/* * Copyright (c) 2000, 2022, Oracle and/or its affiliates. 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. *
*/
//------------------------------is_loop_exit----------------------------------- // Given an IfNode, return the loop-exiting projection or NULL if both // arms remain in the loop.
Node *IdealLoopTree::is_loop_exit(Node *iff) const { if (iff->outcnt() != 2) return NULL; // Ignore partially dead tests
PhaseIdealLoop *phase = _phase; // Test is an IfNode, has 2 projections. If BOTH are in the loop // we need loop unswitching instead of peeling. if (!is_member(phase->get_loop(iff->raw_out(0)))) return iff->raw_out(0); if (!is_member(phase->get_loop(iff->raw_out(1)))) return iff->raw_out(1); return NULL;
}
//------------------------------record_for_igvn---------------------------- // Put loop body on igvn work list void IdealLoopTree::record_for_igvn() { for (uint i = 0; i < _body.size(); i++) {
Node *n = _body.at(i);
_phase->_igvn._worklist.push(n);
} // put body of outer strip mined loop on igvn work list as well if (_head->is_CountedLoop() && _head->as_Loop()->is_strip_mined()) {
CountedLoopNode* l = _head->as_CountedLoop();
Node* outer_loop = l->outer_loop();
assert(outer_loop != NULL, "missing piece of strip mined loop");
_phase->_igvn._worklist.push(outer_loop);
Node* outer_loop_tail = l->outer_loop_tail();
assert(outer_loop_tail != NULL, "missing piece of strip mined loop");
_phase->_igvn._worklist.push(outer_loop_tail);
Node* outer_loop_end = l->outer_loop_end();
assert(outer_loop_end != NULL, "missing piece of strip mined loop");
_phase->_igvn._worklist.push(outer_loop_end);
Node* outer_safepoint = l->outer_safepoint();
assert(outer_safepoint != NULL, "missing piece of strip mined loop");
_phase->_igvn._worklist.push(outer_safepoint);
Node* cle_out = _head->as_CountedLoop()->loopexit()->proj_out(false);
assert(cle_out != NULL, "missing piece of strip mined loop");
_phase->_igvn._worklist.push(cle_out);
}
}
//------------------------------compute_exact_trip_count----------------------- // Compute loop trip count if possible. Do not recalculate trip count for // split loops (pre-main-post) which have their limits and inits behind Opaque node. void IdealLoopTree::compute_trip_count(PhaseIdealLoop* phase) { if (!_head->as_Loop()->is_valid_counted_loop(T_INT)) { return;
}
CountedLoopNode* cl = _head->as_CountedLoop(); // Trip count may become nonexact for iteration split loops since // RCE modifies limits. Note, _trip_count value is not reset since // it is used to limit unrolling of main loop.
cl->set_nonexact_trip_count();
// Loop's test should be part of loop. if (!phase->is_member(this, phase->get_ctrl(cl->loopexit()->in(CountedLoopEndNode::TestValue)))) return; // Infinite loop
// Now compute a loop exit count float loop_exit_cnt = 0.0f; if (_child == NULL) { for (uint i = 0; i < _body.size(); i++) {
Node *n = _body[i];
loop_exit_cnt += compute_profile_trip_cnt_helper(n);
}
} else {
ResourceMark rm;
Unique_Node_List wq;
wq.push(back); for (uint i = 0; i < wq.size(); i++) {
Node *n = wq.at(i);
assert(n->is_CFG(), "only control nodes"); if (n != head) { if (n->is_Region()) { for (uint j = 1; j < n->req(); j++) {
wq.push(n->in(j));
}
} else {
loop_exit_cnt += compute_profile_trip_cnt_helper(n);
wq.push(n->in(0));
}
}
}
} if (loop_exit_cnt > 0.0f) {
trip_cnt = (loop_back_cnt + loop_exit_cnt) / loop_exit_cnt;
} else { // No exit count so use
trip_cnt = loop_back_cnt;
}
} else {
head->mark_profile_trip_failed();
} #ifndef PRODUCT if (TraceProfileTripCount) {
tty->print_cr("compute_profile_trip_cnt lp: %d cnt: %f\n", head->_idx, trip_cnt);
} #endif
head->set_profile_trip_cnt(trip_cnt);
}
//---------------------find_invariant----------------------------- // Return nonzero index of invariant operand for an associative // binary operation of (nonconstant) invariant and variant values. // Helper for reassociate_invariants. int IdealLoopTree::find_invariant(Node* n, PhaseIdealLoop *phase) { bool in1_invar = this->is_invariant(n->in(1)); bool in2_invar = this->is_invariant(n->in(2)); if (in1_invar && !in2_invar) return 1; if (!in1_invar && in2_invar) return 2; return 0;
}
//---------------------is_associative----------------------------- // Return TRUE if "n" is an associative binary node. If "base" is // not NULL, "n" must be re-associative with it. bool IdealLoopTree::is_associative(Node* n, Node* base) { int op = n->Opcode(); if (base != NULL) {
assert(is_associative(base), "Base node should be associative"); int base_op = base->Opcode(); if (base_op == Op_AddI || base_op == Op_SubI) { return op == Op_AddI || op == Op_SubI;
} if (base_op == Op_AddL || base_op == Op_SubL) { return op == Op_AddL || op == Op_SubL;
} return op == base_op;
} else { // Integer "add/sub/mul/and/or/xor" operations are associative. return op == Op_AddI || op == Op_AddL
|| op == Op_SubI || op == Op_SubL
|| op == Op_MulI || op == Op_MulL
|| op == Op_AndI || op == Op_AndL
|| op == Op_OrI || op == Op_OrL
|| op == Op_XorI || op == Op_XorL;
}
}
bool is_int = n1->bottom_type()->isa_int() != NULL;
Node* inv1_c = phase->get_ctrl(inv1);
Node* n_inv1; if (neg_inv1) {
Node* zero; if (is_int) {
zero = phase->_igvn.intcon(0);
n_inv1 = new SubINode(zero, inv1);
} else {
zero = phase->_igvn.longcon(0L);
n_inv1 = new SubLNode(zero, inv1);
}
phase->set_ctrl(zero, phase->C->root());
phase->register_new_node(n_inv1, inv1_c);
} else {
n_inv1 = inv1;
}
Node* inv; if (is_int) { if (neg_inv2) {
inv = new SubINode(n_inv1, inv2);
} else {
inv = new AddINode(n_inv1, inv2);
}
phase->register_new_node(inv, phase->get_early_ctrl(inv)); if (neg_x) { returnnew SubINode(inv, x);
} else { returnnew AddINode(x, inv);
}
} else { if (neg_inv2) {
inv = new SubLNode(n_inv1, inv2);
} else {
inv = new AddLNode(n_inv1, inv2);
}
phase->register_new_node(inv, phase->get_early_ctrl(inv)); if (neg_x) { returnnew SubLNode(inv, x);
} else { returnnew AddLNode(x, inv);
}
}
}
//---------------------reassociate----------------------------- // Reassociate invariant binary expressions with add/sub/mul/ // and/or/xor operators. // For add/sub expressions: see "reassociate_add_sub" // // For mul/and/or/xor expressions: // // inv1 op (x op inv2) => (inv1 op inv2) op x //
Node* IdealLoopTree::reassociate(Node* n1, PhaseIdealLoop *phase) { if (!is_associative(n1) || n1->outcnt() == 0) return NULL; if (is_invariant(n1)) return NULL; // Don't mess with add of constant (igvn moves them to expression tree root.) if (n1->is_Add() && n1->in(2)->is_Con()) return NULL;
int inv1_idx = find_invariant(n1, phase); if (!inv1_idx) return NULL;
Node* n2 = n1->in(3 - inv1_idx); if (!is_associative(n2, n1)) return NULL; int inv2_idx = find_invariant(n2, phase); if (!inv2_idx) return NULL;
if (!phase->may_require_nodes(10, 10)) return NULL;
Node* result = NULL; switch (n1->Opcode()) { case Op_AddI: case Op_AddL: case Op_SubI: case Op_SubL:
result = reassociate_add_sub(n1, inv1_idx, inv2_idx, phase); break; case Op_MulI: case Op_MulL: case Op_AndI: case Op_AndL: case Op_OrI: case Op_OrL: case Op_XorI: case Op_XorL: {
Node* inv1 = n1->in(inv1_idx);
Node* inv2 = n2->in(inv2_idx);
Node* x = n2->in(3 - inv2_idx);
Node* inv = n2->clone_with_data_edge(inv1, inv2);
phase->register_new_node(inv, phase->get_early_ctrl(inv));
result = n1->clone_with_data_edge(x, inv); break;
} default:
ShouldNotReachHere();
}
//---------------------reassociate_invariants----------------------------- // Reassociate invariant expressions: void IdealLoopTree::reassociate_invariants(PhaseIdealLoop *phase) { for (int i = _body.size() - 1; i >= 0; i--) {
Node *n = _body.at(i); for (int j = 0; j < 5; j++) {
Node* nn = reassociate(n, phase); if (nn == NULL) break;
n = nn; // again
}
}
}
//------------------------------policy_peeling--------------------------------- // Return TRUE if the loop should be peeled, otherwise return FALSE. Peeling // is applicable if we can make a loop-invariant test (usually a null-check) // execute before we enter the loop. When TRUE, the estimated node budget is // also requested. bool IdealLoopTree::policy_peeling(PhaseIdealLoop *phase) {
uint estimate = estimate_peeling(phase);
// Perform actual policy and size estimate for the loop peeling transform, and // return the estimated loop size if peeling is applicable, otherwise return // zero. No node budget is allocated.
uint IdealLoopTree::estimate_peeling(PhaseIdealLoop *phase) {
// If nodes are depleted, some transform has miscalculated its needs.
assert(!phase->exceeding_node_budget(), "sanity");
// Peeling does loop cloning which can result in O(N^2) node construction. if (_body.size() > 255) { return 0; // Suppress too large body size.
} // Optimistic estimate that approximates loop body complexity via data and // control flow fan-out (instead of using the more pessimistic: BodySize^2).
uint estimate = est_loop_clone_sz(2);
if (phase->exceeding_node_budget(estimate)) { return 0; // Too large to safely clone.
}
// Check for vectorized loops, any peeling done was already applied. if (_head->is_CountedLoop()) {
CountedLoopNode* cl = _head->as_CountedLoop(); if (cl->is_unroll_only() || cl->trip_count() == 1) { return 0;
}
}
Node* test = tail();
while (test != _head) { // Scan till run off top of loop if (test->is_If()) { // Test?
Node *ctrl = phase->get_ctrl(test->in(1)); if (ctrl->is_top()) { return 0; // Found dead test on live IF? No peeling!
} // Standard IF only has one input value to check for loop invariance.
assert(test->Opcode() == Op_If ||
test->Opcode() == Op_CountedLoopEnd ||
test->Opcode() == Op_LongCountedLoopEnd ||
test->Opcode() == Op_RangeCheck, "Check this code when new subtype is added"); // Condition is not a member of this loop? if (!is_member(phase->get_loop(ctrl)) && is_loop_exit(test)) { return estimate; // Found reason to peel!
}
} // Walk up dominators to loop _head looking for test which is executed on // every path through the loop.
test = phase->idom(test);
} return 0;
}
//------------------------------peeled_dom_test_elim--------------------------- // If we got the effect of peeling, either by actually peeling or by making // a pre-loop which must execute at least once, we can remove all // loop-invariant dominated tests in the main body. void PhaseIdealLoop::peeled_dom_test_elim(IdealLoopTree* loop, Node_List& old_new) { bool progress = true; while (progress) {
progress = false; // Reset for next iteration
Node* prev = loop->_head->in(LoopNode::LoopBackControl); // loop->tail();
Node* test = prev->in(0); while (test != loop->_head) { // Scan till run off top of loop int p_op = prev->Opcode();
assert(test != NULL, "test cannot be NULL");
Node* test_cond = NULL; if ((p_op == Op_IfFalse || p_op == Op_IfTrue) && test->is_If()) {
test_cond = test->in(1);
} if (test_cond != NULL && // Test?
!test_cond->is_Con() && // And not already obvious? // And condition is not a member of this loop?
!loop->is_member(get_loop(get_ctrl(test_cond)))) { // Walk loop body looking for instances of this test for (uint i = 0; i < loop->_body.size(); i++) {
Node* n = loop->_body.at(i); // Check against cached test condition because dominated_by() // replaces the test condition with a constant. if (n->is_If() && n->in(1) == test_cond) { // IfNode was dominated by version in peeled loop body
progress = true;
dominated_by(old_new[prev->_idx]->as_IfProj(), n->as_If());
}
}
}
prev = test;
test = idom(test);
} // End of scan tests in loop
} // End of while (progress)
}
//------------------------------do_peeling------------------------------------- // Peel the first iteration of the given loop. // Step 1: Clone the loop body. The clone becomes the peeled iteration. // The pre-loop illegally has 2 control users (old & new loops). // Step 2: Make the old-loop fall-in edges point to the peeled iteration. // Do this by making the old-loop fall-in edges act as if they came // around the loopback from the prior iteration (follow the old-loop // backedges) and then map to the new peeled iteration. This leaves // the pre-loop with only 1 user (the new peeled iteration), but the // peeled-loop backedge has 2 users. // Step 3: Cut the backedge on the clone (so its not a loop) and remove the // extra backedge user. // // orig // // stmt1 // | // v // loop predicate // | // v // loop<----+ // | | // stmt2 | // | | // v | // if ^ // / \ | // / \ | // v v | // false true | // / \ | // / ----+ // | // v // exit // // // after clone loop // // stmt1 // | // v // loop predicate // / \ // clone / \ orig // / \ // / \ // v v // +---->loop clone loop<----+ // | | | | // | stmt2 clone stmt2 | // | | | | // | v v | // ^ if clone If ^ // | / \ / \ | // | / \ / \ | // | v v v v | // | true false false true | // | / \ / \ | // +---- \ / ----+ // \ / // 1v v2 // region // | // v // exit // // // after peel and predicate move // // stmt1 // | // v // loop predicate // / // / // clone / orig // / // / +----------+ // / | | // / | | // / | | // v v | // TOP-->loop clone loop<----+ | // | | | | // stmt2 clone stmt2 | | // | | | ^ // v v | | // if clone If ^ | // / \ / \ | | // / \ / \ | | // v v v v | | // true false false true | | // | \ / \ | | // | \ / ----+ ^ // | \ / | // | 1v v2 | // v region | // | | | // | v | // | exit | // | | // +--------------->-----------------+ // // // final graph // // stmt1 // | // v // loop predicate // | // v // stmt2 clone // | // v // if clone // / | // / | // v v // false true // | | // | v // | initialized skeleton predicates // | | // | v // | loop<----+ // | | | // | stmt2 | // | | | // | v | // v if ^ // | / \ | // | / \ | // | v v | // | false true | // | | \ | // v v --+ // region // | // v // exit // void PhaseIdealLoop::do_peeling(IdealLoopTree *loop, Node_List &old_new) {
C->set_major_progress(); // Peeling a 'main' loop in a pre/main/post situation obfuscates the // 'pre' loop from the main and the 'pre' can no longer have its // iterations adjusted. Therefore, we need to declare this loop as // no longer a 'main' loop; it will need new pre and post loops before // we can do further RCE. #ifndef PRODUCT if (TraceLoopOpts) {
tty->print("Peel ");
loop->dump_head();
} #endif
LoopNode* head = loop->_head->as_Loop(); bool counted_loop = head->is_CountedLoop(); if (counted_loop) {
CountedLoopNode *cl = head->as_CountedLoop();
assert(cl->trip_count() > 0, "peeling a fully unrolled loop");
cl->set_trip_count(cl->trip_count() - 1); if (cl->is_main_loop()) {
cl->set_normal_loop(); #ifndef PRODUCT if (PrintOpto && VerifyLoopOptimizations) {
tty->print("Peeling a 'main' loop; resetting to 'normal' ");
loop->dump_head();
} #endif
}
}
Node* entry = head->in(LoopNode::EntryControl);
// Step 1: Clone the loop body. The clone becomes the peeled iteration. // The pre-loop illegally has 2 control users (old & new loops). const uint idx_before_clone = Compile::current()->unique();
LoopNode* outer_loop_head = head->skip_strip_mined();
clone_loop(loop, old_new, dom_depth(outer_loop_head), ControlAroundStripMined);
// Step 2: Make the old-loop fall-in edges point to the peeled iteration. // Do this by making the old-loop fall-in edges act as if they came // around the loopback from the prior iteration (follow the old-loop // backedges) and then map to the new peeled iteration. This leaves // the pre-loop with only 1 user (the new peeled iteration), but the // peeled-loop backedge has 2 users.
Node* new_entry = old_new[head->in(LoopNode::LoopBackControl)->_idx];
_igvn.hash_delete(outer_loop_head);
outer_loop_head->set_req(LoopNode::EntryControl, new_entry); for (DUIterator_Fast jmax, j = head->fast_outs(jmax); j < jmax; j++) {
Node* old = head->fast_out(j); if (old->in(0) == loop->_head && old->req() == 3 && old->is_Phi()) {
Node* new_exit_value = old_new[old->in(LoopNode::LoopBackControl)->_idx]; if (!new_exit_value) // Backedge value is ALSO loop invariant? // Then loop body backedge value remains the same.
new_exit_value = old->in(LoopNode::LoopBackControl);
_igvn.hash_delete(old);
old->set_req(LoopNode::EntryControl, new_exit_value);
}
}
// Step 3: Cut the backedge on the clone (so its not a loop) and remove the // extra backedge user.
Node* new_head = old_new[head->_idx];
_igvn.hash_delete(new_head);
new_head->set_req(LoopNode::LoopBackControl, C->top()); for (DUIterator_Fast j2max, j2 = new_head->fast_outs(j2max); j2 < j2max; j2++) {
Node* use = new_head->fast_out(j2); if (use->in(0) == new_head && use->req() == 3 && use->is_Phi()) {
_igvn.hash_delete(use);
use->set_req(LoopNode::LoopBackControl, C->top());
}
}
// Step 4: Correct dom-depth info. Set to loop-head depth.
// Now force out all loop-invariant dominating tests. The optimizer // finds some, but we _know_ they are all useless.
peeled_dom_test_elim(loop,old_new);
loop->record_for_igvn();
}
//------------------------------policy_maximally_unroll------------------------ // Calculate the exact loop trip-count and return TRUE if loop can be fully, // i.e. maximally, unrolled, otherwise return FALSE. When TRUE, the estimated // node budget is also requested. bool IdealLoopTree::policy_maximally_unroll(PhaseIdealLoop* phase) const {
CountedLoopNode* cl = _head->as_CountedLoop();
assert(cl->is_normal_loop(), ""); if (!cl->is_valid_counted_loop(T_INT)) { returnfalse; // Malformed counted loop.
} if (!cl->has_exact_trip_count()) { returnfalse; // Trip count is not exact.
}
uint trip_count = cl->trip_count(); // Note, max_juint is used to indicate unknown trip count.
assert(trip_count > 1, "one iteration loop should be optimized out already");
assert(trip_count < max_juint, "exact trip_count should be less than max_juint.");
// If nodes are depleted, some transform has miscalculated its needs.
assert(!phase->exceeding_node_budget(), "sanity");
// Allow the unrolled body to get larger than the standard loop size limit.
uint unroll_limit = (uint)LoopUnrollLimit * 4;
assert((intx)unroll_limit == LoopUnrollLimit * 4, "LoopUnrollLimit must fit in 32bits"); if (trip_count > unroll_limit || _body.size() > unroll_limit) { returnfalse;
}
if (new_body_size == UINT_MAX) { // Check for bad estimate (overflow). returnfalse;
}
// Fully unroll a loop with few iterations, regardless of other conditions, // since the following (general) loop optimizations will split such loop in // any case (into pre-main-post). if (trip_count <= 3) { return phase->may_require_nodes(new_body_size);
}
// Reject if unrolling will result in too much node construction. if (new_body_size > unroll_limit || phase->exceeding_node_budget(new_body_size)) { returnfalse;
}
// Do not unroll a loop with String intrinsics code. // String intrinsics are large and have loops. for (uint k = 0; k < _body.size(); k++) {
Node* n = _body.at(k); switch (n->Opcode()) { case Op_StrComp: case Op_StrEquals: case Op_StrIndexOf: case Op_StrIndexOfChar: case Op_EncodeISOArray: case Op_AryEq: case Op_CountPositives: { returnfalse;
} #if INCLUDE_RTM_OPT case Op_FastLock: case Op_FastUnlock: { // Don't unroll RTM locking code because it is large. if (UseRTMLocking) { returnfalse;
}
} #endif
} // switch
}
return phase->may_require_nodes(new_body_size);
}
//------------------------------policy_unroll---------------------------------- // Return TRUE or FALSE if the loop should be unrolled or not. Apply unroll if // the loop is a counted loop and the loop body is small enough. When TRUE, // the estimated node budget is also requested. bool IdealLoopTree::policy_unroll(PhaseIdealLoop *phase) {
if (!cl->is_valid_counted_loop(T_INT)) { returnfalse; // Malformed counted loop
}
// If nodes are depleted, some transform has miscalculated its needs.
assert(!phase->exceeding_node_budget(), "sanity");
// Protect against over-unrolling. // After split at least one iteration will be executed in pre-loop. if (cl->trip_count() <= (cl->is_normal_loop() ? 2u : 1u)) { returnfalse;
}
_local_loop_unroll_limit = LoopUnrollLimit;
_local_loop_unroll_factor = 4; int future_unroll_cnt = cl->unrolled_count() * 2; if (!cl->is_vectorized_loop()) { if (future_unroll_cnt > LoopMaxUnroll) returnfalse;
} else { // obey user constraints on vector mapped loops with additional unrolling applied int unroll_constraint = (cl->slp_max_unroll()) ? cl->slp_max_unroll() : 1; if ((future_unroll_cnt / unroll_constraint) > LoopMaxUnroll) returnfalse;
}
constint stride_con = cl->stride_con();
// Check for initial stride being a small enough constant constint initial_stride_sz = MAX2(1<<2, Matcher::max_vector_size(T_BYTE) / 2); // Maximum stride size should protect against overflow, when doubling stride unroll_count times constint max_stride_size = MIN2<int>(max_jint / 2 - 2, initial_stride_sz * future_unroll_cnt); // No abs() use; abs(min_jint) = min_jint if (stride_con < -max_stride_size || stride_con > max_stride_size) returnfalse;
// Don't unroll if the next round of unrolling would push us // over the expected trip count of the loop. One is subtracted // from the expected trip count because the pre-loop normally // executes 1 iteration. if (UnrollLimitForProfileCheck > 0 &&
cl->profile_trip_cnt() != COUNT_UNKNOWN &&
future_unroll_cnt > UnrollLimitForProfileCheck &&
(float)future_unroll_cnt > cl->profile_trip_cnt() - 1.0) { returnfalse;
}
bool should_unroll = true;
// When unroll count is greater than LoopUnrollMin, don't unroll if: // the residual iterations are more than 10% of the trip count // and rounds of "unroll,optimize" are not making significant progress // Progress defined as current size less than 20% larger than previous size. if (UseSuperWord && cl->node_count_before_unroll() > 0 &&
future_unroll_cnt > LoopUnrollMin &&
is_residual_iters_large(future_unroll_cnt, cl) &&
1.2 * cl->node_count_before_unroll() < (double)_body.size()) { if ((cl->slp_max_unroll() == 0) && !is_residual_iters_large(cl->unrolled_count(), cl)) { // cl->slp_max_unroll() = 0 means that the previous slp analysis never passed. // slp analysis may fail due to the loop IR is too complicated especially during the early stage // of loop unrolling analysis. But after several rounds of loop unrolling and other optimizations, // it's possible that the loop IR becomes simple enough to pass the slp analysis. // So we don't return immediately in hoping that the next slp analysis can succeed.
should_unroll = false;
future_unroll_cnt = cl->unrolled_count();
} else { returnfalse;
}
}
Node *init_n = cl->init_trip();
Node *limit_n = cl->limit(); if (limit_n == NULL) returnfalse; // We will dereference it below.
// Non-constant bounds. // Protect against over-unrolling when init or/and limit are not constant // (so that trip_count's init value is maxint) but iv range is known. if (init_n == NULL || !init_n->is_Con() || !limit_n->is_Con()) {
Node* phi = cl->phi(); if (phi != NULL) {
assert(phi->is_Phi() && phi->in(0) == _head, "Counted loop should have iv phi."); const TypeInt* iv_type = phase->_igvn.type(phi)->is_int(); int next_stride = stride_con * 2; // stride after this unroll if (next_stride > 0) { if (iv_type->_lo > max_jint - next_stride || // overflow
iv_type->_lo + next_stride > iv_type->_hi) { returnfalse; // over-unrolling
}
} elseif (next_stride < 0) { if (iv_type->_hi < min_jint - next_stride || // overflow
iv_type->_hi + next_stride < iv_type->_lo) { returnfalse; // over-unrolling
}
}
}
}
// After unroll limit will be adjusted: new_limit = limit-stride. // Bailout if adjustment overflow. const TypeInt* limit_type = phase->_igvn.type(limit_n)->is_int(); if ((stride_con > 0 && ((min_jint + stride_con) > limit_type->_hi)) ||
(stride_con < 0 && ((max_jint + stride_con) < limit_type->_lo))) returnfalse; // overflow
// Rudimentary cost model to estimate loop unrolling // factor. // Adjust body_size to determine if we unroll or not
uint body_size = _body.size(); // Key test to unroll loop in CRC32 java code int xors_in_loop = 0; // Also count ModL, DivL and MulL which expand mightly for (uint k = 0; k < _body.size(); k++) {
Node* n = _body.at(k); switch (n->Opcode()) { case Op_XorI: xors_in_loop++; break; // CRC32 java code case Op_ModL: body_size += 30; break; case Op_DivL: body_size += 30; break; case Op_MulL: body_size += 10; break; case Op_RoundF: case Op_RoundD: {
body_size += Matcher::scalar_op_pre_select_sz_estimate(n->Opcode(), n->bottom_type()->basic_type());
} break; case Op_CountTrailingZerosV: case Op_CountLeadingZerosV: case Op_ReverseV: case Op_RoundVF: case Op_RoundVD: case Op_VectorCastD2X: case Op_VectorCastF2X: case Op_PopCountVI: case Op_PopCountVL: { const TypeVect* vt = n->bottom_type()->is_vect();
body_size += Matcher::vector_op_pre_select_sz_estimate(n->Opcode(), vt->element_basic_type(), vt->length());
} break; case Op_StrComp: case Op_StrEquals: case Op_StrIndexOf: case Op_StrIndexOfChar: case Op_EncodeISOArray: case Op_AryEq: case Op_CountPositives: { // Do not unroll a loop with String intrinsics code. // String intrinsics are large and have loops. returnfalse;
} #if INCLUDE_RTM_OPT case Op_FastLock: case Op_FastUnlock: { // Don't unroll RTM locking code because it is large. if (UseRTMLocking) { returnfalse;
}
} #endif
} // switch
}
if (UseSuperWord) { if (!cl->is_reduction_loop()) {
phase->mark_reductions(this);
}
// Only attempt slp analysis when user controls do not prohibit it if (!cl->range_checks_present() && (LoopMaxUnroll > _local_loop_unroll_factor)) { // Once policy_slp_analysis succeeds, mark the loop with the // maximal unroll factor so that we minimize analysis passes if (future_unroll_cnt >= _local_loop_unroll_factor) {
policy_unroll_slp_analysis(cl, phase, future_unroll_cnt);
}
}
}
int slp_max_unroll_factor = cl->slp_max_unroll(); if ((LoopMaxUnroll < slp_max_unroll_factor) && FLAG_IS_DEFAULT(LoopMaxUnroll) && UseSubwordForMaxVector) {
LoopMaxUnroll = slp_max_unroll_factor;
}
uint estimate = est_loop_clone_sz(2);
if (cl->has_passed_slp()) { if (slp_max_unroll_factor >= future_unroll_cnt) { return should_unroll && phase->may_require_nodes(estimate);
} returnfalse; // Loop too big.
}
// Check for being too big if (body_size > (uint)_local_loop_unroll_limit) { if ((cl->is_subword_loop() || xors_in_loop >= 4) && body_size < 4u * LoopUnrollLimit) { return should_unroll && phase->may_require_nodes(estimate);
} returnfalse; // Loop too big.
}
if (cl->is_unroll_only()) { if (TraceSuperWordLoopUnrollAnalysis) {
tty->print_cr("policy_unroll passed vector loop(vlen=%d, factor=%d)\n",
slp_max_unroll_factor, future_unroll_cnt);
}
}
// Unroll once! (Each trip will soon do double iterations) return should_unroll && phase->may_require_nodes(estimate);
}
void IdealLoopTree::policy_unroll_slp_analysis(CountedLoopNode *cl, PhaseIdealLoop *phase, int future_unroll_cnt) {
// If nodes are depleted, some transform has miscalculated its needs.
assert(!phase->exceeding_node_budget(), "sanity");
// Enable this functionality target by target as needed if (SuperWordLoopUnrollAnalysis) { if (!cl->was_slp_analyzed()) {
SuperWord sw(phase);
sw.transform_loop(this, false);
// If the loop is slp canonical analyze it if (sw.early_return() == false) {
sw.unrolling_analysis(_local_loop_unroll_factor);
}
}
if (cl->has_passed_slp()) { int slp_max_unroll_factor = cl->slp_max_unroll(); if (slp_max_unroll_factor >= future_unroll_cnt) { int new_limit = cl->node_count_before_unroll() * slp_max_unroll_factor; if (new_limit > LoopUnrollLimit) { if (TraceSuperWordLoopUnrollAnalysis) {
tty->print_cr("slp analysis unroll=%d, default limit=%d\n", new_limit, _local_loop_unroll_limit);
}
_local_loop_unroll_limit = new_limit;
}
}
}
}
}
//------------------------------policy_range_check----------------------------- // Return TRUE or FALSE if the loop should be range-check-eliminated or not. // When TRUE, the estimated node budget is also requested. // // We will actually perform iteration-splitting, a more powerful form of RCE. bool IdealLoopTree::policy_range_check(PhaseIdealLoop* phase, bool provisional, BasicType bt) const { if (!provisional && !RangeCheckElimination) returnfalse;
// If nodes are depleted, some transform has miscalculated its needs.
assert(provisional || !phase->exceeding_node_budget(), "sanity");
if (_head->is_CountedLoop()) {
CountedLoopNode *cl = _head->as_CountedLoop(); // If we unrolled with no intention of doing RCE and we later changed our // minds, we got no pre-loop. Either we need to make a new pre-loop, or we // have to disallow RCE. if (cl->is_main_no_pre_loop()) returnfalse; // Disallowed for now.
// check for vectorized loops, some opts are no longer needed // RCE needs pre/main/post loops. Don't apply it on a single iteration loop. if (cl->is_unroll_only() || (cl->is_normal_loop() && cl->trip_count() == 1)) returnfalse;
} else {
assert(provisional, "no long counted loop expected");
}
BaseCountedLoopNode* cl = _head->as_BaseCountedLoop();
Node *trip_counter = cl->phi();
assert(!cl->is_LongCountedLoop() || bt == T_LONG, "only long range checks in long counted loops");
assert(cl->is_valid_counted_loop(cl->bt()), "only for well formed loops");
// Check loop body for tests of trip-counter plus loop-invariant vs // loop-invariant. for (uint i = 0; i < _body.size(); i++) {
Node *iff = _body[i]; if (iff->Opcode() == Op_If ||
iff->Opcode() == Op_RangeCheck) { // Test?
// Comparing trip+off vs limit
Node *bol = iff->in(1); if (bol->req() != 2) { continue; // dead constant test
} if (!bol->is_Bool()) {
assert(bol->Opcode() == Op_Conv2B, "predicate check only"); continue;
} if (bol->as_Bool()->_test._test == BoolTest::ne) { continue; // not RC
}
Node *cmp = bol->in(1);
if (provisional) { // Try to pattern match with either cmp inputs, do not check // whether one of the inputs is loop independent as it may not // have had a chance to be hoisted yet. if (!phase->is_scaled_iv_plus_offset(cmp->in(1), trip_counter, bt, NULL, NULL) &&
!phase->is_scaled_iv_plus_offset(cmp->in(2), trip_counter, bt, NULL, NULL)) { continue;
}
} else {
Node *rc_exp = cmp->in(1);
Node *limit = cmp->in(2);
Node *limit_c = phase->get_ctrl(limit); if (limit_c == phase->C->top()) { returnfalse; // Found dead test on live IF? No RCE!
} if (is_member(phase->get_loop(limit_c))) { // Compare might have operands swapped; commute them
rc_exp = cmp->in(2);
limit = cmp->in(1);
limit_c = phase->get_ctrl(limit); if (is_member(phase->get_loop(limit_c))) { continue; // Both inputs are loop varying; cannot RCE
}
}
if (!phase->is_scaled_iv_plus_offset(rc_exp, trip_counter, bt, NULL, NULL)) { continue;
}
} // Found a test like 'trip+off vs limit'. Test is an IfNode, has two (2) // projections. If BOTH are in the loop we need loop unswitching instead // of iteration splitting. if (is_loop_exit(iff)) { // Found valid reason to split iterations (if there is room). // NOTE: Usually a gross overestimate. // Long range checks cause the loop to be transformed in a loop nest which only causes a fixed number of nodes // to be added return provisional || bt == T_LONG || phase->may_require_nodes(est_loop_clone_sz(2));
}
} // End of is IF
}
returnfalse;
}
//------------------------------policy_peel_only------------------------------- // Return TRUE or FALSE if the loop should NEVER be RCE'd or aligned. Useful // for unrolling loops with NO array accesses. bool IdealLoopTree::policy_peel_only(PhaseIdealLoop *phase) const {
// If nodes are depleted, some transform has miscalculated its needs.
assert(!phase->exceeding_node_budget(), "sanity");
// check for vectorized loops, any peeling done was already applied if (_head->is_CountedLoop() && _head->as_CountedLoop()->is_unroll_only()) { returnfalse;
}
for (uint i = 0; i < _body.size(); i++) { if (_body[i]->is_Mem()) { returnfalse;
}
} // No memory accesses at all! returntrue;
}
//------------------------------clone_up_backedge_goo-------------------------- // If Node n lives in the back_ctrl block and cannot float, we clone a private // version of n in preheader_ctrl block and return that, otherwise return n.
Node *PhaseIdealLoop::clone_up_backedge_goo(Node *back_ctrl, Node *preheader_ctrl, Node *n, VectorSet &visited, Node_Stack &clones) { if (get_ctrl(n) != back_ctrl) return n;
// Only visit once if (visited.test_set(n->_idx)) {
Node *x = clones.find(n->_idx); return (x != NULL) ? x : n;
}
Node *x = NULL; // If required, a clone of 'n' // Check for 'n' being pinned in the backedge. if (n->in(0) && n->in(0) == back_ctrl) {
assert(clones.find(n->_idx) == NULL, "dead loop");
x = n->clone(); // Clone a copy of 'n' to preheader
clones.push(x, n->_idx);
x->set_req(0, preheader_ctrl); // Fix x's control input to preheader
}
// Recursive fixup any other input edges into x. // If there are no changes we can just return 'n', otherwise // we need to clone a private copy and change it. for (uint i = 1; i < n->req(); i++) {
Node *g = clone_up_backedge_goo(back_ctrl, preheader_ctrl, n->in(i), visited, clones); if (g != n->in(i)) { if (!x) {
assert(clones.find(n->_idx) == NULL, "dead loop");
x = n->clone();
clones.push(x, n->_idx);
}
x->set_req(i, g);
}
} if (x) { // x can legally float to pre-header location
register_new_node(x, preheader_ctrl); return x;
} else { // raise n to cover LCA of uses
set_ctrl(n, find_non_split_ctrl(back_ctrl->in(0)));
} return n;
}
Node* PhaseIdealLoop::cast_incr_before_loop(Node* incr, Node* ctrl, Node* loop) {
Node* castii = new CastIINode(incr, TypeInt::INT, ConstraintCastNode::UnconditionalDependency);
castii->set_req(0, ctrl);
register_new_node(castii, ctrl); for (DUIterator_Fast imax, i = incr->fast_outs(imax); i < imax; i++) {
Node* n = incr->fast_out(i); if (n->is_Phi() && n->in(0) == loop) { int nrep = n->replace_edge(incr, castii, &_igvn); return castii;
}
} return NULL;
}
#ifdef ASSERT void PhaseIdealLoop::ensure_zero_trip_guard_proj(Node* node, bool is_main_loop) {
assert(node->is_IfProj(), "must be the zero trip guard If node");
Node* zer_bol = node->in(0)->in(1);
assert(zer_bol != NULL && zer_bol->is_Bool(), "must be Bool");
Node* zer_cmp = zer_bol->in(1);
assert(zer_cmp != NULL && zer_cmp->Opcode() == Op_CmpI, "must be CmpI"); // For the main loop, the opaque node is the second input to zer_cmp, for the post loop it's the first input node
Node* zer_opaq = zer_cmp->in(is_main_loop ? 2 : 1);
assert(zer_opaq != NULL && zer_opaq->Opcode() == Op_OpaqueZeroTripGuard, "must be OpaqueZeroTripGuard");
} #endif
// 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::copy_skeleton_predicates_to_main_loop_helper(Node* predicate, Node* init, Node* stride,
IdealLoopTree* outer_loop, LoopNode* outer_main_head,
uint dd_main_head, const uint idx_before_pre_post, const uint idx_after_post_before_pre, Node* zero_trip_guard_proj_main,
Node* zero_trip_guard_proj_post, const Node_List &old_new) { if (predicate != NULL) { #ifdef ASSERT
ensure_zero_trip_guard_proj(zero_trip_guard_proj_main, true);
ensure_zero_trip_guard_proj(zero_trip_guard_proj_post, false); #endif
IfNode* iff = predicate->in(0)->as_If();
ProjNode* uncommon_proj = iff->proj_out(1 - predicate->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");
predicate = iff->in(0);
Node* current_proj = outer_main_head->in(LoopNode::EntryControl);
Node* prev_proj = current_proj;
Node* opaque_init = new OpaqueLoopInitNode(C, init);
register_new_node(opaque_init, outer_main_head->in(LoopNode::EntryControl));
Node* opaque_stride = new OpaqueLoopStrideNode(C, stride);
register_new_node(opaque_stride, outer_main_head->in(LoopNode::EntryControl));
while (predicate != NULL && predicate->is_Proj() && predicate->in(0)->is_If()) {
iff = predicate->in(0)->as_If();
uncommon_proj = iff->proj_out(1 - predicate->as_Proj()->_con); if (uncommon_proj->unique_ctrl_out() != rgn) break; if (iff->in(1)->Opcode() == Op_Opaque4) {
assert(skeleton_predicate_has_opaque(iff), "unexpected"); // Clone the skeleton predicate twice and initialize one with the initial // value of the loop induction variable. Leave the other predicate // to be initialized when increasing the stride during loop unrolling.
prev_proj = clone_skeleton_predicate_and_initialize(iff, opaque_init, NULL, predicate, uncommon_proj,
current_proj, outer_loop, prev_proj);
assert(skeleton_predicate_has_opaque(prev_proj->in(0)->as_If()), "");
// Rewire any control inputs from the cloned skeleton predicates down to the main and post loop for data nodes that are part of the // main loop (and were cloned to the pre and post loop). for (DUIterator i = predicate->outs(); predicate->has_out(i); i++) {
Node* loop_node = predicate->out(i);
Node* pre_loop_node = old_new[loop_node->_idx]; // Change the control if 'loop_node' is part of the main loop. If there is an old->new mapping and the index of // 'pre_loop_node' is greater than idx_before_pre_post, then we know that 'loop_node' was cloned and is part of // the main loop (and 'pre_loop_node' is part of the pre loop). if (!loop_node->is_CFG() && (pre_loop_node != NULL && pre_loop_node->_idx > idx_after_post_before_pre)) { // 'loop_node' is a data node and part of the main loop. Rewire the control to the projection of the zero-trip guard if node // of the main loop that is immediately preceding the cloned predicates.
_igvn.replace_input_of(loop_node, 0, zero_trip_guard_proj_main);
--i;
} elseif (loop_node->_idx > idx_before_pre_post && loop_node->_idx < idx_after_post_before_pre) { // 'loop_node' is a data node and part of the post loop. Rewire the control to the projection of the zero-trip guard if node // of the post loop that is immediately preceding the post loop header node (there are no cloned predicates for the post loop).
assert(pre_loop_node == NULL, "a node belonging to the post loop should not have an old_new mapping at this stage");
_igvn.replace_input_of(loop_node, 0, zero_trip_guard_proj_post);
--i;
}
}
// Remove the skeleton predicate from the pre-loop
_igvn.replace_input_of(iff, 1, _igvn.intcon(1));
}
predicate = predicate->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);
}
}
staticbool skeleton_follow_inputs(Node* n) { int op = n->Opcode(); return (n->is_Bool() ||
n->is_Cmp() ||
op == Op_AndL ||
op == Op_OrL ||
op == Op_RShiftL ||
op == Op_LShiftL ||
op == Op_LShiftI ||
op == Op_AddL ||
op == Op_AddI ||
op == Op_MulL ||
op == Op_MulI ||
op == Op_SubL ||
op == Op_SubI ||
op == Op_ConvI2L ||
op == Op_CastII);
}
bool PhaseIdealLoop::skeleton_predicate_has_opaque(IfNode* iff) {
uint init;
uint stride;
count_opaque_loop_nodes(iff->in(1)->in(1), init, stride); #ifdef ASSERT
ResourceMark rm;
Unique_Node_List wq;
wq.clear();
wq.push(iff->in(1)->in(1));
uint verif_init = 0;
uint verif_stride = 0; for (uint i = 0; i < wq.size(); i++) {
Node* n = wq.at(i); int op = n->Opcode(); if (!n->is_CFG()) { if (n->Opcode() == Op_OpaqueLoopInit) {
verif_init++;
} elseif (n->Opcode() == Op_OpaqueLoopStride) {
verif_stride++;
} else { for (uint j = 1; j < n->req(); j++) {
Node* m = n->in(j); if (m != NULL) {
wq.push(m);
}
}
}
}
}
assert(init == verif_init && stride == verif_stride, "missed opaque node"); #endif
assert(stride == 0 || init != 0, "init should be there every time stride is"); return init != 0;
}
void PhaseIdealLoop::count_opaque_loop_nodes(Node* n, uint& init, uint& stride) {
init = 0;
stride = 0;
ResourceMark rm;
Unique_Node_List wq;
wq.push(n); for (uint i = 0; i < wq.size(); i++) {
Node* n = wq.at(i); if (skeleton_follow_inputs(n)) { for (uint j = 1; j < n->req(); j++) {
Node* m = n->in(j); if (m != NULL) {
wq.push(m);
}
} continue;
} if (n->Opcode() == Op_OpaqueLoopInit) {
init++;
} elseif (n->Opcode() == Op_OpaqueLoopStride) {
stride++;
}
}
}
// Clone the skeleton predicate bool for a main or unswitched loop: // Main loop: Set new_init and new_stride nodes as new inputs. // Unswitched loop: new_init and new_stride are both NULL. Clone OpaqueLoopInit and OpaqueLoopStride instead.
Node* PhaseIdealLoop::clone_skeleton_predicate_bool(Node* iff, Node* new_init, Node* new_stride, Node* control) {
Node_Stack to_clone(2);
to_clone.push(iff->in(1), 1);
uint current = C->unique();
Node* result = NULL; bool is_unswitched_loop = new_init == NULL && new_stride == NULL;
assert(new_init != NULL || is_unswitched_loop, "new_init must be set when new_stride is non-null"); // Look for the opaque node to replace with the new 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); if (skeleton_follow_inputs(m)) {
to_clone.push(m, 1); continue;
} if (m->is_Opaque1()) { if (n->_idx < current) {
n = n->clone();
register_new_node(n, control);
} int op = m->Opcode(); if (op == Op_OpaqueLoopInit) { if (is_unswitched_loop && m->_idx < current && new_init == NULL) {
new_init = m->clone();
register_new_node(new_init, control);
}
n->set_req(i, new_init);
} else {
assert(op == Op_OpaqueLoopStride, "unexpected opaque node"); if (is_unswitched_loop && m->_idx < current && new_stride == NULL) {
new_stride = m->clone();
register_new_node(new_stride, control);
} if (new_stride != NULL) {
n->set_req(i, new_stride);
}
}
to_clone.set_node(n);
} while (true) {
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 (next->in(j) != cur) {
assert(cur->_idx >= current || next->in(j)->Opcode() == Op_Opaque1, "new node or Opaque1 being replaced"); if (next->_idx < current) {
next = next->clone();
register_new_node(next, control);
to_clone.set_node(next);
}
next->set_req(j, cur);
}
}
} while (result == NULL);
assert(result->_idx >= current, "new node expected");
assert(!is_unswitched_loop || new_init != NULL, "new_init must always be found and cloned"); return result;
}
// Clone a skeleton predicate for the main loop. new_init and new_stride are set as new inputs. Since the predicates cannot fail at runtime, // Halt nodes are inserted instead of uncommon traps.
Node* PhaseIdealLoop::clone_skeleton_predicate_and_initialize(Node* iff, Node* new_init, Node* new_stride, Node* predicate, Node* uncommon_proj,
Node* control, IdealLoopTree* outer_loop, Node* input_proj) {
Node* result = clone_skeleton_predicate_bool(iff, new_init, new_stride, control);
Node* proj = predicate->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, "duplicated predicate failed which is impossible");
_igvn.add_input_to(C->root(), halt);
new_iff->set_req(0, input_proj);
//------------------------------insert_pre_post_loops-------------------------- // Insert pre and post loops. If peel_only is set, the pre-loop can not have // more iterations added. It acts as a 'peel' only, no lower-bound RCE, no // alignment. Useful to unroll loops that do no array accesses. void PhaseIdealLoop::insert_pre_post_loops(IdealLoopTree *loop, Node_List &old_new, bool peel_only) {
#ifndef PRODUCT if (TraceLoopOpts) { if (peel_only)
tty->print("PeelMainPost "); else
tty->print("PreMainPost ");
loop->dump_head();
} #endif
C->set_major_progress();
// Find common pieces of the loop being guarded with pre & post loops
CountedLoopNode *main_head = loop->_head->as_CountedLoop();
assert(main_head->is_normal_loop(), "");
CountedLoopEndNode *main_end = main_head->loopexit();
assert(main_end->outcnt() == 2, "1 true, 1 false path only");
// Need only 1 user of 'bol' because I will be hacking the loop bounds.
Node *bol = main_end->in(CountedLoopEndNode::TestValue); if (bol->outcnt() != 1) {
bol = bol->clone();
register_new_node(bol,main_end->in(CountedLoopEndNode::TestControl));
_igvn.replace_input_of(main_end, CountedLoopEndNode::TestValue, bol);
} // Need only 1 user of 'cmp' because I will be hacking the loop bounds. if (cmp->outcnt() != 1) {
cmp = cmp->clone();
register_new_node(cmp,main_end->in(CountedLoopEndNode::TestControl));
_igvn.replace_input_of(bol, 1, cmp);
}
// Step B1: Clone the loop body. The clone becomes the pre-loop. The main // loop pre-header illegally has 2 control users (old & new loops).
LoopNode* outer_main_head = main_head;
IdealLoopTree* outer_loop = loop; if (main_head->is_strip_mined()) {
main_head->verify_strip_mined(1);
outer_main_head = main_head->outer_loop();
outer_loop = loop->_parent;
assert(outer_loop->_head == outer_main_head, "broken loop tree");
}
uint dd_main_head = dom_depth(outer_main_head);
clone_loop(loop, old_new, dd_main_head, ControlAroundStripMined);
CountedLoopNode* pre_head = old_new[main_head->_idx]->as_CountedLoop();
CountedLoopEndNode* pre_end = old_new[main_end ->_idx]->as_CountedLoopEnd();
pre_head->set_pre_loop(main_head);
Node *pre_incr = old_new[incr->_idx];
// Reduce the pre-loop trip count.
pre_end->_prob = PROB_FAIR;
// Find the pre-loop normal exit.
Node* pre_exit = pre_end->proj_out(false);
assert(pre_exit->Opcode() == Op_IfFalse, "");
IfFalseNode *new_pre_exit = new IfFalseNode(pre_end);
_igvn.register_new_node_with_optimizer(new_pre_exit);
set_idom(new_pre_exit, pre_end, dd_main_head);
set_loop(new_pre_exit, outer_loop->_parent);
if (peel_only) { // Nodes in the peeled iteration that were marked as reductions within the // original loop might not be reductions within their new outer loop. for (uint i = 0; i < loop->_body.size(); i++) {
Node* n = old_new[loop->_body[i]->_idx];
n->remove_flag(Node::Flag_is_reduction);
}
}
// Step B2: Build a zero-trip guard for the main-loop. After leaving the // pre-loop, the main-loop may not execute at all. Later in life this // zero-trip guard will become the minimum-trip guard when we unroll // the main-loop.
Node *min_opaq = new OpaqueZeroTripGuardNode(C, limit, b_test);
Node *min_cmp = new CmpINode(pre_incr, min_opaq);
Node *min_bol = new BoolNode(min_cmp, b_test);
register_new_node(min_opaq, new_pre_exit);
register_new_node(min_cmp , new_pre_exit);
register_new_node(min_bol , new_pre_exit);
// Build the IfNode (assume the main-loop is executed always).
IfNode *min_iff = new IfNode(new_pre_exit, min_bol, PROB_ALWAYS, COUNT_UNKNOWN);
_igvn.register_new_node_with_optimizer(min_iff);
set_idom(min_iff, new_pre_exit, dd_main_head);
set_loop(min_iff, outer_loop->_parent);
// Plug in the false-path, taken if we need to skip main-loop
_igvn.hash_delete(pre_exit);
pre_exit->set_req(0, min_iff);
set_idom(pre_exit, min_iff, dd_main_head);
set_idom(pre_exit->unique_ctrl_out(), min_iff, dd_main_head); // Make the true-path, must enter the main loop
Node *min_taken = new IfTrueNode(min_iff);
_igvn.register_new_node_with_optimizer(min_taken);
set_idom(min_taken, min_iff, dd_main_head);
set_loop(min_taken, outer_loop->_parent); // Plug in the true path
_igvn.hash_delete(outer_main_head);
outer_main_head->set_req(LoopNode::EntryControl, min_taken);
set_idom(outer_main_head, min_taken, dd_main_head);
VectorSet visited;
Node_Stack clones(main_head->back_control()->outcnt()); // Step B3: Make the fall-in values to the main-loop come from the // fall-out values of the pre-loop. for (DUIterator i2 = main_head->outs(); main_head->has_out(i2); i2++) {
Node* main_phi = main_head->out(i2); if (main_phi->is_Phi() && main_phi->in(0) == main_head && main_phi->outcnt() > 0) {
Node* pre_phi = old_new[main_phi->_idx];
Node* fallpre = clone_up_backedge_goo(pre_head->back_control(),
main_head->skip_strip_mined()->in(LoopNode::EntryControl),
pre_phi->in(LoopNode::LoopBackControl),
visited, clones);
_igvn.hash_delete(main_phi);
main_phi->set_req(LoopNode::EntryControl, fallpre);
}
}
// Nodes inside the loop may be control dependent on a predicate // that was moved before the preloop. If the back branch of the main // or post loops becomes dead, those nodes won't be dependent on the // test that guards that loop nest anymore which could lead to an // incorrect array access because it executes independently of the // test that was guarding the loop nest. We add a special CastII on // the if branch that enters the loop, between the input induction // variable value and the induction variable Phi to preserve correct // dependencies.
// CastII for the main loop:
Node* castii = cast_incr_before_loop(pre_incr, min_taken, main_head);
assert(castii != NULL, "no castII inserted");
assert(post_head->in(1)->is_IfProj(), "must be zero-trip guard If node projection of the post loop");
copy_skeleton_predicates_to_main_loop(pre_head, castii, stride, outer_loop, outer_main_head, dd_main_head,
idx_before_pre_post, idx_after_post_before_pre, min_taken, post_head->in(1), old_new);
copy_skeleton_predicates_to_post_loop(outer_main_head, post_head, post_incr, stride);
// Step B4: Shorten the pre-loop to run only 1 iteration (for now). // RCE and alignment may change this later.
Node *cmp_end = pre_end->cmp_node();
assert(cmp_end->in(2) == limit, "");
Node *pre_limit = new AddINode(init, stride);
// Save the original loop limit in this Opaque1 node for // use by range check elimination.
Node *pre_opaq = new Opaque1Node(C, pre_limit, limit);
// Since no other users of pre-loop compare, I can hack limit directly
assert(cmp_end->outcnt() == 1, "no other users");
_igvn.hash_delete(cmp_end);
cmp_end->set_req(2, peel_only ? pre_limit : pre_opaq);
// Special case for not-equal loop bounds: // Change pre loop test, main loop test, and the // main loop guard test to use lt or gt depending on stride // direction: // positive stride use < // negative stride use > // // not-equal test is kept for post loop to handle case // when init > limit when stride > 0 (and reverse).
if (pre_end->in(CountedLoopEndNode::TestValue)->as_Bool()->_test._test == BoolTest::ne) {
BoolTest::mask new_test = (main_end->stride_con() > 0) ? BoolTest::lt : BoolTest::gt; // Modify pre loop end condition
Node* pre_bol = pre_end->in(CountedLoopEndNode::TestValue)->as_Bool();
BoolNode* new_bol0 = new BoolNode(pre_bol->in(1), new_test);
register_new_node(new_bol0, pre_head->in(0));
_igvn.replace_input_of(pre_end, CountedLoopEndNode::TestValue, new_bol0); // Modify main loop guard condition
assert(min_iff->in(CountedLoopEndNode::TestValue) == min_bol, "guard okay");
BoolNode* new_bol1 = new BoolNode(min_bol->in(1), new_test);
register_new_node(new_bol1, new_pre_exit);
_igvn.hash_delete(min_iff);
min_iff->set_req(CountedLoopEndNode::TestValue, new_bol1); // Modify main loop end condition
BoolNode* main_bol = main_end->in(CountedLoopEndNode::TestValue)->as_Bool();
BoolNode* new_bol2 = new BoolNode(main_bol->in(1), new_test);
register_new_node(new_bol2, main_end->in(CountedLoopEndNode::TestControl));
_igvn.replace_input_of(main_end, CountedLoopEndNode::TestValue, new_bol2);
}
// Flag main loop
main_head->set_main_loop(); if (peel_only) {
main_head->set_main_no_pre_loop();
}
// Subtract a trip count for the pre-loop.
main_head->set_trip_count(main_head->trip_count() - 1);
// It's difficult to be precise about the trip-counts // for the pre/post loops. They are usually very short, // so guess that 4 trips is a reasonable value.
post_head->set_profile_trip_cnt(4.0);
pre_head->set_profile_trip_cnt(4.0);
// Now force out all loop-invariant dominating tests. The optimizer // finds some, but we _know_ they are all useless.
peeled_dom_test_elim(loop,old_new);
loop->record_for_igvn();
}
//------------------------------insert_vector_post_loop------------------------ // Insert a copy of the atomic unrolled vectorized main loop as a post loop, // unroll_policy has already informed us that more unrolling is about to // happen to the main loop. The resultant post loop will serve as a // vectorized drain loop. void PhaseIdealLoop::insert_vector_post_loop(IdealLoopTree *loop, Node_List &old_new) { if (!loop->_head->is_CountedLoop()) return;
// Find common pieces of the loop being guarded with pre & post loops
CountedLoopNode *main_head = loop->_head->as_CountedLoop();
CountedLoopEndNode *main_end = main_head->loopexit(); // diagnostic to show loop end is not properly formed
assert(main_end->outcnt() == 2, "1 true, 1 false path only");
// mark this loop as processed
main_head->mark_has_atomic_post_loop();
// In this case we throw away the result as we are not using it to connect anything else.
CountedLoopNode *post_head = NULL;
insert_post_loop(loop, old_new, main_head, main_end, incr, limit, post_head);
copy_skeleton_predicates_to_post_loop(main_head->skip_strip_mined(), post_head, incr, main_head->stride());
// It's difficult to be precise about the trip-counts // for post loops. They are usually very short, // so guess that unit vector trips is a reasonable value.
post_head->set_profile_trip_cnt(cur_unroll);
// Now force out all loop-invariant dominating tests. The optimizer // finds some, but we _know_ they are all useless.
peeled_dom_test_elim(loop, old_new);
loop->record_for_igvn();
}
//-------------------------insert_scalar_rced_post_loop------------------------ // Insert a copy of the rce'd main loop as a post loop, // We have not unrolled the main loop, so this is the right time to inject this. // Later we will examine the partner of this post loop pair which still has range checks // to see inject code which tests at runtime if the range checks are applicable. void PhaseIdealLoop::insert_scalar_rced_post_loop(IdealLoopTree *loop, Node_List &old_new) { if (!loop->_head->is_CountedLoop()) return;
// Find common pieces of the loop being guarded with pre & post loops
CountedLoopNode *main_head = loop->_head->as_CountedLoop();
CountedLoopEndNode *main_end = main_head->loopexit(); // diagnostic to show loop end is not properly formed
assert(main_end->outcnt() == 2, "1 true, 1 false path only");
// In this case we throw away the result as we are not using it to connect anything else.
CountedLoopNode *post_head = NULL;
insert_post_loop(loop, old_new, main_head, main_end, incr, limit, post_head);
copy_skeleton_predicates_to_post_loop(main_head->skip_strip_mined(), post_head, incr, main_head->stride());
// It's difficult to be precise about the trip-counts // for post loops. They are usually very short, // so guess that unit vector trips is a reasonable value.
post_head->set_profile_trip_cnt(4.0);
post_head->set_is_rce_post_loop();
// Now force out all loop-invariant dominating tests. The optimizer // finds some, but we _know_ they are all useless.
peeled_dom_test_elim(loop, old_new);
loop->record_for_igvn();
}
//------------------------------insert_post_loop------------------------------- // Insert post loops. Add a post loop to the given loop passed.
Node *PhaseIdealLoop::insert_post_loop(IdealLoopTree* loop, Node_List& old_new,
CountedLoopNode* main_head, CountedLoopEndNode* main_end,
Node*& incr, Node* limit, CountedLoopNode*& post_head) {
IfNode* outer_main_end = main_end;
IdealLoopTree* outer_loop = loop; if (main_head->is_strip_mined()) {
main_head->verify_strip_mined(1);
outer_main_end = main_head->outer_loop_end();
outer_loop = loop->_parent;
assert(outer_loop->_head == main_head->in(LoopNode::EntryControl), "broken loop tree");
}
//------------------------------ // Step A: Create a new post-Loop.
Node* main_exit = outer_main_end->proj_out(false);
assert(main_exit->Opcode() == Op_IfFalse, ""); int dd_main_exit = dom_depth(main_exit);
// Step A1: Clone the loop body of main. The clone becomes the post-loop. // The main loop pre-header illegally has 2 control users (old & new loops).
clone_loop(loop, old_new, dd_main_exit, ControlAroundStripMined);
assert(old_new[main_end->_idx]->Opcode() == Op_CountedLoopEnd, "");
post_head = old_new[main_head->_idx]->as_CountedLoop();
post_head->set_normal_loop();
post_head->set_post_loop(main_head);
// Build the main-loop normal exit.
IfFalseNode *new_main_exit = new IfFalseNode(outer_main_end);
_igvn.register_new_node_with_optimizer(new_main_exit);
set_idom(new_main_exit, outer_main_end, dd_main_exit);
set_loop(new_main_exit, outer_loop->_parent);
// Step A2: Build a zero-trip guard for the post-loop. After leaving the // main-loop, the post-loop may not execute at all. We 'opaque' the incr // (the previous loop trip-counter exit value) because we will be changing // the exit value (via additional unrolling) so we cannot constant-fold away the zero // trip guard until all unrolling is done.
Node *zer_opaq = new OpaqueZeroTripGuardNode(C, incr, main_end->test_trip());
Node *zer_cmp = new CmpINode(zer_opaq, limit);
Node *zer_bol = new BoolNode(zer_cmp, main_end->test_trip());
register_new_node(zer_opaq, new_main_exit);
register_new_node(zer_cmp, new_main_exit);
register_new_node(zer_bol, new_main_exit);
// Build the IfNode
IfNode *zer_iff = new IfNode(new_main_exit, zer_bol, PROB_FAIR, COUNT_UNKNOWN);
_igvn.register_new_node_with_optimizer(zer_iff);
set_idom(zer_iff, new_main_exit, dd_main_exit);
set_loop(zer_iff, outer_loop->_parent);
// Plug in the false-path, taken if we need to skip this post-loop
_igvn.replace_input_of(main_exit, 0, zer_iff);
set_idom(main_exit, zer_iff, dd_main_exit);
set_idom(main_exit->unique_out(), zer_iff, dd_main_exit); // Make the true-path, must enter this post loop
Node *zer_taken = new IfTrueNode(zer_iff);
_igvn.register_new_node_with_optimizer(zer_taken);
set_idom(zer_taken, zer_iff, dd_main_exit);
set_loop(zer_taken, outer_loop->_parent); // Plug in the true path
_igvn.hash_delete(post_head);
post_head->set_req(LoopNode::EntryControl, zer_taken);
set_idom(post_head, zer_taken, dd_main_exit);
VectorSet visited;
Node_Stack clones(main_head->back_control()->outcnt()); // Step A3: Make the fall-in values to the post-loop come from the // fall-out values of the main-loop. for (DUIterator i = main_head->outs(); main_head->has_out(i); i++) {
Node* main_phi = main_head->out(i); if (main_phi->is_Phi() && main_phi->in(0) == main_head && main_phi->outcnt() > 0) {
Node* cur_phi = old_new[main_phi->_idx];
Node* fallnew = clone_up_backedge_goo(main_head->back_control(),
post_head->init_control(),
main_phi->in(LoopNode::LoopBackControl),
visited, clones);
_igvn.hash_delete(cur_phi);
cur_phi->set_req(LoopNode::EntryControl, fallnew);
}
}
// CastII for the new post loop:
incr = cast_incr_before_loop(zer_opaq->in(1), zer_taken, post_head);
assert(incr != NULL, "no castII inserted");
return new_main_exit;
}
//------------------------------is_invariant----------------------------- // Return true if n is invariant bool IdealLoopTree::is_invariant(Node* n) const {
Node *n_c = _phase->has_ctrl(n) ? _phase->get_ctrl(n) : n; if (n_c->is_top()) returnfalse; return !is_member(_phase->get_loop(n_c));
}
void PhaseIdealLoop::update_main_loop_skeleton_predicates(Node* ctrl, CountedLoopNode* loop_head, Node* init, int stride_con) { if (init->Opcode() == Op_CastII) { // skip over the cast added by PhaseIdealLoop::cast_incr_before_loop() when pre/post/main loops are created because // it can get in the way of type propagation
assert(((CastIINode*)init)->carry_dependency() && loop_head->skip_predicates() == init->in(0), "casted iv phi from pre loop expected");
init = init->in(1);
} // Search for skeleton predicates and update them according to the new stride
Node* entry = ctrl;
Node* prev_proj = ctrl;
LoopNode* outer_loop_head = loop_head->skip_strip_mined();
IdealLoopTree* outer_loop = get_loop(outer_loop_head);
// Compute the value of the loop induction variable at the end of the // first iteration of the unrolled loop: init + new_stride_con - init_inc int new_stride_con = stride_con * 2;
Node* max_value = _igvn.intcon(new_stride_con);
set_ctrl(max_value, C->root());
while (entry != NULL && entry->is_Proj() && entry->in(0)->is_If()) {
IfNode* iff = entry->in(0)->as_If();
ProjNode* proj = iff->proj_out(1 - entry->as_Proj()->_con); if (proj->unique_ctrl_out()->Opcode() != Op_Halt) { break;
} if (iff->in(1)->Opcode() == Op_Opaque4) { // Look for predicate with an Opaque1 node that can be used as a template if (!skeleton_predicate_has_opaque(iff)) { // No Opaque1 node? It's either the check for the first value // of the first iteration or the check for the last value of // the first iteration of an unrolled loop. We can't // tell. Kill it in any case.
_igvn.replace_input_of(iff, 1, iff->in(1)->in(2));
} else { // Add back predicates updated for the new stride.
prev_proj = clone_skeleton_predicate_and_initialize(iff, init, max_value, entry, proj, ctrl, outer_loop,
prev_proj);
assert(!skeleton_predicate_has_opaque(prev_proj->in(0)->as_If()), "unexpected");
}
}
entry = entry->in(0)->in(0);
} if (prev_proj != ctrl) {
_igvn.replace_input_of(outer_loop_head, LoopNode::EntryControl, prev_proj);
set_idom(outer_loop_head, prev_proj, dom_depth(outer_loop_head));
}
}
void PhaseIdealLoop::copy_skeleton_predicates_to_post_loop(LoopNode* main_loop_head, CountedLoopNode* post_loop_head, Node* init, Node* stride) { // Go over the skeleton predicates of the main loop and make a copy for the post loop with its initial iv value and // stride as inputs.
Node* post_loop_entry = post_loop_head->in(LoopNode::EntryControl);
Node* main_loop_entry = main_loop_head->in(LoopNode::EntryControl);
IdealLoopTree* post_loop = get_loop(post_loop_head);
// Rewire any control inputs from the old skeleton predicates above the peeled iteration down to the initialized // skeleton predicates above the peeled loop. for (DUIterator i = predicate->outs(); predicate->has_out(i); i++) {
Node* dependent = predicate->out(i);
Node* new_node = old_new[dependent->_idx];
if (!dependent->is_CFG() &&
dependent->_idx < idx_before_clone && // old node
new_node != nullptr && // cloned
new_node->_idx >= idx_before_clone) { // for peeling // The old nodes from the peeled loop still point to the predicate above the peeled loop. // We need to rewire the dependencies to the newly initialized skeleton predicates.
_igvn.replace_input_of(dependent, 0, input_proj);
--i; // correct for just deleted predicate->out(i)
}
}
}
predicate = next_predicate(predicate);
}
// Remember loop node count before unrolling to detect // if rounds of unroll,optimize are making progress
loop_head->set_node_count_before_unroll(loop->_body.size());
Node *opaq = NULL; if (adjust_min_trip) { // If not maximally unrolling, need adjustment // Search for zero-trip guard.
// Check the shape of the graph at the loop entry. If an inappropriate // graph shape is encountered, the compiler bails out loop unrolling; // compilation of the method will still succeed.
opaq = loop_head->is_canonical_loop_entry(); if (opaq == NULL) { return;
} // Zero-trip test uses an 'opaque' node which is not shared.
assert(opaq->outcnt() == 1 && opaq->in(1) == limit, "");
}
C->set_major_progress();
Node* new_limit = NULL; int stride_con = stride->get_int(); int stride_p = (stride_con > 0) ? stride_con : -stride_con;
uint old_trip_count = loop_head->trip_count(); // Verify that unroll policy result is still valid.
assert(old_trip_count > 1 && (!adjust_min_trip || stride_p <=
MIN2<int>(max_jint / 2 - 2, MAX2(1<<3, Matcher::max_vector_size(T_BYTE)) * loop_head->unrolled_count())), "sanity");
// Adjust loop limit to keep valid iterations number after unroll. // Use (limit - stride) instead of (((limit - init)/stride) & (-2))*stride // which may overflow. if (!adjust_min_trip) {
assert(old_trip_count > 1 && (old_trip_count & 1) == 0, "odd trip count for maximally unroll"); // Don't need to adjust limit for maximally unroll since trip count is even.
} elseif (loop_head->has_exact_trip_count() && init->is_Con()) { // Loop's limit is constant. Loop's init could be constant when pre-loop // become peeled iteration.
jlong init_con = init->get_int(); // We can keep old loop limit if iterations count stays the same: // old_trip_count == new_trip_count * 2 // Note: since old_trip_count >= 2 then new_trip_count >= 1 // so we also don't need to adjust zero trip test.
jlong limit_con = limit->get_int(); // (stride_con*2) not overflow since stride_con <= 8. int new_stride_con = stride_con * 2; int stride_m = new_stride_con - (stride_con > 0 ? 1 : -1);
jlong trip_count = (limit_con - init_con + stride_m)/new_stride_con; // New trip count should satisfy next conditions.
assert(trip_count > 0 && (julong)trip_count < (julong)max_juint/2, "sanity");
uint new_trip_count = (uint)trip_count;
adjust_min_trip = (old_trip_count != new_trip_count*2);
}
if (adjust_min_trip) { // Step 2: Adjust the trip limit if it is called for. // The adjustment amount is -stride. Need to make sure if the // adjustment underflows or overflows, then the main loop is skipped.
Node* cmp = loop_end->cmp_node();
assert(cmp->in(2) == limit, "sanity");
assert(opaq != NULL && opaq->in(1) == limit, "sanity");
// Verify that policy_unroll result is still valid. const TypeInt* limit_type = _igvn.type(limit)->is_int();
assert(stride_con > 0 && ((limit_type->_hi - stride_con) < limit_type->_hi) ||
stride_con < 0 && ((limit_type->_lo - stride_con) > limit_type->_lo), "sanity");
if (limit->is_Con()) { // The check in policy_unroll and the assert above guarantee // no underflow if limit is constant.
new_limit = _igvn.intcon(limit->get_int() - stride_con);
set_ctrl(new_limit, C->root());
} else { // Limit is not constant.
assert(loop_head->unrolled_count() != 1 || has_ctrl(opaq), "should have opaque for first unroll"); if ((stride_con > 0 && (java_subtract(limit_type->_lo, stride_con) < limit_type->_lo)) ||
(stride_con < 0 && (java_subtract(limit_type->_hi, stride_con) > limit_type->_hi))) { // No underflow.
new_limit = new SubINode(limit, stride);
} else { // (limit - stride) may underflow. // Clamp the adjustment value with MININT or MAXINT: // // new_limit = limit-stride // if (stride > 0) // new_limit = (limit < new_limit) ? MININT : new_limit; // else // new_limit = (limit > new_limit) ? MAXINT : new_limit; //
BoolTest::mask bt = loop_end->test_trip();
assert(bt == BoolTest::lt || bt == BoolTest::gt, "canonical test is expected");
Node* adj_max = _igvn.intcon((stride_con > 0) ? min_jint : max_jint);
set_ctrl(adj_max, C->root());
Node* old_limit = NULL;
Node* adj_limit = NULL;
Node* bol = limit->is_CMove() ? limit->in(CMoveNode::Condition) : NULL; if (loop_head->unrolled_count() > 1 &&
limit->is_CMove() && limit->Opcode() == Op_CMoveI &&
limit->in(CMoveNode::IfTrue) == adj_max &&
bol->as_Bool()->_test._test == bt &&
bol->in(1)->Opcode() == Op_CmpI &&
bol->in(1)->in(2) == limit->in(CMoveNode::IfFalse)) { // Loop was unrolled before. // Optimize the limit to avoid nested CMove: // use original limit as old limit.
old_limit = bol->in(1)->in(1); // Adjust previous adjusted limit.
adj_limit = limit->in(CMoveNode::IfFalse);
adj_limit = new SubINode(adj_limit, stride);
} else {
old_limit = limit;
adj_limit = new SubINode(limit, stride);
}
assert(old_limit != NULL && adj_limit != NULL, "");
register_new_node(adj_limit, ctrl); // adjust amount
Node* adj_cmp = new CmpINode(old_limit, adj_limit);
register_new_node(adj_cmp, ctrl);
Node* adj_bool = new BoolNode(adj_cmp, bt);
register_new_node(adj_bool, ctrl);
new_limit = new CMoveINode(adj_bool, adj_limit, adj_max, TypeInt::INT);
}
register_new_node(new_limit, ctrl);
}
assert(new_limit != NULL, ""); // Replace in loop test.
assert(loop_end->in(1)->in(1) == cmp, "sanity"); if (cmp->outcnt() == 1 && loop_end->in(1)->outcnt() == 1) { // Don't need to create new test since only one user.
_igvn.hash_delete(cmp);
cmp->set_req(2, new_limit);
} else { // Create new test since it is shared.
Node* ctrl2 = loop_end->in(0);
Node* cmp2 = cmp->clone();
cmp2->set_req(2, new_limit);
register_new_node(cmp2, ctrl2);
Node* bol2 = loop_end->in(1)->clone();
bol2->set_req(1, cmp2);
register_new_node(bol2, ctrl2);
_igvn.replace_input_of(loop_end, 1, bol2);
} // Step 3: Find the min-trip test guaranteed before a 'main' loop. // Make it a 1-trip test (means at least 2 trips).
// Guard test uses an 'opaque' node which is not shared. Hence I // can edit it's inputs directly. Hammer in the new limit for the // minimum-trip guard.
assert(opaq->outcnt() == 1, "");
_igvn.replace_input_of(opaq, 1, new_limit);
}
// Adjust max trip count. The trip count is intentionally rounded // down here (e.g. 15-> 7-> 3-> 1) because if we unwittingly over-unroll, // the main, unrolled, part of the loop will never execute as it is protected // by the min-trip test. See bug 4834191 for a case where we over-unrolled // and later determined that part of the unrolled loop was dead.
loop_head->set_trip_count(old_trip_count / 2);
// Double the count of original iterations in the unrolled loop body.
loop_head->double_unrolled_count();
// --------- // Step 4: Clone the loop body. Move it inside the loop. This loop body // represents the odd iterations; since the loop trips an even number of // times its backedge is never taken. Kill the backedge.
uint dd = dom_depth(loop_head);
clone_loop(loop, old_new, dd, IgnoreStripMined);
// Make backedges of the clone equal to backedges of the original. // Make the fall-in from the original come from the fall-out of the clone. for (DUIterator_Fast jmax, j = loop_head->fast_outs(jmax); j < jmax; j++) {
Node* phi = loop_head->fast_out(j); if (phi->is_Phi() && phi->in(0) == loop_head && phi->outcnt() > 0) {
Node *newphi = old_new[phi->_idx];
_igvn.hash_delete(phi);
_igvn.hash_delete(newphi);
// If loop is tripping an odd number of times, peel odd iteration if ((cl->trip_count() & 1) == 1) {
do_peeling(loop, old_new);
}
// Now its tripping an even number of times remaining. Double loop body. // Do not adjust pre-guards; they are not needed and do not exist. if (cl->trip_count() > 0) {
assert((cl->trip_count() & 1) == 0, "missed peeling");
do_unroll(loop, old_new, false);
}
}
void PhaseIdealLoop::mark_reductions(IdealLoopTree *loop) { if (SuperWordReductions == false) return;
Node* trip_phi = loop_head->phi(); for (DUIterator_Fast imax, i = loop_head->fast_outs(imax); i < imax; i++) {
Node* phi = loop_head->fast_out(i); if (phi->is_Phi() && phi->outcnt() > 0 && phi != trip_phi) { // For definitions which are loop inclusive and not tripcounts.
Node* def_node = phi->in(LoopNode::LoopBackControl);
if (def_node != NULL) {
Node* n_ctrl = get_ctrl(def_node); if (n_ctrl != NULL && loop->is_member(get_loop(n_ctrl))) { // Now test it to see if it fits the standard pattern for a reduction operator. int opc = def_node->Opcode(); if (opc != ReductionNode::opcode(opc, def_node->bottom_type()->basic_type())
|| opc == Op_MinD || opc == Op_MinF || opc == Op_MaxD || opc == Op_MaxF) { if (!def_node->is_reduction()) { // Not marked yet // To be a reduction, the arithmetic node must have the phi as input and provide a def to it bool ok = false; for (unsigned j = 1; j < def_node->req(); j++) {
Node* in = def_node->in(j); if (in == phi) {
ok = true; break;
}
}
// do nothing if we did not match the initial criteria if (ok == false) { continue;
}
// The result of the reduction must not be used in the loop for (DUIterator_Fast imax, i = def_node->fast_outs(imax); i < imax && ok; i++) {
Node* u = def_node->fast_out(i); if (!loop->is_member(get_loop(ctrl_or_self(u)))) { continue;
} if (u == phi) { continue;
}
ok = false;
}
// iff the uses conform if (ok) {
def_node->add_flag(Node::Flag_is_reduction);
loop_head->mark_has_reductions();
}
}
}
}
}
}
}
}
//------------------------------adjust_limit----------------------------------- // Helper function that computes new loop limit as (rc_limit-offset)/scale
Node* PhaseIdealLoop::adjust_limit(bool is_positive_stride, Node* scale, Node* offset, Node* rc_limit, Node* old_limit, Node* pre_ctrl, bool round) {
Node* sub = new SubLNode(rc_limit, offset);
register_new_node(sub, pre_ctrl);
Node* limit = new DivLNode(NULL, sub, scale);
register_new_node(limit, pre_ctrl);
// When the absolute value of scale is greater than one, the division // may round limit down/up, so add/sub one to/from the limit. if (round) {
limit = new AddLNode(limit, _igvn.longcon(is_positive_stride ? -1 : 1));
register_new_node(limit, pre_ctrl);
}
// Clamp the limit to handle integer under-/overflows by using long values. // We only convert the limit back to int when we handled under-/overflows. // Note that all values are longs in the following computations. // When reducing the limit, clamp to [min_jint, old_limit]: // INT(MINL(old_limit, MAXL(limit, min_jint))) // - integer underflow of limit: MAXL chooses min_jint. // - integer overflow of limit: MINL chooses old_limit (<= MAX_INT < limit) // When increasing the limit, clamp to [old_limit, max_jint]: // INT(MAXL(old_limit, MINL(limit, max_jint))) // - integer overflow of limit: MINL chooses max_jint. // - integer underflow of limit: MAXL chooses old_limit (>= MIN_INT > limit) // INT() is finally converting the limit back to an integer value.
// We use CMove nodes to implement long versions of min/max (MINL/MAXL). // We use helper methods for inner MINL/MAXL which return CMoveL nodes to keep a long value for the outer MINL/MAXL comparison:
Node* inner_result_long; if (is_positive_stride) {
inner_result_long = MaxNode::signed_max(limit, _igvn.longcon(min_jint), TypeLong::LONG, _igvn);
} else {
inner_result_long = MaxNode::signed_min(limit, _igvn.longcon(max_jint), TypeLong::LONG, _igvn);
}
set_subtree_ctrl(inner_result_long, false);
// Outer MINL/MAXL: // The comparison is done with long values but the result is the converted back to int by using CmovI.
Node* old_limit_long = new ConvI2LNode(old_limit);
register_new_node(old_limit_long, pre_ctrl);
Node* cmp = new CmpLNode(old_limit_long, limit);
register_new_node(cmp, pre_ctrl);
Node* bol = new BoolNode(cmp, is_positive_stride ? BoolTest::gt : BoolTest::lt);
register_new_node(bol, pre_ctrl);
Node* inner_result_int = new ConvL2INode(inner_result_long); // Could under-/overflow but that's fine as comparison was done with CmpL
register_new_node(inner_result_int, pre_ctrl);
limit = new CMoveINode(bol, old_limit, inner_result_int, TypeInt::INT);
register_new_node(limit, pre_ctrl); return limit;
}
//------------------------------add_constraint--------------------------------- // Constrain the main loop iterations so the conditions: // low_limit <= scale_con*I + offset < upper_limit // always hold true. That is, either increase the number of iterations in the // pre-loop or reduce the number of iterations in the main-loop until the condition // holds true in the main-loop. Stride, scale, offset and limit are all loop // invariant. Further, stride and scale are constants (offset and limit often are). void PhaseIdealLoop::add_constraint(jlong stride_con, jlong scale_con, Node* offset, Node* low_limit, Node* upper_limit, Node* pre_ctrl, Node** pre_limit, Node** main_limit) {
assert(_igvn.type(offset)->isa_long() != NULL && _igvn.type(low_limit)->isa_long() != NULL &&
_igvn.type(upper_limit)->isa_long() != NULL, "arguments should be long values");
// For a positive stride, we need to reduce the main-loop limit and // increase the pre-loop limit. This is reversed for a negative stride. bool is_positive_stride = (stride_con > 0);
// If the absolute scale value is greater one, division in 'adjust_limit' may require // rounding. Make sure the ABS method correctly handles min_jint. // Only do this for the pre-loop, one less iteration of the main loop doesn't hurt. bool round = ABS(scale_con) > 1;
if ((stride_con^scale_con) >= 0) { // Use XOR to avoid overflow // Positive stride*scale: the affine function is increasing, // the pre-loop checks for underflow and the post-loop for overflow.
// The overflow limit: scale*I+offset < upper_limit // For the main-loop limit compute: // ( if (scale > 0) /* and stride > 0 */ // I < (upper_limit-offset)/scale // else /* scale < 0 and stride < 0 */ // I > (upper_limit-offset)/scale // )
*main_limit = adjust_limit(is_positive_stride, scale, offset, upper_limit, *main_limit, pre_ctrl, false);
// The underflow limit: low_limit <= scale*I+offset // For the pre-loop limit compute: // NOT(scale*I+offset >= low_limit) // scale*I+offset < low_limit // ( if (scale > 0) /* and stride > 0 */ // I < (low_limit-offset)/scale // else /* scale < 0 and stride < 0 */ // I > (low_limit-offset)/scale // )
*pre_limit = adjust_limit(!is_positive_stride, scale, offset, low_limit, *pre_limit, pre_ctrl, round);
} else { // Negative stride*scale: the affine function is decreasing, // the pre-loop checks for overflow and the post-loop for underflow.
// The overflow limit: scale*I+offset < upper_limit // For the pre-loop limit compute: // NOT(scale*I+offset < upper_limit) // scale*I+offset >= upper_limit // scale*I+offset+1 > upper_limit // ( if (scale < 0) /* and stride > 0 */ // I < (upper_limit-(offset+1))/scale // else /* scale > 0 and stride < 0 */ // I > (upper_limit-(offset+1))/scale // )
Node* one = _igvn.longcon(1);
set_ctrl(one, C->root());
Node* plus_one = new AddLNode(offset, one);
register_new_node(plus_one, pre_ctrl);
*pre_limit = adjust_limit(!is_positive_stride, scale, plus_one, upper_limit, *pre_limit, pre_ctrl, round);
// The underflow limit: low_limit <= scale*I+offset // For the main-loop limit compute: // scale*I+offset+1 > low_limit // ( if (scale < 0) /* and stride > 0 */ // I < (low_limit-(offset+1))/scale // else /* scale > 0 and stride < 0 */ // I > (low_limit-(offset+1))/scale // )
*main_limit = adjust_limit(is_positive_stride, scale, plus_one, low_limit, *main_limit, pre_ctrl, false);
}
}
//----------------------------------is_iv------------------------------------ // Return true if exp is the value (of type bt) of the given induction var. // This grammar of cases is recognized, where X is I|L according to bt: // VIV[iv] = iv | (CastXX VIV[iv]) | (ConvI2X VIV[iv]) bool PhaseIdealLoop::is_iv(Node* exp, Node* iv, BasicType bt) {
exp = exp->uncast(); if (exp == iv && iv->bottom_type()->isa_integer(bt)) { returntrue;
}
//------------------------------is_scaled_iv--------------------------------- // Return true if exp is a constant times the given induction var (of type bt). // The multiplication is either done in full precision (exactly of type bt), // or else bt is T_LONG but iv is scaled using 32-bit arithmetic followed by a ConvI2L. // This grammar of cases is recognized, where X is I|L according to bt: // SIV[iv] = VIV[iv] | (CastXX SIV[iv]) // | (MulX VIV[iv] ConX) | (MulX ConX VIV[iv]) // | (LShiftX VIV[iv] ConI) // | (ConvI2L SIV[iv]) -- a "short-scale" can occur here; note recursion // | (SubX 0 SIV[iv]) -- same as MulX(iv, -scale); note recursion // | (AddX SIV[iv] SIV[iv]) -- sum of two scaled iv; note recursion // | (SubX SIV[iv] SIV[iv]) -- difference of two scaled iv; note recursion // VIV[iv] = [either iv or its value converted; see is_iv() above] // On success, the constant scale value is stored back to *p_scale. // The value (*p_short_scale) reports if such a ConvI2L conversion was present. bool PhaseIdealLoop::is_scaled_iv(Node* exp, Node* iv, BasicType bt, jlong* p_scale, bool* p_short_scale, int depth) {
BasicType exp_bt = bt;
exp = exp->uncast(); //strip casts
assert(exp_bt == T_INT || exp_bt == T_LONG, "unexpected int type"); if (is_iv(exp, iv, exp_bt)) { if (p_scale != NULL) {
*p_scale = 1;
} if (p_short_scale != NULL) {
*p_short_scale = false;
} returntrue;
} if (exp_bt == T_LONG && iv->bottom_type()->isa_int() && exp->Opcode() == Op_ConvI2L) {
exp = exp->in(1);
exp_bt = T_INT;
} int opc = exp->Opcode(); int which = 0; // this is which subexpression we find the iv in // Can't use is_Mul() here as it's true for AndI and AndL if (opc == Op_Mul(exp_bt)) { if ((is_iv(exp->in(which = 1), iv, exp_bt) && exp->in(2)->is_Con()) ||
(is_iv(exp->in(which = 2), iv, exp_bt) && exp->in(1)->is_Con())) {
Node* factor = exp->in(which == 1 ? 2 : 1); // the other argument
jlong scale = factor->find_integer_as_long(exp_bt, 0); if (scale == 0) { returnfalse; // might be top
} if (p_scale != NULL) {
*p_scale = scale;
} if (p_short_scale != NULL) { // (ConvI2L (MulI iv K)) can be 64-bit linear if iv is kept small enough...
*p_short_scale = (exp_bt != bt && scale != 1);
} returntrue;
}
} elseif (opc == Op_LShift(exp_bt)) { if (is_iv(exp->in(1), iv, exp_bt) && exp->in(2)->is_Con()) {
jint shift_amount = exp->in(2)->find_int_con(min_jint); if (shift_amount == min_jint) { returnfalse; // might be top
}
jlong scale; if (exp_bt == T_INT) {
scale = java_shift_left((jint)1, (juint)shift_amount);
} elseif (exp_bt == T_LONG) {
scale = java_shift_left((jlong)1, (julong)shift_amount);
} if (p_scale != NULL) {
*p_scale = scale;
} if (p_short_scale != NULL) { // (ConvI2L (MulI iv K)) can be 64-bit linear if iv is kept small enough...
*p_short_scale = (exp_bt != bt && scale != 1);
} returntrue;
}
} elseif (opc == Op_Add(exp_bt)) {
jlong scale_l = 0;
jlong scale_r = 0; bool short_scale_l = false; bool short_scale_r = false; if (depth == 0 &&
is_scaled_iv(exp->in(1), iv, exp_bt, &scale_l, &short_scale_l, depth + 1) &&
is_scaled_iv(exp->in(2), iv, exp_bt, &scale_r, &short_scale_r, depth + 1)) { // AddX(iv*K1, iv*K2) => iv*(K1+K2)
jlong scale_sum = java_add(scale_l, scale_r); if (scale_sum > max_signed_integer(exp_bt) || scale_sum <= min_signed_integer(exp_bt)) { // This logic is shared by int and long. For int, the result may overflow // as we use jlong to compute so do the check here. Long result may also // overflow but that's fine because result wraps. returnfalse;
} if (p_scale != NULL) {
*p_scale = scale_sum;
} if (p_short_scale != NULL) {
*p_short_scale = short_scale_l && short_scale_r;
} returntrue;
}
} elseif (opc == Op_Sub(exp_bt)) { if (exp->in(1)->find_integer_as_long(exp_bt, -1) == 0) {
jlong scale = 0; if (depth == 0 && is_scaled_iv(exp->in(2), iv, exp_bt, &scale, p_short_scale, depth + 1)) { // SubX(0, iv*K) => iv*(-K) if (scale == min_signed_integer(exp_bt)) { // This should work even if -K overflows, but let's not. returnfalse;
}
scale = java_multiply(scale, (jlong)-1); if (p_scale != NULL) {
*p_scale = scale;
} if (p_short_scale != NULL) { // (ConvI2L (MulI iv K)) can be 64-bit linear if iv is kept small enough...
*p_short_scale = *p_short_scale || (exp_bt != bt && scale != 1);
} returntrue;
}
} else {
jlong scale_l = 0;
jlong scale_r = 0; bool short_scale_l = false; bool short_scale_r = false; if (depth == 0 &&
is_scaled_iv(exp->in(1), iv, exp_bt, &scale_l, &short_scale_l, depth + 1) &&
is_scaled_iv(exp->in(2), iv, exp_bt, &scale_r, &short_scale_r, depth + 1)) { // SubX(iv*K1, iv*K2) => iv*(K1-K2)
jlong scale_diff = java_subtract(scale_l, scale_r); if (scale_diff > max_signed_integer(exp_bt) || scale_diff <= min_signed_integer(exp_bt)) { // This logic is shared by int and long. For int, the result may // overflow as we use jlong to compute so do the check here. Long // result may also overflow but that's fine because result wraps. returnfalse;
} if (p_scale != NULL) {
*p_scale = scale_diff;
} if (p_short_scale != NULL) {
*p_short_scale = short_scale_l && short_scale_r;
} returntrue;
}
}
} // We could also recognize (iv*K1)*K2, even with overflow, but let's not. returnfalse;
}
//-------------------------is_scaled_iv_plus_offset-------------------------- // Return true if exp is a simple linear transform of the given induction var. // The scale must be constant and the addition tree (if any) must be simple. // This grammar of cases is recognized, where X is I|L according to bt: // // OIV[iv] = SIV[iv] | (CastXX OIV[iv]) // | (AddX SIV[iv] E) | (AddX E SIV[iv]) // | (SubX SIV[iv] E) | (SubX E SIV[iv]) // SSIV[iv] = (ConvI2X SIV[iv]) -- a "short scale" might occur here // SIV[iv] = [a possibly scaled value of iv; see is_scaled_iv() above] // // On success, the constant scale value is stored back to *p_scale unless null. // Likewise, the addend (perhaps a synthetic AddX node) is stored to *p_offset. // Also, (*p_short_scale) reports if a ConvI2L conversion was seen after a MulI, // meaning bt is T_LONG but iv was scaled using 32-bit arithmetic. // To avoid looping, the match is depth-limited, and so may fail to match the grammar to complex expressions. bool PhaseIdealLoop::is_scaled_iv_plus_offset(Node* exp, Node* iv, BasicType bt, jlong* p_scale, Node** p_offset, bool* p_short_scale, int depth) {
assert(bt == T_INT || bt == T_LONG, "unexpected int type");
jlong scale = 0; // to catch result from is_scaled_iv()
BasicType exp_bt = bt;
exp = exp->uncast(); if (is_scaled_iv(exp, iv, exp_bt, &scale, p_short_scale)) { if (p_scale != NULL) {
*p_scale = scale;
} if (p_offset != NULL) {
Node *zero = _igvn.zerocon(bt);
set_ctrl(zero, C->root());
*p_offset = zero;
} returntrue;
} if (exp_bt != bt) { // We would now be matching inputs like (ConvI2L exp:(AddI (MulI iv S) E)). // It's hard to make 32-bit arithmetic linear if it overflows. Although we do // cope with overflowing multiplication by S, it would be even more work to // handle overflowing addition of E. So we bail out here on ConvI2L input. returnfalse;
} int opc = exp->Opcode(); int which = 0; // this is which subexpression we find the iv in
Node* offset = NULL; if (opc == Op_Add(exp_bt)) { // Check for a scaled IV in (AddX (MulX iv S) E) or (AddX E (MulX iv S)). if (is_scaled_iv(exp->in(which = 1), iv, bt, &scale, p_short_scale) ||
is_scaled_iv(exp->in(which = 2), iv, bt, &scale, p_short_scale)) {
offset = exp->in(which == 1 ? 2 : 1); // the other argument if (p_scale != NULL) {
*p_scale = scale;
} if (p_offset != NULL) {
*p_offset = offset;
} returntrue;
} // Check for more addends, like (AddX (AddX (MulX iv S) E1) E2), etc. if (is_scaled_iv_plus_extra_offset(exp->in(1), exp->in(2), iv, bt, p_scale, p_offset, p_short_scale, depth) ||
is_scaled_iv_plus_extra_offset(exp->in(2), exp->in(1), iv, bt, p_scale, p_offset, p_short_scale, depth)) { returntrue;
}
} elseif (opc == Op_Sub(exp_bt)) { if (is_scaled_iv(exp->in(which = 1), iv, bt, &scale, p_short_scale) ||
is_scaled_iv(exp->in(which = 2), iv, bt, &scale, p_short_scale)) { // Match (SubX SIV[iv] E) as if (AddX SIV[iv] (SubX 0 E)), and // match (SubX E SIV[iv]) as if (AddX E (SubX 0 SIV[iv])).
offset = exp->in(which == 1 ? 2 : 1); // the other argument if (which == 2) { // We can't handle a scale of min_jint (or min_jlong) here as -1 * min_jint = min_jint if (scale == min_signed_integer(bt)) { returnfalse; // cannot negate the scale of the iv
}
scale = java_multiply(scale, (jlong)-1);
} if (p_scale != NULL) {
*p_scale = scale;
} if (p_offset != NULL) { if (which == 1) { // must negate the extracted offset
Node *zero = _igvn.integercon(0, exp_bt);
set_ctrl(zero, C->root());
Node *ctrl_off = get_ctrl(offset);
offset = SubNode::make(zero, offset, exp_bt);
register_new_node(offset, ctrl_off);
}
*p_offset = offset;
} returntrue;
}
} returnfalse;
}
// Helper for is_scaled_iv_plus_offset(), not called separately. // The caller encountered (AddX exp1 offset3) or (AddX offset3 exp1). // Here, exp1 is inspected to see if it is a simple linear transform of iv. // If so, the offset3 is combined with any other offset2 from inside exp1. bool PhaseIdealLoop::is_scaled_iv_plus_extra_offset(Node* exp1, Node* offset3, Node* iv,
BasicType bt,
jlong* p_scale, Node** p_offset, bool* p_short_scale, int depth) { // By the time we reach here, it is unlikely that exp1 is a simple iv*K. // If is a linear iv transform, it is probably an add or subtract. // Let's collect the internal offset2 from it.
Node* offset2 = NULL; if (offset3->is_Con() &&
depth < 2 &&
is_scaled_iv_plus_offset(exp1, iv, bt, p_scale,
&offset2, p_short_scale, depth+1)) { if (p_offset != NULL) {
Node* ctrl_off2 = get_ctrl(offset2);
Node* offset = AddNode::make(offset2, offset3, bt);
register_new_node(offset, ctrl_off2);
*p_offset = offset;
} returntrue;
} returnfalse;
}
// 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, Node* value) { bool overflow = false;
BoolNode* bol = rc_predicate(loop, predicate_proj, scale_con, offset, value, NULL, stride_con, limit, (stride_con > 0) != (scale_con > 0), overflow, false);
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, opaque_bol, PROB_MAX, COUNT_UNKNOWN);
} else {
new_iff = new RangeCheckNode(predicate_proj, opaque_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, "range check predicate failed which is impossible");
register_control(halt, _ltree_root, iffalse);
_igvn.add_input_to(C->root(), 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) { #ifndef PRODUCT if (PrintOpto && VerifyLoopOptimizations) {
tty->print("Range Check Elimination ");
loop->dump_head();
} elseif (TraceLoopOpts) {
tty->print("RangeCheck ");
loop->dump_head();
} #endif
assert(RangeCheckElimination, "");
CountedLoopNode *cl = loop->_head->as_CountedLoop(); // If we fail before trying to eliminate range checks, set multiversion state int closed_range_checks = 1;
// protect against stride not being a constant if (!cl->stride_is_con()) { return closed_range_checks;
} // Find the trip counter; we are iteration splitting based on it
Node *trip_counter = cl->phi(); // Find the main loop limit; we will trim it's iterations // to not ever trip end tests
Node *main_limit = cl->limit();
// Check graph shape. Cannot optimize a loop if zero-trip // Opaque1 node is optimized away and then another round // of loop opts attempted. if (cl->is_canonical_loop_entry() == NULL) { return closed_range_checks;
}
// Need to find the main-loop zero-trip guard
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");
// Find the pre-loop limit; we will expand its iterations to // not ever trip low tests.
Node *p_f = iffm->in(0); // pre loop may have been optimized out if (p_f->Opcode() != Op_IfFalse) { return closed_range_checks;
}
CountedLoopEndNode *pre_end = p_f->in(0)->as_CountedLoopEnd();
assert(pre_end->loopnode()->is_pre_loop(), "");
Node *pre_opaq1 = pre_end->limit(); // Occasionally it's possible for a pre-loop Opaque1 node to be // optimized away and then another round of loop opts attempted. // We can not optimize this particular loop in that case. if (pre_opaq1->Opcode() != Op_Opaque1) { return closed_range_checks;
}
Opaque1Node *pre_opaq = (Opaque1Node*)pre_opaq1;
Node *pre_limit = pre_opaq->in(1);
// Where do we put new limit calculations
Node *pre_ctrl = pre_end->loopnode()->in(LoopNode::EntryControl);
// Ensure the original loop limit is available from the // pre-loop Opaque1 node.
Node *orig_limit = pre_opaq->original_loop_limit(); if (orig_limit == NULL || _igvn.type(orig_limit) == Type::TOP) { return closed_range_checks;
} // Must know if its a count-up or count-down loop
int stride_con = cl->stride_con();
Node* zero = _igvn.longcon(0);
Node* one = _igvn.longcon(1); // Use symmetrical int range [-max_jint,max_jint]
Node* mini = _igvn.longcon(-max_jint);
set_ctrl(zero, C->root());
set_ctrl(one, C->root());
set_ctrl(mini, C->root());
// Count number of range checks and reduce by load range limits, if zero, // the loop is in canonical form to multiversion.
closed_range_checks = 0;
// 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]; if (iff->Opcode() == Op_If ||
iff->Opcode() == Op_RangeCheck) { // Test? // Test is an IfNode, has 2 projections. If BOTH are in the loop // we need loop unswitching instead of iteration splitting.
closed_range_checks++;
Node *exit = loop->is_loop_exit(iff); if (!exit) continue; int flip = (exit->Opcode() == Op_IfTrue) ? 1 : 0;
// Get boolean condition to test
Node *i1 = iff->in(1); if (!i1->is_Bool()) continue;
BoolNode *bol = i1->as_Bool();
BoolTest b_test = bol->_test; // Flip sense of test if exit condition is flipped if (flip) {
b_test = b_test.negate();
} // Get compare
Node *cmp = bol->in(1);
// Look for trip_counter + offset vs limit
Node *rc_exp = cmp->in(1);
Node *limit = cmp->in(2); int scale_con= 1; // Assume trip counter not scaled
Node *limit_c = get_ctrl(limit); if (loop->is_member(get_loop(limit_c))) { // Compare might have operands swapped; commute them
b_test = b_test.commute();
rc_exp = cmp->in(2);
limit = cmp->in(1);
limit_c = get_ctrl(limit); if (loop->is_member(get_loop(limit_c))) { continue; // Both inputs are loop varying; cannot RCE
}
} // Here we know 'limit' is loop invariant
// '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 (is_dominator(ctrl, limit_c)) { continue; // Don't rce this check but continue looking for other candidates.
}
// Check for scaled induction variable plus an offset
Node *offset = NULL;
if (!is_scaled_iv_plus_offset(rc_exp, trip_counter, &scale_con, &offset)) { continue;
}
Node *offset_c = get_ctrl(offset); if (loop->is_member(get_loop(offset_c))) { continue; // Offset is not really loop invariant
} // Here we know 'offset' is loop invariant.
// As above for the 'limit', the 'offset' maybe pinned below the // zero trip test. if (is_dominator(ctrl, offset_c)) { continue; // Don't rce this check but continue looking for other candidates.
} #ifdef ASSERT if (TraceRangeLimitCheck) {
tty->print_cr("RC bool node%s", flip ? " flipped:" : ":");
bol->dump(2);
} #endif // At this point we have the expression as: // scale_con * trip_counter + offset :: limit // where scale_con, offset and limit are loop invariant. Trip_counter // monotonically increases by stride_con, a constant. Both (or either) // stride_con and scale_con can be negative which will flip about the // sense of the test.
// Perform the limit computations in jlong to avoid overflow
jlong lscale_con = scale_con;
Node* int_offset = offset;
offset = new ConvI2LNode(offset);
register_new_node(offset, pre_ctrl);
Node* int_limit = limit;
limit = new ConvI2LNode(limit);
register_new_node(limit, pre_ctrl);
// Adjust pre and main loop limits to guard the correct iteration set if (cmp->Opcode() == Op_CmpU) { // Unsigned compare is really 2 tests if (b_test._test == BoolTest::lt) { // Range checks always use lt // The underflow and overflow limits: 0 <= scale*I+offset < limit
add_constraint(stride_con, lscale_con, offset, zero, limit, pre_ctrl, &pre_limit, &main_limit);
Node* init = cl->init_trip();
Node* opaque_init = new OpaqueLoopInitNode(C, init);
register_new_node(opaque_init, predicate_proj);
// predicate on first value of first iteration
predicate_proj = add_range_check_predicate(loop, cl, predicate_proj, scale_con, int_offset, int_limit, stride_con, init);
assert(!skeleton_predicate_has_opaque(predicate_proj->in(0)->as_If()), "unexpected");
// template predicate so it can be updated on next unrolling
predicate_proj = add_range_check_predicate(loop, cl, predicate_proj, scale_con, int_offset, int_limit, stride_con, opaque_init);
assert(skeleton_predicate_has_opaque(predicate_proj->in(0)->as_If()), "unexpected");
Node* opaque_stride = new OpaqueLoopStrideNode(C, cl->stride());
register_new_node(opaque_stride, predicate_proj);
Node* max_value = new SubINode(opaque_stride, cl->stride());
register_new_node(max_value, predicate_proj);
max_value = new AddINode(opaque_init, max_value);
register_new_node(max_value, predicate_proj); // init + (current stride - initial stride) is within the loop so narrow its type by leveraging the type of the iv Phi
max_value = new CastIINode(max_value, loop->_head->as_CountedLoop()->phi()->bottom_type());
register_new_node(max_value, predicate_proj);
predicate_proj = add_range_check_predicate(loop, cl, predicate_proj, scale_con, int_offset, int_limit, stride_con, max_value);
assert(skeleton_predicate_has_opaque(predicate_proj->in(0)->as_If()), "unexpected");
} else { if (PrintOpto) {
tty->print_cr("missed RCE opportunity");
} continue; // In release mode, ignore it
}
} else { // Otherwise work on normal compares switch(b_test._test) { case BoolTest::gt: // Fall into GE case case BoolTest::ge: // Convert (I*scale+offset) >= Limit to (I*(-scale)+(-offset)) <= -Limit
lscale_con = -lscale_con;
offset = new SubLNode(zero, offset);
register_new_node(offset, pre_ctrl);
limit = new SubLNode(zero, limit);
register_new_node(limit, pre_ctrl); // Fall into LE case case BoolTest::le: if (b_test._test != BoolTest::gt) { // Convert X <= Y to X < Y+1
limit = new AddLNode(limit, one);
register_new_node(limit, pre_ctrl);
} // Fall into LT case case BoolTest::lt: // The underflow and overflow limits: MIN_INT <= scale*I+offset < limit // Note: (MIN_INT+1 == -MAX_INT) is used instead of MIN_INT here // to avoid problem with scale == -1: MIN_INT/(-1) == MIN_INT.
add_constraint(stride_con, lscale_con, offset, mini, limit, pre_ctrl, &pre_limit, &main_limit); break; default: if (PrintOpto) {
tty->print_cr("missed RCE opportunity");
} continue; // Unhandled case
}
}
// Kill the eliminated test
C->set_major_progress();
Node *kill_con = _igvn.intcon(1-flip);
set_ctrl(kill_con, C->root());
_igvn.replace_input_of(iff, 1, kill_con); // Find surviving projection
assert(iff->is_If(), "");
ProjNode* dp = ((IfNode*)iff)->proj_out(1-flip); // Find loads off the surviving projection; remove their control edge for (DUIterator_Fast imax, i = dp->fast_outs(imax); i < imax; i++) {
Node* cd = dp->fast_out(i); // Control-dependent node if (cd->is_Load() && cd->depends_only_on_test()) { // Loads can now float around in the loop // Allow the load to float around in the loop, or before it // but NOT before the pre-loop.
_igvn.replace_input_of(cd, 0, ctrl); // ctrl, not NULL
--i;
--imax;
}
} if (int_limit->Opcode() == Op_LoadRange) {
closed_range_checks--;
}
} // 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 (pre_limit != orig_limit) { // Computed pre-loop limit can be outside of loop iterations range.
pre_limit = (stride_con > 0) ? (Node*)new MinINode(pre_limit, orig_limit)
: (Node*)new MaxINode(pre_limit, orig_limit);
register_new_node(pre_limit, pre_ctrl);
}
_igvn.replace_input_of(pre_opaq, 1, pre_limit);
// Note:: we are making the main loop limit no longer precise; // need to round up based on stride.
cl->set_nonexact_trip_count();
Node *main_cle = cl->loopexit();
Node *main_bol = main_cle->in(1); // Hacking loop bounds; need private copies of exit test if (main_bol->outcnt() > 1) { // BoolNode shared?
main_bol = main_bol->clone(); // Clone a private BoolNode
register_new_node(main_bol, main_cle->in(0));
_igvn.replace_input_of(main_cle, 1, main_bol);
}
Node *main_cmp = main_bol->in(1); if (main_cmp->outcnt() > 1) { // CmpNode shared?
main_cmp = main_cmp->clone(); // Clone a private CmpNode
register_new_node(main_cmp, main_cle->in(0));
_igvn.replace_input_of(main_bol, 1, main_cmp);
}
assert(main_limit == cl->limit() || get_ctrl(main_limit) == pre_ctrl, "wrong control for added limit"); const TypeInt* orig_limit_t = _igvn.type(orig_limit)->is_int(); bool upward = cl->stride_con() > 0; // The new loop limit is <= (for an upward loop) >= (for a downward loop) than the orig limit. // The expression that computes the new limit may be too complicated and the computed type of the new limit // may be too pessimistic. A CastII here guarantees it's not lost.
main_limit = new CastIINode(main_limit, TypeInt::make(upward ? min_jint : orig_limit_t->_lo,
upward ? orig_limit_t->_hi : max_jint, Type::WidenMax));
main_limit->init_req(0, pre_ctrl);
register_new_node(main_limit, pre_ctrl); // Hack the now-private loop bounds
_igvn.replace_input_of(main_cmp, 2, main_limit); // The OpaqueNode is unshared by design
assert(opqzm->outcnt() == 1, "cannot hack shared node");
_igvn.replace_input_of(opqzm, 1, main_limit);
return closed_range_checks;
}
//------------------------------has_range_checks------------------------------- // Check to see if RCE cleaned the current loop of range-checks. void PhaseIdealLoop::has_range_checks(IdealLoopTree *loop) {
assert(RangeCheckElimination, "");
// skip if not a counted loop if (!loop->is_counted()) return;
// skip this loop if it is already checked if (cl->has_been_range_checked()) return;
// Now check for existence of range checks for (uint i = 0; i < loop->_body.size(); i++) {
Node *iff = loop->_body[i]; int iff_opc = iff->Opcode(); if (iff_opc == Op_If || iff_opc == Op_RangeCheck) {
cl->mark_has_range_checks(); break;
}
}
cl->set_has_been_range_checked();
}
//-------------------------multi_version_post_loops---------------------------- // Check the range checks that remain, if simple, use the bounds to guard // which version to a post loop we execute, one with range checks or one without bool PhaseIdealLoop::multi_version_post_loops(IdealLoopTree *rce_loop, IdealLoopTree *legacy_loop) { bool multi_version_succeeded = false;
assert(RangeCheckElimination, "");
CountedLoopNode *legacy_cl = legacy_loop->_head->as_CountedLoop();
assert(legacy_cl->is_post_loop(), "");
// Check for existence of range checks using the unique instance to make a guard with
Unique_Node_List worklist; for (uint i = 0; i < legacy_loop->_body.size(); i++) {
Node *iff = legacy_loop->_body[i]; int iff_opc = iff->Opcode(); if (iff_opc == Op_If || iff_opc == Op_RangeCheck) {
worklist.push(iff);
}
}
// Find RCE'd post loop so that we can stage its guard. if (legacy_cl->is_canonical_loop_entry() == NULL) { return multi_version_succeeded;
}
Node* ctrl = legacy_cl->in(LoopNode::EntryControl);
Node* iffm = ctrl->in(0);
// Now we test that both the post loops are connected
Node* post_loop_region = iffm->in(0); if (post_loop_region == NULL) return multi_version_succeeded; if (!post_loop_region->is_Region()) return multi_version_succeeded;
Node* covering_region = post_loop_region->in(RegionNode::Control+1); if (covering_region == NULL) return multi_version_succeeded; if (!covering_region->is_Region()) return multi_version_succeeded;
Node* p_f = covering_region->in(RegionNode::Control); if (p_f == NULL) return multi_version_succeeded; if (!p_f->is_IfFalse()) return multi_version_succeeded; if (!p_f->in(0)->is_CountedLoopEnd()) return multi_version_succeeded;
CountedLoopEndNode* rce_loop_end = p_f->in(0)->as_CountedLoopEnd(); if (rce_loop_end == NULL) return multi_version_succeeded;
CountedLoopNode* rce_cl = rce_loop_end->loopnode(); if (rce_cl == NULL || !rce_cl->is_post_loop()) return multi_version_succeeded;
CountedLoopNode *known_rce_cl = rce_loop->_head->as_CountedLoop(); if (rce_cl != known_rce_cl) return multi_version_succeeded;
// Then we fetch the cover entry test
ctrl = rce_cl->in(LoopNode::EntryControl); if (!ctrl->is_IfTrue() && !ctrl->is_IfFalse()) return multi_version_succeeded;
#ifndef PRODUCT if (TraceLoopOpts) {
tty->print("PostMultiVersion\n");
rce_loop->dump_head();
legacy_loop->dump_head();
} #endif
// Now fetch the limit we want to compare against
Node *limit = rce_cl->limit(); bool first_time = true;
// If we got this far, we identified the post loop which has been RCE'd and // we have a work list. Now we will try to transform the if guard to cause // the loop pair to be multi version executed with the determination left to runtime // or the optimizer if full information is known about the given arrays at compile time.
Node *last_min = NULL;
multi_version_succeeded = true; while (worklist.size()) {
Node* rc_iffm = worklist.pop(); if (rc_iffm->is_If()) {
Node *rc_bolzm = rc_iffm->in(1); if (rc_bolzm->is_Bool()) {
Node *rc_cmpzm = rc_bolzm->in(1); if (rc_cmpzm->is_Cmp()) {
Node *rc_left = rc_cmpzm->in(2); if (rc_left->Opcode() != Op_LoadRange) {
multi_version_succeeded = false; break;
} if (first_time) {
last_min = rc_left;
first_time = false;
} else {
Node *cur_min = new MinINode(last_min, rc_left);
last_min = cur_min;
_igvn.register_new_node_with_optimizer(last_min);
}
}
}
}
}
// All we have to do is update the limit of the rce loop // with the min of our expression and the current limit. // We will use this expression to replace the current limit. if (last_min && multi_version_succeeded) {
Node *cur_min = new MinINode(last_min, limit);
_igvn.register_new_node_with_optimizer(cur_min);
Node *cmp_node = rce_loop_end->cmp_node();
_igvn.replace_input_of(cmp_node, 2, cur_min);
set_ctrl(cur_min, ctrl);
set_loop(cur_min, rce_loop->_parent);
//-------------------------poison_rce_post_loop-------------------------------- // Causes the rce'd post loop to be optimized away if multiversioning fails void PhaseIdealLoop::poison_rce_post_loop(IdealLoopTree *rce_loop) {
CountedLoopNode *rce_cl = rce_loop->_head->as_CountedLoop();
Node* ctrl = rce_cl->in(LoopNode::EntryControl); if (ctrl->is_IfTrue() || ctrl->is_IfFalse()) {
Node* iffm = ctrl->in(0); if (iffm->is_If()) {
Node* cur_bool = iffm->in(1); if (cur_bool->is_Bool()) {
Node* cur_cmp = cur_bool->in(1); if (cur_cmp->is_Cmp()) {
BoolTest::mask new_test = BoolTest::gt;
BoolNode *new_bool = new BoolNode(cur_cmp, new_test);
_igvn.replace_node(cur_bool, new_bool);
_igvn._worklist.push(new_bool);
Node* left_op = cur_cmp->in(1);
_igvn.replace_input_of(cur_cmp, 2, left_op);
C->set_major_progress();
}
}
}
}
}
//------------------------------DCE_loop_body---------------------------------- // Remove simplistic dead code from loop body void IdealLoopTree::DCE_loop_body() { for (uint i = 0; i < _body.size(); i++) { if (_body.at(i)->outcnt() == 0) {
_body.map(i, _body.pop());
i--; // Ensure we revisit the updated index.
}
}
}
//------------------------------adjust_loop_exit_prob-------------------------- // Look for loop-exit tests with the 50/50 (or worse) guesses from the parsing stage. // Replace with a 1-in-10 exit guess. void IdealLoopTree::adjust_loop_exit_prob(PhaseIdealLoop *phase) {
Node *test = tail(); while (test != _head) {
uint top = test->Opcode(); if (top == Op_IfTrue || top == Op_IfFalse) { int test_con = ((ProjNode*)test)->_con;
assert(top == (uint)(test_con? Op_IfTrue: Op_IfFalse), "sanity");
IfNode *iff = test->in(0)->as_If(); if (iff->outcnt() == 2) { // Ignore dead tests
Node *bol = iff->in(1); if (bol && bol->req() > 1 && bol->in(1) &&
((bol->in(1)->Opcode() == Op_CompareAndExchangeB) ||
(bol->in(1)->Opcode() == Op_CompareAndExchangeS) ||
(bol->in(1)->Opcode() == Op_CompareAndExchangeI) ||
(bol->in(1)->Opcode() == Op_CompareAndExchangeL) ||
(bol->in(1)->Opcode() == Op_CompareAndExchangeP) ||
(bol->in(1)->Opcode() == Op_CompareAndExchangeN) ||
(bol->in(1)->Opcode() == Op_WeakCompareAndSwapB) ||
(bol->in(1)->Opcode() == Op_WeakCompareAndSwapS) ||
(bol->in(1)->Opcode() == Op_WeakCompareAndSwapI) ||
(bol->in(1)->Opcode() == Op_WeakCompareAndSwapL) ||
(bol->in(1)->Opcode() == Op_WeakCompareAndSwapP) ||
(bol->in(1)->Opcode() == Op_WeakCompareAndSwapN) ||
(bol->in(1)->Opcode() == Op_CompareAndSwapB) ||
(bol->in(1)->Opcode() == Op_CompareAndSwapS) ||
(bol->in(1)->Opcode() == Op_CompareAndSwapI) ||
(bol->in(1)->Opcode() == Op_CompareAndSwapL) ||
(bol->in(1)->Opcode() == Op_CompareAndSwapP) ||
(bol->in(1)->Opcode() == Op_CompareAndSwapN) ||
(bol->in(1)->Opcode() == Op_ShenandoahCompareAndExchangeP) ||
(bol->in(1)->Opcode() == Op_ShenandoahCompareAndExchangeN) ||
(bol->in(1)->Opcode() == Op_ShenandoahWeakCompareAndSwapP) ||
(bol->in(1)->Opcode() == Op_ShenandoahWeakCompareAndSwapN) ||
(bol->in(1)->Opcode() == Op_ShenandoahCompareAndSwapP) ||
(bol->in(1)->Opcode() == Op_ShenandoahCompareAndSwapN))) return; // Allocation loops RARELY take backedge // Find the OTHER exit path from the IF
Node* ex = iff->proj_out(1-test_con); float p = iff->_prob; if (!phase->is_member(this, ex) && iff->_fcnt == COUNT_UNKNOWN) { if (top == Op_IfTrue) { if (p < (PROB_FAIR + PROB_UNLIKELY_MAG(3))) {
iff->_prob = PROB_STATIC_FREQUENT;
}
} else { if (p > (PROB_FAIR - PROB_UNLIKELY_MAG(3))) {
iff->_prob = PROB_STATIC_INFREQUENT;
}
}
}
}
}
test = phase->idom(test);
}
}
#ifdef ASSERT static CountedLoopNode* locate_pre_from_main(CountedLoopNode* main_loop) {
assert(!main_loop->is_main_no_pre_loop(), "Does not have a pre loop");
Node* ctrl = main_loop->skip_predicates();
assert(ctrl->Opcode() == Op_IfTrue || ctrl->Opcode() == Op_IfFalse, "");
Node* iffm = ctrl->in(0);
assert(iffm->Opcode() == Op_If, "");
Node* p_f = iffm->in(0);
assert(p_f->Opcode() == Op_IfFalse, "");
CountedLoopNode* pre_loop = p_f->in(0)->as_CountedLoopEnd()->loopnode();
assert(pre_loop->is_pre_loop(), "No pre loop found"); return pre_loop;
} #endif
// Remove the main and post loops and make the pre loop execute all // iterations. Useful when the pre loop is found empty. void IdealLoopTree::remove_main_post_loops(CountedLoopNode *cl, PhaseIdealLoop *phase) {
CountedLoopEndNode* pre_end = cl->loopexit();
Node* pre_cmp = pre_end->cmp_node(); if (pre_cmp->in(2)->Opcode() != Op_Opaque1) { // Only safe to remove the main loop if the compiler optimized it // out based on an unknown number of iterations return;
}
// Can we find the main loop? if (_next == NULL) { return;
}
Node* next_head = _next->_head; if (!next_head->is_CountedLoop()) { return;
}
assert(locate_pre_from_main(main_head) == cl, "bad main loop");
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)); // Remove the OpaqueZeroTripGuardNode of the main loop so it can be optimized out
Node* main_cmp = main_iff->in(1)->in(1);
assert(main_cmp->in(2)->Opcode() == Op_OpaqueZeroTripGuard, "main loop has no opaque node?");
phase->_igvn.replace_input_of(main_cmp, 2, main_cmp->in(2)->in(1));
}
//------------------------------do_remove_empty_loop--------------------------- // We always attempt remove empty loops. The approach is to replace the trip // counter with the value it will have on the last iteration. This will break // the loop. bool IdealLoopTree::do_remove_empty_loop(PhaseIdealLoop *phase) { // Minimum size must be empty loop if (_body.size() > EMPTY_LOOP_SIZE) { returnfalse;
} if (!_head->is_CountedLoop()) { returnfalse; // Dead loop
}
CountedLoopNode *cl = _head->as_CountedLoop(); if (!cl->is_valid_counted_loop(T_INT)) { returnfalse; // Malformed loop
} if (!phase->is_member(this, phase->get_ctrl(cl->loopexit()->in(CountedLoopEndNode::TestValue)))) { returnfalse; // Infinite loop
} if (cl->is_pre_loop()) { // If the loop we are removing is a pre-loop then the main and post loop // can be removed as well.
remove_main_post_loops(cl, phase);
}
#ifdef ASSERT // Ensure only one phi which is the iv.
Node* iv = NULL; for (DUIterator_Fast imax, i = cl->fast_outs(imax); i < imax; i++) {
Node* n = cl->fast_out(i); if (n->Opcode() == Op_Phi) {
assert(iv == NULL, "Too many phis");
iv = n;
}
}
assert(iv == cl->phi(), "Wrong phi"); #endif
// main and post loops have explicitly created zero trip guard bool needs_guard = !cl->is_main_loop() && !cl->is_post_loop(); if (needs_guard) { // Skip guard if values not overlap. const TypeInt* init_t = phase->_igvn.type(cl->init_trip())->is_int(); const TypeInt* limit_t = phase->_igvn.type(cl->limit())->is_int(); int stride_con = cl->stride_con(); if (stride_con > 0) {
needs_guard = (init_t->_hi >= limit_t->_lo);
} else {
needs_guard = (init_t->_lo <= limit_t->_hi);
}
} if (needs_guard) { // Check for an obvious zero trip guard.
Node* inctrl = PhaseIdealLoop::skip_all_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
Node* iff = inctrl->in(0); if (iff->is_If()) {
Node* bol = iff->in(1); if (bol->is_Bool()) {
BoolTest test = bol->as_Bool()->_test; if (maybe_swapped) {
test._test = test.commute();
test._test = test.negate();
} if (test._test == cl->loopexit()->test_trip()) {
Node* cmp = bol->in(1); int init_idx = maybe_swapped ? 2 : 1; int limit_idx = maybe_swapped ? 1 : 2; if (cmp->is_Cmp() && cmp->in(init_idx) == cl->init_trip() && cmp->in(limit_idx) == cl->limit()) {
needs_guard = false;
}
}
}
}
}
}
if (needs_guard) { // Peel the loop to ensure there's a zero trip guard
Node_List old_new;
phase->do_peeling(this, old_new);
}
// Replace the phi at loop head with the final value of the last // iteration (exact_limit - stride), to make sure the loop exit value // is correct, for any users after the loop. // Note: the final value after increment should not overflow since // counted loop has limit check predicate.
Node* phi = cl->phi();
Node* exact_limit = phase->exact_limit(this);
Node* final_iv = new SubINode(exact_limit, cl->stride());
phase->register_new_node(final_iv, cl->in(LoopNode::EntryControl));
phase->_igvn.replace_node(phi, final_iv);
// Set loop-exit condition to false. Then the CountedLoopEnd will collapse, // because the back edge is never taken.
Node* zero = phase->_igvn.intcon(0);
phase->_igvn.replace_input_of(cl->loopexit(), CountedLoopEndNode::TestValue, zero);
phase->C->set_major_progress(); returntrue;
}
//------------------------------do_one_iteration_loop-------------------------- // Convert one iteration loop into normal code. bool IdealLoopTree::do_one_iteration_loop(PhaseIdealLoop *phase) { if (!_head->as_Loop()->is_valid_counted_loop(T_INT)) { returnfalse; // Only for counted loop
}
CountedLoopNode *cl = _head->as_CountedLoop(); if (!cl->has_exact_trip_count() || cl->trip_count() != 1) { returnfalse;
}
#ifndef PRODUCT if (TraceLoopOpts) {
tty->print("OneIteration ");
this->dump_head();
} #endif
Node *init_n = cl->init_trip(); // Loop boundaries should be constant since trip count is exact.
assert((cl->stride_con() > 0 && init_n->get_int() + cl->stride_con() >= cl->limit()->get_int()) ||
(cl->stride_con() < 0 && init_n->get_int() + cl->stride_con() <= cl->limit()->get_int()), "should be one iteration"); // Replace the phi at loop head with the value of the init_trip. // Then the CountedLoopEnd will collapse (backedge will not be taken) // and all loop-invariant uses of the exit values will be correct.
phase->_igvn.replace_node(cl->phi(), cl->init_trip());
phase->C->set_major_progress(); returntrue;
}
//============================================================================= //------------------------------iteration_split_impl--------------------------- bool IdealLoopTree::iteration_split_impl(PhaseIdealLoop *phase, Node_List &old_new) { if (!_head->is_Loop()) { // Head could be a region with a NeverBranch that was added in beautify loops but the region was not // yet transformed into a LoopNode. Bail out and wait until beautify loops turns it into a Loop node. returnfalse;
} // Compute loop trip count if possible.
compute_trip_count(phase);
// Convert one iteration loop into normal code. if (do_one_iteration_loop(phase)) { returntrue;
} // Check and remove empty loops (spam micro-benchmarks) if (do_remove_empty_loop(phase)) { returntrue; // Here we removed an empty loop
}
AutoNodeBudget node_budget(phase);
// Non-counted loops may be peeled; exactly 1 iteration is peeled. // This removes loop-invariant tests (usually null checks). if (!_head->is_CountedLoop()) { // Non-counted loop if (PartialPeelLoop && phase->partial_peel(this, old_new)) { // Partial peel succeeded so terminate this round of loop opts returnfalse;
} if (policy_peeling(phase)) { // Should we peel? if (PrintOpto) { tty->print_cr("should_peel"); }
phase->do_peeling(this, old_new);
} elseif (policy_unswitching(phase)) {
phase->do_unswitching(this, old_new); returnfalse; // need to recalculate idom data
} elseif (phase->duplicate_loop_backedge(this, old_new)) { returnfalse;
} elseif (_head->is_LongCountedLoop()) {
phase->create_loop_nest(this, old_new);
} returntrue;
}
CountedLoopNode *cl = _head->as_CountedLoop();
if (!cl->is_valid_counted_loop(T_INT)) returntrue; // Ignore various kinds of broken loops
// Do nothing special to pre- and post- loops if (cl->is_pre_loop() || cl->is_post_loop()) returntrue;
// Compute loop trip count from profile data
compute_profile_trip_cnt(phase);
// Before attempting fancy unrolling, RCE or alignment, see if we want // to completely unroll this loop or do loop unswitching. if (cl->is_normal_loop()) { if (policy_unswitching(phase)) {
phase->do_unswitching(this, old_new); returnfalse; // need to recalculate idom data
} if (policy_maximally_unroll(phase)) { // Here we did some unrolling and peeling. Eventually we will // completely unroll this loop and it will no longer be a loop.
phase->do_maximally_unroll(this, old_new); returntrue;
} if (StressDuplicateBackedge && phase->duplicate_loop_backedge(this, old_new)) { returnfalse;
}
}
// Counted loops may be peeled, or may need some iterations run up // front for RCE. Thus we clone a full loop up front whose trip count is // at least 1 (if peeling), but may be several more.
// The main loop will start cache-line aligned with at least 1 // iteration of the unrolled body (zero-trip test required) and // will have some range checks removed.
// A post-loop will finish any odd iterations (leftover after // unrolling), plus any needed for RCE purposes.
// If not RCE'ing (iteration splitting), then we do not need a pre-loop. // We may still need to peel an initial iteration but we will not // be needing an unknown number of pre-iterations. // // Basically, if peel_only reports TRUE first time through, we will not // be able to later do RCE on this loop. bool peel_only = policy_peel_only(phase) && !should_rce;
// If we have any of these conditions (RCE, unrolling) met, then // we switch to the pre-/main-/post-loop model. This model also covers // peeling. if (should_rce || should_unroll) { if (cl->is_normal_loop()) { // Convert to 'pre/main/post' loops if (should_rce_long && phase->create_loop_nest(this, old_new)) { returntrue;
}
uint estimate = est_loop_clone_sz(3); if (!phase->may_require_nodes(estimate)) { returnfalse;
}
phase->insert_pre_post_loops(this, old_new, peel_only);
} // Adjust the pre- and main-loop limits to let the pre and post loops run // with full checks, but the main-loop with no checks. Remove said checks // from the main body. if (should_rce) { if (phase->do_range_check(this, old_new) != 0) {
cl->mark_has_range_checks();
} else {
cl->clear_has_range_checks();
}
} elseif (PostLoopMultiversioning) {
phase->has_range_checks(this);
}
if (should_unroll && !should_peel && PostLoopMultiversioning &&
Matcher::has_predicated_vectors()) { // Try to setup multiversioning on main loops before they are unrolled if (cl->is_main_loop() && (cl->unrolled_count() == 1)) {
phase->insert_scalar_rced_post_loop(this, old_new);
}
}
// Double loop body for unrolling. Adjust the minimum-trip test (will do // twice as many iterations as before) and the main body limit (only do // an even number of trips). If we are peeling, we might enable some RCE // and we'd rather unroll the post-RCE'd loop SO... do not unroll if // peeling. if (should_unroll && !should_peel) { if (SuperWordLoopUnrollAnalysis) {
phase->insert_vector_post_loop(this, old_new);
}
phase->do_unroll(this, old_new, true);
}
} else { // Else we have an unchanged counted loop if (should_peel) { // Might want to peel but do nothing else if (phase->may_require_nodes(est_peeling)) {
phase->do_peeling(this, old_new);
}
} if (should_rce_long) {
phase->create_loop_nest(this, old_new);
}
} returntrue;
}
// Look for loop-exit tests with my 50/50 guesses from the Parsing stage. // Replace with a 1-in-10 exit guess. if (!is_root() && is_loop()) {
adjust_loop_exit_prob(phase);
}
// Unrolling, RCE and peeling efforts, iff innermost loop. if (_allow_optimizations && is_innermost()) { if (!_has_call) { if (!iteration_split_impl(phase, old_new)) { returnfalse;
}
} else {
AutoNodeBudget node_budget(phase); if (policy_unswitching(phase)) {
phase->do_unswitching(this, old_new); returnfalse; // need to recalculate idom data
}
}
}
if (_next && !_next->iteration_split(phase, old_new)) { returnfalse;
} returntrue;
}
//============================================================================= // Process all the loops in the loop tree and replace any fill // patterns with an intrinsic version. bool PhaseIdealLoop::do_intrinsify_fill() { bool changed = false; for (LoopTreeIterator iter(_ltree_root); !iter.done(); iter.next()) {
IdealLoopTree* lpt = iter.current();
changed |= intrinsify_fill(lpt);
} return changed;
}
// Examine an inner loop looking for a single store of an invariant // value in a unit stride loop, bool PhaseIdealLoop::match_fill_loop(IdealLoopTree* lpt, Node*& store, Node*& store_value,
Node*& shift, Node*& con) { constchar* msg = NULL;
Node* msg_node = NULL;
store_value = NULL;
con = NULL;
shift = NULL;
// Process the loop looking for stores. If there are multiple // stores or extra control flow give at this point.
CountedLoopNode* head = lpt->_head->as_CountedLoop(); for (uint i = 0; msg == NULL && i < lpt->_body.size(); i++) {
Node* n = lpt->_body.at(i); if (n->outcnt() == 0) continue; // Ignore dead if (n->is_Store()) { if (store != NULL) {
msg = "multiple stores"; break;
} int opc = n->Opcode(); if (opc == Op_StoreP || opc == Op_StoreN || opc == Op_StoreNKlass || opc == Op_StoreCM) {
msg = "oop fills not handled"; break;
}
Node* value = n->in(MemNode::ValueIn); if (!lpt->is_invariant(value)) {
msg = "variant store value";
} elseif (!_igvn.type(n->in(MemNode::Address))->isa_aryptr()) {
msg = "not array address";
}
store = n;
store_value = value;
} elseif (n->is_If() && n != head->loopexit_or_null()) {
msg = "extra control flow";
msg_node = n;
}
}
if (store == NULL) { // No store in loop returnfalse;
}
if (msg == NULL && head->stride_con() != 1) { // could handle negative strides too if (head->stride_con() < 0) {
msg = "negative stride";
} else {
msg = "non-unit stride";
}
}
if (msg == NULL && !store->in(MemNode::Address)->is_AddP()) {
msg = "can't handle store address";
msg_node = store->in(MemNode::Address);
}
// Make sure there is an appropriate fill routine
BasicType t = store->as_Mem()->memory_type(); constchar* fill_name; if (msg == NULL &&
StubRoutines::select_fill_function(t, false, fill_name) == NULL) {
msg = "unsupported store";
msg_node = store;
}
if (msg != NULL) { #ifndef PRODUCT if (TraceOptimizeFill) {
tty->print_cr("not fill intrinsic candidate: %s", msg); if (msg_node != NULL) msg_node->dump();
} #endif returnfalse;
}
// Make sure the address expression can be handled. It should be // head->phi * elsize + con. head->phi might have a ConvI2L(CastII()).
Node* elements[4];
Node* cast = NULL;
Node* conv = NULL; bool found_index = false; int count = store->in(MemNode::Address)->as_AddP()->unpack_offsets(elements, ARRAY_SIZE(elements)); for (int e = 0; e < count; e++) {
Node* n = elements[e]; if (n->is_Con() && con == NULL) {
con = n;
} elseif (n->Opcode() == Op_LShiftX && shift == NULL) {
Node* value = n->in(1); #ifdef _LP64 if (value->Opcode() == Op_ConvI2L) {
conv = value;
value = value->in(1);
} if (value->Opcode() == Op_CastII &&
value->as_CastII()->has_range_check()) { // Skip range check dependent CastII nodes
cast = value;
value = value->in(1);
} #endif if (value != head->phi()) {
msg = "unhandled shift in address";
} else { if (type2aelembytes(store->as_Mem()->memory_type(), true) != (1 << n->in(2)->get_int())) {
msg = "scale doesn't match";
} else {
found_index = true;
shift = n;
}
}
} elseif (n->Opcode() == Op_ConvI2L && conv == NULL) {
conv = n;
n = n->in(1); if (n->Opcode() == Op_CastII &&
n->as_CastII()->has_range_check()) { // Skip range check dependent CastII nodes
cast = n;
n = n->in(1);
} if (n == head->phi()) {
found_index = true;
} else {
msg = "unhandled input to ConvI2L";
}
} elseif (n == head->phi()) { // no shift, check below for allowed cases
found_index = true;
} else {
msg = "unhandled node in address";
msg_node = n;
}
}
if (!found_index) {
msg = "missing use of index";
}
// byte sized items won't have a shift if (msg == NULL && shift == NULL && t != T_BYTE && t != T_BOOLEAN) {
msg = "can't find shift";
msg_node = store;
}
if (msg != NULL) { #ifndef PRODUCT if (TraceOptimizeFill) {
tty->print_cr("not fill intrinsic: %s", msg); if (msg_node != NULL) msg_node->dump();
} #endif returnfalse;
}
// No make sure all the other nodes in the loop can be handled
VectorSet ok;
// store related values are ok
ok.set(store->_idx);
ok.set(store->in(MemNode::Memory)->_idx);
CountedLoopEndNode* loop_exit = head->loopexit();
// Loop structure is ok
ok.set(head->_idx);
ok.set(loop_exit->_idx);
ok.set(head->phi()->_idx);
ok.set(head->incr()->_idx);
ok.set(loop_exit->cmp_node()->_idx);
ok.set(loop_exit->in(1)->_idx);
// Address elements are ok if (con) ok.set(con->_idx); if (shift) ok.set(shift->_idx); if (cast) ok.set(cast->_idx); if (conv) ok.set(conv->_idx);
for (uint i = 0; msg == NULL && i < lpt->_body.size(); i++) {
Node* n = lpt->_body.at(i); if (n->outcnt() == 0) continue; // Ignore dead if (ok.test(n->_idx)) continue; // Backedge projection is ok if (n->is_IfTrue() && n->in(0) == loop_exit) continue; if (!n->is_AddP()) {
msg = "unhandled node";
msg_node = n; break;
}
}
// Make sure no unexpected values are used outside the loop for (uint i = 0; msg == NULL && i < lpt->_body.size(); i++) {
Node* n = lpt->_body.at(i); // These values can be replaced with other nodes if they are used // outside the loop. if (n == store || n == loop_exit || n == head->incr() || n == store->in(MemNode::Memory)) continue; for (SimpleDUIterator iter(n); iter.has_next(); iter.next()) {
Node* use = iter.get(); if (!lpt->_body.contains(use)) {
msg = "node is used outside loop";
msg_node = n; break;
}
}
}
#ifdef ASSERT if (TraceOptimizeFill) { if (msg != NULL) {
tty->print_cr("no fill intrinsic: %s", msg); if (msg_node != NULL) msg_node->dump();
} else {
tty->print_cr("fill intrinsic for:");
}
store->dump(); if (Verbose) {
lpt->_body.dump();
}
} #endif
return msg == NULL;
}
bool PhaseIdealLoop::intrinsify_fill(IdealLoopTree* lpt) { // Only for counted inner loops if (!lpt->is_counted() || !lpt->is_innermost()) { returnfalse;
}
// Must have constant stride
CountedLoopNode* head = lpt->_head->as_CountedLoop(); if (!head->is_valid_counted_loop(T_INT) || !head->is_normal_loop()) { returnfalse;
}
head->verify_strip_mined(1);
// Check that the body only contains a store of a loop invariant // value that is indexed by the loop phi.
Node* store = NULL;
Node* store_value = NULL;
Node* shift = NULL;
Node* offset = NULL; if (!match_fill_loop(lpt, store, store_value, shift, offset)) { returnfalse;
}
#ifndef PRODUCT if (TraceLoopOpts) {
tty->print("ArrayFill ");
lpt->dump_head();
} #endif
// Now replace the whole loop body by a call to a fill routine that // covers the same region as the loop.
Node* base = store->in(MemNode::Address)->as_AddP()->in(AddPNode::Base);
// Build an expression for the beginning of the copy region
Node* index = head->init_trip(); #ifdef _LP64
index = new ConvI2LNode(index);
_igvn.register_new_node_with_optimizer(index); #endif if (shift != NULL) { // byte arrays don't require a shift but others do.
index = new LShiftXNode(index, shift->in(2));
_igvn.register_new_node_with_optimizer(index);
}
Node* from = new AddPNode(base, base, index);
_igvn.register_new_node_with_optimizer(from); // For normal array fills, C2 uses two AddP nodes for array element // addressing. But for array fills with Unsafe call, there's only one // AddP node adding an absolute offset, so we do a NULL check here.
assert(offset != NULL || C->has_unsafe_access(), "Only array fills with unsafe have no extra offset"); if (offset != NULL) {
from = new AddPNode(base, from, offset);
_igvn.register_new_node_with_optimizer(from);
} // Compute the number of elements to copy
Node* len = new SubINode(head->limit(), head->init_trip());
_igvn.register_new_node_with_optimizer(len);
// Build a call to the fill routine constchar* fill_name;
address fill = StubRoutines::select_fill_function(t, aligned, fill_name);
assert(fill != NULL, "what?");
// Convert float/double to int/long for fill routines if (t == T_FLOAT) {
store_value = new MoveF2INode(store_value);
_igvn.register_new_node_with_optimizer(store_value);
} elseif (t == T_DOUBLE) {
store_value = new MoveD2LNode(store_value);
_igvn.register_new_node_with_optimizer(store_value);
}
/* Disable following optimization until proper fix (add missing checks).
// If this fill is tightly coupled to an allocation and overwrites // the whole body, allow it to take over the zeroing. AllocateNode* alloc = AllocateNode::Ideal_allocation(base, this); if (alloc != NULL && alloc->is_AllocateArray()) { Node* length = alloc->as_AllocateArray()->Ideal_length(); if (head->limit() == length && head->init_trip() == _igvn.intcon(0)) { if (TraceOptimizeFill) { tty->print_cr("Eliminated zeroing in allocation"); } alloc->maybe_set_complete(&_igvn); } else { #ifdef ASSERT if (TraceOptimizeFill) { tty->print_cr("filling array but bounds don't match"); alloc->dump(); head->init_trip()->dump(); head->limit()->dump(); length->dump(); } #endif } }
*/
if (head->is_strip_mined()) { // Inner strip mined loop goes away so get rid of outer strip // mined loop
Node* outer_sfpt = head->outer_safepoint();
Node* in = outer_sfpt->in(0);
Node* outer_out = head->outer_loop_exit();
lazy_replace(outer_out, in);
_igvn.replace_input_of(outer_sfpt, 0, C->top());
}
// Redirect the old control and memory edges that are outside the loop. // Sometimes the memory phi of the head is used as the outgoing // state of the loop. It's safe in this case to replace it with the // result_mem.
_igvn.replace_node(store->in(MemNode::Memory), result_mem);
lazy_replace(exit, result_ctrl);
_igvn.replace_node(store, result_mem); // Any uses the increment outside of the loop become the loop limit.
_igvn.replace_node(head->incr(), head->limit());
// Disconnect the head from the loop. for (uint i = 0; i < lpt->_body.size(); i++) {
Node* n = lpt->_body.at(i);
_igvn.replace_node(n, C->top());
}
returntrue;
}
Messung V0.5 in Prozent
¤ Dauer der Verarbeitung: 0.126 Sekunden
(vorverarbeitet am 2026-04-30)
¤
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.