert(opqzm->in(1) == main_limit, "do not understand situation");
@@ -2124,6 +2281,8 @@
// the loop is in canonical form to multiversion.
closed_range_checks = 0;
+ Node* predicate_proj = cl->skip_strip_mined()->in(LoopNode::EntryControl);
+ assert(predicate_proj->is_Proj() && predicate_proj->in(0)->is_If(), "if projection only");
// Check loop body for tests of trip-counter plus loop-invariant vs loop-variant.
for( uint i = 0; i < loop->_body.size(); i++ ) {
Node *iff = loop->_body[i];
@@ -2168,7 +2327,7 @@
// 'limit' maybe pinned below the zero trip test (probably from a
// previous round of rce), in which case, it can't be used in the
// zero trip test expression which must occur before the zero test's if.
- if( limit_c == ctrl ) {
+ if (is_dominator(ctrl, limit_c)) {
continue; // Don't rce this check but continue looking for other candidates.
}
@@ -2186,7 +2345,7 @@
// As above for the 'limit', the 'offset' maybe pinned below the
// zero trip test.
- if( offset_c == ctrl ) {
+ if (is_dominator(ctrl, offset_c)) {
continue; // Don't rce this check but continue looking for other candidates.
}
#ifdef ASSERT
@@ -2209,6 +2368,7 @@
add_constraint( stride_con, scale_con, offset, zero, limit, pre_ctrl, &pre_limit, &main_limit );
// (0-offset)/scale could be outside of loop iterations range.
conditional_rc = true;
+ predicate_proj = add_range_check_predicate(loop, cl, predicate_proj, scale_con, offset, limit, stride_con);
} else {
if (PrintOpto) {
tty->print_cr("missed RCE opportunity");
@@ -2278,6 +2438,10 @@
} // End of is IF
}
+ if (predicate_proj != cl->skip_strip_mined()->in(LoopNode::EntryControl)) {
+ _igvn.replace_input_of(cl->skip_strip_mined(), LoopNode::EntryControl, predicate_proj);
+ set_idom(cl->skip_strip_mined(), predicate_proj, dom_depth(cl->skip_strip_mined()));
+ }
// Update loop limits
if (conditional_rc) {
@@ -2540,7 +2704,7 @@
#ifdef ASSERT
static CountedLoopNode* locate_pre_from_main(CountedLoopNode *cl) {
- Node *ctrl = cl->skip_strip_mined()->in(LoopNode::EntryControl);
+ Node *ctrl = cl->skip_predicates();
assert(ctrl->Opcode() == Op_IfTrue || ctrl->Opcode() == Op_IfFalse, "");
Node *iffm = ctrl->in(0);
assert(iffm->Opcode() == Op_If, "");
@@ -2579,7 +2743,7 @@
}
assert(locate_pre_from_main(main_head) == cl, "bad main loop");
- Node* main_iff = main_head->skip_strip_mined()->in(LoopNode::EntryControl)->in(0);
+ Node* main_iff = main_head->skip_predicates()->in(0);
// Remove the Opaque1Node of the pre loop and make it execute all iterations
phase->_igvn.replace_input_of(pre_cmp, 2, pre_cmp->in(2)->in(2));
@@ -2640,7 +2804,7 @@
}
if (needs_guard) {
// Check for an obvious zero trip guard.
- Node* inctrl = PhaseIdealLoop::skip_loop_predicates(cl->skip_strip_mined()->in(LoopNode::EntryControl));
+ Node* inctrl = PhaseIdealLoop::skip_loop_predicates(cl->skip_predicates());
if (inctrl->Opcode() == Op_IfTrue || inctrl->Opcode() == Op_IfFalse) {
bool maybe_swapped = (inctrl->Opcode() == Op_IfFalse);
// The test should look like just the backedge of a CountedLoop
--- a/src/hotspot/share/opto/loopnode.cpp Thu Mar 22 16:39:02 2018 -0700
+++ b/src/hotspot/share/opto/loopnode.cpp Thu Mar 22 20:21:19 2018 -0700
@@ -1158,9 +1158,9 @@
return NULL;
}
-LoopNode* CountedLoopNode::skip_strip_mined(int expect_opaq) {
+LoopNode* CountedLoopNode::skip_strip_mined(int expect_skeleton) {
if (is_strip_mined()) {
- verify_strip_mined(expect_opaq);
+ verify_strip_mined(expect_skeleton);
return in(EntryControl)->as_Loop();
}
return this;
@@ -1252,6 +1252,20 @@
return l->outer_safepoint();
}
+Node* CountedLoopNode::skip_predicates() {
+ if (is_main_loop()) {
+ Node* ctrl = skip_strip_mined()->in(LoopNode::EntryControl);
+ while (ctrl != NULL && ctrl->is_Proj() && ctrl->in(0)->is_If() &&
+ ctrl->in(0)->as_If()->proj_out(1-ctrl->as_Proj()->_con)->outcnt() == 1 &&
+ ctrl->in(0)->as_If()->proj_out(1-ctrl->as_Proj()->_con)->unique_out()->Opcode() == Op_Halt) {
+ ctrl = ctrl->in(0)->in(0);
+ }
+
+ return ctrl;
+ }
+ return in(LoopNode::EntryControl);
+}
+
void OuterStripMinedLoopNode::adjust_strip_mined_loop(PhaseIterGVN* igvn) {
// Look for the outer & inner strip mined loop, reduce number of
// iterations of the inner loop, set exit condition of outer loop,
@@ -3770,7 +3784,8 @@
if (!cl->is_main_loop() && !cl->is_post_loop()) {
return false;
}
- Node* ctrl = cl->skip_strip_mined()->in(LoopNode::EntryControl);
+ Node* ctrl = cl->skip_predicates();
+
if (ctrl == NULL || (!ctrl->is_IfTrue() && !ctrl->is_IfFalse())) {
return false;
}
--- a/src/hotspot/share/opto/loopnode.hpp Thu Mar 22 16:39:02 2018 -0700
+++ b/src/hotspot/share/opto/loopnode.hpp Thu Mar 22 20:21:19 2018 -0700
@@ -138,7 +138,7 @@
#endif
void verify_strip_mined(int expect_skeleton) const;
- virtual LoopNode* skip_strip_mined(int expect_opaq = 1) { return this; }
+ virtual LoopNode* skip_strip_mined(int expect_skeleton = 1) { return this; }
virtual IfTrueNode* outer_loop_tail() const { ShouldNotReachHere(); return NULL; }
virtual OuterStripMinedLoopEndNode* outer_loop_end() const { ShouldNotReachHere(); return NULL; }
virtual IfFalseNode* outer_loop_exit() const { ShouldNotReachHere(); return NULL; }
@@ -298,6 +298,11 @@
virtual IfFalseNode* outer_loop_exit() const;
virtual SafePointNode* outer_safepoint() const;
+ // If this is a main loop in a pre/main/post loop nest, walk over
+ // the predicates that were inserted by
+ // duplicate_predicates()/add_range_check_predicate()
+ Node* skip_predicates();
+
#ifndef PRODUCT
virtual void dump_spec(outputStream *st) const;
#endif
@@ -724,7 +729,10 @@
return ctrl;
}
- bool cast_incr_before_loop(Node* incr, Node* ctrl, Node* loop);
+ Node* cast_incr_before_loop(Node* incr, Node* ctrl, Node* loop);
+ void duplicate_predicates(CountedLoopNode* pre_head, Node *min_taken, Node* castii,
+ IdealLoopTree* outer_loop, LoopNode* outer_main_head,
+ uint dd_main_head);
public:
@@ -1067,6 +1075,15 @@
// Implementation of the loop predication to promote checks outside the loop
bool loop_predication_impl(IdealLoopTree *loop);
+ ProjNode* insert_skeleton_predicate(IfNode* iff, IdealLoopTree *loop,
+ ProjNode* proj, ProjNode *predicate_proj,
+ ProjNode* upper_bound_proj,
+ int scale, Node* offset,
+ Node* init, Node* limit, jint stride,
+ Node* rng, bool& overflow);
+ Node* add_range_check_predicate(IdealLoopTree* loop, CountedLoopNode* cl,
+ Node* predicate_proj, int scale_con, Node* offset,
+ Node* limit, jint stride_con);
// Helper function to collect predicate for eliminating the useless ones
void collect_potentially_useful_predicates(IdealLoopTree *loop, Unique_Node_List &predicate_opaque1);
--- a/src/hotspot/share/opto/macro.cpp Thu Mar 22 16:39:02 2018 -0700
+++ b/src/hotspot/share/opto/macro.cpp Thu Mar 22 20:21:19 2018 -0700
@@ -2668,8 +2668,7 @@
assert(n->Opcode() == Op_LoopLimit ||
n->Opcode() == Op_Opaque1 ||
n->Opcode() == Op_Opaque2 ||
- n->Opcode() == Op_Opaque3 ||
- n->Opcode() == Op_Opaque4, "unknown node type in macro list");
+ n->Opcode() == Op_Opaque3, "unknown node type in macro list");
}
assert(success == (C->macro_count() < old_macro_count), "elimination reduces macro count");
progress = progress || success;
@@ -2734,9 +2733,6 @@
_igvn.replace_node(n, repl);
success = true;
#endif
- } else if (n->Opcode() == Op_Opaque4) {
- _igvn.replace_node(n, n->in(2));
- success = true;
} else if (n->Opcode() == Op_OuterStripMinedLoop) {
n->as_OuterStripMinedLoop()->adjust_strip_mined_loop(&_igvn);
C->remove_macro_node(n);
--- a/src/hotspot/share/opto/node.cpp Thu Mar 22 16:39:02 2018 -0700
+++ b/src/hotspot/share/opto/node.cpp Thu Mar 22 20:21:19 2018 -0700
@@ -504,6 +504,9 @@
if (cast != NULL && cast->has_range_check()) {
C->add_range_check_cast(cast);
}
+ if (n->Opcode() == Op_Opaque4) {
+ C->add_opaque4_node(n);
+ }
n->set_idx(C->next_unique()); // Get new unique index as well
debug_only( n->verify_construction() );
@@ -612,6 +615,9 @@
if (cast != NULL && cast->has_range_check()) {
compile->remove_range_check_cast(cast);
}
+ if (Opcode() == Op_Opaque4) {
+ compile->remove_opaque4_node(this);
+ }
if (is_SafePoint()) {
as_SafePoint()->delete_replaced_nodes();
@@ -1352,6 +1358,9 @@
if (cast != NULL && cast->has_range_check()) {
igvn->C->remove_range_check_cast(cast);
}
+ if (dead->Opcode() == Op_Opaque4) {
+ igvn->C->remove_range_check_cast(dead);
+ }
igvn->C->record_dead_node(dead->_idx);
// Kill all inputs to the dead guy
for (uint i=0; i < dead->req(); i++) {
--- a/src/hotspot/share/opto/opaquenode.cpp Thu Mar 22 16:39:02 2018 -0700
+++ b/src/hotspot/share/opto/opaquenode.cpp Thu Mar 22 20:21:19 2018 -0700
@@ -60,10 +60,6 @@
return (&n == this); // Always fail except on self
}
-Node* Opaque4Node::Identity(PhaseGVN* phase) {
- return phase->C->major_progress() ? this : in(2);
-}
-
const Type* Opaque4Node::Value(PhaseGVN* phase) const {
return phase->type(in(1));
}
--- a/src/hotspot/share/opto/opaquenode.hpp Thu Mar 22 16:39:02 2018 -0700
+++ b/src/hotspot/share/opto/opaquenode.hpp Thu Mar 22 20:21:19 2018 -0700
@@ -35,14 +35,14 @@
virtual uint hash() const ; // { return NO_HASH; }
virtual uint cmp( const Node &n ) const;
public:
- Opaque1Node( Compile* C, Node *n ) : Node(0,n) {
+ Opaque1Node(Compile* C, Node *n) : Node(NULL, n) {
// Put it on the Macro nodes list to removed during macro nodes expansion.
init_flags(Flag_is_macro);
C->add_macro_node(this);
}
// Special version for the pre-loop to hold the original loop limit
// which is consumed by range check elimination.
- Opaque1Node( Compile* C, Node *n, Node* orig_limit ) : Node(0,n,orig_limit) {
+ Opaque1Node(Compile* C, Node *n, Node* orig_limit) : Node(NULL, n, orig_limit) {
// Put it on the Macro nodes list to removed during macro nodes expansion.
init_flags(Flag_is_macro);
C->add_macro_node(this);
@@ -87,25 +87,23 @@
bool rtm_opt() const { return (_opt == RTM_OPT); }
};
-// Used by GraphKit::must_be_not_null(): input 1 is a check that we
-// know implicitly is always true or false but the compiler has no way
-// to prove. If during optimizations, that check becomes true or
-// false, the Opaque4 node is replaced by that constant true or
-// false. Input 2 is the constant value we know the test takes. After
-// loop optimizations, we replace input 1 by input 2 so the control
-// that depends on that test can be removed and there's no overhead at
-// runtime.
+// Input 1 is a check that we know implicitly is always true or false
+// but the compiler has no way to prove. If during optimizations, that
+// check becomes true or false, the Opaque4 node is replaced by that
+// constant true or false. Input 2 is the constant value we know the
+// test takes. After loop optimizations, we replace input 1 by input 2
+// so the control that depends on that test can be removed and there's
+// no overhead at runtime. Used for instance by
+// GraphKit::must_be_not_null().
class Opaque4Node : public Node {
public:
- Opaque4Node(Compile* C, Node *tst, Node* final_tst) : Node(0, tst, final_tst) {
- // Put it on the Macro nodes list to removed during macro nodes expansion.
- init_flags(Flag_is_macro);
- C->add_macro_node(this);
+ Opaque4Node(Compile* C, Node *tst, Node* final_tst) : Node(NULL, tst, final_tst) {
+ // Put it on the Opaque4 nodes list to be removed after all optimizations
+ C->add_opaque4_node(this);
}
virtual int Opcode() const;
virtual const Type *bottom_type() const { return TypeInt::BOOL; }
virtual const Type* Value(PhaseGVN* phase) const;
- virtual Node* Identity(PhaseGVN* phase);
};
--- a/src/hotspot/share/opto/phaseX.cpp Thu Mar 22 16:39:02 2018 -0700
+++ b/src/hotspot/share/opto/phaseX.cpp Thu Mar 22 20:21:19 2018 -0700
@@ -1421,6 +1421,9 @@
if (cast != NULL && cast->has_range_check()) {
C->remove_range_check_cast(cast);
}
+ if (dead->Opcode() == Op_Opaque4) {
+ C->remove_opaque4_node(dead);
+ }
}
} // while (_stack.is_nonempty())
}
--- a/src/hotspot/share/opto/superword.cpp Thu Mar 22 16:39:02 2018 -0700
+++ b/src/hotspot/share/opto/superword.cpp Thu Mar 22 20:21:19 2018 -0700
@@ -3328,7 +3328,7 @@
return NULL;
}
- Node* p_f = cl->skip_strip_mined()->in(LoopNode::EntryControl)->in(0)->in(0);
+ Node* p_f = cl->skip_predicates()->in(0)->in(0);
if (!p_f->is_IfFalse()) return NULL;
if (!p_f->in(0)->is_CountedLoopEnd()) return NULL;
CountedLoopEndNode* pre_end = p_f->in(0)->as_CountedLoopEnd();
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/test/hotspot/jtreg/compiler/loopopts/IterationSplitPredicateInconsistency.java Thu Mar 22 20:21:19 2018 -0700
@@ -0,0 +1,140 @@
+/*
+ * Copyright (c) 2018, Red Hat, Inc. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ */
+
+/**
+ * @test
+ * @bug 8193130
+ * @summary Bad graph when unrolled loop bounds conflicts with range checks
+ *
+ * @run main/othervm IterationSplitPredicateInconsistency
+ * @run main/othervm -XX:-UseLoopPredicate IterationSplitPredicateInconsistency
+ *
+ */
+
+public class IterationSplitPredicateInconsistency {
+ static volatile int barrier;
+
+ // Pollute profile so loop appears to run for a large number of iterations
+ static boolean test1_helper(int start, int stop, double[] array1, double[] array2, int exit) {
+ for (int i = start; i < stop; i++) {
+ array1[i] = array2[i];
+ if (i == exit) {
+ return true;
+ }
+ barrier = 0x42;
+ }
+ return false;
+ }
+
+ static double[] test1(int start, double[] array2, int exit) {
+ double[] array1 = new double[10];
+ // Predication moves range checks out of loop and
+ // pre/main/post loops are created. The main loop is unrolled
+ // several times to the point where it's never executed but
+ // compiler can't tell from the loop bounds alone. The lower
+ // bound of the loop is negative and would cause range checks
+ // (that were removed from the loop body) to fail.
+ if (test1_helper(start, 5, array1, array2, exit)) {
+ return null;
+ }
+ return array1;
+ }
+
+ // Same as above with other combinations of increasing/decreasing
+ // loops, positive/negative stride
+ static boolean test2_helper(int start, int stop, double[] array1, double[] array2, int exit) {
+ for (int i = start-1; i >= stop; i--) {
+ array1[i] = array2[i];
+ if (i == exit) {
+ return true;
+ }
+ barrier = 0x42;
+ }
+ return false;
+ }
+
+ static double[] test2(int start, double[] array2, int exit) {
+ double[] array1 = new double[10];
+ if (test2_helper(start, 0, array1, array2, exit)) {
+ return null;
+ }
+ return array1;
+ }
+
+ static boolean test3_helper(int start, int stop, double[] array1, double[] array2, int exit) {
+ for (int i = start; i < stop; i++) {
+ array1[stop-i-1] = array2[stop-i-1];
+ if (i == exit) {
+ return true;
+ }
+ barrier = 0x42;
+ }
+ return false;
+ }
+
+ static double[] test3(int start, double[] array2, int exit) {
+ double[] array1 = new double[5];
+ if (test3_helper(start, 5, array1, array2, exit)) {
+ return null;
+ }
+ return array1;
+ }
+
+ static boolean test4_helper(int start, int stop, int from, double[] array1, double[] array2, int exit) {
+ for (int i = start-1; i >= stop; i--) {
+ array1[from-i-1] = array2[from-i-1];
+ if (i == exit) {
+ return true;
+ }
+ barrier = 0x42;
+ }
+ return false;
+ }
+
+ static double[] test4(int start, double[] array2, int exit) {
+ double[] array1 = new double[5];
+ if (test4_helper(start, 0, 5, array1, array2, exit)) {
+ return null;
+ }
+ return array1;
+ }
+
+ public static void main(String[] args) {
+ double[] array2 = new double[10];
+ double[] array3 = new double[1000];
+ for (int i = 0; i < 20_000; i++) {
+ test1_helper(0, 1000, array3, array3, 998);
+ test1(0, array2, 999);
+ test1(0, array2, 4);
+ test2_helper(1000, 0, array3, array3, 1);
+ test2(5, array2, 999);
+ test2(5, array2, 1);
+ test3_helper(0, 1000, array3, array3, 998);
+ test3(0, array2, 999);
+ test3(0, array2, 4);
+ test4_helper(1000, 0, 1000, array3, array3, 1);
+ test4(5, array2, 999);
+ test4(5, array2, 1);
+ }
+ }
+}